Virtual power plant optimization system and method

Abstract

The invention discloses a virtual power plant optimization system and method, and the system comprises a virtualization platform module which is used for integrating distributed energy, energy storage equipment, an electric vehicle, a flexible load and other resources, and achieves the virtualization and centralized management of the resources; the dynamic resource aggregation module is used for dynamically matching resource characteristics with power system requirements; the market transaction module is used for optimizing the earnings of the virtual power plant; and the core algorithm module is used for optimizing the load prediction precision. The market transaction module implements two-stage market bidding based on a multi-agent game model, analyzes the uncertainty of quotation of competitors through a multi-scene simulation technology, optimizes a dynamic transaction strategy, and achieves the maximization of spot market benefits and the controllability of risks. Through cooperative work of a virtualization platform module, a dynamic resource aggregation module, a market transaction module, a core algorithm module, a security management module and the like, efficient management of resources and maximization of market benefits are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of virtual power plants, and in particular to a virtual power plant optimization system and method. Background Art

[0002] As the proportion of renewable energy continues to increase, the power system faces numerous technical challenges. Among them, the integration of distributed energy resources is a key issue. Renewable energy sources such as wind and photovoltaics exhibit significant volatility and intermittency, significantly increasing the complexity of grid dispatch and coordinated control. Existing systems lack the ability to respond in real time to the rapid changes in renewable energy generation, often leading to wasted or undersupplied resources.

[0003] Existing virtual power plant technologies primarily rely on static resource allocation, lack dynamic response capabilities, and struggle to meet multi-timescale control requirements. Distributed resource integration is inefficient, making it difficult to achieve efficient utilization of wide-area resources. Furthermore, market mechanisms have numerous shortcomings. Virtual power plants have a single model for participating in the spot market, making it difficult to cope with market price fluctuations. Existing market mechanisms lack flexibility and are unable to effectively coordinate the collaborative needs of multiple entities. Furthermore, existing research focuses more on the integration and scheduling of single resources, while insufficient research exists on dynamic aggregation methods for wide-area distributed resources and multi-agent market transaction optimization strategies. This limits the effectiveness of virtual power plants in power systems with a high proportion of renewable energy.

[0004] These issues severely restrict the utilization of distributed energy resources and the capacity to absorb new energy, making it impossible to meet the goals of carbon neutrality and the requirements of green energy transformation. Therefore, there is an urgent need for a virtual power plant system that can dynamically aggregate resources and implement real-time control to effectively address the shortcomings of existing technologies. Summary of the Invention

[0005] To address the challenges of low distributed energy integration efficiency, delayed system response, and insufficient market flexibility faced when a high proportion of renewable energy is integrated into the power system, this paper provides a virtual power plant optimization system and method. Through the collaborative work of virtualization platforms, dynamic resource aggregation, market transactions, core algorithms, and security management modules, this system achieves efficient resource management and maximizes market returns.

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

[0007] A virtual power plant optimization system, the system comprising:

[0008] A virtualization platform module is used to integrate distributed energy, energy storage equipment, electric vehicles, and load resources to achieve virtualization and centralized management of resources;

[0009] Dynamic resource aggregation module, used to dynamically match resource characteristics with power system requirements;

[0010] Market trading module, used to optimize the benefits of virtual power plants;

[0011] Core algorithm module, used to optimize load forecasting accuracy.

[0012] Security management module, used to dynamically adjust resource transaction prices.

[0013] In the virtualization platform module, in order to achieve the coordinated optimization of the distributed resources in the virtual power plant (VPP), the output characteristics and operating characteristics of each distributed resource are established, and the corresponding mathematical model is analyzed and established:

[0014] 1) Photovoltaic power generation modeling is as follows:

[0015]

[0016] In formula (1), P PV Represents photovoltaic output power; f PV is the loss factor of the photovoltaic panel; P PV,cap is the output power of the photovoltaic array under standard test conditions; G T,STC is the light intensity under standard test conditions; G T is the actual light intensity; α P is the power temperature coefficient; T cell Indicates the actual operating temperature of the photovoltaic cell; T cell,STC Indicates the temperature under standard test conditions;

[0017] 2) The wind power generation WT model is as follows:

[0018]

[0019] In formula (2), P WT Indicates the actual output power of the wind turbine; v i 、v v Respectively represent the cut-in and cut-out wind speeds; v represents the current wind speed; v r Indicates the cut-out wind speed; v o Indicates rated wind speed;

[0020] 3) The energy storage system charging and discharging model is as follows:

[0021]

[0022] In formula (3), SOC t Indicates the state of charge of the energy storage system at time t; SOC t-1 represents the state of charge of the energy storage system at time t-1; λ b Indicates the self-discharge rate of the energy storage system; Indicates the charging efficiency of the energy storage system; Indicates the discharge efficiency of the energy storage system; represents the charging power of the energy storage system at t-1; represents the discharge power of the energy storage system at t-1; △t represents the time interval between two adjacent time points; SOC min Indicates the lower limit of the state of charge; SOC max Represents the upper limit of the state of charge.

[0023] To achieve safe and stable operation of the energy storage system, it is divided into five charging and discharging action intervals:

[0024] Interval 1: When 0≤SOC t <0.2SOC max When the charge power range is P d =0,P d Indicates the discharge power of the energy storage system in interval 1;

[0025] Interval 2: When 0.2SOC max ≤SOC t <0.35SOC max When the charging power range is:

[0026] 0.2SOC max ≤SOC t <0.35SOC max , the discharge power range is Interval 3: When 0.35SOC max ≤SOC t <0.65SOC max When the charging power range is P C , Interval 4: When 0.65SOC max ≤SOC t <0.8SOC max When the charging power range is Discharge power range is Interval 5: When 0.8SOC max ≤SOC t <SOC max When the charging power is P C =0, which is a non-rechargeable state; the discharge power range is

[0027] At the same time, ensure that the state of charge of the energy storage system is not less than the minimum state of charge and not greater than the maximum state of charge. Then, the energy storage system is divided into free trading energy storage and backup energy storage.

