Capacity optimization allocation method of virtual power plant based on multi-agent game

Abstract

The present invention relates to a virtual power plant capacity optimization configuration method of multi-agent game, which is characterized by comprising the following steps: step one, multi-investor virtual power plant architecture and its adjustment capability analysis; step two, virtual power plant aggregation model and cost-benefit analysis index composition; step three, multi-investor virtual power plant capacity configuration model; step four, cooperative profit distribution correction model based on improved Shapley value method; the cooperative profit distribution correction model based on improved Shapley value method improves the robustness of virtual power plant capacity configuration model. Beneficial effects: The present invention can successfully cope with various game modes of distributed power investors, effectively improve individual investors' benefits, maximize investor group benefits, stabilize the optimal cooperation model of multiple investors, reduce the risks caused by renewable energy generation and electricity price uncertainty, and improve investment risk resistance.

Description

Technical Field

[0001] The present invention belongs to the field of electric power technology, and in particular relates to a virtual power plant capacity optimization configuration method based on multi-agent game. Background Art

[0002] The main feature of the energy internet is the deep integration of a large number of distributed generation grids and Internet technologies. The large number of distributed generation grids makes the smart grid have the characteristics of small distributed capacity, multiple types of distributed generation, large number, and geographical dispersion. It is difficult for the power dispatching center to centrally control distributed generation, which has a significant impact on the reliable and safe operation of the power grid. At the same time, various distributed generation have certain adjustment capabilities and can provide flexible resources for power grid dispatching.

[0003] Among the many feasible technical solutions, virtual power plants (VPPs) can integrate various distributed generation. Through the flexible management and control of a large number of distributed generation, the stability and reliability of its overall output can be improved, and the safety and efficiency of distributed generation grid connection can be achieved. Compared with microgrids, VPPs have smaller geographical limitations and larger scales of distributed generation grid connection. Therefore, VPPs can more effectively connect distributed generation with the power system, realize resource integration and allocation, and are an important way for smart grids to achieve interaction and intelligence on the reference energy supply side, and are also the ultimate configuration of the energy Internet. However, in order to make full use of VPPs, it is very important to reasonably configure the capacity of various distributed generation in the planning stage.

[0004] In recent years, many scholars have studied the capacity configuration of distributed generation, but they mainly focus on microgrids. Compared with microgrids, VPPs participate in the electricity market as a whole and can more flexibly stabilize the fluctuations of renewable energy generation. The optimal configuration of various distributed power sources in VPPs directly affects energy utilization efficiency, power supply reliability and power quality. It is the key to ensure the safe, reliable and economical operation of the distribution network and an important prerequisite for making full use of VPPs. The patent document with patent publication number CN107464010A discloses a virtual power plant capacity optimization configuration method considering risks, using CVaR to measure the risks brought to the planning of VPP by the uncertainty of renewable energy output and market electricity prices; in the objective function of planning and constructing VPPs, CVaR is multiplied by a weight coefficient, and the risk preference level of investors is measured by the size of the weight coefficient. At the same time, various constraints in the real-time operation of VPP are taken into account to establish a virtual power plant multi-power capacity optimization configuration model considering conditional risk value. A patent document with patent publication number CN 110516843 A discloses a virtual power plant capacity optimization method, characterized in that the method includes: generating an output scenario of a distributed power source of a virtual power plant, and constructing a collaborative optimization model for accessing the virtual power plant to a distribution network; constructing a profit distribution model of the virtual power plant based on the output scenario and the collaborative optimization model; and determining a target capacity configuration of the virtual power plant based on the profit distribution model based on a genetic algorithm.

[0005] However, most of the current research on VPP focuses on scheduling and operation, while there are relatively few studies on VPP planning, and there are still some problems. Since there are multiple cooperation modes for the capacity configuration of VPPs containing multiple distributed generation, existing studies lack the necessary modeling and demonstration of the game mode of multiple distributed generation investors, and the obtained VPP capacity configuration is quite different from the actual situation. In an open electricity market, there is direct competition and potential cooperation among investors. In order to maximize their own benefits, investors choose whether to participate in cooperation, with whom to cooperate, and whether to maintain long-term and stable cooperation, which will have an impact on the capacity configuration of multiple distributed generation in the VPP. Analyzing the possibility of cooperation between different types of distributed generation investors can provide theoretical support for the modeling of VPP. Game theory, as a theoretical method for studying the interactive decision-making process and results of multi-agents, is expected to become a powerful tool for the optimal configuration of VPPs. In addition, the research on the optimal configuration of VPPs lacks a quantitative description of the regulation capacity and randomness of VPPs, ignores the economic risks brought about by the uncertainty of wind speed, light intensity and electricity prices, and fails to form a complete VPP optimal configuration system that considers multiple distributed generation, multiple investors and risk value. In addition, there is a lack of dynamic sensitivity analysis of key factors in the capacity optimization configuration of multiple distributed generators. Summary of the invention

[0006] The purpose of the present invention is to overcome the shortcomings of the above-mentioned technology and to provide a virtual power plant capacity optimization configuration method based on multi-agent game, which can successfully cope with various game modes of distributed power investors, effectively improve the individual benefits of investors, maximize the benefits of the investor group, consolidate the optimal cooperation model of multiple investors, reduce the risks caused by the uncertainty of renewable energy generation and electricity prices, and improve the ability to resist investment risks.

[0007] In order to achieve the above object, the present invention adopts the following technical scheme: comprising the following steps:

[0008] Step 1: Analysis of multi-investor virtual power plant architecture and its regulation capabilities;

[0009] Step 2: Virtual power plant aggregation model and cost-benefit analysis indicator composition;

[0010] Step 3: Capacity configuration model of multi-investor virtual power plants;

[0011] Step 4: The cooperative benefit allocation correction model based on the improved Shapley value method has four steps; the cooperative benefit allocation correction model based on the improved Shapley value method improves the robustness of the virtual power plant capacity configuration model.

[0012] The analysis of the multi-investor virtual power plant architecture and its regulatory capabilities in step 1, the analysis of the multi-investor virtual power plant structure organization and its regulatory capabilities based on game theory:

[0013] (1) Multi-investor virtual power plant architecture

[0014] Taking the uncontrollable distributed power sources of wind turbines (WT) and photovoltaics (PV) and the controllable distributed power sources of diesel generators (DG) and energy storage devices (ES) as examples, the structure of the virtual power plant is analyzed when various distributed power sources belong to different investors.

[0015] Referring to the concept of network game, the game model of n investors is shown in equation (1):

[0016] NEG={N,E,G={G ij} ij∈E} (1)

[0017] Where: NEG is the set of all game modes; N = {1, 2, ···, n} is the set of investors; E is the pairwise combination of n investors; Gij represents the cooperation state between investors i and j, the cooperation state Gij = 1, the non-cooperation state Gij = 0, G is a binary code, and the decimal value of G is used to represent the game mode number between investors;

[0018] (2) Virtual power plant strategy: including

[0019] 1) The investment strategy of virtual power plant investors refers to the action strategy taken by each investor in the game. The investment strategy of each investor is based on the installed capacity of distributed power generation. The investment strategy set of each investor is the boundary of the installed capacity of distributed power generation, which is a continuous strategy interval, represented by SWT, SPV, SDG, and SES, as shown in equation (2):

[0020]

[0021] Where: sWT,min, sPV,min, sDG,min, sES,min are the minimum values ​​of wind turbine, photovoltaic, diesel generator and energy storage installed capacity respectively; sWT,max, sPV,max, sDG,max, sES,max are the maximum values ​​of wind turbine, photovoltaic, diesel generator and energy storage installed capacity respectively;

[0022] 2) The operation control strategy of the virtual power plant is to adopt a centralized-decentralized control structure. The virtual power plant control center coordinates the information reported by local control centers and the investment preferences of investors to make capacity allocation decisions;

[0023] (3) Analysis of the regulation capacity of virtual power plants. The differences in the characteristics of distributed power sources under different regulation capacities are used to map the different strategic scenarios in which virtual power plants participate. In virtual power plants, the regulation capacity of wind turbines, photovoltaics, diesel generators, and energy storage units is analyzed.

[0024] The virtual power plant aggregation model and cost-benefit analysis index composition of step 2

[0025] (1) Virtual power plant aggregation model, including NWT wind turbines, NPV photovoltaics, NDG diesel generators, and NES energy storage. The power characteristics of each distributed power source are:

[0026] 1) The power output of the wind turbine is shown in equation (7):

[0027]

[0028] Where: PWT,r is the rated power of the wind turbine; vci, vr, vco are the cut-in wind speed, rated wind speed and cut-out wind speed of the wind turbine, respectively.

[0029] 2) Photovoltaic power generation is closely related to light intensity, so the photovoltaic power generation is as shown in equation (8):

[0030] P PV =r PV A PV η PV (8)

[0031] Where: rPV is the actual light intensity during the period, APV is the area of ​​the photovoltaic array, ηPV is the photoelectric conversion efficiency,

[0032] 3) By configuring diesel generators to smooth out the volatility of renewable energy generation, virtual power plants participate in the electricity market with stable power output;

[0033] 4) Taking advantage of the charging and discharging characteristics of energy storage, participating in demand response to shift peak power consumption to obtain compensation income and participating in the power auxiliary service market to obtain income, comprehensively considering wind turbines, photovoltaics, diesel generators and energy storage, the power generation of the virtual power plant in time period τ and the power output at time t are as shown in equation (9):

[0034]

[0035] Where: PVPP(t), PWT,i(t), PPV,i(t) and PDG,i(t) are the power outputs of VPP, the i-th wind turbine, the i-th photovoltaic, and the i-th diesel generator at time t, respectively; PESD,i(t) and PESC,i(t) are the discharge and charging powers of the i-th energy storage at time t, respectively; μESD,i indicates whether the i-th energy storage is discharged, otherwise it is set to 0, otherwise it is set to 1; μESC,i indicates whether the i-th energy storage is charged, otherwise it is set to 0, otherwise it is set to 1;

[0036] (2) Risk measurement based on VaR and CVaR

[0037] For a given confidence level σ, VaR,σ and CVaR,σ are used to represent the index values ​​of VaR and CVaR, respectively, as shown in equation (12)

[0038]

[0039] Where: f(x,y) is the loss function of the investment plan, x is the decision variable, y is the random variable that causes the loss; ρ(y) is the probability density function of y; κ is the boundary value of the risk value;

[0040] (3) Cost-benefit analysis indicators

[0041] 1) The net present value (NPV) method is used to evaluate the benefits of the virtual power plant over its entire life cycle. The net present value index is expressed as NNPV, as shown in equation (16):

[0042]

[0043] Where: M is the useful life of the system; r is the discount rate; CI(m) is the cash inflow value in the mth year; CO(m) is the cash outflow value in the mth year;

[0044] 2) The internal rate of return (IRR) method is used to measure the risk resistance of virtual power plant investment. The internal rate of return index NIRR is shown in equation (17):

[0045]

[0046] Where: rpos and rneg are the corresponding discount rates when NNPV is a positive indicator and a negative indicator respectively; pos and neg are the values ​​of the positive and negative indicators of NNPV respectively;

[0047] Capacity configuration model of multi-investor virtual power plant in step 3

[0048] (1) Investor cost function: The annual cost of the investor, i.e., the cash outflow value CO(m) in the mth year, consists of the following six parts:

[0049] ① Initial construction cost of equal annual value

[0050]

[0051] Where: CWT,inv, CPV,inv, CDG,inv and CES,inv are the installed costs of a single wind turbine, a single photovoltaic unit, a single diesel generator set and a single lead-acid battery set respectively;

[0052] ②Annual operation and maintenance cost

[0053] C om =N WT C WT,om +N PV C PV,om +N DG C DG,om +N ES C ES,om (19)

[0054] Where: CWT,om, CPV,om, CDG,om and CES,om are the annual operation and maintenance costs of a single wind turbine, a single photovoltaic unit, a single diesel generator set and a single lead-acid battery set respectively;

[0055] ③ Annual fuel cost of diesel generator

[0056]

[0057] Where: prDG is the diesel price coefficient, yuan / kg; QDG is the fuel consumption coefficient, g / kWh; PDG,wpe(t) is the diesel generator power at time t;

[0058] ④ Annual environmental cost of diesel generators

[0059]

[0060] ⑤ Annual power outage penalty costs,

[0061] The penalty Cpen,wpe that should be paid for load outage caused by insufficient VPP power supply is allocated according to the proportion of installed capacity of each distributed power source.

