Virtual power plant optimization operation method for aggregating electric vehicle and carbon capture

Abstract

The invention discloses a virtual power plant optimization operation method for aggregating electric vehicles and carbon capture. The method comprises the following steps: S1, simulating a user charging load by analyzing an electric vehicle user behavior, and proposing a virtual power plant operation strategy; s2, with the peak-valley difference rate and the virtual power plant income as targets, wind power and photovoltaic output and the discharge capacity of the electric vehicle are cleared to be sold in the current market, and a gas turbine is used for bearing the basic load, the wind-solar prediction deviation and the charging demand of the electric vehicle; s3, considering the seasonal property of wind power and photovoltaic output, avoiding prediction deviation penalty through compensation, utilizing a chaos theory to integrate empirical mode decomposition peak frequency bands and a GA to optimize a back-propagation neural network BP, namely, GABP, performing seasonal prediction on the peak frequency bands, and then adopting a carbon capture system to reduce carbon emission and participate in carbon market transaction; and S4, comparing and simulating ordered and unordered charging and discharging through examples, and verifying the effectiveness and practicability of the strategy.

Description

Technical Field

[0001] Based on the impact of electric vehicle charging and discharging on the peak-to-valley difference rate of power load and the revenue of virtual power plants, the present invention utilizes advanced information and communication technology to propose an optimized operation strategy for virtual power plants that aggregate wind power, photovoltaics, electric vehicles, carbon capture and gas turbines. Specifically, it relates to an optimized operation method for virtual power plants that aggregate electric vehicles and carbon capture. Background Art

[0002] As an emerging power system management method, virtual power plant (VPP) is developing rapidly and being widely used around the world, especially for the development of distributed clean energy and electric vehicles (EVs), providing new ideas for solving ecological and resource problems. Distributed clean energy has the characteristics of high energy utilization efficiency, flexible energy supply, and economy. Electric vehicles have the natural advantages of zero pollution and zero emissions. In the context of various clean energy power generation injected into the power grid and the increasing number of electric vehicles, the coordinated control of virtual power plants brings together new power sources, energy storage, and loads, improves the safety and stability of the power system and the new energy absorption capacity, and maintains coordinated contact with the power grid, thereby obtaining certain benefits. However, new energy power generation has uncertainty and randomness, and the disordered charging and discharging of EVs will lead to the accumulation of peak power consumption, making the peak and valley difference of the power grid more obvious. At the same time, the power source will also generate a large amount of carbon emissions when it bears the load. VPP can not only solve the above problems, but also participate in related markets to obtain economic benefits. Therefore, optimizing VPP operations and ensuring economic benefits and costs have become current hot research directions.

[0003] At present, the method for optimizing VPP scheduling mainly uses the typical day fitting output of wind and solar power, without considering short-term prediction, which is not true for VPP participating in grid scheduling; and the prediction deviation will also result in deviation penalty and certain carbon emissions. In this regard, the VPP optimization operation strategy proposed in this paper takes into account the seasonality of wind power and photovoltaic output, and makes seasonal predictions for them, and avoids prediction deviation penalties through compensation; at the same time, it considers the addition of carbon capture equipment; and the operation of this strategy not only obtains revenue from the electricity sales market and the carbon market by coordinating the various parts within the VPP, but also takes into account the peak-to-valley difference rate of the load, ensuring the stability of grid operation. Finally, comparative experiments are used to prove the effectiveness and feasibility of the strategy proposed in this paper. Summary of the invention

[0004] The technical problem to be solved by the present invention is: in view of the deficiencies in existing research and technology, the present invention is an optimized operation strategy for a virtual power plant that aggregates wind power, photovoltaics, electric vehicles and considers carbon capture and gas turbines, solves the deficiencies in the VPP scheduling strategy for aggregated wind power, photovoltaics and electric vehicles in the original research and methods, and proposes an optimized operation method for a virtual power plant that aggregates electric vehicles and carbon capture, utilizes the chaos theory to assemble the peak frequency band of the empirical mode decomposition and GA (Genetic Algorithm) to optimize the back propagation neural network BP (Backpropagation Neural Networks), namely: GABP (Genetic Algorithm Optimization of Backpropagation Neural Networks), to make seasonal predictions, and secondly adopts a carbon capture system to reduce carbon emissions and participate in carbon market transactions. The present invention is simple to implement, optimizes the operating efficiency and stability of the VPP, and enhances effectiveness and practicality.

[0005] The above purpose is achieved through the following technical solutions:

[0006] The present invention provides a virtual power plant optimization operation method combining electric vehicles and carbon capture, the method comprising the following steps:

[0007] S1. By analyzing the behavior of electric vehicle users, we study the charging patterns of electric vehicles, daily mileage, charging time, and charging probability distribution, and use the Monte Carlo method to simulate the charging power of electric vehicles;

[0008] S2. Based on the content studied in step S1, formulate an optimized operation strategy for VPP, starting from the travel patterns of electric vehicle users, with the peak-to-valley difference rate and the income of VPP as the target, and considering relevant constraints; then, make seasonal output forecasts for wind power and photovoltaic power, and compare them with the corresponding seasons after K-Means clustering; finally, use gas turbines to supply energy for basic load, wind and solar forecast deviation, EV charging and carbon capture energy consumption, and settle wind and solar output and EV discharge to the electricity market for sale, while considering the emission reduction effect of carbon capture, and participate in the carbon market to sell carbon quotas for further benefits;

