Virtual power plant robust optimization scheduling method based on source load uncertainty

Abstract

The invention discloses a virtual power plant robust optimization scheduling method based on source load uncertainty, and belongs to the technical field of energy management. The method comprises the following steps: step 1, power supply side uncertainty modeling; step 2, load uncertainty modeling is carried out; 3, carrying out two-stage random robust optimization scheduling modeling; and step 4, solving the two-stage random robust optimization scheduling model. The method has the following beneficial effects: (1) according to the two-stage random robust optimization scheduling method, the economic scheduling of each unit meeting the multi-energy load is realized; and (2) according to a two-stage random robust optimization scheduling strategy, a scheduling scheme with the minimum cost is obtained in the worst scene. And (3) economic dispatching strategies are made for various uncertain conditions by adjusting robust optimization parameters, and the environmental adaptability of the virtual power plant is improved. And (4) in the energy market balance power deviation of the virtual power plant, the economical efficiency is more remarkable.

Description

Technical Field

[0001] The present invention belongs to the technical field of energy management, and in particular relates to a robust optimization scheduling method for a virtual power plant based on source-load uncertainty. Background Art

[0002] Virtual Power Plant (VPP) is not a real power plant, but a smart grid technology. Its core concept is to bring together various decentralized and adjustable power sources and loads through advanced information and communication technology and software systems to form a virtual "power plant" for unified management and dispatch. Virtual power plant is one of the important directions of energy transformation. It can promote the development and utilization of renewable energy, promote the optimization and adjustment of energy structure, and achieve sustainable development of energy. As a distributed energy system with centralized control and operation, virtual power plant integrates, optimizes and coordinates distributed energy systems through technical means, can respond to market demand and stabilize grid operation, has the characteristics of flexibility, scalability and dispatchability, and can achieve optimal configuration and effective utilization of energy. Through virtual power plant, energy utilization can be improved, energy consumption and carbon emissions can be reduced, which is of great significance to future energy development.

[0003] Virtual power plants play an important role in all aspects of the power system. In the construction of new power systems, the proportion of renewable energy power generation has increased rapidly, but renewable energy power generation has great volatility. At the same time, the uncertainty of load demand and fluctuations in market value have brought challenges to the stable operation, safety and economic benefits of the power grid. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a scheduling strategy that can effectively optimize the scheduling of system operation in the virtual power plant control area under complex circumstances and at the lowest cost based on the impact of uncertainty of renewable energy and load on system scheduling.

[0005] In order to solve the above technical problems, the present invention provides a robust optimization scheduling method for a virtual power plant based on source-load uncertainty, and the technical solution is as follows:

[0006] Step 1: Modeling uncertainty on the power supply side

[0007] Robust optimization strategies are used to address the problem that wind power generation and photovoltaic power generation are easily affected by environmental changes, resulting in significant randomness and instability. This is done by summarizing the power generation scenarios of renewable energy into an overall set.

[0008] Step 2 Load uncertainty modeling

[0009] WGAN-GP is used to generate multi-energy load scenarios, and the K-medoids algorithm is used to obtain typical scenarios to solve the problem that load scenarios in different regions and environments are difficult to describe using the same function.

[0010] Step 3: Two-stage stochastic robust optimization scheduling modeling

[0011] A cardinality uncertainty set with robust adjustable coefficients is used to measure the fluctuation of uncertainty in renewable energy output and optimize the uncertainty on the power supply side.

[0012] Step 4: Solving the two-stage stochastic robust optimization scheduling model

[0013] Through the two-stage stochastic robust optimization scheduling model, typical scenarios are obtained, and scheduling strategies under different schemes are provided by changing the robust adjustment coefficient.