[0062]

[0063]

[0064]

[0065] Where: represents the total installed capacity of distributed power sources, i∈{WT,PV,DG,ES}; coefpen is the penalty coefficient, RMB / kWh; Ppen,wpe(t) represents the power shortage at time t; PL(t) represents the load at time t; represents the power output of each distributed power source at time t; Ptie,max represents the tie line limit between the region and the upper grid;

[0066] ⑥ The risk cost caused by the randomness of the virtual power plant is introduced by introducing a virtual variable Mwpe. Mwpe relaxes the inequality shown in equations (25) and (26):

[0067] M wpe ≥0 (25)

[0068] M wpe ≥-(R em,wpe +R e,wpe +R as,wpe )+κ (26)

[0069] (2) Investor income function: The investor’s annual income, i.e., the cash inflow value in the mth year, CI(m), consists of the following three parts:

[0070] ① Annual electricity sales revenue in the market today

[0071] The electricity sales revenue of each distributed power source in the market is shown in equation (27):

[0072]

[0073] Where: Pi,s,wpe(t) is the power sold by each distributed generation to the grid at time t, i∈{WT,PV,DG,ES}; is the electricity price under the day-ahead market price scenario EPk at time t;

[0074] ② The annual environmental income of renewable energy power generation is as shown in equation (28):

[0075]

[0076] Where: PWT,wpe(t) and PPV,wpe(t) are the output power of wind turbines and photovoltaics at time t respectively; coefr is the incentive coefficient, yuan / MWh;

[0077] ③ Annual income from the power auxiliary service market: diesel generators and energy storage can participate in peak load shaving and valley filling auxiliary services to generate revenue.

[0078]

[0079] Where: [τ1,τ2] is the peak electricity price period; [τ3,τ4] is the non-peak electricity price period; θ∈[0,1] is the VPP regulation capacity ratio coefficient, and are the upward and downward adjustable powers of VPP at time t respectively; and are the upward and downward adjustable powers of the i-th diesel generator set at time t respectively; and are the upward and downward adjustable discharge powers of the i-th group of energy storage at time t respectively; and are the upward and downward adjustable charging powers of the i-th group of energy storage at time t, respectively;

[0080] (3) Objective function

[0081] The objective function of the multi-investor virtual power plant capacity allocation model based on maximizing the benefits of each investor is shown in equation (30):

[0082]

[0083] Where: L represents the risk preference of investors, L∈[0,1]. When L<0.1, it means that investors tend to choose high-risk and high-return investment strategies, which belongs to the active type in risk preference. When L>0.5, it means that investors belong to the conservative type in risk preference.

[0084] The modified model of cooperative benefit distribution based on the improved Shapley value method in step 4

[0085] (1) Shapley value method

[0086] The income distributed to investor i from the virtual power plant alliance Ψ is xi(ψ), as shown in equation (37):

[0087]

[0088]

[0089] Where: U is the subset of all virtual power plant alliances Ψ that include investor i; W(U) is the weighting factor; I(U) is the revenue of the virtual power plant alliance; I(U\I) is the revenue of the virtual power plant alliance after investor i is removed; n is the total number of investors;

[0090] (2) Improved Shapley value method: an improved Shapley value method is proposed from three aspects: participation, risk preference and marginal contribution.

[0091] Participation: The participation of each DG investor is measured by installed capacity, that is, the proportion of each DG installed capacity to the total installed capacity, as shown in equation (39):

[0092]

[0093] Where: is the participation coefficient of power source i in the virtual power plant;

[0094] Risk preference: The difference in risk preference coefficient βi is introduced to indicate the attractiveness of the virtual power plant alliance to investors, as shown in equation (40):

[0095]

[0096] Where: Li∈[0,1] is the risk preference coefficient of investor i when he does not participate in the virtual power plant alliance; kj is a state variable, which takes 1 when investor j participates in the virtual power plant alliance, otherwise it takes 0; is the profit of investor j when he does not participate in the virtual power plant alliance;

[0097] The risk preference coefficient difference is normalized as shown in equation (41):

[0098]

[0099] Marginal contribution: The marginal contribution γi(U) of investor i in the virtual power plant alliance U, as shown in equation (42)

[0100] γ i (U)=I(U)-I(U\i) (42)

[0101] The marginal contribution is normalized as shown in equation (43)

[0102]

[0103] (3) Profit distribution correction model

[0104] J = {α, β, γ} is defined as a set of three influencing factors including participation, risk preference and marginal contribution. According to equations (39), (41) and (43), the correction matrix A is established as shown in equation (44):

[0105]

[0106] Further, the step 1 (3) virtual power plant regulation capacity analysis includes regulation capacity analysis of wind turbines, photovoltaics, diesel generators and energy storage units.

[0107] ① The regulation capacity of the diesel generator set: The diesel generator set initially runs at point (0, PDG, n). When t = τ1, the cumulative increase in the diesel generator power is When t = τ2, the cumulative reduction in power of the diesel generator is As shown in equation (3)

[0108]

[0109] Where: PDG,a and PDG,b are the power of the diesel generator set at τ1 and τ2 respectively; and are the upward climbing rate and downward climbing rate of the diesel generator set respectively; PDG,max and PDG,min are the upper and lower limits of the diesel generator set output power respectively. When t = τ3, the diesel generator set power has reached the lower limit PDG,c = PDG,min, and shutdown can further reduce the diesel generator set power output. When t = τ4, the cumulative reduction in power is shown in equation (4)

[0110]

[0111] ② Adjustment capability of energy storage: In the time period 0-τ1, the energy storage increases and decreases the discharge amount, as shown in equation (5)

[0112]

[0113] Where: PESD,max and PESD,min are the maximum and minimum discharge powers of the energy storage, respectively; EES is the remaining capacity of the energy storage; PESD is the discharge power of the energy storage; EES,max is the rated capacity of the energy storage; CDOD is the maximum discharge depth of the energy storage;

[0114] The energy storage is in a charging state during the time period of τ1 to τ2. The energy storage charging amount is adjusted by increasing or decreasing the charging power. During the time period of τ1 to τ2, the energy storage increases or decreases the charging amount, as shown in equation (6).

[0115]

[0116] Where: PESC,max and PESC,min are the maximum and minimum charging powers of the energy storage, respectively, and PESC is the charging power of the energy storage;

[0117] ③Regulation capability of wind turbines and photovoltaics:

[0118] By introducing the random variable x, within a certain period of time, the wind turbine and photovoltaic power generation units output power PWT(x) and PPV(x) with the probability of πWT and πPV respectively.

[0119] Furthermore, PWT,i(t) and PPV,i(t) in equation (9) of step 2 are random. The scenario set method is used to transform the random optimization problem into a deterministic optimization problem for processing. The scenario set of wind turbine output power, photovoltaic output power and electricity market price is shown in equation (10):

[0120]

[0121] Where nWT, nPV, and nEP are the total number of scenario sets for wind turbines, photovoltaics, and electricity prices, respectively. The total number of scenarios is nWT×nPV×nEP. The probability of each scenario for wind turbines, photovoltaics, and electricity prices is π(WTi), π(PVj), and π(EPk), respectively. The regulation capacity of the virtual power plant can compensate for its own randomness. The up- and down-regulation capabilities of the virtual power plant are shown in equation (11):

[0122]

[0123] Where: and They are the adjustable up and down power of VPP respectively; and They are the adjustable power and the adjustable power of the i-th diesel generator set respectively; and They are the adjustable upward and downward capacities when the i-th group of energy storage is discharged; and They are the adjustable amount of power that can be increased and decreased when charging the i-th group of energy storage.

[0124] Further, in the formula (12) in step 2, The expected loss distribution function of the decision plan is φ(x,α), as shown in equation (13):

[0125]

[0126] Beneficial effects: The method for optimizing the capacity of a virtual power plant by considering multi-agent game under the guidance of effectiveness of the present invention is based on game theory. By constructing a game model of multiple investment entities such as wind turbines, photovoltaics, diesel generators and energy storage, a virtual power plant architecture and investment and operation strategy considering multiple investors and multiple distributed power sources are proposed; by establishing a quantitative model of the regulation capacity and randomness of virtual power plants, the risk of uncertainty in renewable energy generation and electricity prices on the capacity configuration of virtual power plants is reduced; a cost-benefit analysis index model for virtual power plant investment is constructed to improve the feasibility and risk resistance of virtual power plant investment; with the maximum benefit of virtual power plants as the goal, comprehensive consideration is given to Taking into account factors such as initial construction cost, operation and maintenance cost, fuel cost, risk cost, environmental cost and income, electricity sales income, and ancillary service income, the installed capacity of each distributed power source in the virtual power plant is optimized; by establishing a cooperative benefit correction model of the improved Shapley value method, the robustness of the virtual power plant capacity configuration model is further improved; the proposed capacity optimization configuration scheme has high universality, good economic benefits, and strong robustness. It can better balance the interests of individual and group investors, reduce the investment risks brought about by renewable energy and electricity price uncertainty, improve the absorption capacity of renewable energy, and promote the low-carbon transformation of electric energy, which is of greater academic significance and engineering value. The present invention can provide technical support for the capacity configuration of virtual power plants, solve the problem of unclear investor game mode in the capacity configuration of virtual power plants, meet the risk-return trade-off principle, and is suitable for scenarios with multiple investment entities, multiple distributed power sources and multiple risk preferences, and has wide practicality; it can fully consider the game relationship of multiple investment entities, optimize the architecture of the virtual power plant, and propose reasonable investment and operation strategies to point out the direction for the capacity configuration of virtual power plants; alleviate the risks brought by the uncertainty of renewable energy power generation and electricity prices, and enhance the feasibility and risk resistance of virtual power plant investment; it can fully consider multiple factors such as construction and operation costs, risk costs, environmental costs and income, electricity market transactions, and ancillary service market income, and propose a method for optimizing the capacity configuration of virtual power plants with multiple investment entities. The obtained capacity optimization configuration scheme has high universality, good economic benefits, and strong robustness. It can better balance the interests of individual and group investors, reduce the investment risks brought by the uncertainty of renewable energy and electricity prices, enhance the absorption capacity of renewable energy, and promote the low-carbon transformation of electric energy, which is of greater academic significance and engineering value. BRIEF DESCRIPTION OF THE DRAWINGS

