A Dynamic Pricing Optimization Method for Virtual Power Plants Based on an Improved Kriging Algorithm

Abstract

Based on the improved Kriging algorithm, the dynamic pricing optimization method of virtual power plants is used to initialize system parameters and input raw data; generate an initial data sample set; bring each group of transaction electricity prices into the upper-level DNO model and the energy management model of the lower-level VPP, and calculate the transaction electricity of each VPP corresponding to each group of transaction electricity prices. The initial sample data set is brought into the upper-level DNO solution model, and the upper-level objective function value corresponding to each group of transaction electricity prices is calculated. The key area is determined according to the upper-level objective function value. In each key area, the genetic algorithm is used to solve the local optimal electricity price. The lower-level VPP energy management model is called to calculate the real transaction electricity, and it is updated to the current sampling point set as an excellent sampling point, so as to calculate the real value in the upper-level DNO model, and the global optimal solution is obtained through repeated iteration and comparison. This method can not only protect the privacy of the lower-level virtual power plant, but also simplify the calculation steps and improve the solution efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of dynamic pricing, and in particular to a virtual power plant dynamic pricing optimization method based on an improved Kriging algorithm. Background Art

[0002] With the rapid development of renewable energy generation, a single power system transaction cannot meet the increasingly complex energy needs. With the opening of the market environment, the reasonable grid connection of renewable energy can reduce operating costs, improve energy efficiency and system power supply reliability. In recent years, virtual power plants (VPPs) including wind, solar, fuel and storage have gradually become new entities in the power market. The traditional power grid has been transformed into an intelligent and decentralized intelligent system, playing an important role in achieving low-carbon economic development.

[0003] Dynamic pricing plays an important role in modern intelligent systems and is also one of the emerging research areas in the retail electricity industry. It can optimize the load on the user side, reduce the pressure on the power supply side, shift demand from peak to off-peak, and effectively solve problems such as mobility management. Relevant research has been conducted on dynamic pricing at home and abroad, and it has been found that in addition to reducing peak demand, dynamic pricing plans also change consumption patterns by changing consumer load demand.

[0004] At present, the main methods for solving the equilibrium solution of the master-slave game are numerical optimization methods based on Karush-Kuhn-Tucker (KKT) conditions and heuristic intelligent algorithms. The KKT condition method can simplify the solution model, but it has high requirements for the lower-level model, and in the solution process, the upper-level model needs all the parameters of the lower-level model, which involves privacy issues; while the upper and lower-level models only need to exchange a small amount of information in the heuristic intelligent algorithm solution, which can protect the privacy of the lower layer, but it requires a large number of calls to the lower-level sub-models, in which there are too many variables and constraints, and the calculation steps are large and complicated. Summary of the invention

[0005] In order to solve the above technical problems, the present invention provides a virtual power plant dynamic pricing optimization method based on an improved Kriging algorithm. For the equilibrium constraint optimization problem, this method can not only protect the privacy of the underlying virtual power plant, but also simplify the calculation steps and improve the solution efficiency.

[0006] The technical solution adopted by the present invention is:

[0007] The virtual power plant dynamic pricing optimization method based on the improved Kriging algorithm includes the following steps:

[0008] Step 1: Initialize system parameters and input original data;

[0009] Step 2: Generate an initial data sample set based on Latin hypercube sampling. The number of sampling points is 10 times the number of variables. Each sample point represents a set of transaction electricity prices, i.e., purchase and sale electricity prices, recorded as They represent the electricity prices in T time periods in the data sample set, that is, the transaction electricity prices sold by the distribution network operator to the VPP. represents the electricity price at 1:00 in the data sample set, represents the electricity price at 2:00 in the data sample set, It represents the electricity price at T:00 in the data sample set. In the present invention, T is 24 times a day. The electricity price here refers to the transaction price at which the distribution network operator sells electricity to the VPP.

[0010] Respectively represent the electricity purchase price of T time periods in the data sample set, that is, the transaction price of electricity purchased by the distribution network operator from the VPP. represents the electricity purchase price at 1:00 in the data sample set, represents the electricity purchase price at 2:00 in the data sample set, It represents the electricity purchase price at T:00 in the data sample set. In the present invention, T is 24 times a day. The electricity purchase price here refers to the transaction price at which the distribution network operator purchases electricity from the VPP.

[0011] Transaction electricity price Satisfy constraints

[0012] represents the on-grid electricity price in the power market at time t, that is, the price at which the distribution network operator purchases electricity from the power market;

[0013] represents the grid electricity price in the power market at time t, that is, the price at which the distribution network operator sells electricity to the power market;

[0014] It represents the electricity purchase price for each time period set by the distribution network operator for the virtual power plant;

[0015] It represents the electricity selling price for each time period set by the distribution network operator for the virtual power plant.