[0028] For multiple virtual power plants (VPPs), a two-layer VPP game model is established. The upper-layer VPP game model takes maximizing the VPP revenue as its objective function, while the lower-layer VPP game model takes minimizing the total power generation cost as its objective function, thereby modeling the internal game of the VPP.

[0029] The upper-level VPP game model maximizes the benefits of the virtual power plant VPP and establishes an objective function, which is expressed as follows:

[0030]

[0031] In formula (4), represents the revenue obtained by the i-th VPP from selling electricity to the load at time t; The gains of the interaction of the i-th VPP with a wider grid are shown; represents the revenue obtained from selling electricity from the i-th VPP to other VPPs;

[0032] For any virtual power plant (VPP), it is necessary to ensure that the system maintains power balance at any point during operation. The sum of the power of all power sources in the VPP interacting with other VPPs and the sum of the power interacting with the main grid equals the load demand. The expression is as follows:

[0033]

[0034] In formula (5), Indicates the load demand power; Indicates photovoltaic output power; Indicates the output power of the wind turbine; Indicates the output power of standby energy storage; Indicates the free energy storage output power; It is the power input from the large power grid to the VPP.

[0035] With the objective function of reducing the overall operating costs of VPP, a lower-level VPP game model is constructed:

[0036]

[0037] In formula (9), represents the total power generation cost of the i-th VPP at time t; represents the photovoltaic power generation cost in the i-th VPP; represents the cost of electricity generated by wind turbines in the i-th VPP; represents the operating cost of the energy storage system in the i-th VPP; N PV represents the number of photovoltaic cells in the i-th VPP; N W N represents the number of wind turbines in the i-th VPP; b represents the amount of energy storage in the i-th VPP; represents the minimum total power generation cost of VPP at time t; represents the photovoltaic power generation cost of the i-th virtual power plant; represents the wind power generation cost of the i-th virtual power plant; represents the cost of the energy storage system in the i-th VPP.

[0038] The upper-level VPP game model includes constraints:

[0039] 1) Wind power and photovoltaic output constraints:

[0040] The production capacity of wind and photovoltaic installations is inherently limited by their maximum output ratings:

[0041]

[0042] In formula (6), represents the actual photovoltaic output power of the i-th virtual power plant at time t; Represents the maximum output power of the PV; represents the actual output power of the wind turbine of the i-th VPP at time t; The maximum power output of the fan.

[0043] 2) The electricity price traded between VPPs should not be higher than the price of electricity purchased from the grid, nor lower than the price of electricity sold from the grid:

[0044] λ da,t ≤λ i,t ≤λ' da,t (7);

[0045] In formula (7), λ i,t is the transaction price of VPP; da,t is the price of electricity sold by VPP to the grid; da,t The price at which VPP purchases electricity from the grid.

[0046] 3) Power trading exists between VPPs and the main grid, and between multiple VPPs with tie-line power constraints:

[0047]

[0048] In formula (8), represents the energy of interaction between the i-th VPP and the large grid; is the maximum allowed interaction power of the contact line between the first VPP and the grid; P ij,t is the energy of interaction between the i-th VPP and the j-th VPP; is the maximum interaction power allowed by the contact line between the i-th VPP and the j-th VPP.

[0049] The lower-level VPP game model satisfies power balance constraints, energy storage charging and discharging power constraints, and wind and solar curtailment constraints. The energy storage system constrains that the energy storage charging and discharging power must be less than the maximum output power and maximum discharge power of the energy storage system.

[0050]

[0051] In formula (10), Indicates the maximum charging power; represents the net charge and discharge power of the energy storage system in the i-th VPP at time t; Indicates the maximum discharge power;

[0052] Abandoning wind power generation WT and photovoltaic power generation output constraints:

[0053]

[0054] In formula (11), ρ pv,t and ρ w,t They represent the constraint coefficients of photovoltaic power generation and wind power generation WT reduction at time t; △P w,t represents the wind curtailment of the ith VPP at time t; △P pv,t represents the amount of abandoned light of the i-th VPP at time t;

[0055] The multiple VPPs in the double-layer VPP game model need to consider both the upper-layer VPP game model and the lower-layer VPP game model, and the two influence each other. The double-layer particle swarm optimization method is used to solve the double-layer VPP game model containing multiple VPPs. In the double-layer particle swarm optimization method, the upper-layer VPP game model is solved by the outer particle swarm method, where the particle position represents the direct transaction power of a single VPP, and the fitness value represents the sum of the economic benefits of all VPPs; the lower-layer VPP game model is solved by the inner particle swarm method, where the particle position represents the output of each unit and the user load power demand, and the fitness value represents the VPP operating cost. The double-layer particle swarm optimization method finally converges to the game equilibrium solution through the iterative coupling of the outer PSO and the inner PSO;

[0056] In the structure and coupling relationship of the two-tier VPP game model, the goal of the upper-tier VPP game model is to maximize the benefits of each VPP:

[0057]

[0058] in: Represents the revenue from electricity sales; Indicates the benefits of interacting with the main network; It indicates that the transaction income between VPPs depends on the power generation cost optimized at the lower layer.

[0059] Decision variable: transaction price λ between VPPs i,t and interactive power P g,t ;

[0060] The cost minimization objective of the lower-level VPP game model is to minimize the internal power generation cost of the VPP;

[0061]

[0062] Decision variables: output P of each unit pv,t ,P w,t ,P b,t and the amount of wind and solar power curtailment △P w,t ,△P pv,t ;

[0063] Coupling Mechanism: The upper-layer VPP game model's revenue calculation requires the lower-layer VPP game model's costs as input, while the lower-layer VPP game model's optimization is subject to the transaction price and power constraints assigned by the upper-layer VPP game model. Both achieve dynamic equilibrium through iterative feedback from the two-layer PSO.

