CN110570014A - A method for forecasting electric vehicle charging load based on Monte Carlo and deep learning - Google Patents

Abstract

The invention provides an electric vehicle charging load prediction method based on Monte Carlo and deep learning. The prediction method comprises the following steps: firstly, dividing the electric automobile into 4 types of electric buses, electric taxis, electric private cars and electric official cars according to the characteristics of the electric automobile, establishing a probability model of load influence factors, and further obtaining calculation models of charging power of different types of electric automobiles; secondly, extracting the initial charge state, the initial charging time and the like of the electric automobile by adopting a Monte Carlo simulation method to calculate the charging load of the electric automobile at each moment according to the reserve prediction result of the electric automobile; and finally, according to the electric vehicle charging load at each moment obtained by Monte Carlo sampling, carrying out deep learning and prediction on the electric vehicle charging load by adopting an LSTM deep learning algorithm, thereby obtaining an electric vehicle charging load curve. The charging load prediction method has better scientificity and objectivity.

Description

A method for forecasting electric vehicle charging load based on Monte Carlo and deep learning

technical field

The invention belongs to the field of power systems, in particular to a method for predicting charging loads of electric vehicles based on Monte Carlo and deep learning.

Background technique

Due to the increasingly prominent problems of energy security and environmental pollution, new energy has been vigorously developed in recent years, among which electric vehicles are developing rapidly, and will gradually replace traditional fuel vehicles and become the main means of transportation in the future, with broad development prospects. Compared with other low-carbon loads, electric vehicles have the basis for large-scale application and will become an important part of grid load. When electric vehicles are applied on a large scale, their impact on the power grid cannot be ignored. Therefore, fully considering the development law and charging law of electric vehicles is of great significance to the long-term development of urban distribution network. Predict its inventory and proportion, model the charging load, and give the future daily load forecast results. The basis for coordinated research between electric vehicles and other energy, transportation and other systems.

At present, some indicators and methods have been proposed for electric vehicle load forecasting. For example, common load forecasting methods include unit consumption method, trend analysis method, elastic coefficient method, regression analysis method, time series method, gray model method, and neural network method. method, Delphi method, expert system method, optimal combination analysis method and other methods.

There are still deficiencies in the existing research on electric vehicle charging load forecasting: the universality is not strong. Most studies only consider simple factors such as charging mileage, charging start time, initial state of charge and other modeling conditions, and there is no theoretical load model covering various key factors and multiple vehicle types. Moreover, the traditional method and the shallow learning method have insufficient adaptive ability, insufficient cognitive ability for nonlinear feature load, and low prediction accuracy.

It can be seen that the electric vehicle charging load forecasting method needs to be improved.

Contents of the invention

In order to solve the above-mentioned technical problems, the present invention proposes a method for predicting electric vehicle charging load based on Monte Carlo and deep learning. This method establishes the probability model of the influence of load factors, builds the charging load calculation model of different types of electric vehicles, uses the Monte Carlo simulation method to obtain the charging load at each time, and then uses the deep learning algorithm to obtain the electric vehicle charging load prediction curve.

The present invention adopts following technical scheme to realize:

A method for forecasting electric vehicle charging load based on Monte Carlo and deep learning, comprising the following steps:

First of all, according to the characteristics of electric vehicles, electric vehicles are divided into four types: electric buses, electric taxis, electric private cars and electric official vehicles, and the probability model of load influencing factors is established, and then the calculation model of charging power of different types of electric vehicles is obtained .

Secondly, according to the prediction results of electric vehicle ownership, Monte Carlo simulation method is used to extract the initial state of charge and initial charging time of electric vehicles to calculate the charging load of electric vehicles, and obtain the charging load at each moment.

Finally, according to the charging load of electric vehicles at each time obtained by Monte Carlo sampling, the LSTM deep learning algorithm is used to predict the charging load of electric vehicles, so as to obtain the charging load curve of electric vehicles.

In the above technical solution, further, step 1) establishes a probability model of load influencing factors to obtain a calculation model of electric vehicle charging power, including:

(1) The charging load of electric vehicles is affected by many factors, the main influencing factors are initial charging time, daily mileage, charging time, electric vehicle ownership, initial state of charge (SOC), battery capacity, etc.