[0016] Step 3: Bring each group of transaction electricity prices into the upper-level distribution network operator (DNO) model and the lower-level VPP energy management model to calculate the transaction electricity of each VPP corresponding to each group of transaction electricity prices.

[0017] Step 4: Initial sample data set B jSubstitute the upper DNO solution model maxC in step 3 DNO Then calculate the upper objective function value C corresponding to each group of transaction electricity prices DNO , according to the upper objective function value C DNO Determine the key areas and divide the optimization strategy space of the upper DNO model into N l key areas, and the optimal value of each key area is In this way, the area with a low probability of optimal value is excluded, the search range of excellent sampling points is narrowed, and the solution efficiency is improved. In each key area, the genetic algorithm is used to solve the local optimal electricity price.

[0018] Step 5: Call the lower-level VPP energy management model in step 3 Calculate the real transaction electricity and update it to the current sampling point set as an excellent sampling point, so as to calculate the real value in the upper DNO model and obtain the global optimal solution through repeated iteration and comparison;

[0019] Step 6: Determine whether the output global optimal solution meets the convergence conditions. If so, stop the iteration and output the Nash equilibrium solution. If not, return to step 2.

[0020] In step 1, the original data includes the on-grid electricity price of the power market and grid electricity prices Parameters related to multiple sources (wind, solar, fuel, storage and load) in virtual power plants, parameters related to the Kriging metamodel, and parameters related to the genetic algorithm.

[0021] The multi-source (wind, solar, gas, storage and load) related parameters specifically include: wind power, photovoltaic, and load forecast data, upper and lower limits of output power of wind turbines and photovoltaic generators, maximum output power of gas turbines, lower limit of downward climbing rate and upper limit of upward climbing rate, upper and lower limits of energy storage charging and discharging power, upper and lower limits of energy storage charge state, and the maximum interruption amount of interruptible load on the demand side.

[0022] The relevant parameters of the Kriging metamodel specifically include: the initial values ​​of theta, lob and upb in the dacefit function.

[0023] The genetic algorithm related parameters specifically include: population size, crossover rate, mutation rate and number of iterations.

[0024] In step 2, the transaction electricity price includes the electricity selling price and the electricity purchasing price, and the number of initial sample points is generally 9-10 times the number of variables.

[0025] In step 3, the upper-level DNO sets the electricity purchase price and electricity sales price for each time period with the VPP to maximize economic benefits, with electricity price constraints and electricity purchase and sales constraints as constraints;

[0026] The objective function expression of the upper DNO model is as follows:

[0027]

[0028] Where: represents the on-grid electricity price in the power market at time t, represents the grid electricity price in the power market at time t; It indicates the electricity purchase price for each period set by DNO for VPP. P represents the electricity price for each period set by DNO for VPP; t DNO,s represents the amount of electricity sold by DNO to the electricity market, P t DNO,b It means the amount of electricity purchased by DNO from the electricity market; represents the electricity sold by the i-th VPP to the DNO, represents the amount of electricity purchased by the ith VPP from the DNO; Num represents the total number of VPPs.

[0029] The constraints of the upper DNO model are specifically expressed as follows:

[0030] (1) Electricity price constraints:

[0031]

[0032] Where: represents the on-grid electricity price in the power market at time t, that is, the price at which the distribution network operator purchases electricity from the power market; represents the grid electricity price in the power market at time t, that is, the price at which the distribution network operator sells electricity to the power market; It represents the electricity purchase price for each time period set by the distribution network operator for the virtual power plant. It represents the electricity selling price for each time period set by the distribution network operator for the virtual power plant.

[0033] (2) Constraints on electricity purchase and sales:

[0034]

[0035] Where: P t DNO After each VPP reports the amount of electricity purchased and sold, the DNO aggregates the total amount of electricity traded with the power market. A value greater than 0 indicates that the DNO has purchased electricity from the power market, and a value less than 0 indicates that the DNO has sold electricity to the power market.

[0036] represents the amount of electricity purchased by the i-th VPP from the DNO at time period t, represents the electricity sold by the i-th VPP to the DNO in time period t;

[0037] P t DNO,b P represents the amount of electricity purchased by DNO from the electricity market during period t, t DNO,s It represents the amount of electricity sold by DNO to the electricity market in period t.

[0038] In step 3, the lower-level VPP takes into account multiple costs, including the cost of purchasing electricity, with the goal of minimizing operating costs. Gas Turbine Cost Operating costs of energy storage power stations Wind and solar power generator operating costs and the interruptible load costs on the demand side The constraints are electricity consumption constraint, power balance constraint and internal output constraint.