[0127] Figure 1 It is a flow chart of the virtual power plant capacity optimization configuration method;

[0128] Figure 2a This is a graph of electricity price data from 2017 to 2020;

[0129] Figure 2b This is a graph of wind speed data from 2017 to 2020;

[0130] Figure 2c This is a data chart of light intensity from 2017 to 2020;

[0131] Figure 2d This is the average load data chart from 2017 to 2020;

[0132] Figure 3 It is the installed capacity diagram of distributed power generation under various game modes;

[0133] Figure 4a is the relationship between the randomness of the virtual power plant and θ;

[0134] Figure 4b is a plot of conditional value at risk versus θ;

[0135] Figure 5 It is a graph of NNPV accumulated over time;

[0136] Figure 6 This is a diagram of the day-ahead output plan of a virtual power plant based on a balancing solution. DETAILED DESCRIPTION

[0137] The specific implementation of the present invention is described in detail below in conjunction with preferred embodiments.

[0138] See attached for details Figure 1 , this embodiment provides a virtual power plant capacity optimization configuration method considering multi-agent game under the guidance of results.

[0139] At present, there are few studies on the capacity configuration of virtual power plants (VPPs) involving multiple distributed generation investors, and the research on the optimal configuration of VPPs lacks a quantitative description of the VPP's adjustment ability and randomness, ignores the economic risks brought by the uncertainty of wind speed, light intensity and electricity price, and fails to form a complete VPP optimization configuration system that considers multiple distributed generation, multiple investors and risk value. In addition, in the capacity optimization configuration of multiple distributed generators, there is a lack of dynamic sensitivity analysis of key factors. To this end, the present invention first constructs the organizational structure of the virtual power plant, analyzes the strategy of the virtual power plant and the adjustability of the virtual power plant under different strategies. Secondly, four distributed generation modes of wind turbines, photovoltaic power generation, diesel power generation and energy storage are modeled, and the power generation capacity of the VPP comprehensive model of the day-ahead market, the adjustment ability of the auxiliary service market and the risk resistance benefit index are quantitatively evaluated. Then, based on the portfolio theory, an optimal configuration model with the goal of maximizing VPP benefits is established, and the game theory is used to analyze the installed capacity of distributed generation of virtual power plants under different game modes. On this basis, the shortcomings of the traditional Shapley value method are analyzed. From the three aspects of participation, risk preference and marginal contribution, a cooperative profit distribution correction model based on the improved Shapley value method is proposed, which makes the distribution method more practical, ensures stable cooperation among investors, and ensures the long-term stability of the virtual power plant structure under different game modes, and realizes the virtual power plant capacity optimization configuration method considering multi-agent game under the results-oriented approach.

[0140] The effectiveness-oriented virtual power plant capacity optimization configuration method considering multi-agent game in the present invention is divided into four steps, including 1) multi-investor virtual power plant architecture and its regulation capability analysis, 2) virtual power plant aggregation model and cost-benefit analysis indicator composition, 3) capacity configuration model of multi-investor virtual power plant, and 4) cooperative benefit distribution correction model based on improved Shapley value method, which are described in detail as follows.

[0141] 1) Analysis of multi-investor virtual power plant architecture and its regulation capabilities

[0142] Game theory is a theory as follows: when the behaviors of multiple decision makers interact with each other, decision makers make decisions that are beneficial to themselves based on the information they have and their perception of their own abilities. This paper assumes that the game between investors belongs to a complete information static game, that is, all investors make decisions at the same time and fully understand the strategies and investments in various situations in the game. The structure organization of a multi-investor virtual power plant based on game theory and its regulatory capabilities are analyzed as follows.

[0143] (1) Multi-investor virtual power plant architecture

[0144] A typical virtual power plant (VPP) consists of uncontrollable distributed power sources such as wind turbines (WT) and photovoltaics (PV), and controllable distributed power sources such as diesel generators (DG) and energy storage devices (ES). Diesel generators are used to smooth power fluctuations caused by the uncertainty of renewable energy generation. Energy storage devices can play a peak-shaving role in the daily operation of virtual power plants. They jointly stabilize the output of virtual power plants to participate in the electricity market. Taking the above four types of distributed power sources as an example, the present invention analyzes the structure of virtual power plants when various distributed power sources belong to different investors. Wind turbines, photovoltaics, diesel generators and energy storage investors are represented by WT, PV, DG and ES respectively.

[0145] Referring to the concept of network game, the game model of n investors is shown in equation (1).

[0146] NEG={N,E,G={G ij} ij∈E} (1)

[0147] Where: NEG is the set of all game modes; N = {1, 2, ···, n} is the set of investors; E is the pairwise combination of n investors; Gij represents the cooperation state of investors i and j, the cooperation state Gij = 1, and the non-cooperation state Gij = 0. The present invention regards G as a binary code, and uses the decimal value of G to represent the game mode number between investors. Some examples are given below.

[0148] The game model of multiple investors such as wind turbines, photovoltaics, diesel generators and energy storage is as follows: N = {WT, PV, DG, ES}, E = {WT&PV, WT&DG, WT&ES, PV&DG, PV&ES, DG&ES}. If any investor chooses to cooperate with other investors, G = {1,1,1,1,1}, and the current game mode number is 63, which is the decimal value of G. According to the cooperation between investors, the virtual power plant structure corresponding to the 15 game modes is shown in Table 1.

[0149] Table 1 Comparison between virtual power plant structure and game model

[0150]

[0151]

[0152] It is worth noting that the assumption of the present invention is one of the scenarios of various distributed power sources and their investors in the virtual power plant. In practice, the types and quantities of distributed power sources can be increased according to the needs of the power grid, and various scenarios can be constructed according to the needs of investors, and the corresponding objective functions can be changed, which does not affect the applicability of the virtual power plant capacity optimization configuration method proposed by the present invention.

[0153] (2) Virtual power plant strategy

[0154] The virtual power plant strategy includes two parts: the investment strategy of the virtual power plant investors and the operation and control strategy of the virtual power plant.

[0155] The investment strategy of investors refers to the action strategy that each investor may take according to his or her own preferences in the game. Si is a set of investment strategies that investor i can choose, si is the investment strategy selected by investor i, si∈Si. In the present invention, the investment strategy of each investor is based on its own distributed power installed capacity, represented by sWT, sPV, sDG, and sES. The investment strategy set of each investor is the boundary of the distributed power installed capacity, which is a continuous strategy interval, represented by SWT, SPV, SDG, and SES, as shown in equation (2).

[0156]

[0157] Where: sWT,min, sPV,min, sDG,min, sES,min are the minimum values ​​of the installed capacity of wind turbines, photovoltaics, diesel generators and energy storage respectively; sWT,max, sPV,max, sDG,max, sES,max are the maximum values ​​of the installed capacity of wind turbines, photovoltaics, diesel generators and energy storage respectively.

[0158] The operation control strategy of the virtual power plant is of great significance to the capacity configuration of the virtual power plant. The virtual power plant constructed by the present invention adopts a centralized-decentralized control structure. The virtual power plant control center coordinates the information reported by the control centers in various places and the investment preferences of investors to make capacity configuration decisions. The specific situation is as follows:

[0159] First, when the output power of wind power and photovoltaic power is greater than the regional load demand, the diesel generator set will not be started, and renewable energy generation will meet the power load demand. During peak electricity prices, the energy storage will not be charged, and the remaining electricity will be sold to the grid first to obtain maximum benefits. During off-peak electricity prices, the energy storage will be charged first. If the remaining capacity of the energy storage reaches the rated installed capacity, the remaining electricity will be sold to the grid.

[0160] Second, when the output power of wind power and photovoltaic power is less than the regional load demand, the load demand is met by discharging energy storage. If the load demand is still not met, consider purchasing electricity from the grid or starting diesel generators. During peak electricity prices, diesel generators are started first to meet the load demand. During non-peak electricity prices, electricity is purchased from the grid first to meet the load demand.

[0161] (3) Analysis of Virtual Power Plant Regulation Capacity

[0162] Distributed power generation is a basic component of virtual power plants. The differences in the characteristics of distributed power generation under different regulation capacities will reflect the different strategic scenarios in which virtual power plants participate. In virtual power plants, the regulation capacity analysis of wind turbines, photovoltaics, diesel generators and energy storage units is detailed as follows.

[0163] First, the regulation capacity of diesel generator sets

[0164] The adjustable power of the diesel generator set varies with the time scale. Assuming that the diesel generator set initially operates at point (0, PDG, n), when t = τ1, the cumulative adjustable power of the diesel generator is When t = τ2, the total amount of electricity that can be reduced by the diesel generator is As shown in equation (3).

[0165]

[0166] Where: PDG,a and PDG,b are the power of the diesel generator set at τ1 and τ2 respectively; and are the upward climbing rate and downward climbing rate of the diesel generator set respectively; PDG,max and PDG,min are the upper and lower limits of the output power of the diesel generator set respectively;

[0167] When t = τ3, the power of the diesel generator set has reached the lower limit PDG,c = PDG,min, and shutdown can further reduce the power output of the diesel generator set. When t = τ4, the cumulative power reduction is shown in equation (4),

[0168]

[0169] According to the above analysis, in the early stage of regulation, when the power of the diesel generator set is low, the adjustable power is a fixed value, which is limited by the climbing rate. When the power of the diesel generator gradually increases and the adjustment space for the upward power is less than the climbing capacity, the adjustment capacity is limited by the upper limit of the diesel generator power. The diesel generator's adjustment capacity index reflects the maximum upward and downward adjustment power of the diesel generator, representing the climbing rate of the diesel generator power.

[0170] Second, the regulation capability of energy storage

[0171] Assuming that the energy storage is in a discharging state during the 0-τ1 time period, the energy storage discharge amount can be adjusted by increasing or decreasing the discharge power. During the 0-τ1 time period, the energy storage can be adjusted up or down in discharge, as shown in equation (5),

[0172]

[0173] Where: PESD,max and PESD,min are the maximum and minimum discharge powers of the energy storage, respectively; EES is the remaining capacity of the energy storage; PESD is the discharge power of the energy storage; EES,max is the rated capacity of the energy storage; CDOD is the maximum discharge depth of the energy storage. The energy storage is in a charging state during the τ1-τ2 time period, and the energy storage charge can be adjusted by increasing or decreasing the charging power. During the τ1-τ2 time period, the energy storage can increase or decrease the charge, as shown in equation (6)

[0174]

[0175] Where: PESC,max and PESC,min are the maximum and minimum charging powers of the energy storage, respectively; PESC is the charging power of the energy storage.