According to the characteristics of electric vehicles, electric vehicles are divided into four types: electric buses, electric taxis, electric private cars and electric official vehicles, and the initial charging time model is constructed, and the initial charging time of electric vehicles is obtained through data fitting processing The normal distribution given by:

In the formula: t is the initial charging time, that is, the end time of the last trip; μ _a and σ _a are the expectation and standard deviation of the initial charging time, respectively, since the positive The results of the state distribution are different, so the expectations and standard deviations of different types of electric vehicles are different;

(2) Construct a daily mileage model. The daily mileage is an important indicator of the driving characteristics of electric vehicles, reflecting the power consumption of the vehicle in a day, which in turn affects the charging time of electric vehicles. Similarly, using the travel characteristics of traditional fuel vehicles instead of electric vehicles for analysis, the daily mileage of electric vehicles follows a lognormal distribution, and its probability density function is:

In the formula: s is the daily mileage in km; μ _b and σ _b are the expectation and variance of the logarithm lns of the mileage s, respectively, which vary with the driving characteristics of different types of electric vehicles.

(3) Construct the initial state-of-charge model. In a charging cycle, the electric power demand for electric vehicle charging changes with time. To determine the charging load of electric vehicles, it is necessary to obtain the state of charge at the beginning of battery charging, that is, the initial SOC. The starting SOC is a random function of the distance traveled by the EV since the last charge, with values ranging from 0 to 100%. Assuming that the state of charge of an electric vehicle decreases linearly with the mileage, the state of charge at the initial charging time can be estimated from the mileage of the vehicle, expressed by a probability density function:

S _OC ＝(S _OC0 -s/s _max )×100%

In the formula: S _OC represents the initial SOC of battery charging; S _OC0 represents the state of charge value of the battery after the latest charge. Since the charging time and location of electric vehicles are more scattered, S _OC0 is often not 1; s _max indicates the maximum mileage that can be driven after the battery is fully charged, in km.

(4) Build a charging time model. At present, most electric vehicles use lithium batteries, and the charging process of lithium batteries is a two-stage charging process of constant voltage and constant current. Assuming that the entire charging process is a constant power, the charging time can be calculated in the following two ways:

To calculate the charging time based on the state of charge, there is the following formula:

In the formula: Te is the charging time, the unit is h; U is the battery capacity, the unit is kW h; P is the charging power, the unit is kW; η is the charging efficiency;

Calculate the charging time based on the daily mileage, as shown in the following formula:

In the formula: W ₁₀₀ is the power consumption of the car every 100km, and the unit is (kW·h)/100km.

Furthermore, step 2) calculates the charging load at each moment, including:

(1) Divide the travel characteristics of different types of electric vehicles: divide electric vehicles into four categories according to different purposes: buses, taxis, private cars, and official vehicles, and analyze the travel characteristics of different types of electric vehicles in order to obtain the load forecasting model. parameters, as follows:

electric bus

The driving characteristics of the bus are relatively fixed, and the operation system is mainly adopted in shifts. According to the data in the literature "Electric Vehicle Charging Load Prediction Based on Monte Carlo Simulation", the daily mileage of the bus is about 70km. Considering safe operation, charging once a day is difficult to meet the operating needs of electric buses, and charging twice a day is required. The operating hours and routes of buses are relatively concentrated, and centralized charging can be carried out. Fast charging at noon and regular charging at night after get off work. Generally speaking, the charging period of the bus is from 9.30 to 16.00, and from 23:00 to 05:00 the next day, which respectively obey the normal distribution. The specific distribution parameters can be obtained according to the urban research data.

electric taxi

The operating hours of taxis are roughly from 06:00 to 24:00. According to the data in the document "Travel Characteristics and Development Strategies of Private Motor Vehicles in Beijing", the daily mileage of taxis is about 400km. Like electric buses, electric taxis generally adopt the mode of charging twice a day, and the charging time is selected at noon and night. Since the rest time of the taxi is limited, but the battery needs to be replenished in time, the electric taxi chooses the fast charging mode. According to the above analysis, the charging time of taxis is 02:00-05:00 and 11:30-14:30, respectively obeying the normal distribution.