[0039] The objective function expression of the lower-level VPP energy management model is as follows:

[0040]

[0041] in:

[0042]

[0043] Where: M 1,n represents the quadratic term coefficient of the gas turbine cost model; M 2,n represents the first-order coefficient of the gas turbine cost model; M 3,n represents the constant term coefficient of the gas turbine cost model;

[0044] T represents the scheduling period, which is 24 hours; t = 1: T represents the time period;

[0045] It represents the electricity purchase price for each time period set by the distribution network operator for the virtual power plant. It represents the electricity selling price for each time period set by the distribution network operator for the virtual power plant.

[0046] represents the output power of the gas turbine; Indicates the charging and discharging power of the energy storage power station; represents the dispatch cost coefficient of energy storage; Indicates the charging and discharging power of energy storage; a n represents the wind turbine operating cost coefficient; is the output power of the wind turbine; b n Represents the operating cost coefficient of the photovoltaic power generation unit; Indicates the output power of the photovoltaic generator set; represents the demand side interruption compensation electricity price; It represents the interruption power of the interruptible load on the demand side;

[0047] The constraints of the lower-level VPP energy management model are specifically expressed as follows:

[0048] ①Power Constraint:

[0049]

[0050] Where: represents the transaction volume between the ith VPP and the operator; represents the amount of electricity purchased by the i-th VPP from the DNO in time period t; represents the electricity sold by the i-th VPP to the DNO in time period t; represents the maximum amount of electricity traded between the ith VPP and the DNO;

[0051] θ i,t It represents a Boolean variable. When the value is 1, it means that the i-th VPP sells electricity to the DNO at time t. When the value is 0, it means that the i-th VPP buys electricity from the DNO at time t.

[0052] ②Power balance constraints:

[0053]

[0054] Where: represents the predicted value of the load at time t; N represents the set of distributed power sources contained in the i-th VPP; j∈N, represents the j-th distributed power source.

[0055] ③Internal output constraints:

[0056]

[0057] Where: Indicates the maximum output power of the gas turbine;

[0058] represents the gas turbine downward ramp rate; represents the gas turbine upward ramp rate;

[0059] represents the output power of the gas turbine at time t-1;

[0060] Indicates the upper limit of energy storage charging and discharging power; Indicates the lower limit of energy storage charging and discharging power;

[0061] Indicates the state of charge of the energy storage at time t; Indicates the state of charge of the energy storage at time t-1;

[0062] △t represents the time difference; E n,max Indicates the maximum energy capacity value of energy storage;

[0063] Indicates the upper limit of the energy storage charge state; Indicates the lower limit of the energy storage charge state;

[0064] In the above formula, the relevant constraints do not take into account the system network loss;

[0065] represents the maximum output power of the wind turbine at time t; Indicates the maximum output power of the photovoltaic generator set at time t; is the maximum interruption amount of the interruptible load on the incentive demand side; Represents the interruption power of the incentive-type demand-side interruptible load.

[0066] In step 3, the transaction power of each virtual power plant corresponding to each group of transaction power prices is determined according to the following method:

[0067] The Kriging metamodel is selected to simulate the lower-level VPP energy management model. The Kriging metamodel is combined with the genetic algorithm to optimize and solve the energy management model of each VPP in the MATLAB environment through the Yalmip toolbox and CPLEX toolkit. The solution flow chart is shown in the following figure. Figure 4 shown.

[0068] Transaction electricity price Trading electricity with virtual power plants Construct the initial sample dataset: represents the initial data sample set; Represents any distributed power source in the virtual power plant; n represents the number of distributed power sources included in the VPP.

[0069] i=1,...,N, where N indicates that there are N virtual power plants in the system block diagram.

[0070] In step 4, according to the upper layer objective function value C DNO Determine the key area. The key area here refers to: according to the division steps, eliminate the area with low probability of containing the optimal solution, narrow the sampling range of excellent sampling points, and improve the optimization efficiency.

[0071] In step 4, the l The key areas include the following steps:

[0072] Step a: Let l = 1, where l represents the number of the partitioned area. The smaller the number, the greater the probability that the area contains the optimal solution.

[0073] make F represents all electricity price sample points and the corresponding upper-level objective function value set.

[0074] in, Indicates all sampling points of the current electricity price, It represents the optimal value of the objective function of the corresponding DNO, and puts all the electricity price sample point sets into the set X.

[0075] Step b: Take the maximum objective function value y in F l , the expression is as follows:

[0076]

[0077] in, is the electricity price sample point corresponding to the maximum objective function value in area l.

[0078] Step c: Put in F l middle, Put in X l middle.

[0079] F l represents all the electricity price sample points in the divided area l and the corresponding upper layer objective function value set; X l Represents the set of all current electricity price sample points in area l.