[0176] Third, the regulation capacity of wind turbines and photovoltaics

[0177] The regulation capability of controllable distributed power sources represented by diesel generators and energy storage is deterministic, while the power output of uncontrollable distributed power sources such as wind turbines and photovoltaics is intermittent and random. By introducing the random variable x, randomness is added to the regulation capability of both, that is, in a certain period of time, wind turbines and photovoltaics output power PWT(x) and PPV(x) with the probability of πWT and πPV respectively.

[0178] 2) Virtual power plant aggregation model and cost-benefit analysis index composition

[0179] In VPP, in addition to controllable distributed power sources, the power output of uncontrollable distributed power sources such as wind turbines and photovoltaics is not only related to the installed capacity, but also to the environmental scenario. Therefore, in the process of VPP capacity configuration, it is necessary to fully consider the coordination between the regulation capacity of controllable distributed power sources and the randomness of renewable energy generation. Reasonable allocation of the capacity of each distributed power source to participate in the electricity market is a typical asset allocation problem in portfolio theory. Therefore, portfolio theory can be applied to the capacity configuration of VPP.

[0180] (1) Virtual Power Plant Aggregation Model

[0181] Assuming that the VPP includes NWT wind turbines, NPV photovoltaics, NDG diesel generators, and NES energy storage, the power characteristics of each distributed power source are detailed as follows.

[0182] First, it is generally believed that the wind speed follows the Weibull distribution, and the power output of the wind turbine is shown in equation (7).

[0183]

[0184] Where: PWT,r is the rated power of the wind turbine; vci, vr, vco are the cut-in wind speed, rated wind speed and cut-out wind speed of the wind turbine respectively.

[0185] Second, the photovoltaic power generation is closely related to the light intensity, so the photovoltaic power generation is as shown in equation (8),

[0186] P PV =r PV A PV η PV (8)

[0187] Where: rPV is the actual light intensity during the period; APV is the area of ​​the photovoltaic array; ηPV is the photoelectric conversion efficiency.

[0188] Third, by configuring diesel generators to smooth out the volatility of renewable energy generation, VPP can participate in the electricity market with a more stable power output.

[0189] Fourth, energy storage has flexible charging and discharging characteristics, can participate in demand response and peak shifting to obtain compensation income, and can also participate in the power auxiliary service market to obtain income. In consideration of cost, the present invention uses large-capacity and long-life lead-acid batteries for energy storage.

[0190] Taking into account wind turbines, photovoltaics, diesel generators and energy storage, the power generation of VPP in time period τ and the power output at time t are as shown in equation (9),

[0191]

[0192] Where: PVPP(t), PWT,i(t), PPV,i(t) and PDG,i(t) are the power outputs of VPP, the i-th wind turbine, the i-th photovoltaic, and the i-th diesel generator at time t, respectively; PESD,i(t) and PESC,i(t) are the discharge and charge powers of the i-th energy storage at time t, respectively; μESD,i indicates whether the i-th energy storage is discharged, otherwise it is set to 0, otherwise it is set to 1; μESC,i indicates whether the i-th energy storage is charged, otherwise it is set to 0, otherwise it is set to 1.

[0193] PWT,i(t) and PPV,i(t) in equation (9) are random. The present invention uses a scenario set method to transform the random optimization problem into a deterministic optimization problem for processing. The scenario set of wind turbine output power, photovoltaic output power and electricity market price is shown in equation (10).

[0194]

[0195] Where: nWT, nPV, nEP are the total number of scenario sets for wind turbines, photovoltaics, and electricity prices, respectively, and the total number of scenarios is nWT×nPV×nEP. The probability of each scenario for wind turbines, photovoltaics, and electricity prices is π(WTi), π(PVj), π(EPk), respectively. In the subsequent analysis, the variable subscript wpe indicates the form of the variable under the wind turbine scenario WTi, the photovoltaic scenario PVj, and the electricity price scenario EPk.

[0196] The regulation ability of VPP can compensate for its randomness. The up-regulation and down-regulation ability of VPP is shown in equation (11).

[0197]

[0198] Where: and are the adjustable power of VPP, respectively; and are the adjustable power of the i-th diesel generator set, respectively; and are the adjustable power of the i-th energy storage group when discharging, respectively; and are the adjustable power of the i-th energy storage group when charging, respectively.

[0199] From the above analysis, it can be seen that by aggregating distributed power sources with different regulation capabilities, VPP participates in the power market with power generation EVPP in the [0, τ1] time period, and the power that VPP can increase and decrease in the [τ1, τ2] time period is and respectively. On the one hand, the regulation capability of VPP can compensate for its own randomness and participate in the power market with relatively stable power output. On the other hand, it can directly participate in the power auxiliary service market.

[0200] (2) Risk measurement based on VaR and CVaR

[0201] Value at risk (VaR) and conditional value at risk (CVaR) are commonly used risk measurement techniques in investment decisions and have been widely used in various fields. Based on the scenario set method, the present invention uses historical data to estimate the conditional value at risk. By using CVaR to measure the economic losses caused by the uncertainty of renewable energy generation and electricity prices, combined with the capacity configuration model of virtual power plants, a capacity configuration plan for multi-investor virtual power plants under different risk preferences can be obtained.

[0202] For a given confidence level σ, VaR,σ and CVaR,σ are used to represent the index values ​​of VaR and CVaR, respectively, as shown in equation (12),

[0203]

[0204] Where: f(x,y) is the loss function of the investment plan, x is the decision variable, y is the random variable that causes the loss; ρ(y) is the probability density function of y; κ is the boundary value of the risk value. The expected loss distribution function of the decision plan φ(x,α) is shown in equation (13).

[0205]

[0206] Since it is difficult to find the analytical expression of CVaR,σ(x), an auxiliary function Fσ(x,κ) is usually constructed to solve CVaR, as shown in equation (14),

[0207]

[0208] According to the probability of each scenario of wind turbines, photovoltaics and electricity prices, the auxiliary function Fσ(x,κ) is transformed and represented by δ, as shown in equation (15),

[0209]

[0210] Where: Mwpe is a virtual coefficient, indicating the amount of loss exceeding the VaR value. Then, C VaR,σ =min{δ}.

[0211] (3) Cost-benefit analysis indicators

[0212] In order to quantitatively analyze the investment returns of multi-investor virtual power plants over the entire life cycle, the present invention first uses the net present value method (NPV) to evaluate the returns of virtual power plants over the entire life cycle, thereby measuring the feasibility of virtual power plant investment. The net present value refers to the amount of income of a virtual power plant over the entire life cycle, taking into account the time value of money. The larger the net present value, the better the investment benefits and the more feasible the plan. The net present value index value is represented by NNPV, as shown in equation (16)

[0213]

[0214] Where: M is the useful life of the system; r is the discount rate; CI(m) is the cash inflow value in the mth year; CO(m) is the cash outflow value in the mth year.

[0215] Then, the internal rate of return (IRR) method is used to measure the risk resistance of virtual power plant investment. The internal rate of return index value is represented by NIRR, which is closely related to the net present value and is the discount rate when NNPV = 0. The internal rate of return refers to the maximum currency depreciation rate that the virtual power plant investment can withstand during the entire life cycle of the virtual power plant, taking into account the time value of money, that is, the risk resistance of the virtual power plant investment. NIRR is shown in equation (17),

[0216]

[0217] Where: rpos and rneg are the corresponding discount rates when NNPV is a positive indicator and a negative indicator respectively; pos and neg are the values ​​of the positive and negative indicators of NNPV respectively.

[0218] 3) Capacity allocation model of multi-investor virtual power plants

[0219] The income I obtained by each investor in the game is a function of the investment strategy s. In the present invention, the income of each investor is expressed as IWT, IPV, IDG and IES, that is, the difference between income and cost. The cost of investor i includes initial construction cost, operation and maintenance cost, fuel cost, environmental cost, power outage penalty cost and risk cost. The income of investor i includes the income from electricity sales in the day-ahead market, the environmental income from renewable energy power generation and the income from the power auxiliary service market. Depending on whether there is a cooperative relationship between investors, it can be divided into cooperative games and non-cooperative games. Cooperative games focus on group rationality and study how to maximize group benefits through cooperation, as well as how to distribute group benefits to form a stable investment portfolio strategy. Non-cooperative games focus on individual rationality and study how to maximize individual benefits. The present invention aims to maximize the benefits of each investor, and based on a variety of virtual power plant structures of cooperative games and non-cooperative games, proposes a method for constructing a capacity configuration model of a multi-investor virtual power plant.

[0220] (1) Investor cost function

[0221] The investor's annual cost, that is, the cash outflow value CO(m) in the mth year, consists of the following six parts.

[0222] First, the initial construction cost of equal annual value

[0223]

[0224] Where: CWT,inv, CPV,inv, CDG,inv and CES,inv are the installed costs of a single wind turbine, a single photovoltaic unit, a single diesel generator set and a single lead-acid battery unit respectively.

[0225] Second, annual operation and maintenance costs

[0226] C om =N WT C WT,om +N PV C PV,om +N DG C DG,om +N ES C ES,om (19)

[0227] Where: CWT,om, CPV,om, CDG,om and CES,om are the annual operation and maintenance costs of a single wind turbine, a single photovoltaic unit, a single diesel generator set and a single lead-acid battery set respectively;

[0228] Third, the annual fuel cost of diesel generators

[0229]

[0230] Where: prDG is the diesel price coefficient, yuan / kg; QDG is the fuel consumption coefficient, g / kWh; PDG,wpe(t) is the diesel generator power at time t;

[0231] Fourth, the annual environmental cost of diesel generators

[0232]

[0233] Where: ne is the total number of pollutant emissions from diesel generators; m(e), Y(e), and V(e) are the emission, fine magnitude, and environmental value of the e-th pollutant, respectively;

[0234] Fifth, annual power outage penalty costs

[0235] The penalty Cpen,wpe that should be paid for load outage caused by insufficient VPP power supply is distributed according to the installed capacity of each distributed generation. The penalty that distributed generation i should pay is shown in equation (24):

[0236]

[0237]

[0238]

[0239] Where: ∑s i represents the total installed capacity of distributed generation, i∈{WT,PV,DG,ES}; coefpen is the penalty coefficient, RMB / kWh; Ppen,wpe(t) represents the power shortage at time t; PL(t) represents the load at time t; ∑P i represents the power output of each distributed power source at time t; Ptie,max represents the tie line limit between the area and the upper power grid,

[0240] Sixth, the risk cost caused by VPP randomness

[0241] The randomness of renewable energy generation such as wind turbines and photovoltaics in VPP will cause certain risks to their participation in the electricity market. The risk cost caused by the randomness of VPP is described by the product of the CvaR value and the risk preference coefficient L as shown in equation (15), and a virtual variable Mwpe is introduced. For the convenience of solution, let Mwpe = max{f(x,y)-κ,0}, and the loss function f(x,y) takes the negative value of the income function. For the convenience of calculation, Mwpe is relaxed to the inequality shown in equations (25) and (26)

[0242] M wpe ≥0 (25)

[0243] M wpe ≥-(R em,wpe +R e,wpe +R as,wpe )+κ (26)

[0244] (2) Investor income function

[0245] The investor's annual income, that is, the cash inflow value CI(m) in the mth year, consists of the following three parts.