electric private car

Compared with buses and taxis, the driving characteristics of private cars are more random and arbitrary. Electric private cars are charged once a day, and the charging time is divided into 09:00-12:00 in the morning, 14:00-17:00 in the afternoon, and 19:00 in the evening to 07:00 the next day. Charge in the parking lot of the workplace in the morning and afternoon, and charge in the parking lot of the residential area after get off work in the evening. Choose the fast charging mode for charging in the unit parking lot, and choose the regular charging mode for the parking lot in the residential area. According to the data in the literature, regardless of long-distance travel, the daily mileage of private cars is 40km.

Electric commercial vehicle

Official vehicles are mainly used for daily business trips. If long-distance travel is not considered, their driving characteristics are similar to private cars. The charging time of official vehicles is generally charged in the parking lot of the unit after get off work at night. Official vehicles can be charged once a day, and the regular charging mode is adopted, and the charging period is from 19:00 to 07:00 the next day.

(2) Predict the number of various types of electric vehicles: Based on the data in 2020 pointed out in the "China Automobile Industry Development Report (2012)", the number of various types of vehicles is predicted according to the proportion of different types of electric vehicles ,

Calculate the charging load of electric vehicles based on Monte Carlo simulation:

Make reasonable assumptions:

a. The initial charging time, daily driving mileage and charging power of various types of electric vehicles are independent random variables;

b. The charging power of various types of electric vehicles is regarded as a constant power model;

c. All vehicles are fully charged every time;

d. The end time of the last trip of all vehicles is the initial charging time of the vehicle.

(3) Choose the Monte Carlo simulation method to sample different types of electric vehicles, and obtain the charging time and charging power of a single electric vehicle by extracting the initial charging time and daily mileage. Then add up the charging power of all electric vehicles to obtain the charging load; take time as the abscissa and the charging load at each moment as the ordinate to obtain the charging load curve. Then the charging load at time j is:

In the formula: P _{total, j} is the charging load at time j, the unit is kw; N _b , N _t , N _s , N _w are the total number of electric buses, taxis, private cars and commercial vehicles respectively, and the unit is vehicle ; Divide a day into T time periods; P _ibj , P _ibj , P _ibj , P _ibj are the charging power of different types of electric vehicles at time j, and the unit is kW.

(4) Since the charging power is constant, the charging load is only related to the charging time. Taking the charging time of the electric vehicle as the sampling constraint of the initial charging time, the sampling range of the initial charging time can be selected, and then the initial charging time can be extracted , so as to obtain the charging load at the corresponding time of each sampling point; the calculation method of charging time is as follows:

Electric vehicles have two modes of regular charging and fast charging, and the specific charging situation is more complicated. For the convenience of modeling, it is assumed that the car chooses conventional charging first. During conventional charging, the charging time is calculated based on the daily driving mileage. First, the Monte Carlo method is used to extract the daily driving mileage and calculate the required charging time; the constraints to meet the required charging time are The premise is to select the sampling range of the initial charging time, and then extract the initial charging time for load calculation. When the fast charging mode is selected, the charging time is calculated based on the state of charge, and the initial charging time is also extracted based on the Monte Carlo method, and the remaining charging time in the charging time period and the charging time to meet the driving demand for the next charging are calculated. The shorter of the two times is the actual charging time.

Furthermore, step 3) predicts the electric vehicle charging load with the LSTM deep learning algorithm, including the following steps:

(1) It is necessary to normalize the input data before electric vehicle charging load prediction to ensure better results in network training and application. Using min-max standardization, the formula is as follows:

In the formula: X _o represents the standardized electric vehicle load matrix; X represents the original electric vehicle load matrix, which has only one row, corresponding to the load value at each moment; x _min and x _max respectively represent the Load minimum and maximum.

(2) Determine the storage of historical charging load information. That is, the output of the forget gate:

f _t ＝sigmoid(W _f ·[h _t-1 ,x _t ]+b _f )

In the formula: f _t represents the output of the forget gate; W _f represents the weight coefficient from the input to the forget gate; h _t-1 represents the output at the previous moment; x _t represents the input of the network at the current moment; b _f represents the bias of the forget gate.