[0014] Furthermore, the step 1 comprises:

[0015] The core of solving the robust optimization problem is to select uncertainty sets, which describe wind and solar energy output according to the characteristics of uncertain parameters and reduce the uncertainty sets that excessively increase the computational complexity of the model. The cardinality uncertainty set with robust adjustable coefficients can effectively regulate the impact range of uncertainty deviation and improve the accuracy of the prediction of uncertainty fluctuation range. RE The element P in RE,t The following constraints are met:

[0016]

[0017] Where: The time set for optimal scheduling is T = {1, 2, ..., T, ..., T max}, represents the predicted output of renewable energy at time t; and P RE,u,t With P RE,d,t The corresponding maximum forecast deviation values ​​of renewable energy output in period t are set based on historical forecast errors; Z RE,u,t and Z RE,d,t is a variable between 0 and 1, used to describe the fluctuation range; Γ RE is a robust adjustment factor, which defines the number of cycles during which the renewable energy generation reaches the fluctuation range during the dispatch period. RE,u,t and P RE,d,t describes the volatility of output, while Γ RE It determines the persistence of the impact of this fluctuation on the scheduling plan, and its value balances the conservatism and economic efficiency of the plan.

[0018] Furthermore, the step 2 comprises the following steps:

[0019] Generative Adversarial Networks (GAN) are used for image restoration and image generation, including generative networks (GN) and discriminant networks (DN).

[0020] will satisfy any distribution p test The training data x of (x) is input into GN to train the network parameters so that it is as close as possible to the real electrical and thermal load data. The probability distribution of

[0021] WGAN-GP uses Wasserstein (optimal transmission) distance to measure the distance between the generated data distribution and the real data distribution, solving the problem of unstable training of GAN and improving the accuracy of generated scenes.

[0022] The gradient penalty (GP) term of the discriminant network function is introduced into the definition domain to ensure that the discriminant network satisfies the first-order Lipschitz function constraint. DN can fully describe the Wasserstein distance and solve the problems of gradient vanishing and gradient exploding in training.

[0023] The objective function of WGAN-GP is shown in (2).

[0024]

[0025] Where: the mapping spaces of DN and GN are represented by D and G respectively; the value function is represented by V(D,G); the discriminant network output function is represented by D(x); the GN generated data is represented by G(x); the expected distribution is represented by E; Generate homogeneous sampling points on the link between real data and GN; P pena (x) means The probability distribution of is the gradient of DN; λ is the gradient penalty coefficient.

[0026] The composite scenarios generated by WGAN-GP are complex and diverse, and the calculation is difficult. The K-medoids method is used to reduce the energy load scenarios and select typical scenarios. The probability calculation of various scenarios follows formula (3). N is the number of electric and thermal load scenarios, N k is the number of scenes in the kth category.

[0027]

[0028] Furthermore, the step 3 comprises the following steps:

[0029] On the load side, WGAN-GP is used to generate multi-energy load scenarios for load forecast values ​​and historical values, and then the typical load scenarios are obtained through the K-medoids method. For the optimal scheduling problem of virtual power plants with multiple uncertainties, a two-stage stochastic robust optimization model is established, and the decision process is as follows: Figure 2 shown.

[0030] ① Constraints

[0031] a. According to formula (1), the specific expressions of photovoltaic and wind power in time period t are obtained:

[0032]

[0033] b. Energy and power balance constraint, which ensures the balance of heat and electricity supply and demand in VPP during time period t; where the subscripts ele and ther represent electrical energy and thermal energy, respectively.

[0034]

[0035] c. Operating constraints of energy conversion equipment.

[0036]

[0037] Where: P max,m and P min,m Respectively represent the output upper and lower limits of the mth device; ΔP max,m and ΔP min,m are the upper and lower limits of the climbing power of the mth device respectively.

[0038] d. Operational constraints of energy storage equipment

[0039]

[0040] Where: P ch,max,S,e and P dis,max,S,e Respectively represent the maximum charging and discharging power allowed by the energy storage device; U S,w,e,t is an indicator variable used to indicate the charge and discharge status of the e-type energy storage system. When its value is equal to 1, the device is in the discharge mode. If its value is equal to 0, it indicates the charging state. S,w,e,t E is defined as the energy state of the e-type energy storage device at time t and scenario w; max,S,e and E min,S,e They are respectively set as the upper and lower limits of the energy of the energy storage device during the entire scheduling period.