[0246] First, the annual electricity sales revenue in the day-ahead market

[0247] The electricity sales revenue of each distributed generation in the day-ahead market is as shown in equation (27):

[0248]

[0249] Where: Pi,s,wpe(t) is the power sold by each distributed generation to the grid at time t, i∈{WT,PV,DG,ES}; is the electricity price under the day-ahead market price scenario EPk at time t;

[0250] Second, the annual environmental income of renewable energy generation

[0251] In addition to environmental penalties for diesel generators, environmental rewards should also be given to renewable energy generation such as wind and light in VPPs to increase the use of renewable energy in VPPs. The annual environmental income of renewable energy generation is shown in equation (28):

[0252]

[0253] Where: PWT,wpe(t) and PPV,wpe(t) are the output power of wind turbines and photovoltaics at time t respectively; coefr is the incentive coefficient, yuan / MWh;

[0254] Third, the annual revenue of the power auxiliary service market

[0255] In addition to participating in the day-ahead market, VPP can also participate in auxiliary services such as peak load shaving and valley filling through the regulation capabilities of diesel generators and energy storage to generate revenue. The auxiliary service income of wind turbines and photovoltaics is 0.

[0256]

[0257] Where: [τ1,τ2] is the peak electricity price period; [τ3,τ4] is the non-peak electricity price period; θ∈[0,1] is the VPP regulation capacity ratio coefficient, which indicates the proportion of VPP participating in auxiliary services through its own regulation capacity; and are the upward and downward adjustable powers of VPP at time t respectively; and are the upward and downward adjustable powers of the i-th diesel generator set at time t respectively; and are the upward and downward adjustable discharge powers of the i-th group of energy storage at time t respectively; and are the upward and downward adjustable charging powers of the i-th group of energy storage at time t respectively;

[0258] (3) Objective function

[0259] In summary, the objective function of the multi-investor virtual power plant capacity allocation model based on maximizing the benefits of each investor is as shown in equation (30):

[0260]

[0261] Where: L represents the investor's risk preference, L∈[0,1]. When L<0.1, it means that the investor tends to choose a high-risk, high-return investment strategy, is not afraid of risk, and is a positive risk preference. When L>0.5, it means that the investor is afraid of risk, but hopes to gain certain returns on the basis of investment preservation, and is a conservative risk preference. The risk preference coefficient should be in line with the actual situation of the investor and should be selected reasonably.

[0262] The net profit functions IWT, IPV, IDG, and IES of each investor can refer to equation (30) and select some contents for construction. The virtual power plant capacity configuration model of multi-investor cooperative game and non-cooperative game is described in detail as follows.

[0263] First, the cooperative game model

[0264] As shown in Table 1, the wind turbine-photovoltaic-diesel generator-energy storage alliance of game mode 63. In this mode, each distributed power source investor forms a fully cooperative game to maximize the group's benefits. It is a single-objective optimization model that can be solved using the particle swarm algorithm.

[0265] Participants: {WT, PV, DG, ES},

[0266] Strategy set: SWT&PV&DG&ES=[sWT,min,sWT,max,sPV,min,sPV,max,sDG,min,sDG,max,sES,min,sES,max],

[0267] Objective function: max IWT&PV&DG&ES(sWT,sPV,sDG,sES)=IWT+IPV+IDG+IES

[0268] Balanced strategy: (s*WT, s*PV, s*DG, s*ES) = argmax IWT&PV&DG&ES(sWT, sPV, sDG, sES);

[0269] Second, non-cooperative game model

[0270] Taking the game between the wind turbine-diesel generator alliance and the photovoltaic-energy storage alliance in game mode 18 in Table 1 as an example, the capacity optimization problem of multiple virtual power plant alliances is explained. In game mode 18, there is a non-cooperative game between virtual power plant alliance models, and a cooperative game is formed within each virtual power plant alliance to seek the maximization of group benefits.

[0271] Participants: {WT,DG}, {PV,ES}.

[0272] Policy set: SWT&DG=[sWT,min,sWT,max,sDG,min,sDG,max], SPV&ES=[sPV,min,sPV,max,sES,min,sES,max].

[0273] Objective function: max IWT&DG(sWT,sDG)=IWT+IDG,max IPV&ES(sPV,sES)=IPV+IES

[0274] Balance strategy: (s*WT,s*DG)=argmax IWT&DG(sWT,s*PV,sDG,s*ES),(s*PV,s*ES)=argmax IPV&ES(s*WT,sPV,s*DG,sES)

[0275] The specific solution process of the non-cooperative game model is briefly described as follows: Step 1, input the original data and initialize the capacity configuration. Step 2, use the particle swarm algorithm to optimize the benefits of each virtual power plant alliance. Step 3, determine whether the two adjacent configuration results are consistent. If not, return to step 2; if consistent, output the final configuration result.

[0276] (4) Constraints

[0277] First, diesel generator power constraints

[0278] 0≤P DG,wpe (t)≤N DG P DG,max (31)

[0279] Where: PDG,max is the rated power of the diesel generator set;

[0280] Second, energy storage remaining capacity and power constraints

[0281] E ES,wpe (t) = E ES,wpe (t-1)+η ESC μ ESC (t)P ESC,wpe (t)-μ ESD (t)P ESD,wpe (t) / η ESD (32)

[0282] (1-C DOD )N ES E ES,max ≤E ES,wpe (t)≤N ES E ES,max (33)

[0283] P ESC,wpe (t)≤N ES P ESC,max (34)

[0284] P ESD,wpe (t)≤N ES P ESD,max (35)

[0285] Where: EES,wpe(t), PESC,wpe(t) and PESD,wpe(t) are the remaining capacity, charging and discharging power of energy storage in each scenario at time t, respectively; ηESC and ηESD are the charging and discharging efficiencies of energy storage, respectively;

[0286] Third, power balance constraints

[0287]

[0288] Where: PWT,WTi(t) is the output of a single wind turbine under the wind power scenario WTi at time t; PPV,PVj(t) is the output of a single photovoltaic group under the photovoltaic scenario PVj at time t;

[0289] 4) A revised model of cooperative benefit distribution based on the improved Shapley value method

[0290] Although cooperative games focus on maximizing the benefits of the virtual power plant alliance, investors still focus on their own benefits. An effective cooperative benefit allocation strategy is the key to the stable existence of the virtual power plant alliance. For a stable virtual power plant alliance, its cooperative benefit allocation strategy must satisfy individual rationality and group rationality. Individual rationality requires that the benefits allocated to each investor in the virtual power plant alliance should not be lower than the benefits when not participating in the virtual power plant alliance. Group rationality requires that investors generate more benefits when cooperating, that is, the benefits of the virtual power plant alliance are greater than the sum of the benefits of each investor.

[0291] At present, there are many allocation strategies, such as average allocation strategy, capacity allocation strategy, allocation scheme based on Shapley value, etc. The allocation strategy based on Shapley value has been proven to satisfy individual rationality and group rationality, and is a more reasonable method to solve the problem of multi-investor cooperative income distribution. Therefore, the academic community often uses the allocation strategy based on Shapley value to study the problem of cooperative income distribution. The allocation strategy based on Shapley value is referred to as Shapley value method in the following text.

[0292] (1) Shapley value method

[0293] The Shapley value method is based on the contribution of investors in the virtual power plant alliance. The benefit allocated to investor i from the virtual power plant alliance Ψ is xi(ψ), as shown in equation (37):

[0294]

[0295]

[0296] Where: U is the subset of all virtual power plant alliances Ψ that include investor i; W(U) is the weighting factor; I(U) is the revenue of the virtual power plant alliance; I(U\I) is the revenue of the virtual power plant alliance after investor i is removed; n is the total number of investors;

[0297] (2) Shortcomings of the Shapley value method

[0298] Although the Shapley value method has been widely used in solving income distribution problems, there are still three problems.

[0299] First, the average marginal contribution value of various possible subsets of virtual power plant alliances is used as the only basis for investors to distribute profits. This method is currently controversial. For example, in a virtual power plant alliance, although the profits created by investors are not much, they help to improve the quality of the entire virtual power plant alliance and are of great help in maintaining the long-term stability of the virtual power plant alliance. Therefore, using the contribution rate as the only basis for distributing profits is not conducive to the long-term stability of the virtual power plant alliance.

[0300] Second, using the difference between revenue and cost as the sole basis for measuring investor contribution is unfair in some cases. For example, the returns of high-revenue and high-cost investors may be similar to those of low-revenue and low-cost investors, but the party with more investment has a higher degree of participation in the virtual power plant alliance. Therefore, using revenue alone as the basis for measuring contribution will, to a certain extent, harm the interests of investors with high participation.

[0301] Third, each investor has different risk tolerance and risk appetite. Although a virtual power plant alliance with a higher risk factor can create more profits, it is less attractive to conservative investors. This allocation method that does not consider risk appetite is likely to suppress the enthusiasm of investors with different risk appetites to join the virtual power plant alliance.

[0302] At present, existing literature has confirmed that there are indeed some problems in the application of Shapley value method in income distribution, so it has been modified to varying degrees. Based on the existing research, this paper proposes an improved Shapley value method to address the three problems of the Shapley value method, and establishes a profit distribution correction model to promote the long-term stability of the virtual power plant alliance.

[0303] (3) Improved profit distribution model

[0304] First, the basic assumption

[0305] In the virtual power plant alliance, each investor forms a cooperative game relationship and distributes cooperative benefits. Cooperative game emphasizes group rationality and pursues efficient and fair benefits. The profit distribution correction model proposed in this invention is based on the following two assumptions: each investor is absolutely rational, independent of each other, and makes the best decision with the goal of maximizing its own benefits; each investor has group rationality and works 100% hard to complete the work, so that all virtual power plant alliances can achieve the expected benefits.

[0306] Second, the improved Shapley value method

[0307] Aiming at the three problems of allocating cooperative benefits based on Shapley value method, an improved Shapley value method is proposed from three aspects: participation, risk preference and marginal contribution.