(3) Determining the storage of new charging load information. The first part is the input gate output obtained by the sigmoid function; the other part is the new candidate vector established by tanh:

i _t = sigmoid(W _i ·[h _t-1 ,x _t ]+b _i )

In the formula: W _i represents the weight parameter of the input gate; b _i represents the bias of the input gate; Indicates the candidate value of the cell charging load state, W _C represents the weight parameter for calculating the candidate value of the cell charging load state; b _C represents the calculation bias for the candidate value of the cell charging load state. The cell represents the unit at this moment, which can be determined according to the previous Monte Carlo sampling point interval;

(4) Update the cell charging load status:

(5) Determine the current charging load output. In LSTM, the cell charging load state is processed by the tanh function and filtered by the output gate to obtain the final output:

o _t ＝sigmoid(W _o ·[h _t-1 ,x _t ]+b _o )

h _t ＝o _t *tanh(C _t )

In the formula: o _t represents the output of the output gate; W _o represents the weight parameter input to the output gate; b _o represents the bias of the output gate; h _t represents the output at the current moment.

(6) The sample data of electric vehicle charging load obtained by Monte Carlo simulation is processed by the above steps (1)-(5), and finally the prediction result of electric vehicle charging load of LSTM network is obtained.

(7) After the electric vehicle charging load prediction is completed, the root mean square error prediction accuracy evaluation index is used:

The beneficial effects of the present invention are:

The present invention establishes a method for predicting the charging load of electric vehicles based on Monte Carlo and deep learning. According to the method, the calculation models of charging power of different types of electric vehicles and the charging load of electric vehicles at each moment can be obtained. The present invention predicts the charging load of the electric vehicle in the coming day based on the Monte Carlo simulation and the deep learning algorithm, and obtains the overall trend of the charging load of the electric vehicle in the future. The simulation results show that the method proposed by the invention can accurately predict the charging load at each moment, which effectively proves the scientificity and objectivity of the charging load prediction method of the invention.

Description of drawings

Figure 1 Charging methods and charging time distribution of different types of electric vehicles;

Figure 2 The 10-day electric vehicle charging load data obtained by Monte Carlo sampling;

Figure 3 LSTM structure diagram;

Figure 4 uses the data of the first 9 days to predict the data of the 10th day after learning;

Figure 5. The number of nodes in the hidden layer is the 400LSTM load prediction result and actual value;

Fig. 6 is a flowchart of a charging load forecasting method based on Monte Carlo and deep learning of the present invention.

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

A kind of electric vehicle charging load prediction method based on Monte Carlo and deep learning of the present invention comprises the following steps:

S01. Build the initial charging time model, and obtain the initial charging time of the electric vehicle through data fitting processing to meet the normal distribution shown in the following formula:

In the formula: t is the initial charging time, that is, the end time of the last trip; μ _a and σ _a are the expectation and variance of the initial charging time, respectively, and the expectations and variances of different types of electric vehicles are different.

S02. Construct a daily mileage model. The daily mileage is an important indicator of the driving characteristics of an electric vehicle, reflecting the power consumption of the vehicle in a day, which in turn affects the charging time of the electric vehicle. Similarly, using the travel characteristics of traditional fuel vehicles instead of electric vehicles for analysis, the daily mileage of electric vehicles follows a lognormal distribution, and its probability density function is:

In the formula: s is the daily mileage in km; μ _b and σ _b are the expectation and variance of the logarithm lns of the mileage s, respectively, which vary with the driving characteristics of different types of electric vehicles.

S03. Construct the initial state-of-charge model. In a charging cycle, the power demand required for charging an electric vehicle changes with time. To determine the charging load of electric vehicles, it is necessary to obtain the state of charge at the beginning of battery charging, that is, the initial SOC. The starting SOC is a random function of the distance traveled by the electric vehicle since the last charge, with values ranging from 0 to 100%. Assuming that the state of charge of an electric vehicle decreases linearly with the mileage, the state of charge at the initial charging time can be estimated from the mileage of the vehicle, expressed by a probability density function:

S _OC ＝(S _OC0 -s/s _max )×100%

In the formula: S _OC represents the initial SOC of battery charging; S _OC0 represents the state of charge value of the battery after the latest charge, and s is the daily mileage. Since the charging time and location of electric vehicles are more scattered, S _OC0 is often not 1; s _max indicates the maximum mileage that can be driven after the battery is fully charged, in km.