[0080] by As the center, with r as the radius, the set X The electricity price sample points with a distance less than or equal to r are put into the set X l , and the corresponding Put in F l In which, the expression of r is as follows:

[0081]

[0082] in, Indicates the highest transaction electricity price; represents the minimum transaction electricity price; k represents the number of iterations of the current game model; k max Indicates the maximum number of iterations.

[0083] Step d: Let F = F\F l , X=X\X l ,if Then let l=l+1 and return to step b.

[0084] in: Indicates that the current set of all electricity price sample points is an empty set.

[0085] In step 4, a genetic algorithm is used to solve the local optimal electricity price. The following steps are involved:

[0086] Step ①: For the divided key areas, the final optimal benefit value of the Kriging model in each area is set as a zero matrix with one row and one column in the MATLAB program, and the corresponding optimal solution is set as a zero matrix with one row and forty-eight columns in the MATLAB program.

[0087] Step ②: Set the population size, crossover rate, mutation rate and maximum number of iterations.

[0088] Step ③: Calculate the fitness of each individual in the group, and then solve the game model G.

[0089] G = {(γ D,s ,γ D,b ),{P i VPP},{minf i VPP}{maxC DNO}} (11);

[0090] Step ④: Apply the crossover and mutation operators to the population to generate a population representing a new solution set. Calculate the individual fitness of the new population.

[0091] Step ⑤: Determine whether the new population obtained in step ④ meets the convergence conditions and outputs the local optimal electricity price, otherwise return to step ④.

[0092] In step 5, the actual value of the objective function of the upper DNO model is calculated. and For comparison, if Then By comparing the optimal solutions of all regions, the maximum value is taken as the global optimal solution of the current iteration.

[0093]

[0094] Where: l represents the area number; N l Indicates the total number of partitioned areas.

[0095] The present invention provides a virtual power plant dynamic pricing optimization method based on an improved Kriging algorithm, and the technical effects are as follows:

[0096] 1) The present invention aims at the complexity of the interests of distribution network operators and multiple VPPs and the flexibility of the internal entities of the energy system, constructs a two-level game optimization model of DNO and multiple VPPs, and proposes a solution by combining the Kriging metamodel with the genetic optimization algorithm. Compared with the existing line technology, the present invention can be used to solve the model of different stakeholders in the energy system. In the solution process, it has an effective correction mechanism to generate an excellent sampling point set, and in the iterative process, new sampling points are added to continuously correct the model, which can not only simplify the calculation steps, but also improve the solution efficiency.

[0097] 2) The present invention improves an optimization algorithm based on the Kriging metamodel by establishing a two-layer game optimization model of DNO and multiple VPPs. In the process of solving the master-slave game model, the improved algorithm uses the Kriging model to fit and replace the energy management model of the lower virtual power plant (Virtual Power Plants, VPP). In the iterative process, the genetic algorithm is used to perform roulette selection and cross-mutation on the upper layer (Distribution Network Operator, DNO) model to obtain the local optimal sampling point. The energy management model of the lower layer VPP is called according to the optimization result, and the actual transaction power situation and the value of the upper layer model are calculated. The Nash equilibrium solution is obtained through multiple iterations. Compared with the existing methods, this method can not only protect the privacy of the lower virtual power plant for the equilibrium constraint optimization problem, but also simplify the calculation steps and improve the solution efficiency.

[0098] 3) The present invention divides the optimization strategy space of the upper DNO model into multiple regions, avoiding regions with low probability of the optimal solution appearing, narrowing the search range, and improving calculation efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0099] Figure 1 This is the system operation structure diagram.

[0100] Figure 2 This is the system transaction block diagram.

[0101] Figure 3 Schematic diagram of the optimization method of the present invention.

[0102] Figure 4 This is a flow chart of the strategy for solving the hierarchical scheduling model of the present invention.

[0103] Figure 5 Topology diagram of the master-slave game system.

[0104] Figure 6 This is a curve chart of electricity price optimization strategy results.

[0105] Figure 7 This is a comparison chart of the sum of electricity purchased and sold by DNO and VPP.

[0106] Figure 8 This is a simulation comparison chart of the convergence process of each algorithm. DETAILED DESCRIPTION

[0107] The Kriging metamodel is a simplified mathematical model that can be used to approximate and replace the simulation model. It has good approximation ability and unique error estimation function for nonlinear models. Therefore, the algorithm of the present invention selects the Kriging metamodel to simulate the underlying energy management model and simply displays the relationship between the transaction price and the transaction volume.