[0009] S3. Considering the seasonality of wind power and photovoltaic power output, compensation is used to avoid prediction deviation penalties. The peak frequency band of the chaotic theory set empirical mode decomposition and GA (Genetic Algorithm) are used to optimize the back propagation neural network BP (Backpropagation Neural Networks), namely: GABP (Genetic Algorithm Optimization of Backpropagation Neural Networks) to make seasonal predictions. Secondly, the carbon capture system is used to reduce carbon emissions and participate in carbon market transactions;

[0010] S4. Finally, the effectiveness and practicality of the proposed strategy are verified by comparing the ordered and disordered charging and discharging simulations.

[0011] Furthermore, step S1 specifically includes the following:

[0012] S1-1. The daily driving time t of EV follows the log-normal distribution, the charging time of EV follows the piecewise normal distribution, and the return time follows the normal distribution, which is: Its probability density function F(t) is:

[0013]

[0014] Where: μ t is the average daily driving time of EV, μ t Take 17.6, σ t is the variance of EV daily driving time, σ t Take 3.4;

[0015] S1-2. EV daily mileage s follows a log-normal distribution, that is: Its probability density F(s) is:

[0016]

[0017] Where: μ s is the mean daily driving distance, μ s Take 3.2, σ s is the variance of daily driving distance, σ s Take 0.88;

[0018] S1-3. For the study of electric vehicle charging time periods, the charging area is divided into four typical areas: residential areas, office areas, industrial areas, and commercial areas. Then, orderly charging, disorderly charging, delayed charging, and peak-shaving charging are selected as research scenarios. Taking the residential area under the disorderly charging scenario as an example, the peak and valley time periods of time-of-use electricity prices under the model are analyzed to determine the peak-valley relationship of each time period in the residential area. The determination method is as follows:

[0019]

[0020] Where: x is the load demand at the selected time point; P(x) is the peak-valley state; m and n represent the highest and lowest points on the load curve in the area in a day, respectively;

[0021] S1-4. The power consumption of EV is proportional to the mileage. The power consumption of EV is determined according to the mileage. That is, the power consumption of EV is obtained by subtracting the power consumption from the original power. The charging time T is obtained from the initial power. The calculation is as follows:

[0022]

[0023] Where: E SOC The amount of electricity when the electric vehicle starts to charge; is the original power of the electric vehicle; x D is the mileage; A is the battery capacity of the electric vehicle; W 100 is the energy consumption of electric vehicles per 100 kilometers; is the EV charging target value; φ is the charging efficiency; P charge is the charging power;

[0024] S1-5. The charging behavior of EVs is predicted through probability distribution: first, the number of EVs, battery capacity, and charging power are determined; then the remaining power of the EVs at the start of charging is calculated based on the mileage; then the time to start charging is determined based on the probability distribution of EVs starting charging; finally, the charging load of each EV is calculated, and the charging load curves of all EVs are accumulated and output.

[0025] Furthermore, step S2 specifically includes:

[0026] The revenue of VPP comes from wind and solar units participating in the electricity sales market and EV discharge and selling carbon quotas in the carbon market. VPP requires gas turbines to bear the base load, forecast deviation, EV charging demand, carbon capture operation, and battery loss of EV discharge. The objective function of the operation strategy model is as follows:

[0027]

[0028] Where: R WT , R PV , R EV,DIS , R C They are wind power sales revenue, photovoltaic power sales revenue, EV discharge revenue, and carbon emission revenue; R BASE , R BIAS , R CHARGE , R EV,BATTER , R FC , REC They are base load, compensation for the deviation between wind and solar power forecast and actual value undertaken by gas turbine, cost of EV charging demand undertaken by gas turbine, EV battery loss cost, CO2 storage cost, and carbon capture energy consumption cost;

[0029] Among them, the revenue from wind power sales, photovoltaic power sales, and EV discharge revenue are calculated based on the time-of-use electricity price:

[0030]

[0031] Where: Q WT , Q PV , Q EV,DIS It is divided into wind power forecast power generation, photovoltaic forecast power generation, and electric vehicle discharge. P is the time-of-use electricity price.

[0032] The carbon emission benefits are calculated as follows:

[0033]

[0034] Where: k C , E, E C 、E P They are carbon trading price, carbon quota, carbon capture, and carbon emissions respectively; Q, B, and F are the power supply of the unit, the power supply benchmark value of the category to which it belongs, and the heat correction coefficient of the unit respectively;

[0035] Wind and solar forecast deviations and EV charging use gas turbines to compensate and bear:

[0036]

[0037] In the formula, Q BIAS , Q CHARGE , Q GAS They are the deviation compensation power, EV charging power, and gas turbine power cost;

[0038] EV battery depletion costs are:

[0039]

[0040] Where: B is the linear relationship between battery life and cycle number; γ_t is the cycle power of EV within charging time t; C is the battery replacement cost; and The EV power at time T and T-1. When the electric vehicle is discharged, the battery life will be consumed. If there is no discharge, there will be no life consumption.