[0064] The dynamic resource aggregation module combines the responsiveness of distributed resources with market supply and demand dynamics, and uses a data-driven clustering algorithm to dynamically match resource characteristics with power system requirements to optimize peak and frequency regulation performance.

[0065] Distributed resources are divided into energy storage and power sources. Energy storage includes distributed energy storage and electric vehicles, which have energy accumulation capabilities. Power sources include fully controlled power sources (FCPs) such as micro gas turbines, and semi-controlled power sources (HCPs) such as distributed wind power and distributed photovoltaics.

[0066] The constraints of the fully controlled power supply (FCP) include power, capacity, and ramp constraints, namely:

[0067]

[0068] In formula (14): and are the active and reactive outputs of FCPi at time t respectively; and are the minimum and maximum active output of FCPi respectively; is the rated capacity of FCPi; and They are the upper and lower limits of FCPi’s climbing ability, respectively; represents the active and reactive output of FCPi at time t+1;

[0069] A semi-controlled power supply HCP resource output model considering the uncertainty of wind and solar power output is constructed, and the chance constraints containing random variables are converted into deterministic constraints through the Gaussian mixture model (GMM) to achieve the solution of the virtual power plant feasible region (VFR).

[0070] The resource output model of the semi-controlled power supply HCP is:

[0071]

[0072] In formula (13): and are the active and reactive outputs of HCPi at time t respectively; and are the minimum and maximum power factor angles of HCPi respectively; represents the maximum active power output limit of FCPi at time t;

[0073] The objective function of the economic optimization model of multiple VPPs is to optimize the total cost of each aggregated VPP. The total cost of VPPi includes the active power cost. and reactive power costs Right now:

[0074]

[0075] In formula (16): P i and Q i are the active and reactive power inputs of VPPs, respectively; I is the number of VPPs included in the distribution network; represents the minimum total cost of all VPPs; C represents the total cost of all VPPs;

[0076] Among them, active cost The piecewise quadratic function aggregation cost model is adopted, which is the superposition of the costs of various resources within the VPP. The cost functions of various internal resources are shown in formula (17);

[0077] Reactive power cost Including fixed costs and opportunity costs, the specific expressions of the active power cost model and reactive power cost model of VPP are:

[0078]

[0079] In the above formula: are the costs of FCP, HCP and energy storage resources respectively; and is the cost coefficient of FCPm; and is the cost coefficient of HCPn; and is the cost coefficient of energy storage resource l; α i,n , βi,n and γ i,n are the secondary, primary and constant cost coefficients of VPPi output active power in segment n, usually n∈[2,5], which is determined by the internal resource type of VPP; and are the active powers of FCPm, HCPn and energy storage resource l respectively; D i,n is the boundary value of VPPi output active power; Q,i is the discounted fixed cost; is the cost coefficient; P i,max and S i,max are the maximum active power and apparent power of VPPi respectively; α i represents the secondary cost coefficient of VPPi output active power; β i represents the primary cost coefficient of VPPi output active power; γ i represents the constant cost coefficient of VPPi output active power; P i Indicates the active power of VPPi; Q i Indicates the reactive power of VPPi; D i1 Indicates VPP i Minimum technical output; D i2 Indicates the output limit or operation mode switching point of the first type of resources. i3 Indicates the output upper limit of the second type of resources or the maximum output allowed by the system.

[0080] Combining the cluster analysis of load current field and electrical distance, the cluster center is determined by the load current field, and then clustering is completed based on the electrical distance, avoiding the problem of repeated iteration in traditional clustering algorithms.

[0081] (1). Calculation of load current field and virtual potential:

[0082] The virtual potential φ of the load current at node i is defined as i :

[0083]

[0084] Among them, Z ij is the matrix Z u Elements, I j is the load current at node j. This formula reflects the distribution characteristics of node voltage affected by load current.

[0085] Describe the relationship between the voltage drop at node i and the load current:

[0086] △U i =Z ij I j (twenty four);

[0087] Among them, △U i represents the voltage drop at node i; Equation (24) is used to equivalently represent the voltage impact between nodes.

[0088] (2) Definition of electrical distance:

[0089] The electrical distance matrix D between nodes is defined based on the inverse matrix of the Jacobian matrix J, namely the susceptance matrix B;

[0090] D ij =|(J -1 ) ii +(J -1 ) jj -2(J -1 ) ij | (25);

[0091] Formula (25) is used to quantify the degree of electrical connection between nodes and is the basis of cluster analysis.

[0092] (3) Determination of cluster center nodes

[0093] The cluster center node is determined by the maximum point of the local potential value of the load current field:

[0094] For node i, if its potential value φ i If the potential value of i is greater than that of all directly connected nodes, then i is the point with the maximum local potential value. The number of points with the maximum local potential value is the number of partitions, and these points serve as cluster center nodes.

[0095] (4) Cluster analysis process:

[0096] Clustering method: Calculate the electrical distance D between each node and the cluster center node ij , divide the nodes into the partitions to which the cluster centers with the smallest distance belong.

[0097] Partition optimization: By exchanging boundary nodes, optimization is performed with the minimum active network loss as the objective function:

[0098]

[0099] in, and are the active output and active load of the node respectively.

[0100] The market transaction module implements a two-stage market bidding based on a multi-agent game model to optimize the benefits of the virtual power plant; the details are as follows:

[0101] Assuming that the supply and demand entities do not fully understand the decision-making process and decision-making information of their competitors, the operation process of the electricity market is established as a Markov decision process in an uncertain environment. The Markov decision process model of this uncertain environment has various subject objects participating in the market, environmental objects, the subject's state set S and action set A, and rewards R. In transactions at a specific time scale, the subject determines its own state S at each discrete time step t by observing the information of environmental objects. t ∈S, select and execute action A according to a certain strategy t ∈A, after executing the action, the agent The state transition probability migrates to the new state S t+1 ∈S, get instant reward R t+1 ∈R and update the policy parameters, and finally generate or perform the contract to update the environment information.