[0108] A virtual power plant dynamic pricing optimization method based on the improved Kriging algorithm first establishes an initial Kriging meta-model based on an appropriate sample data set. The specific steps are as follows:

[0109] Step 1: Initialize system parameters, input raw data, and then randomly generate the initial population:

[0110] Related parameters of multi-source (wind, solar, fuel, storage and load) in virtual power plants and grid-connected electricity prices in the power market and grid electricity prices Relevant parameters of the Kriging metamodel and relevant parameters of the genetic algorithm.

[0111] Step 2: Generate initial sample data set:

[0112] To ensure the uniformity and rationality of sampling, Latin hypercube sampling is used to generate n initial sample points;

[0113] Each sample point coordinate represents a set of transaction electricity prices, including the electricity sales price and the electricity purchase price, recorded as

[0114] in: Satisfy constraints The number of initial sampling points is generally 9-10 times the number of variables.

[0115] By bringing each group of electricity prices into the upper-layer DNO model and the lower-layer VPP energy management model, the transaction power of each VPP corresponding to each group of transaction electricity prices can be obtained through calculation. Trading electricity price and transaction volume Constructing the initial sample data set Then we start to build the Kriging initial model of multiple VPPs.

[0116] Step 3: Construction and modification of Kriging model:

[0117] The current sample data set B jBring it into the solution model and calculate the upper objective function value C corresponding to each group of transaction electricity prices DNO , according to the objective function value C DNO Determine the key areas and divide the DNO optimization strategy space into N l key areas, and the optimal value for each area is In this way, the area with a low probability of optimal value is excluded, the search range of excellent sampling points is narrowed, and the solution efficiency is improved. In each key area, the genetic algorithm is used to solve the local optimal electricity price.

[0118] (1) Divide key areas:

[0119] a: Let l = 1, where l represents the number of the partitioned area. The smaller the number, the greater the probability that the area contains the optimal solution.

[0120] make

[0121] in: They represent the optimal values ​​of the objective function of all sampling points of the current electricity price and their corresponding DNO, and put all the electricity price sample point sets into the set X.

[0122] b: Take the maximum objective function value y in F l , the expression is as follows:

[0123]

[0124] in, is the electricity price sample point corresponding to the maximum objective function value in area l.

[0125] c: Put in F l middle, Put in X l In. As the center, with r as the radius, the set X The electricity price sample points with a distance less than or equal to r are put into the set X l , and the corresponding Put in F l In which r is expressed as follows:

[0126]

[0127] in, Indicates the highest transaction electricity price; represents the minimum transaction electricity price; k represents the number of iterations of the current game model; k max Indicates the maximum number of iterations.

[0128] d: Let F = F\F l, X=X\X l ,if Then let l=l+1 and return to b.

[0129] (2) Generate excellent sampling points using genetic algorithm:

[0130] In each key area of ​​the division, a genetic algorithm is used to optimize the game model G, i.e., formula (11).

[0131] G = {(γ D,s ,γ D,b ),{P i VPP},{minf i VPP}{maxC DNO}} (11)

[0132] First, the genetic algorithm is used to calculate the fitness of each individual in the population, and the roulette wheel selection method is used to randomly select a certain number of excellent individuals to generate a direct genetic sub-population; a certain number of individuals are randomly selected to form an evolutionary sub-population. Then, the adaptive crossover operation and mutation operation are applied to the evolutionary sub-population, and the evolutionary sub-population is merged with the direct genetic sub-population to obtain the next generation population. Finally, if the shutdown condition is met, the operation is terminated and the current optimal solution is output, otherwise the individual fitness is recalculated.

[0133] After the optimal solution is output, the Kriging metamodel is used to quickly lock in the local optimal transaction price and transaction volume. However, the transaction volume at this time is not the actual value, but the sample point reflects the changing trend of the objective function to a certain extent. Then the original lower-level model is optimized to calculate the actual optimal transaction volume, and added to the original sample data set to correct the Kriging metamodel.

[0134] The Kriging metamodel is an initial Kriging model established in MATLAB based on an appropriate amount of sample data sets by setting parameters theta, lob and upb. It fits and replaces the underlying VPP energy management model in the MATLAB solution process, effectively simplifying the calculation steps and improving the solution efficiency.

[0135] Step 4: Compare the optimal solutions of all regions to obtain the global optimal solution for the current number of iterations.

[0136] Calculate the actual value of the objective function of the upper DNO model and For comparison, if Then By comparing the optimal solutions of all regions, the maximum value is taken as the global optimal solution of the current iteration.

[0137]

[0138] Step 5: Determine whether the output global optimal solution meets the convergence conditions. If so, stop the iteration and output the optimal electricity price strategy set of the upper-level game subject, that is, the Nash equilibrium solution of the master-slave game system. If not, return to step 2.