S04. Build a charging time model. At present, most electric vehicles use lithium batteries, and the charging process of lithium batteries is a two-stage charging process of constant voltage and constant current. Assume that the entire charging process is constant power. The charging time is restricted by many factors. To calculate the charging time based on the state of charge, there is the following formula:

In the formula: T _e is the charging time, the unit is h; U is the battery capacity, the unit is kW h; P is the charging power, the unit is kW; η is the charging efficiency.

S05. Calculate the charging time based on the daily mileage, as shown in the following formula:

In the formula: W ₁₀₀ is the power consumption of the car every 100km, and the unit is (kW·h)/100km.

S06. Divide the travel characteristics of different types of electric vehicles: divide electric vehicles into four categories according to different purposes: buses, taxis, private cars, and official vehicles, and analyze the travel characteristics of different types of electric vehicles in detail to obtain the parameters of the load forecasting model . The charging methods and charging time distribution of the four types of electric vehicles are shown in Figure 1.

S07. Predict the number of various types of electric vehicles: Based on the data of 2020 pointed out in the "China Automobile Industry Development Report (2012)", the number of various types of vehicles is predicted according to the proportion of different types of electric vehicles. The forecast of ownership of different types of electric vehicles is shown in Table 1.

Table 1 Prediction of ownership of different types of electric vehicles

Vehicle Type proportion(%) Vehicle Type proportion(%) electric bus 65.52 electric private car 11.24 electric taxi 15.52 Electric commercial vehicle 7.72

S08. Calculate the charging load of electric vehicles based on Monte Carlo simulation:

Make reasonable assumptions:

a. The initial charging time, daily mileage and charging power of various types of electric vehicles are independent random variables;

b. The charging power of various types of electric vehicles is regarded as a constant power model, and there are no two stages of constant voltage and constant current charging;

c. All vehicles are fully charged every time;

d. The end time of the last trip of all vehicles is the starting charging time of the vehicle.

The Monte Carlo simulation method is selected to sample different types of electric vehicles, and the charging time and charging power of a single electric vehicle are obtained by extracting the initial charging time and daily mileage. Then add up the charging power of all electric vehicles to get the charging load at each moment. Taking time as the abscissa and the charging load at each moment as the ordinate, the charging load curve can be obtained. Then the charging load at time j is:

In the formula: among them, P _{total, j} is the charging load at time j, the unit is kw; N _b , N _t , N _s , N _w are the total number of electric buses, taxis, private cars and commercial vehicles, respectively, and the unit is is a vehicle; divide a day into T time periods; P _ibj , P _ibj , P _ibj , P _ibj are the charging power of different types of electric vehicles at time j, and the unit is kW.

S09. Since the charging power is constant, the charging load is only related to the charging time. Taking the charging time of the electric vehicle as the sampling constraint of the initial charging time, the sampling range of the initial charging time can be selected, and then the initial charging time can be extracted. In this way, the charging load at the corresponding time of each sampling point is obtained; the calculation method of the charging time is as follows:

Assume that the car chooses conventional charging first. During conventional charging, the charging time is calculated based on the daily driving mileage. First, the Monte Carlo method is used to extract the daily driving mileage and calculate the required charging time; the starting point is selected on the premise of satisfying the constraints of the required charging time. The sampling range of the charging time, and then extract the initial charging time for load calculation. When the fast charging mode is selected, the charging time is calculated based on the state of charge, and the initial charging time is also extracted based on the Monte Carlo method, and the remaining charging time in the charging time period and the charging time to meet the driving demand for the next charging are calculated. The shorter of the two times is the actual charging time. The 10-day electric vehicle charging load data obtained by Monte Carlo simulation is shown in Figure 2.

S10. It is necessary to normalize the input data before electric vehicle charging load prediction, so as to ensure better results in network training and application. Using min-max standardization, the formula is as follows:

In the formula: X _o represents the standardized electric vehicle load matrix; X represents the original electric vehicle load matrix, which has only one row, corresponding to the load value at each moment; x _min and x _max respectively represent the Load minimum and maximum.