[0308] Participation: The more an investor invests in the virtual power plant alliance, the higher its participation and the greater its expected return. The present invention measures the participation of each distributed power source investor by installed capacity, that is, the proportion of each distributed power source installed capacity to the total installed capacity, as shown in equation (39):

[0309]

[0310] Where: is the participation coefficient of power source i in the virtual power plant;

[0311] Risk preference: The power generation of renewable energy such as wind and light is greatly affected by the natural environment, which makes its returns have certain risks. Compared with diesel generators and energy storage, its risk preference is lower. Therefore, the overall risk preference coefficient of the virtual power plant alliance is different from the risk preference coefficient of each investor. Investors whose personal risk preference coefficient is higher than that of the virtual power plant alliance are in an advantageous position. The difference in risk preference coefficient βi is introduced to indicate the attractiveness of the virtual power plant alliance to investors, as shown in equation (40)

[0312]

[0313] Where: Li∈[0,1] is the risk preference coefficient of investor i when he does not participate in the virtual power plant alliance; kj is a state variable, which takes 1 when investor j participates in the virtual power plant alliance, otherwise it takes 0; is the profit of investor j when he does not participate in the virtual power plant alliance;

[0314] The risk preference coefficient difference is normalized as shown in equation (41):

[0315]

[0316] Marginal contribution: Marginal contribution is the main basis for applying the Shapley value method to profit distribution. The marginal contribution γi(U) of investor i in the virtual power plant alliance U is shown in equation (42):

[0317] γ i (U)=I(U)-I(U\i) (42)

[0318] The marginal contribution is normalized as shown in equation (43)

[0319]

[0320] Third, the income distribution correction model

[0321] J = {α, β, γ} is defined as the set of three influencing factors, including participation, risk preference and marginal contribution. According to equations (39), (41) and (43), the correction matrix A is established, as shown in equation (44):

[0322]

[0323] Assume that μ = [μα, μβ, μγ] is the weight vector of the three influencing factors in J, and μα + μβ + μγ = 1. It should be noted that the values ​​of μα, μβ, and μγ will affect the virtual power plant alliance model chosen by investors and should be determined before the virtual power plant alliance is formed. The return of any investor in the virtual power plant alliance should be higher than that of other virtual power plant alliance models, otherwise investors will pursue higher returns, which will lead to instability in the virtual power plant alliance. Therefore, the optimization model of μα, μβ, and μγ in the virtual power plant alliance U is shown in equation (45)

[0324]

[0325] Where: xi(U) is the profit obtained by investor i after participating in the virtual power plant alliance U; xi(U) is the profit obtained by investor i under any other game mode;

[0326] After determining the modified matrix A and weight vector μ of the influencing factors, the improved comprehensive coefficient vector of the influencing factors of each investor ω=AμT=[ωWT,ωPV,ωDG,ωES]T. At this time, the benefit xi′(U) obtained by investor i from the virtual power plant alliance U is shown in equation (46):

[0327] x i ′(U)=ω i (U)×I(U) (46)

[0328] Where: ωi(U) represents the comprehensive coefficient of the influencing factors of investor i in the virtual power plant alliance U, i∈{WT,PV,DG,ES};

[0329] The game model and profit distribution model proposed in this invention are solved on MATLAB R2018b.

[0330] 5) The virtual power plant capacity optimization configuration method considering multi-agent game under the guidance of the effectiveness of the present invention will build a capacity configuration game model of a multi-agent virtual power plant based on the MATLAB R2018b platform in an environment where the CPU is Inter(R)Core(TM)i7-8250U and the main frequency is 1.8GHz, and the cooperative profit distribution correction model based on the improved Shapley value method proposed in the present invention is used to improve the robustness of the virtual power plant capacity configuration model.

[0331] Best Mode

[0332] The best implementation example takes a virtual power plant to be planned in a certain place in my country as an example, establishes a virtual power plant capacity optimization configuration model for multiple investment entities, and uses a cooperative benefit correction model based on the improved Shapley value method to further improve the robustness of the virtual power plant capacity configuration model, thereby obtaining the virtual power plant capacity optimization configuration method that considers multi-agent game under the effectiveness orientation proposed by the present invention.

[0333] 1) Basic data of the example

[0334] Based on the wind speed, light intensity, electricity price and load data of a certain region in my country, the virtual power plant capacity optimization configuration model considering multi-agent game under the performance orientation proposed in this invention is used for calculation. The electricity price, wind speed and light intensity data from 2017 to 2020 are selected as different scenarios of the electricity price scenario set, wind power scenario set and photovoltaic scenario set, namely n WT =n PV =n EP =4, the probability of each scenario is 0.25, and the total number of scenarios is 64, that is, n WT ×n PV ×n EP =64. Considering that the electricity load from 2017 to 2020 is relatively stable, the load data selects the average load value from 2017 to 2020. The annual electricity price, wind and solar environment conditions and load data curves are shown in Figure 2. The diesel price is 6.47 yuan / kg.

[0335] Considering that the average annual load in the area is 0.3MW, a wind turbine with a rated power of 0.2MW and a diesel generator with a rated power of 0.1MW are selected. According to the power level requirements of the power system, a 2.8kW photovoltaic array with 14 photovoltaic cells rated at 0.2kW in series is selected. Within the grid-connected voltage range, 125 lead-acid batteries rated at 400Ah are selected in series for 0.1MWh energy storage. The detailed parameters of the wind turbine, photovoltaic cell, diesel generator and lead-acid battery selected by the present invention are shown in Table 2. The pollutant emissions, environmental values ​​and fines of the diesel generator are shown in Table 3.

[0336] Table 2 Related parameters of each distributed power source

[0337]

[0338]

[0339] Table 3 Pollutant emission data of diesel generator sets

[0340]

[0341] 2) Optimization of distributed power generation installation capacity

[0342] This section selects parameters that are relatively balanced between net income and conditional risk value, that is, assuming that the risk preference coefficient L is 0.2 and the VPP adjustable capacity ratio coefficient θ is 0.5, and uses game theory to optimize the capacity of each distributed power source in the virtual power plant. The configuration of wind turbines, photovoltaics, diesel generators and energy storage capacity under various game modes, such as Figure 3 shown.

[0343] from Figure 3 It can be seen that under various game modes, the installed capacity of distributed power generation shows certain similarities and differences. Under any game mode, wind power installed capacity is the largest, followed by diesel generators and energy storage, and photovoltaic installed capacity is the smallest.

[0344] The difference is reflected in that when wind turbines and photovoltaics form an alliance, as shown in game modes 32, 33, 42, 52, and 63, the installed capacity of both decreases significantly, reflecting the complementary characteristics of wind turbines and photovoltaics. The installed capacity of diesel generators and energy storage is greatly reduced, further proving that the unified planning of wind power and photovoltaics can effectively improve the quality of electricity, reduce construction costs, and avoid idleness and waste of resources.

[0345] 3) Distribution of cooperative benefits of the virtual power plant alliance

[0346] Based on the best embodiment 2), the benefits of investors under various game modes are shown in Table 4. In most game models, when investors choose to cooperate with other investors, they can obtain more benefits. As shown in Table 1, cooperative game model 63, the wind turbine-photovoltaic-diesel generator-energy storage alliance has the largest benefits.

[0347] Table 4 Total revenue of virtual power plants under various game modes

[0348] serial number Income / 10,000 yuan serial number Income / 10,000 yuan serial number Income / 10,000 yuan 0 469.22 8 461.91 32 486.48 1 474.71 12 462.17 33 489.13 2 459.78 16 476.26 42 494.62 4 452.47 18 465.27 52 490.68 7 444.58 25 486.22 63 497.72

[0349] pass Figure 3 From the analysis of Table 4, it can be seen that the joint planning of multiple distributed generation sources can create more benefits at a lower cost and more easily maximize the group benefits.

[0350] Non-cooperative games emphasize individual rationality. Each investor only pursues the maximization of his own benefits, which may lead to non-optimal group benefits. Any investor can cooperate with other investors to maximize the benefits of the virtual power plant alliance, and then further improve his own benefits by reasonably distributing the group benefits. A stable cooperative game emphasizes group rationality, that is, cooperation between investors will obtain greater group benefits than non-cooperation. Investors choose whether to withdraw from the current virtual power plant alliance based on their own benefits, forming a new game model. The choice of any investor will affect the decision-making of other investors. There are six cooperation modes that meet group rationality, as shown in game modes 25, 32, 42, 52, and 63. Among them, the group benefits of investors in mode 63 are the largest, so mode 63 is the preferred mode for investors.

[0351] according to Figure 3 The correction matrix A is constructed by combining the relevant data in Table 4. When the weights of the influencing factors change, the final game mode also changes. In order to maximize the group benefits, the relevant weights can be adjusted to encourage each investor to achieve game mode 63 as much as possible. α ,μ β ,μ γ ]=[0.059,0.013,0.928], each distributed power source investor can obtain the maximum benefit, achieve a fully cooperative game, and form a virtual power plant alliance of wind turbines, photovoltaics, diesel generators and energy storage. The comparison of the profit distribution results of the cooperative game model based on the improved Shapley value method 63 and the cooperative game and non-cooperative game based on the traditional Shapley value method is shown in Table 5.

[0352] Table 5 Comparison of virtual power plant benefits under different allocation modes

[0353]

[0354] The profit distribution results in Table 5 show that in non-cooperative games, investors do not participate in cooperation, resulting in poor group benefits. According to the traditional Shapley value method, distributed power investors often form diesel generator units, energy storage units, and wind turbine-photovoltaic alliances to participate in the power system through cooperation. The former reflects the regulatory ability of controllable distributed generation to participate in the power auxiliary service market, while the latter reflects the complementary characteristics of wind power and photovoltaics. Compared with non-cooperative games, group benefits have increased, but have not yet reached the optimal level.

[0355] Since wind turbines have lower costs and higher returns after government subsidies, they have a competitive advantage over other distributed power generation in non-cooperative games and higher returns, making the improvement in returns after participating in the virtual power plant alliance not obvious. Under the improved Shapley value method, the returns of wind turbine investors are lower than those of the traditional Shapley value method. However, the returns obtained by other distributed power investors participating in the virtual power plant are higher than those of the non-cooperative game and the traditional Shapley value method, and the group returns are maximized, so the virtual power plant alliance can remain stable.

[0356] 4) Dynamic effect analysis of key parameters under the fully cooperative game model 63

[0357] (1) Impact of VPP risk preference coefficient on capacity allocation

[0358] Assume that the proportional coefficient of VPP regulation capacity θ = 0.5. Under different risk preference coefficients L, the capacity configuration results of virtual power plants under the fully cooperative game model 63 are shown in Table 6. As can be seen from Table 6, due to the increase in the risk preference coefficient L, the income I of VPP decreases, and the CVaR value also decreases accordingly. The above situation shows that investors' risk preferences are gradually changing from active to stable, and the capacity configuration strategy of virtual power plants tends to obtain income on the basis of value preservation.