[0102] In each state, the agent will adopt a behavioral strategy that maximizes the expected benefit, and the decision-making process is represented by the Bellman optimality equation.

[0103]

[0104] Where: V * (s) is the optimal state value function, which represents the cumulative maximum expected return that can be obtained in state s; Q * (s,a) is the optimal action value function, which represents the cumulative maximum expected reward that can be obtained by performing action a in state s; s is the state of the agent; a is the optimal action in the current state s; A represents the action space; a∈A, s∈S, S represents the state space; R(s,a) represents the immediate reward after performing the action; γ is the depreciation factor; P(s'|s,a) represents the probability of transitioning to the next state s'; V * (s') represents the optimal state value function of state s; s' represents the state after state s is transferred; s'∈S; π * (s) is the optimal strategy in state s;

[0105] The intelligent agent reinforcement learning decision model uses the Q learning algorithm to estimate the optimal action value function, that is, for each possible action-state pair (s, a), the action value function is defined as the Q-Value function, which gives the expected utility of taking action a in state s, and is used to estimate Q * (s,a), thus obtaining the optimal strategy π * .

[0106] The intelligent agent first initializes the Q-value lookup table through random strategies or historical learning information. In each state s, it adopts the ε-greedy strategy, performs random actions with a probability of ε, and performs the greedy strategy with a probability of 1-ε; that is, it finds an optimal strategy π based on the maximum Q-Value. * (s)∈A, after executing the strategy, the state s←s' is updated at the same time, and the Q value lookup table is updated as follows:

[0107]

[0108] In formula (22), α is the update step size; Q'(s,a) represents the updated action value function, which is the weighted result of the current Q value and the new experience; R(s,a) represents the direct benefit obtained after executing action a in state s; It represents the maximum Q value among all possible actions a' in the next state s', which represents the estimation of the optimal strategy in the future; Q(s',a') represents the maximum Q value that can be obtained by choosing the optimal action a' in the future starting from the new state s';

[0109] Bidding revenue optimization analysis based on historical market data and simulation environment:

[0110] In order to maximize profits and measure the risks caused by electricity price uncertainty, risk factors are introduced into the objective function of power generation enterprises.

[0111] maxf=(1-ρ)B+ρV CVaR (33);

[0112] In formula (23), maxf represents the maximum value of the optimization objective; ρ is the risk factor, ρ∈[0,1], and its value is positively correlated with the conservative degree of the power generation enterprise's bidding strategy; B is the expected profit value of the power generation enterprise in participating in the day-ahead energy and ancillary service market; V CVaR is the risk indicator; f is the optimization target.

[0113] The expected profit value B of a power generation enterprise is the difference between the revenue and cost in the market, where the revenue from participating in the market includes the revenue from selling electricity in the energy market. and the benefits of participating in secondary frequency modulation Cost includes unit cost and carbon emission trading costs

[0114]

[0115] In formula (24): S is the number of scenes; π s is the probability corresponding to scene s.

[0116] The revenue from electricity sales of power generation enterprises is:

[0117]

[0118] In formula (35): is the electricity price at time t in scenario s; p s,t is the winning bid power of the power generation enterprise at time t under scenario s; T is the total operating time;

[0119] Secondary frequency modulation benefits It includes two aspects: opportunity cost compensation and frequency modulation call benefits;

[0120]

[0121] In formula (36): is the compensation price given by the market operator at time t under scenario s; k is the declared capacity corresponding to the opportunity cost of the unit at time t under scenario s; s,t is the actual calling coefficient of the system for the unit at time t under scenario s.

[0122] The unit cost consists of two parts: start-up and shutdown cost and operating cost. SU 、c SD They are the unit start-up and shutdown costs, and the unit operating costs, respectively, using the quadratic function express.

[0123]

[0124] In formula (37): x s,t 、y s,t They are 0-1 variables describing the start and stop status of the unit. s,t =1,y s,t =0, when the machine stops x s,t =0,y s,t =1; is the total output of the power generation enterprise, and the calculation formula is:

[0125]

[0126] In formula (38): p s,t represents the total output at time t under scenario s; k s,t The actual calling function in time and space t under scenario s;

[0127] represents the secondary frequency modulation capacity at time t in scenario s.

[0128] The calculation formula for carbon emission rights trading costs is:

[0129]

[0130] In formula (39): is the carbon emission rights trading price; r G is the carbon emission intensity; R G is the carbon emission quota corresponding to the unit.

[0131] The constraints of the power generation enterprise joint bidding optimization model are as follows:

[0132] The unit start and stop constraints are:

[0133]

[0134] x s,t +y s,t ≤1 (42);

[0135] z s,t -z s,t-1 =x s,t -y s,t (43);

[0136] In the above formula: z s,t 、z s,i are the status of the unit at time t and time i respectively. When the value is 1, it means the unit is on, and when the value is 0, it means it is off; are the shortest start-up and shutdown time of the unit respectively. s,t-1 represents the operating status of the unit at time t-1 under scenario s; x s,t Indicates the startup status of the unit at time t in scenario s; y s,t Indicates the shutdown status of the unit at time t in scenario s; Represents the minimum continuous running time constraint;

[0137] The output constraints are:

[0138] P min z s,t ≤p s,t ≤P max z s,t -P t AGC (44);

[0139] In formula (44): P max 、P min are the upper and lower limits of the unit output respectively; P t AGC Reserve frequency modulation capacity for time t; p s,t represents the actual output of the generator set at time t under scenario s;

[0140] The climbing constraint is:

[0141]