S11. The structure diagram of LSTM is shown in Figure 3. Determine the storage of historical charging load information. That is, the output of the forget gate:

f _t ＝sigmoid(W _f ·[h _t-1 ,x _t ]+b _f )

In the formula: f _t represents the output of the forget gate; W _f represents the weight coefficient from the input to the forget gate; x _t represents the input of the network at the current moment; b _f represents the bias of the forget gate.

S12. Determining storage of new charging load information. The first part is the input gate output obtained by the sigmoid function; the other part is the new candidate vector established by tanh:

i _t = sigmoid(W _i ·[h _t-1 ,x _t ]+b _i )

In the formula: W _i represents the weight parameter of the input gate; b _i represents the bias of the input gate; Indicates the candidate value of the cell charging load state; W _C represents the calculation weight parameter of the cell charging load state candidate value; b _C represents the calculation bias of the cell charging load state candidate value.

S13. Update the cell charging load state:

S14. Determine the charging load output at the current moment. In LSTM, the cell charging load state is processed by the tanh function and filtered by the output gate to obtain the final output:

o _t ＝sigmoid(W _o ·[h _t-1 ,x _t ]+b _o )

h _t ＝o _t *tanh(C _t )

In the formula: o _t represents the output of the output gate; W _o represents the weight parameter input to the output gate; b _o represents the bias of the output gate; h _t represents the output at the current moment.

S15. After the sample data of the electric vehicle charging load obtained by the Monte Carlo simulation undergoes the above steps, the prediction result of the electric vehicle charging load of the LSTM network is finally obtained.

S16. After the electric vehicle charging load prediction is completed, the root mean square error prediction accuracy evaluation index is used:

Application example:

There are a total of 125,000 electric vehicles in a city, and the proportions of different types of electric vehicles are shown in Table 1. Among them, electric buses are charged twice a day, and the charging periods follow the normal distribution N(13, ^{1 2} ) and N(23, ^{1 2} ), respectively, and the powers of fast charging and conventional charging are 108kW and 60kW, respectively. The daily driving mileage obeys lognormal distribution N(4.3, 0.34 ² ). Taking "BYD K9" as an example, the battery capacity of the electric bus is 324kW h, the cruising range is 250km, and the power consumption for driving 100km is 140(kWh)/100km. Electric taxis are charged twice a day, and the charging periods follow the normal distribution N(3.5, ^{1 2} ) and N(13, ^{1 2} ), respectively, and the fast charging power is 60kW. The daily driving mileage obeys the logarithmic normal distribution N(5.2, 0.34 ² ). Take the "BYD E6" electric vehicle as the analysis object of electric taxis, private cars and commercial vehicles. The battery capacity of this type of electric vehicle is 82kW h, the cruising range is 400km, the power consumption for driving 100km is 20.5(kW h)/100km, the charging power of the fast charging mode is 60kW, and the charging power of the conventional charging mode It is 7kW. ^The charging probabilities of electric private cars in the morning, afternoon and evening are set to ^0.2 , 0.1 and 0.7 ^respectively . ), the daily mileage obeys lognormal distribution N(3.1, O.86 ² ). The charging period of official vehicles obeys the normal distribution N(19, ^{2 2} ), and the daily mileage obeys the logarithmic normal distribution N(3.1, O.86 ² ).

Taking 15 minutes as a sampling point, Monte Carlo simulation is used to extract a total of 96 points a day to simulate the charging load of all electric vehicles, and a total of 9 days of load data are taken as sample data. The load data on the 10th day predicted by the proposed LSTM deep learning algorithm is shown in Figure 4, and the comparison with the sampling data on the 10th day obtained through Monte Carlo simulation is shown in Figure 5.

It can be seen that the electric vehicle charging load predicted by LSTM deep learning is basically consistent with the charging load obtained by Monte Carlo sampling, and the error is relatively small. The LSTM prediction result error and the BP neural network prediction result error are shown in Table 2. It can be seen that the result of deep learning is better than that of shallow learning such as BP neural network.