[0139] (1) Proof of the existence of Nash equilibrium solution:

[0140] The master-slave game model of the present invention is a special form of non-cooperative game, namely, generalized Nash strategy or generalized Nash equilibrium. The generalized Nash equilibrium problem is expressed as the decision-making of all layers is coupled through shared constraints, and the interests of each other influence and restrain each other. In formula (11), let γ D =(γ D,s ,γ D,b ), according to the Stackelberg Nash equilibrium definition, if there exists a strategy (γ Do ,p o ), and satisfy C DNO (γ Do ,p o )≥C DNO (γ D ,p o ),and Where p -i represents the strategies of the remaining VPPs except the i-th VPP, then the strategy (γ D° ,p o ) is the Nash equilibrium solution of the model of the present invention.

[0141] According to the existence theorem of Nash equilibrium point in the master-slave game model, as recorded in the literature: Liu Lianguang, Pan Mingming, Tian Shiming, et al. Non-cooperative game method for power sales competition considering multiple entities of source, grid and load [J]. Proceedings of the CSEE, 2017, 37(06): 1618-1626.

[0142] The proof is mainly divided into three aspects: in the game process, the strategy set of the leader and the follower is a non-empty compact convex subset, and each utility function is a continuous quasi-convex function; when the upper leader gives an optimization strategy, the game Nash equilibrium solution of the lower followers exists; when the lower followers give an optimization strategy, the optimal solution of the optimization model of the upper leader exists.

[0143] In the non-cooperative game model of the present invention, the optimization strategies of the leader and the participants are determined by linear constraints and are non-empty compact convex subsets; the optimization model C of the upper leader DNO DNO is a linear function, which is obviously a quasi-convex function; in the optimization model of the lower follower VPP and is a linear function, which is obviously a quasi-convex function, and and are nonlinear functions, and their quadratic partial derivatives are respectively obtained to be greater than zero, so the optimization model of the follower is also a quasi-convex function. Therefore, the master-slave game model established by the present invention has a Nash equilibrium solution.

[0144] In order to verify the effectiveness of the dynamic electricity price strategy and energy management between DNO and multiple VPPs, the following two strategies are used to explore the system optimization operation effect:

[0145] Strategy 1 is that DNO does not adopt electricity price optimization, and the electricity purchase and sale price in the power market is the transaction price between DNO and VPP; Strategy 2 is that DNO adopts the electricity price optimization strategy proposed in the present invention.

[0146] Under the electricity price optimization strategy proposed in the present invention, the electricity purchase and sale prices set by DNO are as follows: Figure 6 As shown, Figure 6 The Nash equilibrium solution is shown when the game reaches the point where the leader's benefit is maximized and the follower's cost is minimized. Figure 7 In the figure, VPP has three states: transaction power is greater than 0, equal to 0, and less than 0, which respectively represent VPP purchasing power from DNO, VPP internal supply and demand balance, and VPP selling power to DNO.

[0147] according to Figure 8 The simulation results show that the Nash equilibrium solution of the game model is solved by the strategy of the present invention, and the solution efficiency and convergence speed are fast, and the number of calls to the lower model is much less than that of the particle swarm optimization algorithm and the genetic algorithm. This is mainly because the proposed algorithm reasonably samples and fits the energy management model of the lower VPP to generate the Kriging metamodel. In the iterative process, the Kriging metamodel and the genetic algorithm are combined to quickly lock the local optimal solution, and then the energy management model of the lower VPP is called to calculate the real transaction value. The point set is added to modify the Kriging metamodel, which avoids blind sampling points and improves the calculation speed and accuracy.

Claims

1. A virtual power plant dynamic pricing optimization method based on an improved Kriging algorithm, characterized by The following steps are involved: Step 1: Initialize system parameters and input original data; Step 2: Generate an initial data sample set based on Latin hypercube sampling. Each sample point represents a set of transaction electricity prices, denoted as They represent the electricity prices in T time periods in the data sample set respectively; They represent the electricity purchase prices in T time periods in the data sample set; Step 3: Substitute each group of transaction electricity prices into the upper-layer DNO model and the lower-layer VPP energy management model to calculate the transaction electricity volume of each VPP corresponding to each group of transaction electricity prices. Step 4: Bring the initial sample data set into the upper-level DNO solution model in step 3, and then calculate the upper-level objective function value C corresponding to each group of transaction electricity prices DNO , according to the upper objective function value C DNO Determine the key areas and divide the optimization strategy space of the upper DNO model into N l key areas, and the optimal value of each key area is In each key area, a genetic algorithm is used to solve the local optimal electricity price. Step 5: Call the lower-level VPP energy management model in step 3 to calculate the actual transaction power, and update it to the current sampling point set as an excellent sampling point, so as to calculate the real value in the upper-level DNO model, and obtain the global optimal solution through repeated iteration and comparison; Step 6: Determine whether the output global optimal solution meets the convergence conditions. If so, stop the iteration and output the Nash equilibrium solution. If not, return to step 2.