[0359] Table 6 Virtual power plant capacity configuration under different risk preference coefficients L

[0360]

[0361] At present, the improvement of energy storage technology and benefits is not enough to replace diesel generators to smooth the fluctuations of renewable energy generation. When the risk preference coefficient L is small, low-cost diesel generators are used to smooth the fluctuations of renewable energy generation and sell a large amount of electricity to the power market. At this time, the installed capacity of energy storage is small and the installed capacity of diesel generators is large. When the risk preference coefficient L increases, the virtual power plant reduces the amount of electricity sold while meeting the power load demand to avoid the risks caused by price fluctuations, and the installed capacity of wind power and photovoltaic power will decrease accordingly. When the output of renewable energy generation is insufficient and the power of the interconnection line reaches its limit, the load power requirements are met by dispatching diesel generators and energy storage. Given the flexibility of energy storage charging and discharging, it can better reduce the risks brought by the randomness of renewable energy generation output, and the installed capacity of energy storage will increase at this time.

[0362] As can be seen from Table 6, as the risk preference coefficient L changes, the combination of the total return and conditional risk value of the VPP alliance also changes, and the optimization configuration results of distributed power in the VPP alliance are also different. Select the scheme with a relatively balanced return and conditional risk value (L = 0.2), hereinafter referred to as the balanced scheme. According to Table 2 and Table 6, the configuration number of distributed power in the VPP alliance [NWT ,N PV ,N DG ,N ES ]=[10,73,8,8].

[0363] Based on the balanced solution, the following chapters will analyze the impact of factors such as VPP regulation capability, environmental cost-effectiveness, and data relevance on capacity configuration, as well as the cost-effectiveness of the capacity configuration solution.

[0364] (2) Impact of VPP regulation capability on capacity configuration

[0365] Based on the balancing scheme, the randomness of VPP changes with the adjustable capacity ratio coefficient θ, as shown in Figure 4(a). When θ = 0, the adjustable capacity of VPP is all used to smooth the random fluctuations of renewable energy generation. At this time, the external regulation capacity of VPP is 0, but its randomness is minimal. As θ increases, the proportion of VPP adjustable capacity used for peak shaving and valley filling increases, and the proportion of smoothing random fluctuations of renewable energy generation decreases. At this time, the external regulation capacity of VPP gradually increases, and its randomness gradually increases. When θ = 1, the external regulation capacity of VPP is the largest, and its randomness is also the largest. At this time, the output power error of VPP reaches the maximum value. C VaR As θ changes, it is shown in Figure 4(b).

[0366] As shown in Figure 4(b), as θ increases, the randomness of VPP gradually increases. VaR Therefore, if different θ values ​​are selected, VPP has different randomness and external adjustment capabilities, and the capacity configuration of each distributed power source will also change, as shown in Table 7.

[0367] Table 7 Capacity configuration of virtual power plant under different adjustable capacity ratio coefficients θ

[0368]

[0369]

[0370] As shown in Table 7, as θ increases, the proportion of VPP regulation capacity used to smooth the random fluctuations of renewable energy gradually decreases, and the randomness of VPP gradually increases. In order to reduce the risk value brought by randomness, the installed capacity s of controllable distributed generation in the optimization results is DG and ES Gradually increase, random renewable energy generation installed capacity s WT and PV Gradually decrease.

[0371] (3) Impact of VPP environmental cost-effectiveness on capacity allocation

[0372] According to the diesel generator annual environmental cost equation (21) and the renewable energy environmental benefit equation (28), certain penalties and rewards are given to diesel generators and intermittent renewable energy generation respectively. e,wpe , without considering environmental benefits R e,wpe , while considering C e,wpe and R e,wpe The optimized configuration results of distributed power sources within the VPP alliance based on the balancing solution under the three scenarios are shown in Table 8.

[0373] Table 8 Virtual power plant capacity configuration under various environmental cost benefits

[0374]

[0375] As shown in Table 8, without considering the environmental costs of diesel generators, diesel generators account for a larger proportion in the VPP optimization configuration due to their stable and controllable characteristics, while intermittent renewable energy generation accounts for a smaller proportion. When only environmental costs are considered without considering the environmental benefits of renewable energy generation, the installed capacity of diesel generators will decrease and the installed capacity of renewable energy generation will increase. On the premise of taking into account both environmental costs and environmental benefits, the use of renewable energy for power generation can be encouraged to a certain extent. The further increase in the installed capacity of renewable energy power generation is in line with the background of energy structure transformation and expanding the scale of renewable energy utilization to support low-carbon economic growth and achieve carbon peak.

[0376] (4) Impact of wind, solar and load correlation on capacity configuration

[0377] The differences in the relationship between wind, light and load in different regions will also affect the capacity configuration results of the virtual power plant. Based on the balancing solution, the capacity configuration results of distributed power sources in the virtual power plant corresponding to the following three typical scenarios are shown in Table 9.

[0378] Scenario 1: Select electricity market price scenario EP1, wind power scenario WT1, and photovoltaic scenario PV1.

[0379] Scenario 2: Select electricity market price scenario EP2, wind power scenario WT1, and photovoltaic scenario PV2.

[0380] Scenario 3: Select electricity market price scenario EP1, wind power scenario WT4, and photovoltaic scenario PV3.

[0381] Table 9 Virtual power plant capacity configuration under different wind, light and load correlations

[0382]

[0383] Through the comparative analysis of scenario 1 and scenario 2, the results show that the higher the positive correlation between light intensity and load, the lower the risk of load loss caused by PV output fluctuations. Therefore, the installed capacity of PV is increased. Through the comparative analysis of scenario 3 and scenario 1, the results show that when wind speed and light intensity are negatively correlated, the random fluctuations of renewable energy generation are alleviated to a certain extent. Therefore, the installed capacity of diesel generators and energy storage is reduced, and the proportion of renewable energy generation capacity in the total installed capacity of VPP is increased.

[0384] (5) Cost-benefit analysis of VPP capacity configuration scheme

[0385] Based on the balanced solution, the cost-effectiveness of the VPP capacity configuration solution is analyzed. When the discount rate r = 8%, according to the net present value indicator equation (16), the cumulative N NPV Change curve, such as Figure 5 shown.

[0386] from Figure 5 It can be seen that the total life cycle of the virtual power plant based on the balancing solution is N NPV When the virtual power plant replaces energy storage in the 10th year of operation and replaces diesel generators in the 15th year, the cumulative N NPV However, virtual power plants are still profitable.

[0387] According to the discount rate r = 8%, we can calculate Figure 5 Based on the balancing scheme, the final cumulative N NPV is greater than 0, indicating that the N of the balance scheme IRR >8%. According to the internal rate of return equation (17), the N of the equilibrium solution is IRR =24%. N IRR The larger the value is, the higher the currency depreciation rate that the virtual power plant investment can withstand, and the stronger the risk resistance is. For the balanced solution, when the currency depreciation rate is equal to 24%, the investment in the virtual power plant is in a break-even state. This further illustrates that the wind turbine-photovoltaic-diesel generator-energy storage alliance has a strong risk resistance. In order to reduce investment risks, the joint planning of distributed power sources such as wind power, photovoltaics, diesel generators and energy storage should be considered.

[0388] (6) Analysis of VPP day-ahead output scheme

[0389] As a power plant, the day-ahead output plan of a VPP is crucial. For given variables (L = 0.2, θ = 0.5), the model proposed in this invention can not only give the distributed power optimization configuration strategy of the balancing solution [N WT ,NPV ,N DG ,N ES ]=[10,73,8,8], and the day-ahead output plan of VPP can be obtained according to the balancing plan, such as Figure 6 shown.

[0390] like Figure 6 As shown in the figure, during the non-peak electricity price period, when the load power consumption is less than the power output of renewable energy generation, energy storage charging is preferred, and the remaining power is sold to the grid. During the peak electricity price period, by dispatching energy storage discharge, increasing the power output of diesel generator sets, within the power limit of the tie line, electricity is sold to the grid as much as possible according to the grid requirements to increase VPP revenue. Figure 6 During the two peak load periods, in order to reduce environmental pollution, the VPP output is mainly stabilized by energy storage to meet the load demand, while the output power of the diesel generator set is less than the rated power.

[0391] The detailed description of the virtual power plant capacity optimization configuration method based on multi-agent game in the above reference embodiment is illustrative rather than restrictive, and several embodiments can be listed according to the limited scope. Therefore, changes and modifications without departing from the overall concept of the present invention should fall within the scope of protection of the present invention.