Table 2 Comparison of LSTM and BP prediction errors

The peak of the total charging load of the four types of electric vehicles is between 14:00 and 15:00, which is about 3900MW. This is because there are more electric vehicles that choose to charge electric vehicles during this period, and most of them use fast charging mode for charging. The charging load curve shows that the charging load fluctuates violently within a day, which poses a potential threat to the stability of the power grid. Therefore, the result of the invention provides support for the power grid to consider corresponding guiding charging strategies to guide and regulate the charging behavior of users.

Claims (4)

1. the method for predicting the charging load of the electric automobile based on Monte Carlo and deep learning is characterized by comprising the following steps of:

s1: dividing the electric automobiles into 4 types of electric buses, electric taxis, electric private cars and electric official cars according to the characteristics of the electric automobiles, establishing probability models of load influence factors, and further obtaining calculation models of charging power of different types of electric automobiles;

S2: according to the reserve prediction result of the electric automobile, extracting the initial charge state, the initial charging time and the like of the electric automobile by adopting a Monte Carlo simulation method to calculate the charging load of the electric automobile at each moment;

S3: according to the charging load of the electric automobile at each moment obtained by Monte Carlo sampling, deep learning and prediction are carried out on the charging load of the electric automobile by adopting an LSTM deep learning algorithm, so that a charging load curve of the electric automobile is obtained.

2. The method for predicting the charging load of the electric vehicle based on the monte carlo and the deep learning according to claim 1, wherein the step S1 is specifically as follows:

The method comprises the following steps of dividing the electric automobile into 4 types of electric buses, electric taxis, electric private cars and electric business cars according to the characteristics of the electric automobile, constructing an initial charging time model, and obtaining the initial charging time of the electric automobile through data fitting processing, wherein the initial charging time meets the following normal distribution:

In the formula, t is initial charging time, namely the end time of the last trip; mu.s_aAnd σ_arespectively, the expected and standard deviation of the initial charge time;

The daily mileage model is constructed, the trip characteristics of the traditional fuel vehicle are used for replacing the trip characteristics of the electric vehicle for analysis, the daily mileage of the electric vehicle obeys log-normal distribution, and the probability density function is as follows:

In the formula: s is the daily mileage in km; mu.s_band σ_bRespectively, the expectation and standard deviation of the logarithm lns of the driving range s;

Constructing an initial state of charge model, wherein in a charging cycle, the power demand required by charging of the electric automobile changes along with time, so that the charging load of the electric automobile needs to obtain the state of charge at the battery charging starting moment, namely the initial SOC, wherein the initial SOC is a random function of the driving distance of the electric automobile after the last charging, and the value range is 0-100%; assuming that the state of charge of the electric vehicle linearly decreases with the driving range, the state of charge at the initial charging time can be estimated from the driving range of the vehicle, and is expressed as follows by a probability density function:

S_OC＝(S_OC0-s/s_max)×100％

In the formula: s_OCrepresents the starting SOC of the battery charge; s_OC0represents the state of charge value of the battery after the last charge, s is the daily mileage, s_maxthe maximum mileage that can be driven after the battery is fully charged is expressed in km;

a charging time period model is constructed, and if the whole charging process of the electric automobile is constant power, the charging time period has the following two calculation modes:

calculating the charge duration based on the state of charge, the following equation applies:

in the formula: t is_eIs the charging time length with the unit of h; u is the battery capacity, and the unit is kW.h; p is charging power, and the unit is kW; eta is charging efficiency;

Calculating the charging time based on the daily mileage, as shown in the following formula:

in the formula: w₁₀₀The unit is (kW.h)/hundred kilometers for the power consumption of the automobile per 100 km.