[0041] Energy interaction power constraints:

[0042]

[0043] Where: P M,max,lThe maximum transaction volume of the VPP energy supply network; U M,w,l,t It is the transaction status between VPP and the upper energy supply network.

[0044] ②Objective function

[0045] The optimal scheduling solution can be obtained by constructing a two-stage stochastic robust optimization scheduling model. The objective function of the model is as follows:

[0046]

[0047] Where: C G (t) represents the operating cost of the controllable equipment at time t; C E (t) represents the cost of the energy storage device during the operation period t; and C M (t) is the transaction cost between the virtual power plant and the superior energy network during period t.

[0048]

[0049]

[0050] Where: Δt is the optimized time interval; c d,m is the operating cost factor of the mth device; M is the number of devices within the control range of VPP; E is the number of energy storage devices within the supervision range of VPP; L is the number of energy supply networks; P w,m,t represents the power output of the mth power generation unit for scenario w in time period t; c s,e represents the operating cost coefficient of type e energy storage device; P ch,w,e,t and P dis,w,e,t are the charging and discharging powers of the e-type energy storage device in scenario w at time t; buy,l,t and λ sell,l,t is the corresponding price of buying and selling energy when the VPP trades with the upper energy network at time t; P buy,w,l,t and P sell,w,l,t They are respectively the type l purchase and sale energy in scenario w during period t when VPP trades with the upper energy grid.

[0051] Furthermore, the step 4 includes the following contents:

[0052] The VPP two-stage stochastic robust optimization scheduling model is expressed as:

[0053]

[0054] In the objective function structure, "min" is outside the curly braces, which is an outer first-order optimization task with variable x, while the "max min" structure nested in the curly braces is an inner second-order optimization problem with variables y and u. The coefficients of the objective function are represented by the column vector C, while in the constraints, the coefficient matrices are described by A, B, C, D, and Eu, respectively, and the constant columns are represented by f and g. Ω(x,u) is the available domain of variable y given (x,u).

[0055]

[0056] The two-stage stochastic robust optimization scheduling contains a min-max-min structure and uncertainty parameters, which makes direct solution complicated. The C&CG algorithm decomposes the original problem into two key parts: the master problem (MP) and the sub-problem (SP). In the initial stage, the initial worst scenario is set by the value of the uncertainty variable u. First, by solving SP, the uncertainty value of the source end in the worst case is obtained, and then MP is solved based on this scenario and the lower limit of the optimal solution is obtained, as shown in formula (17). SP not only provides the variable value in the worst case, but also provides the upper bound of the optimal solution. (19) The variable value in the worst scenario is passed to MP, which further strengthens the constraint of the problem. For the maximum minimization subproblem, the duality theory is used to transform it into a single-level problem.

[0057]

[0058] Where: θ represents the newly added auxiliary variable; y k represents the scheduling strategy obtained after the kth iteration; N is the current iteration count; The worst case determined at the kth iteration, α, β, γ and η are the dual expressions of the constraints of the second-stage optimization problem; Defined as the upper limit of η, a sufficiently large positive real number is chosen; and are additional auxiliary variables introduced.

[0059] Through the above design scheme, the present invention can bring the following beneficial effects:

[0060] (1) Based on the two-stage stochastic robust optimization scheduling method, economic scheduling of each unit to meet multi-energy loads is achieved.

[0061] (2) Based on the two-stage stochastic robust optimization scheduling strategy, the scheduling solution with minimum cost is obtained under the worst scenario.

[0062] (3) By adjusting the robust optimization parameters, economic dispatch strategies are formulated for various uncertain conditions to improve the environmental adaptability of the virtual power plant.

[0063] (4) The economic benefits are more significant in balancing power deviations in the energy market of virtual power plants. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] The present invention is further described below with reference to the accompanying drawings and specific embodiments:

[0065] Figure 1 This is a step diagram of a robust optimization scheduling method for a virtual power plant based on source-load uncertainty of the present invention.