2. The virtual power plant dynamic pricing optimization method based on the improved Kriging algorithm according to claim 1 is characterized by: In step 1, the original data includes the on-grid electricity price of the power market and grid electricity prices Multi-source related parameters in virtual power plants, Kriging metamodel related parameters, and genetic algorithm related parameters; The multi-source related parameters specifically include: wind power, photovoltaic, and load forecast data, upper and lower limits of wind turbine and photovoltaic generator output power, maximum output power of gas turbines, lower limit of ramp rate and upper limit of ramp rate, upper and lower limits of energy storage charging and discharging power, upper and lower limits of energy storage charge state, and maximum interruption amount of interruptible load on the demand side; The parameters related to the Kriging metamodel specifically include: the initial values ​​of theta, lob, and upb in the dacefit function; The genetic algorithm related parameters specifically include: population size, crossover rate, mutation rate and number of iterations.

3. The virtual power plant dynamic pricing optimization method based on the improved Kriging algorithm according to claim 1 is characterized by: In step 2, the transaction electricity price includes the electricity selling price and the electricity purchasing price, and the number of initial sample points is generally 9-10 times the number of variables.

4. The virtual power plant dynamic pricing optimization method based on the improved Kriging algorithm according to claim 1 is characterized by: In step 3, The upper-level DNO sets the electricity purchase and sales prices for each period with the VPP to maximize economic benefits, with electricity price constraints and electricity purchase and sales constraints as constraints; The objective function expression of the upper DNO model is as follows: Where: represents the on-grid electricity price in the power market at time t, represents the grid electricity price in the power market at time t; It indicates the electricity purchase price for each period set by DNO for VPP. P represents the electricity price for each period set by DNO for VPP; t DNO,s represents the amount of electricity sold by DNO to the electricity market, P t DNO,b It means the amount of electricity purchased by DNO from the electricity market; represents the electricity sold by the i-th VPP to the DNO, represents the amount of electricity purchased by the ith VPP from the DNO; Num represents the total number of VPPs; The constraints of the upper DNO model are specifically expressed as follows: (1) Electricity price constraints: Where: represents the on-grid electricity price in the power market at time t; represents the grid electricity price in the power market at time t; It represents the electricity purchase price for each time period set by the distribution network operator for the virtual power plant. It represents the electricity selling price for each time period set by the distribution network operator for the virtual power plant; (2) Constraints on electricity purchase and sales: Where: P t DNO After each VPP reports the amount of electricity purchased and sold, the DNO aggregates the total amount of electricity traded with the power market. A value greater than 0 indicates that the DNO has purchased electricity from the power market, and a value less than 0 indicates that the DNO has sold electricity to the power market. represents the amount of electricity purchased by the i-th VPP from the DNO at time period t, represents the electricity sold by the i-th VPP to the DNO in time period t; P t DNO,b P represents the amount of electricity purchased by DNO from the electricity market during period t, t DNO,s It represents the amount of electricity sold by DNO to the electricity market in period t.

5. The virtual power plant dynamic pricing optimization method based on the improved Kriging algorithm according to claim 4 is characterized by: In step 3, The lower-level VPP aims to minimize operating costs and takes into account multiple costs, including the cost of purchasing electricity. Gas Turbine Cost Operating costs of energy storage power stations Wind and solar power generator operating costs and the interruptible load costs on the demand side The constraints are electricity consumption constraint, power balance constraint and internal output constraint; The objective function expression of the lower-level VPP energy management model is as follows: in: Where: M 1,n represents the quadratic term coefficient of the gas turbine cost model; M 2,n represents the first-order coefficient of the gas turbine cost model; M 3,n represents the constant term coefficient of the gas turbine cost model; T represents the scheduling period; t represents the time period; It represents the electricity purchase price for each time period set by the distribution network operator for the virtual power plant. It represents the electricity selling price for each time period set by the distribution network operator for the virtual power plant; represents the output power of the gas turbine; Indicates the charging and discharging power of the energy storage power station; represents the dispatch cost coefficient of energy storage; Indicates the charging and discharging power of energy storage; a n represents the wind turbine operating cost coefficient; is the output power of the wind turbine; b n Represents the operating cost coefficient of the photovoltaic power generation unit; Indicates the output power of the photovoltaic generator set; represents the demand-side interruption compensation price; It represents the interruption power of the interruptible load on the demand side; The constraints of the lower-level VPP energy management model are specifically expressed as follows: ①Power Constraint: Where: represents the transaction volume between the ith VPP and the operator; represents the amount of electricity purchased by the i-th VPP from the DNO in time period t; represents the electricity sold by the i-th VPP to the DNO in time period t; represents the maximum amount of electricity traded between the ith VPP and the DNO; θ i,t It represents a Boolean variable. When the value is 1, it means that the i-th VPP sells electricity to the DNO at time t. When the value is 0, it means that the i-th VPP buys electricity from the DNO at time t. ②Power balance constraints: Where: represents the predicted value of the load at time t; N represents the set of distributed power sources contained in the i-th VPP; j∈N, represents the j-th distributed power source; ③Internal output constraints: Where: Indicates the maximum output power of the gas turbine; represents the gas turbine downward ramp rate; represents the gas turbine upward ramp rate; represents the output power of the gas turbine at time t-1; Indicates the upper limit of energy storage charging and discharging power; Indicates the lower limit of energy storage charging and discharging power; Indicates the state of charge of the energy storage at time t; Indicates the state of charge of the energy storage at time t-1; △t represents the time difference; E n,max Indicates the maximum energy capacity value of energy storage; Indicates the upper limit of the energy storage charge state; Indicates the lower limit of the energy storage charge state; represents the maximum output power of the wind turbine at time t; Indicates the maximum output power of the photovoltaic generator set at time t; is the maximum interruption amount of the interruptible load on the incentive demand side; Represents the interruption power of the incentive-type demand-side interruptible load.