[0041] The cost of carbon storage and the energy consumption of carbon capture is:

[0042]

[0043] Where: k CQ is the storage coefficient of carbon capture equipment; P B , P C 、w C They are the baseline energy consumption, operating energy consumption, and operating energy consumption required to process unit carbon dioxide of the carbon capture equipment;

[0044] The charging and discharging of EVs in VPP will change the peak-to-valley difference of the load. In order to avoid the phenomenon of adding peaks to peaks and to achieve peak shaving and valley filling, the load fluctuation and the power consumption during peaks and valleys must be taken into account. The change value of the peak-to-valley difference rate F2 is set as the objective function. The peak-to-valley difference rate η of the objective function F2 load is as follows:

[0045]

[0046] Where: F2 represents the change value of the peak-to-valley difference rate; P SUM,MIN , P SUM,MAX is the maximum and minimum value of the sum of the original load and the EV’s charging and discharging power; P SUM is the sum of the original load and the charging and discharging power of the EV; P initial is the original load; P i,EV is the charging and discharging power of the i-th EV; the total number of EVs is N;

[0047] Under the premise of ensuring the benefits of virtual power plants and taking into account the load changes of stable power grids, the two objective functions are normalized using the linear weighted method:

[0048]

[0049] Where: F is a multi-objective optimization function; and is the maximum value of the two objective functions; α1 and α2 are the optimization weights of objective functions F1 and F2 respectively;

[0050] Secondly, we need to consider the relevant constraints, as follows:

[0051] 1) Constraints on wind turbines and photovoltaic units

[0052]

[0053] Where: Represent the upper and lower limits of wind power and photovoltaic power generation units respectively; P WT , P PV For wind power and photovoltaic power output;

[0054] 2) Electric vehicle constraints

[0055] 0≤P CHARGE ≤P EV,MAX

[0056] 0≤P DIS ≤P EV,MAX

[0057] Where: P CHARGE and P DIS is the charging and discharging power of EV; P EV,MAX is the maximum value of charge and discharge power;

[0058] 3) Battery available capacity constraints

[0059]

[0060] Where: E EV The remaining available battery power; E RATED is the rated capacity of the battery; SOC EV E is the ratio of the available capacity of the battery to the rated capacity; CHARGE,DIS SOC is the charge and discharge capacity of EV; CHARGE,DIS SOC is the ratio of EV charge and discharge to rated capacity; EV,LAST SOC is the percentage of electricity after charging and discharging; MIN and SOC MAX They are the minimum and maximum values ​​of the battery capacity respectively.

[0061] 4) User expected power constraints

[0062] E i，MIN ≤E i，EV ≤E i,MAX

[0063] Where: E i，EV is the expected power consumption of the i-th EV user; E i,MIN and E i,MAX are the maximum and minimum values ​​of the power expected by the user of the i-th EV.

[0064] Furthermore, the specific process of step S3 is:

[0065] When the input is (X=X1,X2…X j …X m ), the output Y of the BP neural network wind power model is:

[0066]

[0067] In the above formula, w jk is the weight of input layer node j and hidden layer node k, θ j is the threshold of hidden layer node j; w j is the threshold between hidden layer node j and output layer node; θ is the threshold of output layer node; the total number of input layer nodes is M, the total number of hidden layer nodes is Np ;

[0068] BP has two stages in prediction: model training and prediction. During training, weights and thresholds are continuously adjusted to achieve parameter mapping; during prediction, the prediction samples are substituted into the trained prediction model to output the results. Follow the steps below:

[0069] 1) Forward transmission of signals

[0070] The input of hidden node j under the pth sample and output for:

[0071]

[0072] The input of the output node at the pth sample Output y p for:

[0073]

[0074] 2) Error reverse transmission:

[0075] Through gradient descent, the weights and thresholds are adjusted to make the error meet the accuracy requirements. The error calculation formula is:

[0076]

[0077] Where: d p and p are the actual value and output value under sample P;

[0078] In the process of BP neural network training, the weight threshold is updated by forward propagation of data and error backward transmission. After the initial parameters are selected, gradient descent takes them as the starting point to optimize and update the parameters. However, the convergence value is not the global optimum because the reliability and stability are affected by the initial values. The genetic algorithm GA has a good global search capability, so GA is introduced to solve this problem, that is, the parameters are used as the decision variables of the problem, and the accuracy of the model is used as the objective function of the problem. The specific process is as follows:

[0079] ⑦Set the population size S P , crossover probability P c , mutation probability P c , Genetic Evolution Algebra G max parameter;

[0080] ⑧ Initialize the population, encode, and initialize individuals;

[0081] ⑨ Determine the fitness function for individual evaluation, and use the inverse of the mean square error as the fitness function H;

[0082]

[0083] Where: N is the total number of training samples; y p is the predicted output; d p is the actual output;

[0084] ⑩Selection, where node j is selected with probability p j for:

[0085]

[0086] Where: N0 is the number of individuals; H i is the fitness function of the i-th individual;

[0087] Crossover, cross two individuals according to the crossover probability:

[0088]

[0089] Where: a pq represents the qth gene of p individuals; a zx represents the x-th gene of individual z; γ0 is between 0 and 1;

[0090] Mutation, mutate individuals according to the probability of mutation:

[0091]

[0092] Where: a min and a max for a ij The lower and upper limits of ; G is the number of evolutions; γ1 is a random number between 0 and 1; γ2 is a random number;

[0093] Determine whether the termination conditions are met. If so, decode and output the weights and thresholds. If not, adaptively adjust the crossover probability and mutation probability and execute again from ④.

[0094] Compared with the prior art, the advantages of the present invention are:

[0095] 1) Based on the behavior of electric vehicle users, the Monte Carlo method is used to simulate the charging load of electric vehicles based on parameters such as daily mileage, starting charging time, and charging time. On this basis, the peak-to-peak and related cost issues caused by disorderly charging of electric vehicles are taken into account.

[0096] 2) Considering the seasonality of wind and solar power output, the GABP model is used to make seasonal predictions, and the seasons are clustered by K-Means as a reference for the prediction results, which shows the feasibility and accuracy of the prediction model. This avoids the situation where the fitting prediction of distributed clean energy use in virtual power plants has no real and practical significance.

[0097] 3) Taking virtual power plant revenue and load peak-valley difference rate as two goals, taking into account the charging and discharging costs and revenues of electric vehicles and battery losses, as well as related constraints, experiments were conducted using ordered and disordered comparisons, highlighting that the proposed method can not only participate in the electricity sales market to increase the economic benefits of VPP, but also effectively reduce the peak-valley difference of the power grid, improve the economy and safety of the power grid, reduce carbon emissions through carbon capture, and participate in carbon market transactions. The effectiveness of the proposed strategy was verified through simulation. BRIEF DESCRIPTION OF THE DRAWINGS

[0098] Figure 1 is a modeling flow chart of simulating EV charging load using the Monte Carlo method in this embodiment;

[0099] Figure 2 This is the flow chart of GA optimizing BP neural network in this embodiment;

[0100] Figure 3 is a wind power clustering result diagram in this embodiment;

[0101] Figure 4 is a photovoltaic clustering result diagram in this embodiment;

[0102] Figure 5 This is a wind power prediction comparison chart in this embodiment;

[0103] Figure 6 is a photovoltaic prediction comparison diagram in this embodiment;

[0104] Figure 7 is the Pareto front curve diagram of the dual objective function in this embodiment;

[0105] Figure 8 This is a comparison diagram of base load, orderly charging, and disorderly charging in this embodiment;

[0106] Fig. 9 It is a curve diagram of EV charging, discharging and load variation in this embodiment. DETAILED DESCRIPTION

[0107] The present invention is further described below in conjunction with the accompanying drawings and specific preferred embodiments, but the protection scope of the present invention is not limited thereby.

[0108] In this embodiment, the virtual power plant optimization operation method combining electric vehicles and carbon capture includes the following steps:

[0109] S1. Analyze the behavior of electric vehicle users, study the charging rules of electric vehicles, simulate user charging loads, and propose virtual power plant operation strategies;

[0110] S2. Based on the travel patterns of electric vehicle users studied above, with the peak-to-valley difference rate and VPP revenue as the target, and considering the relevant constraints, seasonal output forecasts are made for wind power and photovoltaic power, and compared with the corresponding seasons after K-Means clustering. Finally, the wind and photovoltaic output and EV discharge are cleared and sold in the electricity market;

[0111] S3. Considering the seasonality of wind power and photovoltaic power output, compensation is used to avoid prediction bias penalties, and the peak frequency band of the chaotic theory set empirical mode decomposition and GA optimization back propagation neural network BP, namely: GABP, are used to make seasonal predictions. Secondly, carbon capture systems are used to reduce carbon emissions and participate in carbon market transactions;

[0112] S4. Finally, the effectiveness and practicality of the proposed strategy were verified through a comparative simulation of ordered and disordered charging and discharging.

[0113] In this embodiment, the generation of new energy is uncertain and random, and the disordered charging and discharging of EVs will lead to the accumulation of peak power consumption, making the peak-valley difference of the power grid more obvious. At the same time, the load borne by the power source will also generate a large amount of carbon emissions. VPP can not only solve the above problems, but also participate in related markets to obtain economic benefits. Therefore, optimizing VPP operations and ensuring economic benefits and costs have become the current hot research directions.

[0114] In this embodiment, various behaviors of electric vehicle users are studied in step 1, mainly the charging rules, daily mileage, charging time period, and charging probability distribution of electric vehicles. Finally, the Monte Carlo method is used to simulate the charging power of electric vehicles. The research model is as follows:

[0115] ① According to the survey and statistics of the US Highway Safety Administration, the daily driving time of EVs follows a log-normal distribution.

[0116] The charging time of EV at the beginning follows a piecewise normal distribution, and the return time also follows a normal distribution, which is t~

[0117] Its probability density function is:

[0118]

[0119] Where: μ t is the average daily driving time of EV, μ t Take 17.6, σ t is the variance of EV daily driving time, σ t Take 3.4;

[0120] ②EV daily mileage s follows a log-normal distribution, that is: Its probability density is:

[0121]

[0122] Where: μ s is the mean daily driving distance, μ s Take 3.2, σ s is the variance of daily driving distance, σ s Take 0.88;

[0123] ③ For the study of electric vehicle charging time periods, the charging area can be divided into four typical areas: residential areas, office areas, industrial areas, and commercial areas, and then orderly charging, disorderly charging, delayed charging, peak-shaving charging and other scenarios can be selected for research. Taking the residential area under the disorderly charging scenario as an example, the peak and valley time periods of time-of-use electricity prices under this model are analyzed to determine the peak-valley relationship of each time period in the residential area. The determination method is as follows.

[0124]

[0125] Where: x is the load demand at the selected time point; P(x) is the peak-valley state; m and n represent the highest and lowest points on the load curve in the area within a day, respectively.

[0126] ④ The power consumption of EV is proportional to the mileage. The power consumption of EV can be determined based on the mileage. That is, the power consumption of EV can be obtained by subtracting the power consumption from the original power. The charging time T can be obtained from the initial power. The calculation is as follows:

[0127]

[0128] Where: ESOC is the power of the electric vehicle when it starts charging; is the original power of the electric vehicle; x D is the mileage; A is the battery capacity of the electric vehicle; W 100 It is the energy consumption of electric vehicle per 100 kilometers. EV charging target value; is the charging efficiency; P charge is the charging power.

[0129] ⑤ The charging behavior of EVs can be predicted through probability distribution. First, determine the number of EVs, battery capacity, and charging power; then calculate the remaining power of EVs based on the mileage; and then determine the time to start charging based on the probability distribution of EVs starting to charge;

[0130] Finally, the charging load of each EV is calculated, and the charging load curve of all EVs is accumulated and output. Figure 1 shown.

[0131] In this embodiment, the BP neural network is a multi-layer feedforward neural network. The basic idea is to use the gradient reduction method to adjust the model parameters in the negative gradient direction of the objective function so that the deviation between the actual and expected output values ​​of the network is minimized. It consists of three parts: input layer, hidden layer, and output layer. It includes two processes: forward transmission and error reverse transmission.

[0132] When the input is (X=X1,X2…X j …X m ), the output Y of the BP neural network wind power model is:

[0133]

[0134] In the above formula, w jk is the weight of input layer node j and hidden layer node k, θ j is the threshold of hidden layer node j; w j is the threshold between hidden layer node j and output layer node; θ is the threshold of output layer node; the total number of input layer nodes is M, the total number of hidden layer nodes is N p .

[0135] There are two stages in BP prediction: model training and prediction. During training, the weights and thresholds are constantly adjusted to achieve parameter mapping. During prediction, the prediction samples are substituted into the trained prediction model to output the results. Follow the steps below:

[0136] 1) Forward transmission of signals

[0137] The input of hidden node j under the pth sample and output for:

[0138]

[0139] The input of the output node at the pth sample Output y p for:

[0140]

[0141] 2) Error reverse transmission:

[0142] Through gradient descent, the weights and thresholds are adjusted to make the error meet the accuracy requirements. The error calculation formula is:

[0143]

[0144] Where: d p and p are the actual value and output value under the sample.

[0145] During the BP neural network training process, the weight threshold is updated by forward propagation of data and error backward transmission. After the initial parameters are selected, gradient descent uses them as the starting point to optimize and update the parameters, but the convergence value is not the global optimum because the reliability and stability are affected by the initial values. The genetic algorithm GA has good global search capabilities, so GA is introduced to solve this problem. That is, the parameters are used as the decision variables of the problem, and the accuracy of the model is used as the objective function of the problem. The specific process is as follows:

[0146] ①Set the population size S P , crossover probability P c , mutation probability P c , Genetic Evolution Algebra G max parameter.

[0147] ② Initialize the population, perform encoding, and initialize individuals.

[0148] ③ Determine the fitness function for individual evaluation and take the inverse of the mean square error as the fitness function H.

[0149]

[0150] Where: N is the total number of training samples; y p is the predicted output; d p is the actual output.

[0151] ④Selection, node j selection probability p j for:

[0152]

[0153] Where: N0 is the number of individuals; H j is the fitness function of the jth individual.

[0154] ⑤Crossover, cross two individuals according to the crossover probability:

[0155]

[0156] Where: a pq represents the qth gene of p individuals; a zx Represents the x-th gene of individual z; γ0 is between 0 and 1.

[0157] ⑥ Mutation, according to the probability of mutation, individual mutation:

[0158]

[0159] Where: a min and a max for a ij The lower and upper limits of ; G is the number of evolutions; γ1 is a random number between 0 and 1; γ2 is a random number.

[0160] ⑦ Determine whether the termination conditions are met. If so, decode and output the weights and thresholds. If not, adaptively adjust the crossover probability and mutation probability and execute again from ④.

[0161] The specific flow chart is as follows Figure 2 shown.

[0162] Therefore, an embodiment is selected to verify the effectiveness of the present invention. Under the premise of Monte Carlo simulation of basic load, EV disordered and orderly charging and discharging are set for comparative simulation. The wind power and photovoltaic power station power and meteorological data of a certain place in Xinjiang Uygur Autonomous Region from 2018 to 2022 are used as samples. The simulation time is 24h and the step length is 1h. The YALMIP toolkit in MATLAB calls the CPLEX solver for solution.

[0163] The parameters are set as follows:

[0164] The VPP is formed by selecting photovoltaic and wind turbines with a total rated capacity of 10MW, 1,000 electric vehicles, and gas turbines whose equipment parameters and "gas consumption per kilowatt-hour" are calculated by Siemens. The calorific value of natural gas is 8400Cal / m 3 , calculated based on 85% combustion efficiency and 40% power generation efficiency, that is, 3.321 kWh of electricity can be generated per cubic meter of natural gas. The price of natural gas is 1.5 yuan / m 3 , the cost of generating 1 kWh is 0.452 yuan. The carbon trading price is 92.168 yuan / t; the power supply benchmark value of the unit is 0.3901 (tCO2 / MW h); the unit heat correction factor is 0.88; the carbon capture equipment storage factor is 35 yuan / t; the benchmark energy consumption is 3MW; the unit processing energy consumption of carbon dioxide is 0.269MW h / t; the carbon capture efficiency is calculated as 90%; the unit output carbon emission of the gas turbine is 0.3tCO2 / MW h.

[0165] The electricity sales price is calculated based on the time-of-use electricity price. The time-of-use electricity price and electric vehicle related parameters are shown in Table 1 and Table 2.

[0166] Table 1 Time-of-use electricity price

[0167]

[0168] Table 2 Electric vehicle parameters

[0169]

[0170] First, K-Means clustering is performed on the wind and solar power output from March to May 2021, and then the output scenes with higher probability are selected as real data and compared with the wind and solar power data predicted by GABP. Because wind power and photovoltaic power output scenes are diverse, representative scenes are selected as real output scenes. The wind and solar power clustering results are as follows: Figure 3 , 4 shown.

[0171] The output scenario with the highest probability is selected as the actual output scenario, and GABP is used to make seasonal forecasts of wind power and photovoltaic output from March to May from 2021 to 2024. For the part where the predicted value is smaller than the actual value, gas turbines are used to compensate for the output, while for the part where the predicted value is larger than the actual value, no compensation is required. The comparison results of wind power and photovoltaic output forecasts and actual values ​​are shown in the figure below. Figure 5 , 6 shown.

[0172] Since the two objective functions have different dimensions, the appropriate weight is determined by the Pareto frontier. EV is the main cause of load fluctuation, so the EV discharge benefit and peak-to-valley difference rate are used as reference. Figure 7 It can be seen that when the income weight becomes larger, the EV discharge volume increases, and the peak-to-valley difference rate gradually increases. When the peak-to-valley difference reaches 61.78%, the income growth is slow, while the peak-to-valley difference rate rises rapidly. When the peak-to-valley difference rate reaches 65.74%, a new peak will appear. Therefore, the weight range of F1 is selected to be 0.45-0.5, and the weight range of F2 is selected to be 0.5-0.55.

[0173] The original load obtained by Monte Carlo simulation is as follows Figure 8 It can be seen that the original load has two peaks at 10:00-13:00 and 18:00-21:00, with a peak-to-valley difference of 74.52%. On this basis, when electric vehicles are charged and discharged disorderly, they will be charged at the two peaks of the basic load and discharged at the valley. That is, peaks are added to peaks and valleys are even more valleys. Disorderly charging and discharging will also affect the economic benefits obtained by the virtual power plant from the discharge of electric vehicles.

[0174] When orderly charging and discharging takes into account the peak-to-valley difference and the charging and discharging benefits and costs, Fig. 9 It can be seen that it will avoid charging during peak hours, discharge to obtain the profit of high-price sales, and charge during off-peak hours to reduce the electricity price expenditure. It can not only suppress the fluctuation of the power grid, but also improve the stable operation of the power system.

[0175] Under the VPP operation with EV and carbon capture device, the total revenue of VPP operation includes wind power and photovoltaic output revenue, carbon market revenue, and EV discharge revenue. Among them, the wind and solar output is 164.3317MW, and the electricity sales revenue is 107,127.835 yuan; the carbon emission quota is 19.035t, the carbon capture is 14.971t, and the carbon emission is 1.663t. The carbon trading volume that can participate in the carbon market is 32.323t, which can earn 2,980.99 yuan; the EV discharge revenue is 5,377.266 yuan. The total revenue is 110,108.825 yuan. The costs include wind and solar deviation compensation, basic load energy supply, EV battery replacement, carbon capture energy consumption cost, carbon storage cost, and EV charging cost. The total is 32,851.9366 yuan.

[0176] Among them, the wind and solar forecast deviation is 8.877MW, and the deviation penalty is calculated as 6715.41 yuan based on the time-of-use electricity price. However, the operation strategy in this paper uses gas turbines for compensation, and the deviation electricity cost is 4012.404 yuan, which is 40.25% lower than the cost of the deviation penalty case.

[0177] When EVs participate in an unordered manner, the discharge income and charging cost are 1102.259 yuan and 1291.816 yuan respectively; the peak-to-valley difference is 77.57%, which is 3.02% higher than the base load. When EVs participate in an orderly manner, the discharge income and charging cost are 5377.266 yuan and 3200.16 yuan respectively, which is 387.84% higher than the unordered income. The peak-to-valley difference is 61.78%, which is 11.79% lower than the base load.