[0066] Figure 2 This is a two-stage robust optimization decision flow chart in a robust optimization scheduling method for a virtual power plant based on source-load uncertainty of the present invention. DETAILED DESCRIPTION

[0067] In order to make the purpose, technical solution and advantages of the present invention more clear, the following is a further detailed description of a method for robust optimization scheduling of a virtual power plant based on source-load uncertainty in conjunction with the accompanying drawings. It should be understood that the specific embodiments described herein are only used to explain the present invention and do not constitute a limitation of the present invention.

[0068] Example

[0069] like Figure 1 As shown, a robust optimization scheduling method for a virtual power plant based on source-load uncertainty includes the following steps:

[0070] Step 1: Modeling uncertainty on the power supply side

[0071] Robust optimization strategies are used to address the problem that wind power generation and photovoltaic power generation are easily affected by environmental changes, resulting in significant randomness and instability. This is done by summarizing the power generation scenarios of renewable energy into an overall set.

[0072] The core of solving the robust optimization problem is to select uncertainty sets, which describe wind and solar energy output according to the characteristics of uncertain parameters and reduce the uncertainty sets that excessively increase the computational complexity of the model. The cardinality uncertainty set with robust adjustable coefficients can effectively regulate the impact range of uncertainty deviation and improve the accuracy of the prediction of uncertainty fluctuation range. RE The element P in RE,t The following constraints are met:

[0073]

[0074] Where: The time set for optimal scheduling is T = {1, 2, ..., T, ..., T max}, represents the predicted output of renewable energy at time t; and P RE,u,t With P RE,d,t The corresponding maximum forecast deviation values ​​of renewable energy output in period t are set based on historical forecast errors; Z RE,u,t and Z RE,d,t is a variable between 0 and 1, used to describe the fluctuation range; Γ RE is a robust adjustment factor, which defines the number of cycles during which the renewable energy generation reaches the fluctuation range during the dispatch period. RE,u,t and P RE,d,t describes the volatility of output, while Γ RE It determines the persistence of the impact of this fluctuation on the scheduling plan, and its value balances the conservatism and economic efficiency of the plan.

[0075] Step 2 Load uncertainty modeling

[0076] WGAN-GP is used to generate multi-energy load scenarios, and the K-medoids algorithm is used to obtain typical scenarios to solve the problem that load scenarios in different regions and environments are difficult to describe using the same function.

[0077] Generative Adversarial Networks (GAN) are used for image restoration and image generation, including generative networks (GN) and discriminant networks (DN).

[0078] will satisfy any distribution p test The training data x of (x) is input into GN to train the network parameters so that it is as close as possible to the real electrical and thermal load data. The probability distribution of

[0079] WGAN-GP uses Wasserstein distance to measure the distance between the generated data distribution and the real data distribution, solving the problem of unstable training of GAN and improving the accuracy of generated scenarios.

[0080] The gradient penalty (GP) term of the discriminant network function is introduced into the definition domain to ensure that the discriminant network satisfies the first-order Lipschitz function constraint. DN can fully describe the Wasserstein distance and solve the problems of gradient vanishing and gradient exploding in training.

[0081] The objective function of WGAN-GP is shown in (2).

[0082]

[0083] Where: the mapping spaces of DN and GN are represented by D and G respectively; the value function is represented by V(D,G); the discriminant network output function is represented by D(x); the GN generated data is represented by G(x); the expected distribution is represented by E; Generate homogeneous sampling points on the link between real data and GN; P pena (x) means The probability distribution of is the gradient of DN; λ is the gradient penalty coefficient.

[0084] The composite scenarios generated by WGAN-GP are complex and diverse, and the calculation is difficult. The K-medoids method is used to reduce the energy load scenarios and select typical scenarios. The probability calculation of various scenarios follows formula (3). N is the number of electric and thermal load scenarios, N k is the number of scenes in the kth category.

[0085]

[0086] Step 3: Two-stage stochastic robust optimization scheduling modeling

[0087] A cardinality uncertainty set with robust adjustable coefficients is used to measure the fluctuation of uncertainty in renewable energy output and optimize the uncertainty on the power supply side.