6. The virtual power plant dynamic pricing optimization method based on the improved Kriging algorithm according to claim 5 is characterized by: In step 3, the transaction price Trading electricity with virtual power plants Construct the initial sample dataset: represents the initial data sample set; represents any distributed power source in the virtual power plant; n represents the number of distributed power sources included in the VPP; i=1,...,N, where N indicates that there are N virtual power plants in the system block diagram.

7. The virtual power plant dynamic pricing optimization method based on the improved Kriging algorithm according to claim 1 is characterized by: In step 4, according to the upper layer objective function value C DNO Determine the key area, where the key area refers to: according to the division steps, eliminate the area with a low probability of containing the optimal solution.

8. The virtual power plant dynamic pricing optimization method based on the improved Kriging algorithm according to claim 1 is characterized by: In step 4, the l The key areas include the following steps: Step a: Let l = 1, where l represents the number of the partitioned area. The smaller the number, the greater the probability that the area contains the optimal solution. make F represents all electricity price sample points and the corresponding upper-level objective function value set; in, Indicates all sampling points of the current electricity price, represents the optimal value of the objective function of the corresponding DNO, and puts all the electricity price sample point sets into the set X; Step b: Take the maximum objective function value y in F l , the expression is as follows: in, is the electricity price sample point corresponding to the maximum objective function value in area l; Step c: Put in F l middle, Put in X l middle; F l represents all the electricity price sample points in the divided area l and the corresponding upper layer objective function value set; X l represents the set of all current electricity price sample points in area l; by As the center, with r as the radius, the set X The electricity price sample points with a distance less than or equal to r are put into the set X l , and the corresponding Put in F l In which, the expression of r is as follows: in, Indicates the highest transaction electricity price; represents the minimum transaction electricity price; k represents the number of iterations of the current game model; k max Indicates the maximum number of iterations; Step d: Let F = F\F l , X=X\X l ,if Then let l=l+1 and return to step b; in: Indicates that the current set of all electricity price sample points is an empty set.

9. The virtual power plant dynamic pricing optimization method based on the improved Kriging algorithm according to claim 1 is characterized by: In step 4, a genetic algorithm is used to solve the local optimal electricity price. The following steps are involved: Step ①: For the divided key areas, the final optimal benefit value of the Kriging model of each area is set as a zero matrix with one row and one column in the MATLAB program, and the corresponding optimal solution is set as a zero matrix with one row and forty-eight columns in the MATLAB program; Step ②: Set the population size, crossover rate, mutation rate and maximum number of iterations; Step ③: Calculate the fitness of each individual in the group, and then solve the game model G; G={(γ D,s ,c D,b ),{P i VPP },{minf i VPP }{maxC DNO }} (11); Step ④: Apply crossover and mutation operators to the population to generate a population representing a new solution set; calculate the individual fitness of the new population; Step ⑤: Determine whether the new population obtained in step ④ meets the convergence conditions and outputs the local optimal electricity price, otherwise return to step ④.

10. The virtual power plant dynamic pricing optimization method based on the improved Kriging algorithm according to claim 1 is characterized by: In step 5, the actual value of the objective function of the upper DNO model is calculated. and For comparison, if Then By comparing the optimal solutions of all regions, the maximum value is taken as the global optimal solution of the current iteration. Where: l represents the area number; N l Indicates the total number of partitioned areas.

Images