[0142] In the above formula: p on、p off are respectively the starting and stopping climbing power; p ru 、p rd are the climbing power for upward and downward operation respectively; p s,t-1 represents the actual output of the generator set at time t-1 under scenario s; k s,t k represents the actual call coefficient of the frequency modulation capacity at time t under scenario s; s,t-1 represents the actual call coefficient of the frequency modulation capacity at time t-1 under scenario s; represents the FM capacity at time t-1 under scenario s;

[0143] The auxiliary variable constraints are:

[0144] x s,t ,y s,t ,z s,t ∈{0,1}(47);

[0145] The quotation curve is required to be monotonically increasing, and the constraints are shown in equations (48) to (48). When the conventional thermal power unit reserves the frequency regulation capacity, the minimum value of the increase is set to 15% of the unit's rated capacity, as shown in equation (50);

[0146]

[0147]

[0148] In the above formula: represents the market price of electricity at time t+1 under scenario s; represents the market price of electricity at time t under scenario s; p s,t+1 represents the actual output of the generator set at time t+1 under scenario s; P G is the rated capacity of the unit.

[0149] The core algorithm module uses a method that combines time series analysis with a deep learning model to optimize load forecasting accuracy and improve response speed.

[0150] The security management module builds a multi-party trust system through blockchain technology and dynamically adjusts resource transaction prices based on smart contracts to achieve transparency in resource management and credibility in resource transactions.

[0151] The present invention provides a virtual power plant optimization system and method, with the following technical effects:

[0152] 1) The virtual power plant optimization system of the present invention realizes efficient resource management and maximization of market benefits through the collaborative work of modules such as virtualization platform, dynamic resource aggregation, market transaction, core algorithm and security management.

[0153] 2) The virtualization platform module of the virtual power plant optimization system of the present invention realizes the virtualization and centralized management of resources by integrating distributed energy, energy storage equipment, electric vehicles and flexible loads, and uses edge computing technology to reduce management pressure and improve response speed.

[0154] 3) The dynamic resource aggregation module of the virtual power plant optimization system of the present invention adopts a data-driven clustering algorithm, combines resource responsiveness and market supply and demand dynamics, optimizes resource integration efficiency, and improves peak-shaving and frequency regulation performance.

[0155] 4) The market transaction module of the virtual power plant optimization system of the present invention implements a two-stage market bidding based on a multi-agent game model, analyzes the uncertainty of competitors' quotations through multi-scenario simulation technology, optimizes dynamic trading strategies, and realizes the maximization of spot market returns and controllability of risks.

[0156] 5) The core algorithm modules of the virtual power plant optimization system include load forecasting, power generation forecasting, and dynamic control optimization algorithms, leveraging deep learning technology to improve forecast accuracy and response speed. The security management module builds a multi-party trust system based on blockchain technology, dynamically adjusting resource transaction prices through smart contracts to ensure transparency and trustworthiness in resource management and transactions.

[0157] 6) Through modular design and a unified interface, this invention achieves efficient system integration and flexible expansion, making it applicable to a variety of application scenarios. Its innovation lies in combining artificial intelligence, edge computing, and blockchain technologies with the optimized management of virtual power plants, significantly improving the integration efficiency of distributed resources, the flexibility of market transactions, and the security of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0158] The present invention will be further described below with reference to the accompanying drawings and examples:

[0159] Figure 1 This is the algorithm flow chart of the two-layer game model;

[0160] Figure 2 Flowchart for partitioning virtual power plants based on clustering characteristics;

[0161] Figure 3 This is a diagram of the reinforcement learning decision process;

[0162] Figure 4 This is a decision-making flowchart for the intelligent agent. DETAILED DESCRIPTION

[0163] The present invention provides a virtual power plant optimization system, comprising the following modules:

[0164] A virtualization platform module is used to integrate distributed energy, energy storage equipment, electric vehicles, flexible loads and other resources to achieve virtualization and centralized management of resources;

[0165] Dynamic resource aggregation module, used to dynamically match resource characteristics with power system requirements;

[0166] Market trading module, used to optimize the benefits of virtual power plants;

[0167] Core algorithm module, used to optimize load forecasting accuracy;

[0168] Security management module, used to dynamically adjust resource transaction prices.

[0169] The virtualization platform module utilizes edge computing units to be distributed and deployed on resource access nodes to reduce the pressure of centralized management and improve resource response speed; the details are as follows:

[0170] To achieve the coordinated and optimized operation of distributed resources in a virtual power plant (VPP), it is necessary to establish the output and operating characteristics of each distributed resource, analyze them, and establish the corresponding mathematical model:

[0171] 1) The generation of electrical energy through photovoltaics is known as the photovoltaic effect. This process involves using solar cells to convert sunlight into electrical energy. Photovoltaic power generation is modeled as follows:

[0172]

[0173] In formula (1), P PV Represents photovoltaic output power; f PV is the loss factor of the photovoltaic panel; P PV,cap is the output power of the photovoltaic array under standard test conditions; G T,STC is the light intensity under standard test conditions; G T is the actual light intensity; α P is the power temperature coefficient; T cell Indicates the actual operating temperature of the photovoltaic cell; T cell,STC Indicates the temperature under standard test conditions;

[0174] 2) The wind power generation WT model is as follows:

[0175]

[0176] In formula (2), P WT Indicates the actual output power of the wind turbine; v i 、v v Respectively represent the cut-in and cut-out wind speeds; v represents the current wind speed; v r Indicates the cut-out wind speed; v o Indicates rated wind speed;

[0177] 3) The energy storage system charging and discharging model is as follows:

[0178]

[0179] In formula (3), SOC t Indicates the state of charge of the energy storage system at time t; SOC t-1 represents the state of charge of the energy storage system at time t-1; λ b Indicates the self-discharge rate of the energy storage system; Indicates the charging efficiency of the energy storage system; Indicates the discharge efficiency of the energy storage system; represents the charging power of the energy storage system at t-1; represents the discharge power of the energy storage system at t-1; △t represents the time interval between two adjacent time points; SOC min Indicates the lower limit of the state of charge; SOC max Represents the upper limit of the state of charge.