[0088] On the load side, WGAN-GP is used to generate multi-energy load scenarios for load forecast values ​​and historical values, and then the typical load scenarios are obtained through the K-medoids method. For the optimal scheduling problem of virtual power plants with multiple uncertainties, a two-stage stochastic robust optimization model is established, and the decision process is as follows: Figure 2 shown.

[0089] ①1①Constraints

[0090] a. According to formula (1), the specific expressions of photovoltaic and wind power in time period t are obtained:

[0091]

[0092]

[0093] b. Energy and power balance constraint, which ensures the balance of heat and electricity supply and demand in VPP during time period t; where the subscripts ele and ther represent electrical energy and thermal energy, respectively.

[0094]

[0095] c. Operating constraints of energy conversion equipment.

[0096]

[0097] Where: P max,m and P min,m Respectively represent the output upper and lower limits of the mth device; ΔP max,m and ΔP min,m are the upper and lower limits of the climbing power of the mth device respectively.

[0098] d. Operational constraints of energy storage equipment

[0099]

[0100] Where: P ch,max,S,e and P dis,max,S,e Respectively represent the maximum charging and discharging power allowed by the energy storage device; U S,w,e,t is an indicator variable used to indicate the charge and discharge status of the e-type energy storage system. When its value is equal to 1, the device is in the discharge mode. If its value is equal to 0, it indicates the charging state. S,w,e,t E is defined as the energy state of the e-type energy storage device at time t and scenario w; max,S,e and E min,S,e They are respectively set as the upper and lower limits of the energy of the energy storage device during the entire scheduling period.

[0101] Energy interaction power constraints:

[0102]

[0103] Where: P M,max,l The maximum transaction volume of the VPP energy supply network; U M,w,l,t It is the transaction status between VPP and the upper energy supply network.

[0104] ②Objective function

[0105] The optimal scheduling solution can be obtained by constructing a two-stage stochastic robust optimization scheduling model. The objective function of the model is as follows:

[0106]

[0107] Where: C G (t) represents the operating cost of the controllable equipment at time t; C E (t) represents the cost of the energy storage device during the operation period t; and C M (t) is the transaction cost between the virtual power plant and the superior energy network during period t.

[0108]

[0109] Where: Δt is the optimized time interval; c d,mis the operating cost factor of the mth device; M is the number of devices within the control range of VPP; E is the number of energy storage devices within the supervision range of VPP; L is the number of energy supply networks; P w,m,t represents the power output of the mth power generation unit for scenario w in time period t; c s,e represents the operating cost coefficient of type e energy storage device; P ch,w,e,t and P dis,w,e,t are the charging and discharging powers of the e-type energy storage device in scenario w at time t; buy,l,t and λ sell,l,t is the corresponding price of buying and selling energy when the VPP trades with the upper energy network at time t; P buy,w,l,t and P sell,w,l,t They are respectively the type l purchase and sale energy in scenario w during period t when VPP trades with the upper energy grid.

[0110] Step 4: Solving the two-stage stochastic robust optimization scheduling model

[0111] Through the two-stage stochastic robust optimization scheduling model, typical scenarios are obtained, and scheduling strategies under different schemes are provided by changing the robust adjustment coefficient.

[0112] The VPP two-stage stochastic robust optimization scheduling model is expressed as:

[0113]

[0114] In the objective function structure, "min" is outside the curly braces and is an outer first-order optimization task with x as the variable. The "max min" structure nested in the curly braces is an inner second-order optimization problem with y and u as the variables. The coefficients of the objective function are represented by the column vector C, while in the constraints, the coefficient matrices are described by A, B, C, D, and Eu, respectively, and the constant columns are represented by f and g. Ω(x,u) is the available domain of the variable y given (x,u).

[0115]

[0116] The two-stage stochastic robust optimization scheduling contains a min-max-min structure and uncertainty parameters, which makes direct solution complicated. The C&CG algorithm decomposes the original problem into two key parts: the master problem (MP) and the sub-problem (SP). In the initial stage, the initial worst scenario is set by the value of the uncertainty variable u. First, by solving SP, the uncertainty value of the source end in the worst case is obtained, and then MP is solved based on this scenario and the lower limit of the optimal solution is obtained, as shown in formula (17). SP not only provides the variable value in the worst case, but also provides the upper bound of the optimal solution. (19) The variable value in the worst scenario is passed to MP, which further strengthens the constraint of the problem. For the maximum minimization subproblem, the duality theory is used to transform it into a single-level problem.