Claims

1. A virtual power plant capacity optimization configuration method based on multi-agent game, characterized by: The following steps are involved: Step 1: Analysis of multi-investor virtual power plant architecture and its regulation capabilities; Step 2: Virtual power plant aggregation model and cost-benefit analysis indicator composition; Step 3: Capacity configuration model of multi-investor virtual power plants; Step 4: Four steps of the revised model of cooperative benefit distribution based on the improved Shapley value method; The cooperative revenue allocation correction model based on the improved Shapley value method improves the robustness of the virtual power plant capacity configuration model; The analysis of the multi-investor virtual power plant architecture and its regulatory capabilities in step 1, the analysis of the multi-investor virtual power plant structure organization and its regulatory capabilities based on game theory: (1) Multi-investor virtual power plant architecture Taking the uncontrollable distributed power sources of wind turbines (WT) and photovoltaics (PV) and the controllable distributed power sources of diesel generators (DG) and energy storage devices (ES) as examples, the structure of the virtual power plant is analyzed when various distributed power sources belong to different investors. Referring to the concept of network game, the game model of n investors is shown in equation (1): NEG={N,E,G={G ij } ij∈E } (1) Where: NEG is the set of all game modes; N = {1, 2, ···, n} is the set of investors; E is a pairwise combination of n investors; G ij Indicates the cooperation status between investors i and j, cooperation status G ij =1, non-cooperative state G ij =0, G is a binary code, and the decimal value of G is used to represent the game mode number between investors; (2) Virtual power plant strategy: including 1) The investment strategy of virtual power plant investors refers to the action strategy taken by each investor in the game. The investment strategy of each investor is based on the installed capacity of distributed power generation. The investment strategy set of each investor is the boundary of the installed capacity of distributed power generation, which is a continuous strategy interval with S as the target. WT , S PV , S DG , S ES It is expressed as shown in equation (2) Where: S WT,min , S PV,min , S DG,min , S ES,min They are the minimum values ​​of installed capacity of wind turbines, photovoltaics, diesel generators and energy storage respectively; S WT,max , S PV,max , S DG,max , S ES,max They are the maximum installed capacities of wind turbines, photovoltaics, diesel generators and energy storage; 2) The operation control strategy of the virtual power plant is to adopt a centralized-decentralized control structure. The virtual power plant control center coordinates the information reported by local control centers and the investment preferences of investors to make capacity allocation decisions; (3) Analysis of the regulation capacity of virtual power plants, mapping the differences in the characteristics of distributed power sources under different regulation capacities, and virtual power plants participating in different strategic scenarios. In virtual power plants, the regulation capacity analysis includes wind turbines, photovoltaics, diesel generators, and energy storage units. The virtual power plant aggregation model and cost-benefit analysis indicators of step 2 are as follows: (1) Virtual Power Plant Aggregation Model Including N WT Typhoon generator set, N PV Photovoltaic group, N DG Diesel generator sets, N ES The power characteristics of each distributed power source are as follows: 1) The power output of the wind turbine is shown in equation (7): Where: P WT,r is the rated power of the wind turbine; v ci , v r , v co are the cut-in wind speed, rated wind speed and cut-out wind speed of the wind turbine. 2) Photovoltaic power generation is closely related to light intensity, so the photovoltaic power generation is as shown in equation (8): P PV =r PV A PV or PV (8) Where: r PV is the actual light intensity during this period, A PV is the area of ​​the photovoltaic array, η PV is the photoelectric conversion efficiency, 3) By configuring diesel generators to smooth out the volatility of renewable energy generation, virtual power plants participate in the electricity market with stable power output; 4) Taking advantage of the charging and discharging characteristics of energy storage, participating in demand response to shift peak power consumption to obtain compensation income and participating in the power auxiliary service market to obtain income, comprehensively considering wind turbines, photovoltaics, diesel generators and energy storage, the power generation of the virtual power plant in time period τ and the power output at time t are as shown in equation (9): Where: P VPP(t) , P WT,i (t), P PV,i (t) and P DG,i (t) are the power outputs of VPP, the i-th wind turbine, the i-th photovoltaic group, and the i-th diesel generator at time t; P ESD,i (t) and P ESC,i (t) are the discharge and charge power of the i-th group of energy storage at time t; μ ESD,i Indicates whether the i-th group of energy storage is discharged, otherwise it is set to 0, and if it is, it is set to 1; μ ESC,i Indicates whether the i-th group of energy storage is charged, otherwise it is set to 0, and if it is, it is set to 1; (2) Based on V aR and C VaR Risk measurement For a given confidence level σ, V aR,σ and C VaR,σ Used to represent V aR and C VaR The exponential value of is shown in equation (12) Where: f(x,y) is the loss function of the investment plan, x is the decision variable, y is the random variable that causes the loss; ρ(y) is the probability density function of y; κ is the boundary value of the risk value; (3) Cost-benefit analysis indicators 1) The net present value (NPV) method is used to evaluate the benefits of virtual power plants over their entire life cycle. The NPV index is expressed as NPV It is expressed as shown in equation (16) Where: M is the service life of the system; r is the discount rate; C I (m) is the cash inflow value in the mth year; C O (m) is the cash outflow value in year m; 2) The internal rate of return (IRR) method is used to measure the risk resistance of virtual power plant investment. The internal rate of return index value N IRR As shown in equation (17) Where: r pos and r neg When N NPV The corresponding discount rates for positive and negative indicators; pos and neg are N NPV The values ​​of positive and negative indicators; the capacity configuration model of the multi-investor virtual power plant in step 3 (1) Investor cost function: the annual cost of the investor, i.e. the cash outflow value C in the mth year O (m), consists of the following six parts: ① Initial construction cost of equal annual value Where: C WT,inv , C PV,inv , C DG,inv and C ES,inv They are the installation costs of a single wind turbine, a single photovoltaic unit, a single diesel generator set, and a single lead-acid battery unit; ②Annual operation and maintenance cost C om =N WT C WT,om +N PV C PV,om +N DG C DG,om +N ES C ES,om (19) Where: C WT,om , C PV,om , C DG,om and C ES,om They are the annual operation and maintenance costs of a single wind turbine, a single photovoltaic unit, a single diesel generator set, and a single lead-acid battery unit; ③ Annual fuel cost of diesel generator Where: pr DG is the diesel price coefficient, yuan / kg; Q DG is the fuel consumption coefficient, g / kWh; P DG,wpe (t) is the power of the diesel generator set at time t; ④ Annual environmental cost of diesel generators ⑤ Annual power outage penalty costs, The penalty Cpen,wpe that should be paid for load outage caused by insufficient VPP power supply is allocated according to the proportion of installed capacity of each distributed power source. The penalty that distributed power source i should pay is: Where: ∑s i represents the total installed capacity of distributed generation, i∈{WT,PV,DG,ES}; coef pen is the penalty coefficient, yuan / kWh; P pen,wpe (t) represents the power shortage at time t; P L (t) represents the load at time t; represents the power output of each distributed power source at time t; P tie,max Indicates the tie line limit between the area and the upper grid; ⑥ The risk cost caused by the randomness of the virtual power plant is introduced by introducing a virtual variable M wpe , M wpe Relaxing the inequalities shown in Eq. (25) and Eq. (26) M wpe ≥0 (25) M wpe ≥-(R em,wpe +R e,wpe +R as,wpe )+k (26) (2) Investor income function: the investor’s annual income, i.e., the cash inflow value C in the mth year I (m) includes the following three parts ① Annual electricity sales revenue in the market today The electricity sales revenue of each distributed power source in the market is shown in equation (27): Where: P i,s,wpe (t) is the power sold by each distributed generation to the grid at time t, i∈{WT,PV,DG,ES}; λ PEk is the PE of the day-ahead market price scenario at time t k The electricity price under ② The annual environmental income of renewable energy power generation is as shown in equation (28): Where: P WT,wpe (t) and P PV,wpe (t) are the output power of wind turbine and photovoltaic at time t; coef r is the incentive coefficient, yuan / MWh; ③ Annual income from the power auxiliary service market: diesel generators and energy storage can participate in peak load shaving and valley filling auxiliary services to generate revenue. Where: [τ1, τ2] is the peak electricity price period; [τ3, τ4] is the non-peak electricity price period; θ∈[0,1] is the VPP regulation capacity proportionality coefficient, and are the upward and downward adjustable powers of VPP at time t respectively; and are the upward and downward adjustable powers of the i-th diesel generator set at time t respectively; and are the upward and downward adjustable discharge powers of the i-th group of energy storage at time t respectively; and are the upward and downward adjustable charging powers of the i-th group of energy storage at time t, respectively; (3) Objective function The objective function of the multi-investor virtual power plant capacity allocation model based on maximizing the benefits of each investor is shown in equation (30): Where: L represents the risk preference of investors, L∈[0,1]. When L<0.1, it means that investors tend to choose high-risk and high-return investment strategies, which belongs to the active type in risk preference. When L>0.5, it means that investors belong to the conservative type in risk preference.

2. The virtual power plant capacity optimization configuration method based on multi-agent game according to claim 1 is characterized by: The modified model of cooperative benefit distribution based on the improved Shapley value method in step 4 (1) Shapley value method The income distributed to investor i from the virtual power plant alliance Ψ is x i (ψ), as shown in equation (37) Where: U is the subset of all virtual power plant alliances Ψ that include investor i; W(U) is the weighting factor; I(U) is the revenue of the virtual power plant alliance; I(U\i) is the revenue of the virtual power plant alliance after investor i is removed; n is the total number of investors; (2) Improved Shapley value method: an improved Shapley value method is proposed from three aspects: participation, risk preference and marginal contribution. Participation: The participation of each DG investor is measured by installed capacity, that is, the proportion of each DG installed capacity to the total installed capacity, as shown in equation (39): Where: is the participation coefficient of power source i in the virtual power plant; Risk preference: Introducing the risk preference coefficient β i The difference between the two is used to indicate the attractiveness of the virtual power plant alliance to investors, as shown in equation (40): Where: L i ∈[0,1] is the risk preference coefficient when investor i does not participate in the virtual power plant alliance; k j is a state variable, which takes the value 1 when investor j participates in the virtual power plant alliance, otherwise it takes the value 0; is the profit of investor j when he does not participate in the virtual power plant alliance; The risk preference coefficient difference is normalized as shown in equation (41): Marginal contribution: marginal contribution of investor i in virtual power plant alliance U γ i (U), as shown in equation (42) γ i (U)=I(U)-I(U\i) (42) The marginal contribution is normalized as shown in equation (43) (3) Profit distribution correction model J = {α, β, γ} is defined as a set of three influencing factors including participation, risk preference and marginal contribution. According to equations (39), (41) and (43), the correction matrix A is established as shown in equation (44):

3. The virtual power plant capacity optimization configuration method based on multi-agent game according to claim 1 is characterized by: The step 1 virtual power plant regulation capacity analysis includes regulation capacity analysis of wind turbines, photovoltaics, diesel generators and energy storage units. ① Diesel generator set adjustment capability: Diesel generator set initially operates at (0,P DG ,n), when t=τ1, the accumulated increase in power of the diesel generator is When t = τ2, the cumulative reduction in power of the diesel generator is As shown in equation (3) Where: P DG,a and P DG,b are the power of the diesel generator set at τ1 and τ2 respectively; and are the upward climbing rate and downward climbing rate of the diesel generator set; P DG,max and P DG,min They are the upper and lower limits of the diesel generator set output power. When t = τ3, the diesel generator set power has reached the lower limit P DG,c =P DG,min , shutdown can further reduce the power output of the diesel generator set. When t = τ4, the cumulative power reduction is shown in equation (4): ② Adjustment capability of energy storage: In the time period 0-τ1, the energy storage increases and decreases the discharge amount, as shown in equation (5) Where: P ESD,max and P ESD,min are the maximum and minimum discharge powers of the energy storage, respectively; E ES is the remaining amount of energy stored; P ESD is the discharge power of the energy storage; E ES,max is the rated capacity of energy storage; C DOD is the maximum discharge depth of the stored energy; The energy storage is in a charging state during the time period of τ1 to τ2. The energy storage charging capacity is adjusted by increasing or decreasing the charging power. During the time period of τ1 to τ2, the energy storage increases or decreases the charging capacity, as shown in equation (6): Where: P ESC,max and P ESC,min are the maximum and minimum charging power of energy storage, P ESC is the charging power of the energy storage; ③Regulation capability of wind turbines and photovoltaics: By introducing the random variable x, in a certain period of time, the wind turbine and photovoltaic are respectively WT , π PV The probability output power P WT (x), P PV (x).

4. The virtual power plant capacity optimization configuration method based on multi-agent game according to claim 1 is characterized by: P in equation (9) of step 2 WT,i (t) and P PV,i (t) is random, and the scenario set method is used to transform the random optimization problem into a deterministic optimization problem for processing. The scenario set of wind turbine output power, photovoltaic output power and electricity market price is shown in equation (10): Where: n WT 、n PV 、n EP are the total number of scenario sets for wind turbines, photovoltaics and electricity prices, and the total number of scenarios is n WT ×n PV ×n EP The probability of each scenario for wind turbines, photovoltaics and electricity prices is π(WT i )、π(PV j )、π(EP k ), where the regulation capability of the virtual power plant can compensate for its own randomness, and the up- and down-regulation capabilities of the virtual power plant are shown in equation (11) Where: and They are the adjustable up and down power of VPP respectively; and They are the adjustable power and the adjustable power of the i-th diesel generator set respectively; and They are the adjustable upward and downward capacities when the i-th group of energy storage is discharged; and They are the adjustable amount of power that can be increased and decreased when charging the i-th group of energy storage.

5. The virtual power plant capacity optimization configuration method based on multi-agent game according to claim 1 is characterized by: In the formula (12) in step 2, Expected loss distribution function of decision options As shown in equation (13)

Images