[0180] To achieve safe and stable operation of the energy storage system, it is divided into five charging and discharging action intervals:

[0181] Interval 1: When 0≤SOC t <0.2SOC max When the charge power range is P d =0,P d Indicates the discharge power of the energy storage system in interval 1; when the energy storage system is extremely low (SOC t <20%), discharging is prohibited for the sake of protecting battery life and system safety. This can prevent battery performance degradation or damage caused by excessive discharge, and ensure that the energy storage system retains a minimum amount of power to cope with sudden demand or maintain system stability;

[0182] Interval 2: When 0.2SOC max ≤SOC t <0.35SOC max When the charging power range is:

[0183] 0.2SOC max ≤SOC t <0.35SOC max , the discharge power range is Interval 3: When 0.35SOC max ≤SOC t <0.65SOC max When the charging power range is P C , Interval 4: When 0.65SOC max ≤SOC t <0.8SOC maxWhen the charging power range is Discharge power range is Interval 5: When 0.8SOC max ≤SOC t <SOC max When the charging power is P C =0, which is a non-rechargeable state; the discharge power range is

[0184] At the same time, it is ensured that the state of charge of the energy storage system is not less than the minimum state of charge and not greater than the maximum state of charge. Then, the energy storage system ESS is divided into free trading energy storage and backup energy storage.

[0185] For multiple virtual power plants (VPPs), a two-layer VPP game model is established. The upper-layer VPP game model takes maximizing the VPP revenue as its objective function, while the lower-layer VPP game model takes minimizing the total power generation cost as its objective function, thereby modeling the internal game of the VPP.

[0186] (I): The upper-level VPP game model maximizes the benefits of the virtual power plant VPP and establishes an objective function, which is expressed as follows:

[0187]

[0188] In formula (4), represents the revenue obtained by the i-th VPP from selling electricity to the load at time t; The gains of the interaction of the i-th VPP with a wider grid are shown; represents the revenue obtained from selling electricity from the i-th VPP to other VPPs;

[0189] For any virtual power plant (VPP), it is necessary to ensure that the system maintains power balance at any point during operation. The sum of the power of all power sources in the VPP interacting with other VPPs and the sum of the power interacting with the main grid equals the load demand. The expression is as follows:

[0190]

[0191] In formula (5), Indicates the load demand power; Indicates photovoltaic output power; Indicates the output power of the wind turbine; Indicates the output power of standby energy storage; Indicates the free energy storage output power; It is the power input from the large power grid to the VPP.

[0192] 1) Wind power and photovoltaic output constraints:

[0193] The production capacity of wind and photovoltaic installations is inherently limited by their maximum output ratings:

[0194]

[0195] In formula (6), represents the actual photovoltaic output power of the i-th virtual power plant at time t; Represents the maximum output power of the PV; represents the actual output power of the wind turbine of the i-th VPP at time t; The maximum power output of the fan.

[0196] 2) The electricity price traded between VPPs should not be higher than the price of electricity purchased from the grid, nor lower than the price of electricity sold from the grid:

[0197] λ da,t ≤λ i,t ≤λ' da,t (7);

[0198] In formula (7), λ i,t is the transaction price of VPP; da,t is the price of electricity sold by VPP to the grid; da,t The price at which VPP purchases electricity from the grid.

[0199] 3) There is a certain amount of power trading between VPPs and the main grid, and between multiple VPPs with tie-line power constraints:

[0200]

[0201] In formula (8), represents the energy of interaction between the i-th VPP and the large grid; is the maximum interaction power allowed by the contact line between the first VPP and the grid; P ij,t is the energy of interaction between the i-th VPP and the j-th VPP; is the maximum interaction power allowed by the contact line between the i-th VPP and the j-th VPP.

[0202] (II): With the objective function of reducing the overall operating costs of VPP, a lower-level VPP game model is constructed:

[0203]

[0204] In formula (9), represents the total power generation cost of the i-th VPP at time t; represents the photovoltaic power generation cost in the i-th VPP; represents the cost of electricity generated by wind turbines in the i-th VPP; represents the operating cost of the energy storage system in the i-th VPP; N PV represents the number of photovoltaic cells in the i-th VPP; N W N represents the number of wind turbines in the i-th VPP; b represents the amount of energy storage in the i-th VPP; represents the minimum total power generation cost of VPP at time t; represents the photovoltaic power generation cost of the i-th virtual power plant; represents the wind power generation cost of the i-th virtual power plant; represents the cost of the energy storage system in the i-th VPP.

[0205] The lower-level VPP game model needs to meet power balance constraints, energy storage charging and discharging power constraints, and wind and solar curtailment constraints. The energy storage system constrains that the energy storage charging and discharging power must be less than the maximum output power and maximum discharge power of the energy storage system.

[0206]

[0207] In formula (10), Indicates the maximum charging power; represents the net charge and discharge power of the energy storage system in the i-th VPP at time t; Indicates the maximum discharge power;

[0208] Abandoning wind power generation WT and photovoltaic power generation output constraints:

[0209]

[0210] In formula (11), ρ pv,t and ρ w,t They represent the constraint coefficients of photovoltaic power generation and wind power generation WT reduction at time t; △P w,t represents the wind curtailment of the ith VPP at time t; △P pv,t represents the amount of abandoned light of the i-th VPP at time t;

[0211] The multiple VPPs in a two-layer VPP game model require consideration of both the upper-layer VPP game model and the lower-layer VPP game model, and the two interact with each other. A two-layer particle swarm optimization method is used to solve the two-layer VPP game model containing multiple VPPs.

[0212] In the two-layer particle swarm optimization method, the upper-layer VPP game model is solved using the outer-layer particle swarm method, where the particle position represents the directly traded electricity of a single VPP, and the fitness value represents the sum of the economic benefits of all VPPs. The lower-layer VPP game model is solved using the inner-layer particle swarm method, where the particle position represents the output of each unit, user load electricity demand, etc., and the fitness value represents the VPP operating cost.

[0213] An improved two-layer particle swarm optimization method is used to solve the two-layer VPP game model for virtual power plants (VPPs). This method iteratively couples the outer layer PSO (solving the upper layer non-cooperative game) with the inner layer PSO (solving the lower layer cost optimization) to ultimately converge to the game equilibrium solution. The following describes the model formula and algorithm flow in a step-by-step manner:

[0214] In the structure and coupling relationship of the two-layer VPP game model, the goal of the upper-layer VPP game model (non-cooperative game) is to maximize the benefits of each VPP:

[0215]

[0216] in: Represents the revenue from electricity sales; Indicates the benefits of interacting with the main network; It indicates that the transaction income between VPPs depends on the power generation cost optimized at the lower layer.

[0217] Decision variable: transaction price λ between VPPs i,t and interactive power P g,t , see formula (7) and formula (8);

[0218] The cost minimization objective of the lower-level VPP game model is to minimize the internal power generation cost of the VPP, as described in Equation (13);

[0219]

[0220] Decision variables: output P of each unit pv,t ,P w,t ,P b,t and the amount of wind and solar power curtailment △P w,t ,△P pv,t ;

[0221] Coupling mechanism: The upper-layer VPP game model's revenue calculation requires the lower-layer VPP game model's cost as input, while the lower-layer VPP game model optimization must accept the transaction price and power constraints assigned by the upper-layer VPP game model, such as the power balance in formula (5). The two achieve dynamic equilibrium through iterative feedback of the two-layer PSO.

[0222] The algorithm flow is as follows Figure 1 Said method comprises the following steps:

[0223] Initialization: predict wind and solar power output (Equation (1)-Equation (2)) and load demand; set parameters such as the population size and number of iterations of the outer PSO (upper game) and inner PSO (lower optimization).

[0224] Outer PSO (upper game): particle position, representing the trading strategy λ of each VPP i,t ,Pg,t ; Fitness function, total benefits of all VPPs The lower layer PSO needs to be called to calculate the cost; constraint processing is to ensure that the electricity price and power consumption meet the constraints of formula 7-8 (such as ).

[0225] Inner PSO (lower layer optimization): particle position, representing the VPP internal unit output P pv,t ,P w,t ,P b,t and abandoned power (△P w,t ,△P pv,t ; Fitness function, total power generation cost The ESS and curtailment constraints of the following equations (10) and (11) must be satisfied.

[0226] Two-layer iteration and convergence: After the outer PSO generates a trading strategy, the inner PSO calculates the corresponding cost and returns the profit value to the outer layer. The particle speed is adjusted by the inertia weight and learning factor (reference parameters: inertia factor 0.8, learning factor 0.2), gradually approaching the Nash equilibrium.

[0227] Termination condition: the rate of change of benefit and cost is less than the threshold, or the maximum number of iterations (such as 50) is reached.

[0228] The final balance is achieved through continuous iteration between the two layers.

[0229] The dynamic resource aggregation module combines the responsiveness of distributed resources with market supply and demand dynamics, and uses a data-driven clustering algorithm to dynamically match resource characteristics with power system requirements to optimize peak and frequency regulation performance.

[0230] Distributed resources can be divided into energy storage and power supply. Energy storage mainly includes distributed energy storage and electric vehicles, which have energy accumulation. Power supply includes full control power supply type (FCP) represented by micro gas turbines, and half control power supply type (HCP) represented by distributed wind power and distributed photovoltaics. The power flow analysis of power network topology is based on the DistFlow power flow model.

[0231] The constraints of the fully controlled power supply (FCP) mainly consider power, capacity, and ramp constraints, namely:

[0232]

[0233] In formula (14): and are the active and reactive outputs of FCPi at time t respectively; and are the minimum and maximum active output of FCPi respectively; is the rated capacity of FCPi; and They are the upper and lower limits of FCPi’s climbing ability, respectively; represents the active and reactive output of FCPi at time t+1;

[0234] Semi-controlled power supply HCPs improve output flexibility through power reduction. This paper constructs a semi-controlled power supply HCP resource output model that takes into account the uncertainty of wind and solar power output, and uses the Gaussian mixture model (GMM) to transform the chance constraints containing random variables into deterministic constraints to achieve the solution of the virtual power plant feasible region (VFR). The virtual power plant feasible region (VFR) refers to the set of all possible combinations of active power (P) and reactive power (Q) that a virtual power plant (VPP) can provide at the point of common coupling (PCC) while satisfying internal resource constraints and grid operating conditions.

[0235] The resource output model of the semi-controlled power supply HCP is:

[0236]

[0237] In formula (13): and are the active and reactive outputs of HCPi at time t respectively; and are the minimum and maximum power factor angles of HCPi respectively; represents the maximum active power output limit of FCPi at time t;

[0238] Collaborative optimization of multiple VPPs for economic optimization:

[0239] Based on this optimization problem, the present invention designs a binary consistency algorithm that solves the problem involving active-reactive binary consistency variables. The difference between the designed binary consistency algorithm and the unary consistency algorithm is that the optimization model solved by the binary consistency algorithm contains two types of decision variables, which correspond to the binary consistency variables in the binary consistency algorithm. When both binary consistency variables converge, the system completes convergence; the details are as follows:

[0240] The binary consensus algorithm uses two consensus variables, λ and w, to coordinate the allocation of active and reactive power, respectively, minimizing total cost while satisfying grid constraints. These two consensus variables are introduced: λ represents the marginal cost of active power (similar to the price of electricity); and w represents the marginal cost of reactive power. Distributed iterations are performed on each VPP, updating λ and w based on local information. Ultimately, all VPPs reach consensus on λ and w, and the system converges to the optimal solution.