[0117]

[0118] Where: θ represents the newly added auxiliary variable; y k represents the scheduling strategy obtained after the kth iteration; N is the current iteration count; The worst case determined at the kth iteration, α, β, γ and η are the dual expressions of the constraints of the second-stage optimization problem; Defined as the upper limit of η, a sufficiently large positive real number is chosen; and are additional auxiliary variables introduced.

Claims

1. A robust optimization scheduling method for virtual power plants based on source-load uncertainty, characterized by: The technical solution is as follows: Step 1: Power supply uncertainty modeling The renewable energy generation scenarios are summarized as a whole set using a robust optimization strategy; Step 2: Load uncertainty modeling Use WGAN-GP to generate multi-energy load scenarios, and use the K-medoids algorithm to obtain typical scenarios; Step 3: Two-stage stochastic robust optimization scheduling modeling A cardinality uncertainty set with robust adjustable coefficients is used to measure the fluctuation of uncertainty in renewable energy output and optimize the uncertainty on the power supply side. Step 4: Solving the two-stage stochastic robust optimization scheduling model A typical scenario is obtained through a two-stage stochastic robust optimization scheduling model, and the scheduling strategies under different schemes are implemented by changing the robust adjustment coefficient.

2. The method for robust optimization scheduling of virtual power plants based on source-load uncertainty according to claim 1 is characterized in that: The step 1 comprises: The core of solving the robust optimization problem is to select uncertainty sets, describe wind and solar output according to the characteristics of uncertain parameters, reduce the uncertainty sets that excessively increase the computational complexity of the model, and select the uncertainty set P of the renewable energy output base. RE The element P in RE,t The following constraints are met: Where: The time set for optimal scheduling is T = {1, 2, ..., T, ..., T max }, represents the predicted output of renewable energy at time t; and P RE,u,t With P RE,d,t They correspond to the maximum forecast deviation of renewable energy output in period t; Z RE,u,t and Z RE,d,t is a variable between 0 and 1; Γ RE is a robust adjustment factor.

3. The method for robust optimization scheduling of virtual power plants based on source-load uncertainty according to claim 1 is characterized in that: The step 2 comprises: Generative adversarial networks are used in image restoration and image generation. Generative adversarial networks are referred to as GANs, which include generative networks and discriminative networks. Generative networks are referred to as GN and discriminative networks are referred to as DN. will satisfy any distribution p test The training data x of (x) is input into GN to train the network parameters so that it is as close to the real electrical and thermal load data as possible. The probability distribution of WGAN-GP uses Wasserstein distance to measure the distance between the generated data distribution and the true data distribution; The gradient penalty term of the discriminant network function is introduced into the definition domain. The gradient penalty is referred to as GP. DN satisfies the first-order Lipschitz function constraint. The objective function of WGAN-GP is shown in (2). Where: the mapping space of DN and GN is represented by D, G; the value function is represented by V(D, G); the discriminant network output function is represented by D(x); the GN generated data is represented by G(x); the expected distribution is represented by E; Generate homogeneous sampling points on the data link between real data and GN; refer to The probability distribution of is the gradient of DN; λ is the gradient penalty coefficient; The K-medoids method is used to reduce energy load scenarios, and typical scenarios are selected. The probability calculation of various scenarios follows the following formula: Where N is the number of electric heating load scenarios, N k is the number of scenes in the kth category.

4. According to claim 1, a method for robust optimization scheduling of virtual power plants based on source-load uncertainty is characterized by: The step 3 comprises the following steps: On the load side, WGAN-GP is used to generate multi-energy load scenarios for load forecast values ​​and historical values, and then the K-medoids method is used to obtain typical load scenarios. For the optimal scheduling problem of virtual power plants with multiple uncertainties, a two-stage stochastic robust optimization model is established. ① Constraints a. According to formula (1), the specific expressions of photovoltaic and wind power in time period t are obtained: b. Energy and power balance constraint, which ensures the balance of heat and electricity supply and demand in VPP within time period t; where the subscripts ele and ther represent electrical energy and thermal energy respectively; c. Operational constraints of energy conversion equipment, Where: P max,m and P min,m Respectively represent the output upper and lower limits of the mth device; ΔP max,m and ΔP min,m are the upper and lower limits of the climbing power of the mth device, respectively. d. Operational constraints of energy storage equipment Where: P ch,max,S,e and P dis,max,S,e Respectively represent the maximum charging and discharging power allowed by the energy storage device; U S,w,e,t It is an indicator variable used to indicate the charge and discharge status of the E-type energy storage system. When its value is equal to 1, the device is in the discharge mode. If its value is equal to 0, it indicates the charging state. S,w,e,t E is defined as the energy state of the e-type energy storage device at time t and scenario w; max,S,e and E min,S,e are respectively set as the upper and lower limits of the energy of the energy storage device during the entire scheduling period, Energy interaction power constraints: Where P M,max,l The maximum transaction volume of the VPP energy supply network; U M,w,l,t The transaction status between VPP and the upper energy supply network; ②Objective function The optimal scheduling scheme is obtained by constructing a two-stage stochastic robust optimization scheduling model. The objective function of the model is as follows: Where: C G (t) represents the operating cost of the controllable equipment at time t; C E (t) represents the cost of the energy storage device during the operation period t; and C M (t) is the transaction cost between the virtual power plant and the upper-level energy network during period t; Where: Δt is the optimized time interval; c d,m is the operating cost factor of the mth device; M is the number of devices within the control range of VPP; E is the number of energy storage devices within the supervision range of VPP; L is the number of energy supply networks; P w,m,t represents the power output of the mth power generation unit for scenario w in time period t; c s,e P represents the operating cost coefficient of the e-type energy storage device; ch,w,e,t and P dis,w,e,t are the charging and discharging powers of the e-type energy storage device in scenario w at time t; buy,l,t and λ sell,l,t is the corresponding price of buying and selling energy when the VPP trades with the upper energy network at time t; P buy,w,l,t and P sell,w,l,t They are respectively the type l purchase and sale energy in scenario w during period t when VPP trades with the upper energy grid.

5. According to claim 1, a method for robust optimization scheduling of virtual power plants based on source-load uncertainty is characterized by: The step 4 comprises the following steps: The VPP two-stage stochastic robust optimization scheduling model is expressed as: In the objective function structure, "min" is outside the curly braces and is an outer first-order optimization task with variable x. The "max min" structure nested in the curly braces is an inner second-order optimization problem with variables y and u. The coefficients of the objective function are represented by the column vector C. In the constraints, the coefficient matrices are described by A, B, C, D, and Eu, respectively. The constant columns are represented by f and g. Ω(x,u) is the available domain of variable y given (x,u). The two-stage stochastic robust optimization scheduling includes a min-max-min structure and uncertainty parameters. The C&CG algorithm decomposes the original problem into two key parts: the main problem and the sub-problem. The main problem is referred to as MP and the sub-problem is referred to as SP. In the initial stage, the initial worst scenario is set by the value of the uncertainty variable u. First, by solving SP, the worst-case source uncertainty value is obtained. Then, based on this scenario, MP is solved and the lower limit of the optimal solution is obtained, as shown in formula (17). Formula (19) passes the variable value in the worst scenario to MP, further strengthening the constraint of the problem. For the maximum-minimization sub-problem, the duality theory is used to transform it into a single-level problem. Where: θ represents the newly added auxiliary variable; y k represents the scheduling strategy obtained after the kth iteration; N is the current iteration count; The worst case determined at the kth iteration, α, β, γ and η are the dual expressions of the constraints of the second-stage optimization problem; Defined as the upper limit of η, select a sufficiently large positive real number; and Z = [Z PV,d,t ,Z WT,d,t ] T ,Z`=[Z` PV,d,t ,Z` WT,d,t ] T are additional auxiliary variables introduced.

Images