3. The method for predicting the charging load of the electric vehicle based on monte carlo and deep learning according to claim 2, wherein the method for calculating the charging load of the electric vehicle at each moment in step S2 comprises:

specifically analyzing travel characteristics of different types of electric automobiles to obtain parameters of a load prediction model;

predicting the holding capacity of various types of electric automobiles: the data in 2020 shown in Chinese automobile industry development report (2012) is used as a cardinal number, the quantity of the electric automobiles of various types is predicted according to the proportion of the electric automobiles of different types,

Calculating the charging load of the electric automobile based on Monte Carlo simulation:

Making reasonable assumptions:

a. The initial charging time, the daily mileage and the charging power of each type of electric automobile are mutually independent random variables;

b. The charging power of each type of electric automobile is regarded as a constant power model;

c. all vehicles are fully charged each time;

d. The last trip ending time of all the vehicles is the initial charging time of the vehicle;

Selecting a Monte Carlo simulation method to sample different types of electric automobiles, obtaining the charging duration and the charging power of a single electric automobile by extracting initial charging time and daily driving mileage, then accumulating the charging power of all the electric automobiles at each moment to obtain the charging load of the electric automobiles, wherein the charging load at the moment j is as follows:

in the formula: p_total，jThe charging load at time j is in kw; n is a radical of_b，N_t，N_s，N_wthe total number of the electric bus, the taxi, the private car and the commercial car is respectively, and the unit is a vehicle; dividing a day into T time periods;the unit of charging power of different types of electric automobiles at the moment j is kW;

because the charging power is constant, the charging load is only related to the charging time, the charging time of the electric automobile is taken as a sampling constraint condition of the initial charging time, namely, the sampling range of the initial charging time can be selected, and then the initial charging time is extracted, so that the charging load at the corresponding moment of each sampling point is obtained; the calculation method of the charging time is as follows:

The electric automobile has two modes of conventional charging and quick charging, the conventional charging is supposed to be preferentially selected by the automobile, and the charging time is calculated based on the daily mileage during the conventional charging; the method comprises the steps of selecting an electric automobile in a quick charging mode, calculating charging time based on a charge state, extracting initial charging time based on a Monte Carlo simulation method, calculating residual charging time in a charging time period and charging time meeting the driving requirement at the next charging, and taking the short time of the residual charging time and the charging time as actual charging time.

4. The method for predicting the charging load of the electric vehicle based on the monte carlo and the deep learning as claimed in claim 3, wherein the step S3 of predicting the charging load of the electric vehicle by using the LSTM deep learning algorithm comprises the following steps:

before the electric vehicle charging load prediction is carried out, normalization processing is carried out on input data, min-max standardization is adopted, and the formula is as follows:

In the formula: x_oRepresenting the electric vehicle load matrix after standardization; x represents an original electric vehicle load matrix, the matrix has only one row, and the matrix corresponds to the load value at each moment; x is the number of_min、x_maxRespectively representing the minimum value and the maximum value of the load in the original electric vehicle load matrix;

Determining the storage of the historical charging load information, namely the output of the forgetting gate:

f_t＝sigmoid(W_f·[h_t-1,x_t]+b_f)

In the formula: f. of_tindicating a forgotten gate output; w_fa weight coefficient representing the weight from the input to the forgetting gate; h is_t-1An output representing a previous time; x is the number of_trepresenting the current time input of the network; b_fIndicating a forgotten gate bias;

Determining storage of new charging load information, wherein the first part is input gate output obtained through a sigmoid function; the other part is a new candidate vector established by tanh:

i_t＝sigmoid(W_i·[h_t-1,x_t]+b_i)

in the formula: w_iRepresenting an input gate weight parameter; b_irepresenting input gate bias;Representing a cell charge load state candidate; w_CCalculating a weight parameter representing a candidate value of the cell charging load state; b_CRepresents the cell charging load state candidate calculation bias,

updating the cell charge load state:

Determining the charging load output at the current moment, wherein in the LSTM, the cell charging load state is processed by a tanh function and filtered by an output gate to obtain the final output:

o_t＝sigmoid(W_o·[h_t-1,x_t]+b_o)

h_t＝o_t*tanh(C_t)

In the formula: o_tRepresents the output of the output gate; w_oRepresenting a weight parameter input to the output gate; b_oRepresents the output gate offset; h is_tan output representing a current time;

Processing the charging load sample data of the electric vehicle obtained by Monte Carlo simulation by adopting the steps to finally obtain a charging load prediction result of the electric vehicle of the LSTM network;

after the charging load of the electric automobile is predicted, the root mean square error prediction accuracy evaluation index is adopted: