A distributed energy aggregation and coordinated control method for virtual power plants

Abstract

The present invention belongs to the field of virtual power plant control technology, and specifically relates to a method for coordinated control of distributed energy aggregation in a virtual power plant, including: S1, constructing a virtual power plant system with multiple distributed energy sources including charging piles, selecting multiple groups of operating status data and charging loads of each energy subsystem equipment as samples; S2, dividing the control categories of multiple groups of operating status data based on an improved Kohonen neural network; S3, controlling the load output under different control categories based on a simulated annealing immune control algorithm network. The urban virtual power plant system of the present invention provides the daily required power load through the distribution network and the photovoltaic power generation system in the system, and utilizes the regulating effect of the energy storage system and the charging and swapping load of electric vehicles to achieve the effect of peak shaving and valley filling and smoothing load fluctuations.

Description

Technical Field

[0001] The present invention belongs to the technical field of virtual power plant control, and in particular relates to a distributed energy aggregation and coordinated control method for a virtual power plant. Background Art

[0002] Due to the extensive use of fossil fuels, environmental problems and energy crises are becoming increasingly severe. It is urgent to build a new power system based on renewable energy. The so-called new power system is an upgrade of the traditional power system. It is to achieve clean, intelligent and low-carbon development on the basis of the original power system. Its architecture is as follows: Figure 1 As shown; at the same time, with the large-scale access of electric vehicles, the demand-side load has changed from the traditional rigid and consumer-oriented to the flexible and both production and consumption-oriented, and its flexibility has been greatly improved.

[0003] As a new type of green, clean, zero-emission transportation, electric vehicles have penetrated into the transportation network and power network on a large scale. They can no longer be simply regarded as independent transportation tools, but as a group of bridges that can connect transportation and energy. Each electric vehicle can become a virtual energy carrier. At the same time, with the popularization of large-scale electric vehicles, their influence on the power system is also increasing. If their behavior is not controlled, it will have a huge adverse impact on the safety and economy of the power system. In particular, the superposition of electric vehicle charging load and grid load will further aggravate the peak-to-valley difference of the grid, thereby leading to serious problems such as over-limit of distribution network line flow and transformer overload.

[0004] The disordered charging characteristics of electric vehicles connected on a large scale will cause an imbalance in system voltage and frequency, affecting the quality of power and further increasing operating costs. Predicting and guiding the disordered charging behavior of electric vehicles, so that electric vehicles can be connected to charge when the grid is at low load as much as possible, reducing the peak-to-valley difference of the grid, and rationally and scientifically controlling the charging behavior of electric vehicles can greatly reduce grid fluctuations. Electric vehicles have flexibility and dispatchability that other types of loads do not have. Some electric vehicles have the ability to feed battery power back to the grid, that is, such electric vehicles have the dual identity of load and energy storage. Using electric vehicles as energy storage devices can well compensate for the volatility of wind and solar power generation, creating greater space for the development of user-side energy storage.

[0005] In order to optimize the orderly charging of electric vehicles, it is proposed to integrate the distributed energy of the distribution network through virtual power plants and formulate reasonable power sales strategies to guide the orderly access of electric vehicles to the grid to achieve coordinated complementarity and overall optimization among multiple energy sources. Virtual power plants integrate large-scale renewable energy and flexible loads, aggregate the supply side and demand side through advanced acquisition, communication, control and other technologies, and achieve coordinated and optimized operation among multiple subjects through high-level data analysis and intelligent algorithms, so as to achieve energy optimization allocation and provide flexible support for economic dispatch. At the same time, the demand-side management of virtual power plants smoothes the volatility of renewable energy generation. With the help of a large number of electric vehicles in the distribution network and the characteristics of two-way energy flow when electric vehicles are connected to the power grid, the economic dispatch of distributed generators in virtual power plants is realized, while the pollution generated by the operation of virtual power plants is reduced, the resource allocation effect is fully utilized, the intermittent nature of renewable energy generation is smoothed, and the coordinated and optimized dispatch of distributed generation units is realized.

[0006] As a special type of power plant, the virtual power plant achieves coordinated optimization of distributed power generation, energy storage systems, controllable loads, electric vehicles, etc. through advanced information and communication technologies and software systems, so as to participate in the coordinated management of power supply in the power market and power grid operation. From a technical perspective, the virtual power plant with large-scale charging piles aggregates different types of distributed power sources such as distributed power sources, controllable flexible loads, energy storage systems, electric vehicles, etc. through advanced control, metering, communication and other technologies, and achieves coordinated and optimized operation of multiple distributed energy sources through a higher-level software architecture, enabling it to participate in the operation of the power market and ancillary service market, and fully tap the value and benefits that distributed energy brings to the power grid and users.

[0007] Existing technology 1 "Duan Pian, Zhu Jianquan, Liu Mingbo. Optimal scheduling of virtual power plants based on double-layer fuzzy chance constrained programming [J]. Transactions of the Chinese Society of Electrotechnical Engineering, 2016, 31(09): 58-67, using fuzzy parameters to describe the uncertainty of distributed renewable energy generation and load" optimizes the scheduling of virtual power plants through double-layer fuzzy chance constrained programming, and conducts a detailed analysis of carbon emission costs, charging and discharging costs of electric vehicles, and supply and demand imbalance costs between the power grid and the virtual power plant.

[0008] Existing technology 2 "Lou Wei, Zhai Haibao, Xu Ling, et al. Research on wind power-energy storage-electric vehicle joint frequency regulation control strategy [J]. Renewable Energy, 2021, 39(12): 1648-1654" proposes a wind power-energy storage-electric vehicle joint frequency regulation control strategy, using the energy storage system to make up for the insufficient wind power frequency regulation, and proposes a frequency regulation strategy that takes into account the response of electric vehicle clusters.

[0009] Existing Technology 3 "Tan Qinliang, Dai Mei, Mei Shufan. Research on power system dispatch considering electric vehicle carbon quota and demand response [J]. Power Grid & Clean Energy, 2021, 37(07): 79-86." Based on the disorderly charging scenario of electric vehicles, the impact of the synergistic effect of carbon quota and demand response on the energy cost of electric vehicles and system power generation dispatch was explored. There is a lack of research on the power balance management of systems with high operating flexibility such as large-scale virtual power plants. Therefore, an aggregated and efficient power balance management model of a virtual power plant with large-scale charging piles is established, and the carbon emission cost is introduced as a boundary condition. The output balance of wind turbines, photovoltaics, gas turbines, energy storage systems and electric vehicles is studied to control the output of different energy system equipment, thereby achieving the purpose of peak shaving and valley filling and smoothing load fluctuations, and finally realizing the power balance management of virtual power plants that takes into account both economic and environmental benefits. Summary of the invention

[0010] In order to solve the technical problem, that is, the output of different energy system equipment in the virtual power plant, to achieve the purpose of peak shaving and valley filling and smoothing load fluctuations, the present invention provides the following technical solutions:

[0011] A virtual power plant distributed energy aggregation coordinated control method, comprising:

[0012] S1. Construct a virtual power plant system with multiple distributed energy sources including charging piles, and select multiple groups of operating status data and charging loads of various energy subsystem equipment in the virtual power plant system as samples;

[0013] S2. Classifying the control categories of the multiple groups of operating status data based on an improved Kohonen neural network, wherein the improved Kohonen neural network is a Kohonen neural network structure with an output layer added thereto, and the Kohonen neural network structure includes an input layer and a competition layer;

[0014] S3. Control the load output under different control categories based on the simulated annealing immune control algorithm network. The simulated annealing immune control algorithm is a control algorithm that combines the simulated annealing algorithm with the immune control algorithm.

[0015] Furthermore, the virtual power plant system includes multiple distributed energy subsystems such as wind turbines, photovoltaics, gas turbines, energy storage systems, charging piles and electric vehicles.

[0016] Further, the specific method of step S2 is:

[0017] S21, perform normalized clustering processing on multiple groups of running status data input in the sample to obtain x i =(x 1 ,x 2 ,..,x n), the input operating status data features include system time, number of electric vehicles charged, remaining power of energy storage system and battery replacement battery, n is the total number of input data samples, and i is the input layer node number after normalization and clustering of input data;

[0018] S22, connect the input layer and the competition layer weight ω ij Initialization, i is the input layer node number, j is the competition layer node number, ω ij Located in the middle area of ​​the input data, let time count t = 0 and determine the initial learning rate ξ 1 , 2 ;

[0019] S23. Calculate x i After initialization, ω ij The Euclidean distance d j ,

[0020] n=1,2,3...., n is the total number of input data samples;

[0021] S24, select the competition layer neuron with the smallest Euclidean distance as the winning neuron to replace the original smallest neuron;

[0022] S25, adjusting the weight ω between the input data of the input layer and the competition layer ij , adjust the weights ω between the competition layer and the output layers of different control categories jk ,ω jk is the weight connecting the jth node in the competition layer and the kth node in the output layer, where k is the node number in the output layer;

[0023] S26, according to the weight ω ij ,ω jk , data of different categories move closer to corresponding different areas, completing the clustering of the categories to which the input data belongs;

[0024] S27. If there are still input data samples, then t=t+1, return to step S23, and continue to execute until the algorithm ends.

[0025] Further, the specific calculation method of step S25 is:

[0026] ω ij (t+1)=ω ij (t)+ξ 1 (x i -ω ij ) Formula 2

[0027] ω jk (t+1)=ω jk (t)+ξ 2 (y k-ω jk ) Formula 3

[0028] y k The category to which the input data belongs.

[0029] Further, in step S26, the data is clustered into 4 categories.

[0030] Further, step S3 includes:

[0031] S31, taking the system operation status, the number of electric vehicle charging requirements, and the system time as antigens and initializing;

[0032] S32, taking wind turbines, photovoltaics, gas turbines, energy storage, and charging piles as the initial antibody group objects, calculate the actual load value P of each antibody i1 , i.e. wind turbine output, photovoltaic output, gas turbine output, energy storage output, and electric vehicle charging power, i1 represents the serial number of the antibody, which determines the initial annealing temperature;

[0033] S33, calculating the affinity between the system operation requirements and the control output based on the initial annealing temperature;

[0034] S34, immune selection, crossover, and mutation operations to determine the type of control strategy;

[0035] S35, obtaining the initial population of the simulated annealing algorithm, and searching the load output of each subsystem;

[0036] S36, generating new load data according to the simulated annealing algorithm state generation function, and receiving the new load data according to probability;

[0037] S37. Determine whether the new load data meets the convergence criteria. If so, perform the cooling operation, output the control strategy result, and end the algorithm. Otherwise, return to step S34.

[0038] Further, the affinity calculation method in step S33 is formula 4:

[0039]

[0040] Among them, TF(P i1 ) is affinity, the total population is N, P g is the load value of the g-th memory unit, P i1 is the loading value of the i1th antibody.

[0041] Further, the basis for receiving the new load data in step S36 is that the annealing temperature at the moment satisfies Formula 5:

[0042] min[1,exp(-ΔT / T sk )]≥random[0,1] Formula 5

[0043] Among them, T sk is the annealing temperature at time sk, sk is the hourly time, and ΔT is the change in annealing temperature.

[0044] Further, the cooling operation is performed according to Formula 6:

[0045] T sk =CgT sk-1 ,C∈(0,1), Formula 6

[0046] Among them, T sk-1 is the annealing temperature at time sk-1, sk-1 is the hour, C is the temperature cooling coefficient, g is the number of memory units, C g is the temperature cooling coefficient of the g-th memory unit.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] The present invention is based on an improved Kohonen neural network, which divides the control categories of multiple groups of operating status data of subsystem equipment of a virtual power plant, and controls the load output under different control categories based on a simulated annealing immune control algorithm network. The urban virtual power plant system provides the daily power load required by the distribution network and the photovoltaic power generation system in the system, and utilizes the regulating effect of the energy storage system and the charging and swapping load of electric vehicles to achieve the effect of peak shaving and valley filling and smoothing load fluctuations. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is a schematic diagram of the process of transformation from traditional power system to new power system;

[0050] Figure 2 It is a framework diagram of a multi-distributed energy virtual power plant with large-scale charging piles;

[0051] Figure 3 It is the topological structure diagram of Kohonen neural network;

[0052] Figure 4 It is a flowchart of the improved Kohonen neural network training clustering process;

[0053] Figure 5 It is the flow chart of simulated annealing immune control algorithm program;

[0054] Figure 6 It is a multi-agent intelligent collaborative control strategy flow chart based on improved Kohonen and SAIA;

[0055] Figure 7 shows the energy flow values ​​of the subsystem at each time period based on SAIA control solution. Figure 7a It is the fan system. Figure 7bIt is a photovoltaic system. Figure 7c It is a gas turbine system. Figure 7d It is an energy storage system. Figure 7e It's a battery replacement system. Figure 7f It is the charging system;

[0056] Figure 8 This is a comparison chart of system loads before and after intelligent collaborative control on a typical summer day;

[0057] Fig. 9 This is a comparison chart of system loads before and after intelligent collaborative control on a typical winter day. DETAILED DESCRIPTION

[0058] The technical solution of the present invention is clearly described below. Obviously, the described embodiments are not all embodiments of the present invention. All other embodiments obtained by ordinary technicians in this field without making creative work are within the protection scope of the present invention.

[0059] The present invention provides a method for coordinated control of distributed energy aggregation in a virtual power plant, comprising:

[0060] S1. Construct a virtual power plant system with multiple distributed energy sources including large-scale charging piles, and select multiple groups of operating status data and charging loads of each energy subsystem equipment as samples;

[0061] The multi-source virtual power plant model with large-scale charging piles aggregates multiple distributed energy sources such as wind turbines, photovoltaics, gas turbines, energy storage systems, large-scale charging piles and electric vehicles. The user-side load is coordinated with the distributed power source and energy storage system. The overall framework is as follows Figure 2 shown.

[0062] The main sources of energy for virtual power plants are power grids and gas companies. New energy power generation units are wind turbines and photovoltaic power stations. Gas turbines consume gas to supply power to user-side loads. Gas turbines and energy storage systems have the ability to adjust quickly, which complements the non-adjustable output of wind and solar power. They can effectively smooth out the volatility of wind and solar power output and improve the consumption of new energy power generation. Through the technology of electric vehicle battery power feedback to the grid, electric vehicles are dispatched to smooth out the fluctuations in new energy power generation, reduce the impact of the randomness of renewable energy output on the power system, and enable the power system to operate efficiently, stably, and safely.

[0063] The control strategies for various load outputs of multiple distributed energy sources have different characteristics, but the external manifestations are all changes in the performance parameters of the access system equipment. The operating status data and charging load of each energy subsystem equipment under specific constraints such as power balance constraints, power generation equipment output constraints, and electricity price constraints are selected as samples for testing experiments. 4320 groups of data are selected as the total number of samples, of which 3900 groups are used as training samples and 420 groups are used as test samples.

[0064] S2, classifying the control categories of multiple groups of operating status data based on the improved Kohonen neural network;

[0065] Kohonen neural network has the characteristics of unsupervised learning and self-organization. It can automatically cluster input data by training samples. Its network structure is usually composed of input layer and competition layer. The upper competition layer constitutes a two-dimensional node matrix, assuming there are m nodes; the input layer nodes are in the lower layer, assuming there are n nodes in total. Use weights to connect the two types of nodes, and the corresponding variable weight is ω ij (i=1,2,..,n;j=1,2,...,m), the topological structure of Kohonen neural network is as follows Figure 3 shown.

[0066] The improved Kohonen neural network adds an output layer to the original Kohonen neural network structure to give it a supervisory function and realize supervised learning of data clustering. The improved Kohonen neural network training clustering process is as follows: Figure 4 As shown, the specific method is:

[0067] S21, normalize and cluster the input data in the sample into x i =(x 1 , x 2 , ..., x n) , the characteristics of the input data include system time, number of electric vehicles charged, remaining power of energy storage system and battery replacement battery, n is the total number of input data samples, and i is the node number after normalization and clustering of input data;

[0068] S22, connect the input layer and the competition layer weight ω ij Initialization, the connection weight is located in the middle area of ​​the input data as much as possible, i is the input layer node number, j is the competition layer node number; let the time count t = 0, determine the initial learning rate ξ 1 , 2 , the Euclidean distance d between neurons under the initial weight j ;

[0069] S23. Calculate x i After initialization, ω ij The Euclidean distance d j , the specific calculation method is formula 1:

[0070] n=1,2,3...., n is the total number of input data samples;

[0071] S24, select the competition layer neuron with the smallest Euclidean distance as the winning neuron to replace the original smallest neuron;

[0072] S25, adjusting the weight ω between the input data and the competition layer according to Formula 2 ij , according to formula 3, adjust the weights ω between the competition layer and different control categories jk ,ω jk is the weight connecting the jth node in the competition layer and the kth node in the output layer, where k is the node number in the output layer;

[0073] ω ij (t+1)=ω ij (t)+ξ 1 (x i -ω ij ) Formula 2

[0074] ω jk (t+1)=ω jk (t)+ξ 2 (y k -ω jk ) Formula 3

[0075] y k The category to which the input data belongs;

[0076] S26, according to the weight ω ij ,ω jk , data of different categories move closer to corresponding different areas, completing the clustering of the categories to which the input data belongs;

[0077] S27. If there are still input data samples, then t=t+1, and return to step S23 until the algorithm ends.

[0078] The historical operating status data is classified by improving the Kohonen neural network. After processing the system time, the number of electric vehicles charged, the remaining power of the energy storage system and the battery replacement battery, the weights between the neurons in each layer are calculated after iteration, and the samples with different Euclidean distances are classified. According to the different operating characteristics and the similarity of the data itself, the final results are mapped to the neurons in each layer, so that the control categories are divided into 4 categories (i.e., 1, 2, 3, and 4). According to the clustering characteristics of various types of data after classification by the improved Kohonen neural network, the specific control categories and their classification feature criteria can be obtained, as shown in Table 1, where EV represents electric vehicle and t represents the hourly time.

[0079] Table 1 is the control strategy generation characteristic criterion of each distributed energy in the virtual power plant S3. Based on the simulated annealing immune control algorithm network, the output load under different control categories is controlled.

[0080] Simulated Annealing Immune Algorithm (SAIA) is a control algorithm that combines the Simulated Annealing Algorithm (SA) and the Immune Algorithm (IA). The mutual mixing of SA and IA improves the algorithm in terms of control mechanism, control structure, control operation, etc., greatly improving the efficiency of model solving, and controlling the output of different energy system equipment according to the characteristics of different system operating parameters, thereby achieving the purpose of peak shaving and valley filling.

[0081] The program flow chart of step S3 is as follows: Figure 5 As shown, the specific method is:

[0082] Let the total population be N and the number of memory unit antibodies be N m , the crossover probability is P c , the mutation probability is P m , the temperature cooling coefficient is C, the number of memory units is g, and the initial annealing temperature is T.

[0083] S31, taking the system operation status, the number of electric vehicle charging requirements, and the system time as antigens and initializing;

[0084] S32, taking wind turbines, photovoltaics, gas turbines, energy storage, and charging piles as the initial antibody group objects, calculate the actual load value P of each antibody i , i.e. wind turbine output, photovoltaic output, gas turbine output, energy storage output, and electric vehicle charging power, determine the initial annealing temperature;

[0085] Randomly generate NN m antibodies, mainly wind turbines, photovoltaics, gas turbines, energy storage, charging piles, etc. as the initial antibody group objects, and N in the memory unit m antibodies together form the initial population; if the memory unit is empty, N antibodies are randomly generated to form the initial population. At k = 0, the actual load value P of each antibody is calculated. i , namely wind turbine output, photovoltaic system output, gas turbine output, energy storage system output, electric vehicle charging power, battery replacement battery charging power, etc.

[0086] S33, based on the initial annealing temperature, calculate the affinity TF (P between the system operation requirements and the control output) i1 ), put N m The antibodies are stored in the memory unit P g ; The affinity is calculated by formula 4:

[0087]

[0088] Among them, TF(P i1 ) is affinity, the total population is N, P g is the load value of the g-th memory unit, P i1 is the loading value of the i1th antibody.

[0089] S34, immune selection, crossover, and mutation operations to determine the type of control strategy;

[0090] S35, obtaining the initial population of the simulated annealing algorithm, and searching the load output of each subsystem;

[0091] S36, generating new load data according to the simulated annealing algorithm state generation function, and receiving the new load data according to probability;

[0092] The simulated annealing operation of fixed-step sampling is performed on the parameters of photovoltaic, energy storage, charging and swapping energy equipment in the population, and new antibodies are generated for the population. If the annealing temperature T at time sk sk The result satisfies Formula 5,

[0093] min[1,exp(-ΔT / T sk )]≥random[0,1] Formula 5

[0094] Then update and calculate the new load value of each subsystem, where T sk is the annealing temperature at time sk, sk is the hourly time, and ΔT is the change in annealing temperature; otherwise, the load value remains unchanged.

[0095] S37. Whether the new load meets the convergence criteria, if yes, execute the cooling operation, output the control strategy result, and end the algorithm; otherwise, return to step S34.

[0096] According to formula 6, the cooling operation is performed.

[0097] T sk =CgT sk-1 ,C∈(0,1) Formula 6

[0098] Among them, T sk-1 is the annealing temperature at time sk-1, sk-1 is the hour, C is the temperature cooling coefficient, g is the number of memory units, C g is the temperature cooling coefficient of the g-th memory unit.

[0099] Based on a virtual power plant with large-scale charging piles, the above-mentioned virtual power plant distributed energy aggregation and coordinated control method is used to verify and analyze the intelligent collaborative control strategy. Taking into account uncontrollable factors such as weather conditions and sudden changes in charging and swapping demand, the energy storage system can be used as backup capacity and charged and discharged in a constant power form; the source of electricity required for charging is limited to wind turbine power generation, photovoltaic power generation, gas turbine output, energy storage discharge and grid power supply, and the charging output load is adjusted according to the control strategy; the replaced batteries are sent to the charging warehouse for centralized charging. If there are fully charged batteries that have not been replaced in the swapping area, they can be used as system backup capacity to play the role of energy storage.

[0100] The virtual power plant with large-scale charging piles provides the daily power load through the distribution network and the wind turbine power generation, photovoltaic power generation, and gas turbine power generation systems in the system, and uses the regulation of the energy storage system and the charging and swapping load of electric vehicles to control the system load output, so as to achieve the effect of peak shaving and valley filling and load fluctuation smoothing. At the same time, in the regulation process, the simulated annealing immune control algorithm is combined to transform the non-convex and nonlinear problems involved in the system into linear problems, so as to realize the rapid search for the global energy coordinated regulation solution, and finally obtain the specific energy flow direction and numerical size of the six subsystems of wind, light, fuel, storage, charging, and swapping in different time periods. Table 2 shows the specific energy flow direction of the energy storage system, charging system, and battery swapping system in different time periods. Figure 7 shows the respective energy flow values. Figure 7 is the energy flow value of the subsystem in each time period based on SAIA regulation. Figure 7a is the energy flow value of the fan system at each time period, Figure 7b is the energy flow value of the photovoltaic system at each time period, Figure 7c is the energy flow value of the gas turbine system at each time period, Figure 7d is the energy flow value of the energy storage system at each time period, Figure 7e is the energy flow value of the battery swap system at each time period, Figure 7f It is the energy flow value of the charging system at each time period.

[0101] Table 2 shows the energy flow of each subsystem in each period based on the simulated annealing immune control algorithm.

[0102]

[0103]

[0104] Based on the differences in solar irradiance, electric vehicle charging load characteristics, etc., the virtual power plant system load is defined to typical days in summer and winter for modeling and analysis. The urban virtual power plant system provides the daily power load required through the distribution network and the photovoltaic power generation system within the system, and uses the regulating effect of the energy storage system and electric vehicle charging and swapping load to control the system load output, thereby achieving the effect of peak shaving and valley filling and smoothing load fluctuations.

[0105] On a typical day, the urban virtual power plant system is managed through coordinated intelligent collaborative control based on the capacity configuration and load conditions of the energy equipment in each subsystem within the system. Figure 8 and Fig. 9 The results of coordinated control of energy subsystem output and electric vehicle charging and swapping load on different typical days in summer and winter are shown respectively. Before the intelligent coordinated control of each energy subsystem, the system load was high during the two time periods of 10:00-14:00 and 17:00-21:00, while the system load dropped rapidly during the period of 22:00-24:00, which will lead to an increase in the volatility of the system equivalent load. Figure 8 and Fig. 9 It can be seen that during the peak electricity price period from 10:00 to 14:00, the maximum output of photovoltaic power is guaranteed, the purchase of electricity from the distribution network is reduced, charging is mainly carried out at night or during the day when the system load is less, and the power generated by photovoltaic power is used first.

[0106] In addition, some vehicles that are charged during peak hours (10:00-14:00 or 17:00-21:00) are transferred to valley hours (22:00-1:00 the next day) for charging, and the battery swapping batteries are charged centrally during the valley hours through chargers. The participation of the energy storage system and the battery swapping system will increase the load peak, but the two also play a key role in collaborative control. For a typical summer day, the system load peak before and after control decreased by 202.6kW, and the load peak-to-valley difference rate decreased from 50.8% to 40.6%; for a typical summer day, the system load peak before and after control decreased by 222.3kW, and the load peak-to-valley difference rate decreased from 51.4% to 41.6%. Overall, the above-mentioned intelligent collaborative control strategy has reduced the load peak-to-valley difference of the virtual power plant system.

[0107] Finally, it should be noted that the above content is only used to illustrate the technical solution of the present invention, rather than to limit the scope of protection of the present invention. Simple modifications or equivalent substitutions of the technical solution of the present invention by ordinary technicians in this field do not deviate from the essence and scope of the technical solution of the present invention.

Claims

1. A method for coordinated control of distributed energy aggregation in a virtual power plant, characterized in that: include: S1. Construct a virtual power plant system with multiple distributed energy sources including charging piles, and select multiple groups of operating status data and charging loads of various energy subsystem equipment in the virtual power plant system as samples; S2. Classifying the control categories of the multiple groups of operating status data based on an improved Kohonen neural network, wherein the improved Kohonen neural network is a Kohonen neural network structure with an output layer added thereto, and the Kohonen neural network structure includes an input layer and a competition layer; S3, controlling the load output under different control categories based on the simulated annealing immune control algorithm network, wherein the simulated annealing immune control algorithm is a control algorithm formed by combining the simulated annealing algorithm with the immune control algorithm; The specific method of step S2 is: S21, perform normalized clustering processing on multiple groups of running status data input in the sample to obtain x i =(x1,x2,..,x n) , the input operation status data features include system time, number of electric vehicles charged, remaining power of energy storage system and battery replacement battery, n is the total number of input data samples, i is the input layer node number after normalization and clustering of input data; S22, connect the input layer and the competition layer weight ω ij Initialization, i is the input layer node number, j is the competition layer node number, ω ij Located in the middle area of ​​the input data, let the time count t = 0 and determine the initial learning rates ξ1 and ξ2; S23. Calculate x i After initialization, ω ij The Euclidean distance d j , n=1,2,3...., n is the total number of input data samples; S24, select the competition layer neuron with the smallest Euclidean distance as the winning neuron to replace the original smallest neuron; S25, adjusting the weight ω between the input data of the input layer and the competition layer, adjusting the competition layer and different ij Control the weights ω between the category output layers jk ,ω jk is the weight connecting the jth node in the competition layer and the kth node in the output layer, where k is the node number in the output layer; S26, according to the weight ω ij ,ω jk , data of different categories move closer to corresponding different areas, completing the clustering of the categories to which the input data belongs; S27. If there are still input data samples, then t=t+1, return to step S23, and continue to execute until the algorithm ends.

2. The method according to claim 1, characterized in that The virtual power plant system includes multiple distributed energy subsystems such as wind turbines, photovoltaics, gas turbines, energy storage systems, charging piles and electric vehicles.

3. The method according to claim 1, characterized in that: The specific calculation method of step S25 is: ω ij (t + 1) = ω ij (t) + ξ1(x i -ω ij Formula Two ω jk (t + 1) = ω jk (t) + ξ2(y k -ω jk Formula Three y k The category to which the input data belongs.

4. The method according to claim 1, characterized in that: Step S26: cluster the data into 4 categories.

5. The method according to claim 1, characterized in that Step S3 includes: S31, taking the system operation status, the number of electric vehicle charging requirements, and the system time as antigens and initializing; S32, taking wind turbines, photovoltaics, gas turbines, energy storage, and charging piles as the initial antibody group objects, calculate the actual load value P of each antibody i1 , i.e. wind turbine output, photovoltaic output, gas turbine output, energy storage output, and electric vehicle charging power, i1 represents the serial number of the antibody, which determines the initial annealing temperature; S33, calculating the affinity between the system operation requirements and the control output based on the initial annealing temperature; S34, immune selection, crossover, and mutation operations to determine the type of control strategy; S35, obtaining the initial population of the simulated annealing algorithm, and searching the load output of each subsystem; S36, generating new load data according to the simulated annealing algorithm state generation function, and receiving the new load data according to probability; S37. Determine whether the new load data meets the convergence criteria. If so, perform the cooling operation, output the control strategy result, and end the algorithm. Otherwise, return to step S34.

6. The method according to claim 5, characterized in that The affinity calculation method in step S33 is formula 4: Among them, TF(P i1 ) is affinity, the total population is N, P g is the load value of the g-th memory unit, P i1 is the loading value of the i1th antibody.

7. The method according to claim 6, characterized in that The basis for receiving the new load data in step S36 is that the annealing temperature at the moment satisfies Formula 5: min[1,exp(-ΔT / T sk )]≥random[0,1]Formula 5 Among them, T sk is the annealing temperature at time sk, sk is the hourly time, and ΔT is the change in annealing temperature.

8. The method according to claim 7, characterized in that Perform cooling operation according to formula 6. T sk =CgT sk-1 ,C∈(0,1), Formula 6 Among them, T sk-1 is the annealing temperature at time sk-1, sk-1 is the hour, C is the temperature cooling coefficient, g is the number of memory units, C g is the temperature cooling coefficient of the g-th memory unit.

Images