Claims

1. A virtual power plant optimization operation method combining electric vehicles and carbon capture, It is characterized in that The method comprises the following steps: S1. By analyzing the behavior of electric vehicle users, we study the charging patterns of electric vehicles, daily mileage, charging time, and charging probability distribution, and use the Monte Carlo method to simulate the charging power of electric vehicles; S2. Based on the content studied in step S1, formulate an optimized operation strategy for VPP, starting from the travel patterns of electric vehicle users, with the peak-to-valley difference rate and the income of VPP as the target, and considering relevant constraints; then, make seasonal output forecasts for wind power and photovoltaic power, and compare them with the corresponding seasons after K-Means clustering; finally, use gas turbines to supply energy for basic load, wind and solar forecast deviation, EV charging and carbon capture energy consumption, and settle wind and solar output and EV discharge to the electricity market for sale, while considering the emission reduction effect of carbon capture, and participate in the carbon market to sell carbon quotas for further benefits; S3. Considering the seasonality of wind power and photovoltaic power output, compensation is used to avoid prediction deviation penalties. The peak frequency band of the chaotic theory set empirical mode decomposition and GA (Genetic Algorithm) are used to optimize the back propagation neural network BP (Backpropagation Neural Networks), namely: GABP (Genetic Algorithm Optimization of Backpropagation Neural Networks) to make seasonal predictions. Secondly, the carbon capture system is used to reduce carbon emissions and participate in carbon market transactions; S4. Finally, the effectiveness and practicality of the proposed strategy are verified by comparing the ordered and disordered charging and discharging simulations.

2. The method for optimizing operation of a virtual power plant combining electric vehicles and carbon capture according to claim 1, Features: Step S1 specifically includes the following: S1-1. The daily driving time t of EV follows the log-normal distribution, the charging time of EV follows the piecewise normal distribution, and the return time follows the normal distribution, which is: Its probability density function F(t) is: Where: μ t is the average daily driving time of EV, μ t Take 17.6, σ t is the variance of EV daily driving time, σ t Take 3.4; S1-2. EV daily mileage s follows a log-normal distribution, that is: Its probability density F(s) is: Where: μ s is the mean daily driving distance, μ s Take 3.2, σ s is the variance of daily driving distance, σ s Take 0.88; S1-3. For the study of electric vehicle charging time periods, the charging area is divided into four typical areas: residential areas, office areas, industrial areas, and commercial areas. Then, orderly charging, disorderly charging, delayed charging, and peak-shaving charging are selected as research scenarios. Taking the residential area under the disorderly charging scenario as an example, the peak and valley time periods of time-of-use electricity prices under the model are analyzed to determine the peak-valley relationship of each time period in the residential area. The determination method is as follows: Where: x is the load demand at the selected time point; P(x) is the peak-valley state; m and n represent the highest and lowest points on the load curve in the area in a day, respectively; S1-4. The power consumption of EV is proportional to the mileage. The power consumption of EV is determined according to the mileage. That is, the power consumption of EV is obtained by subtracting the power consumption from the original power. The charging time T is obtained from the initial power. The calculation is as follows: Where: E SOC The amount of electricity when the electric vehicle starts to charge; is the original power of the electric vehicle; x D is the mileage; A is the battery capacity of the electric vehicle; W 100 is the energy consumption of electric vehicles per 100 kilometers; EV charging target value; is the charging efficiency; P charge is the charging power; S1-5. The charging behavior of EVs is predicted through probability distribution: first, the number of EVs, battery capacity, and charging power are determined; then the remaining power of the EVs at the start of charging is calculated based on the mileage; then the time to start charging is determined based on the probability distribution of EVs starting charging; finally, the charging load of each EV is calculated, and the charging load curves of all EVs are accumulated and output.

3. The method for optimizing operation of a virtual power plant combining electric vehicles and carbon capture according to claim 2, Features: Step S2 specifically includes: The income of VPP comes from wind and solar units participating in the electricity sales market and EV discharge and selling carbon quotas in the carbon market. VPP requires gas turbines to bear the base load, forecast deviation, EV charging demand, carbon capture operation, and battery loss of EV discharge. The objective function F of the operation strategy model is 1 as follows: Where: R WT , R PV , R EV,DIS , R C They are wind power sales revenue, photovoltaic power sales revenue, EV discharge revenue, and carbon emission revenue; R BASE , R BIAS , R CHARGE , R EV,BATTER , R FC , R EC They are base load, compensation for the deviation between wind and solar power forecast and actual value undertaken by gas turbine, cost of EV charging demand undertaken by gas turbine, EV battery loss cost, CO2 storage cost, and carbon capture energy consumption cost; Among them, the revenue from wind power sales, photovoltaic power sales, and EV discharge revenue are calculated based on the time-of-use electricity price: Where: Q WT , Q PV , Q EV,DIS It is divided into wind power forecast power generation, photovoltaic forecast power generation, and electric vehicle discharge. P is the time-of-use electricity price. The carbon emission benefits are calculated as follows: Where: k C , E, E C 、E P They are carbon trading price, carbon quota, carbon capture, and carbon emissions respectively; Q, B, and F are the power supply of the unit, the power supply benchmark value of the category to which it belongs, and the heat correction coefficient of the unit respectively; Wind and solar forecast deviations and EV charging use gas turbines to compensate and bear: In the formula, Q BIAS , Q CHARGE , Q GAS They are the deviation compensation power, EV charging power, and gas turbine power cost; EV battery depletion costs are: Where: B is the linear relationship between battery life and cycle number; γ_t is the cycle power of EV within charging time t; C is the battery replacement cost; and The EV power at time T and T-1. When the electric vehicle is discharged, the battery life will be consumed. If there is no discharge, there will be no life consumption. The cost of carbon storage and the energy consumption of carbon capture is: Where: k CQ is the storage coefficient of carbon capture equipment; P B , P C 、w C They are the baseline energy consumption, operating energy consumption, and operating energy consumption required to process unit carbon dioxide of the carbon capture equipment; The charging and discharging of EVs in VPP will change the peak-to-valley difference of the load. In order to avoid the phenomenon of adding peaks to peaks and to achieve peak shaving and valley filling, it is necessary to take into account the load fluctuation and the power consumption during peaks and valleys. 2 Set as the objective function, then the objective function F 2 The peak-to-valley rate η of the load is as follows: Where: F 2 Indicates the change value of the peak-to-valley difference rate; P SUM,MIN , P SUM,MAX is the maximum and minimum value of the sum of the original load and the EV’s charging and discharging power; P SUM is the sum of the original load and the charging and discharging power of the EV; P initial is the original load; P i,EV is the charging and discharging power of the i-th EV; the total number of EVs is N; Under the premise of ensuring the benefits of virtual power plants and taking into account the load changes of stable power grids, the two objective functions are normalized using the linear weighted method: Where: F is a multi-objective optimization function; and is the maximum value of the two objective functions; α 1 and α 2 The objective function F 1 and F 2 The optimization weight of Secondly, we need to consider the relevant constraints, as follows: 1) Constraints on wind turbines and photovoltaic units Where: Represent the upper and lower limits of wind power and photovoltaic power generation units respectively; P WT , P PV For wind power and photovoltaic power output; 2) Electric vehicle constraints 0≤P CHARGE ≤P EV,MAX 0≤P DIS ≤P EV,MAX Where: P CHARGE and P DIS is the charging and discharging power of EV; P EV,MAX is the maximum value of charge and discharge power; 3) Battery available capacity constraints Where: E EV The remaining available battery power; E RATED is the rated capacity of the battery; SOC EV E is the ratio of the available capacity of the battery to the rated capacity; CHARGE,DIS SOC is the charge and discharge capacity of EV; CHARGE,DIS SOC is the ratio of EV charge and discharge to rated capacity; EV,LAST SOC is the percentage of electricity after charging and discharging; MIN and SOC MAX They are the minimum and maximum values ​​of the battery capacity respectively. 4) User expected power constraints AND i，MIN ≤E i，EV ≤E i,MAX Where: E i，EV is the expected power consumption of the i-th EV user; E i,MIN and E i,MAX are the maximum and minimum values ​​of the power expected by the user of the i-th EV.

4. The method for optimizing operation of a virtual power plant combining electric vehicles and carbon capture according to claim 3, Features: The specific process of step S3 is: When the input is (X=X 1 ,X 2 …X j …X m ), the output Y of the BP neural network wind power model is: In the above formula, w jk is the weight of input layer node j and hidden layer node k, θ j is the threshold of hidden layer node j; w j is the threshold between hidden layer node j and output layer node; θ is the threshold of output layer node; the total number of input layer nodes is M, the total number of hidden layer nodes is N p ; BP has two stages in prediction: model training and prediction. During training, weights and thresholds are continuously adjusted to achieve parameter mapping; during prediction, the prediction samples are substituted into the trained prediction model to output the results. Follow the steps below: 1) Forward transmission of signals The input of hidden node j under the pth sample and output for: The input of the output node at the pth sample Output y p for: 2) Error reverse transmission: Through gradient descent, the weights and thresholds are adjusted to make the error meet the accuracy requirements. The calculation formula of the error E is: Where: d p and p are the actual value and output value under sample P; In the process of BP neural network training, the weight threshold is updated by forward propagation of data and error backward transmission. After the initial parameters are selected, gradient descent takes them as the starting point to optimize and update the parameters. However, the convergence value is not the global optimum because the reliability and stability are affected by the initial values. The genetic algorithm GA has a good global search capability, so GA is introduced to solve this problem, that is, the parameters are used as the decision variables of the problem, and the accuracy of the model is used as the objective function of the problem. The specific process is as follows: ①Set the population size S P , crossover probability P c , mutation probability P c , Genetic Evolution Algebra G max parameter; ② Initialize the population, encode, and initialize individuals; ③ Determine the fitness function for individual evaluation, and use the inverse of the mean square error as the fitness function H; Where: N is the total number of training samples; y p is the predicted output; d p is the actual output; ④Selection, where node j is selected with probability p j for: Where: N 0 is the number of individuals; H i is the fitness function of the i-th individual; ⑤Crossover, cross two individuals according to the crossover probability: Where: a pq represents the qth gene of p individuals; a zx represents the x-th gene of individual z; γ0 is between 0 and 1; ⑥ Mutation, according to the probability of mutation, individual mutation: Where: a min and a max for a ij The lower and upper limits of ; G is the number of evolutions; γ 1 is a random number between 0 and 1; 2 is a random number; Determine whether the termination conditions are met. If so, decode and output the weights and thresholds. If not, adaptively adjust the crossover probability and mutation probability and execute again from ④.

Images