[0241] The objective function of the economic optimization model of multiple VPPs is to optimize the total cost of each aggregated VPP. The total cost of VPPi includes the active power cost. and reactive power costs Right now:

[0242]

[0243] In formula (16): P i and Q i are the active and reactive power inputs of VPPs, respectively; I is the number of VPPs included in the distribution network; represents the minimum total cost of all VPPs; C represents the total cost of all VPPs;

[0244] Among them, active cost The piecewise quadratic function aggregation cost model is adopted, which is the superposition of the costs of various resources within the VPP. The cost functions of various internal resources are shown in formula (17);

[0245] Reactive power cost This includes fixed costs and opportunity costs. Fixed costs are the unit time costs calculated based on the planned service life of the equipment, while opportunity costs are the portion of the costs lost in active power generation due to the VPP providing reactive power services. Because various power generation resources within the VPP are subject to corresponding power constraints during actual operation, reactive power output reaching a certain level will inevitably affect active power generation revenue.

[0246] Since various power generation resources within a VPP have corresponding power constraints during actual operation, when reactive power output reaches a certain level, it will inevitably affect the active power generation revenue. The specific expressions of the active power cost model and reactive power cost model of the VPP are:

[0247]

[0248] In the above formula: are the costs of FCP, HCP and energy storage resources respectively; and is the cost coefficient of FCPm; and is the cost coefficient of HCPn; and is the cost coefficient of energy storage resource l; α i,n , β i,n and γ i,n are the secondary, primary and constant cost coefficients of VPPi output active power in segment n, usually n∈[2,5], which is determined by the internal resource type of VPP; and P l E are the active powers of FCPm, HCPn and energy storage resource l respectively; Di,n is the boundary value of VPPi output active power; Q,i is the discounted fixed cost; is the cost coefficient; P i,max and S i,max are the maximum active power and apparent power of VPPi respectively;

[0249] α i represents the secondary cost coefficient of VPPi output active power; β i represents the primary cost coefficient of VPPi output active power; γ i represents the constant cost coefficient of VPPi output active power; P i Indicates the active power of VPPi; Q i Indicates the reactive power of VPPi; D i1 Indicates VPP i The minimum technical output (usually 0 or the minimum operating point of the resource); D i2 Indicates the output limit or operation mode switching point of the first type of resource (such as the maximum adjustable power of photovoltaic). i3 Indicates the output upper limit of the second type of resource or the maximum output allowed by the system (such as the rated capacity of a gas turbine).

[0250] Constraints:

[0251]

[0252] Where: P Pr and Q Pr are the total active and reactive power requirements of the system respectively; P Loss and Q Loss are the total active and reactive losses of the system respectively; △P i and △Q i are the active and reactive power increments of VPP respectively; U j is the voltage at node j; U min and U max are the minimum and maximum values of the node voltage respectively; R ij and X ij are the resistance and reactance of the ij line respectively; P i N and are the injected active and reactive power of node i respectively; N is the total number of nodes; Ω VFR represents the feasible region of VPPi; I represents the total number of virtual power plants in the distribution network system; represents the active voltage product of N nodes j; represents the reactive voltage product of N nodes j;

[0253] The data-driven clustering algorithm uses machine learning techniques, including but not limited to K-means clustering and Gaussian mixture models. The details are as follows:

[0254] The proposed method combines cluster analysis based on load current fields (data-driven) and electrical distances (model-driven), achieving efficient grid partitioning through clear physical meaning and mathematical formulas. Its core approach is to determine cluster centers based on load current fields and then perform clustering based on electrical distances, avoiding the repeated iterations of traditional clustering algorithms.

[0255] (1). Calculation of load current field and virtual potential:

[0256] The virtual potential φ of the load current at node i is defined as i :

[0257]

[0258] Among them, Z ij is the matrix Z u Elements, I j is the load current at node j. This formula reflects the distribution characteristics of node voltage affected by load current.

[0259] Describe the relationship between the voltage drop at node i and the load current:

[0260] △U i =Z ij I j (twenty four);

[0261] Among them, △U i represents the voltage drop at node i; Equation (24) is used to equivalently represent the voltage impact between nodes.

[0262] (2) Definition of electrical distance:

[0263] The electrical distance matrix D between nodes is defined based on the inverse matrix of the Jacobian matrix J, namely the susceptance matrix B;

[0264] D ij =|(J -1 ) ii +(J -1 ) jj -2(J -1 ) ij | (25);

[0265] Formula (25) is used to quantify the degree of electrical connection between nodes and is the basis of cluster analysis.

[0266] (3) Determination of cluster center nodes

[0267] The cluster center node is determined by the maximum point of the local potential value of the load current field:

[0268] For node i, if its potential value φ i If the potential value of i is greater than that of all directly connected nodes, then i is the point with the maximum local potential value. The number of points with the maximum local potential value is the number of partitions, and these points serve as cluster center nodes.

[0269] (4) Cluster analysis process:

[0270] Clustering method: Calculate the electrical distance D between each node and the cluster center node ij , divide the nodes into the partitions to which the cluster centers with the smallest distance belong.

[0271] Partition optimization: By exchanging boundary nodes, optimization is performed with the minimum active network loss as the objective function:

[0272]

[0273] in, and are the active output and active load of the node respectively.

[0274] (5) Constraints:

[0275] Including node voltage constraints, branch current constraints, and short-circuit current constraints, to ensure that the partitioning results meet the safety and economy of power grid operation.

[0276] A power grid optimization partitioning method that combines load current field and cluster analysis, determines the number of partitions and cluster centers through the load current field, uses electrical distance to perform clustering, and finally optimizes the boundary node division with active network loss as the optimization target to achieve power grid optimization partitioning. The process of virtual power plant partitioning based on clustering characteristics is as follows Figure 2 As shown, the following steps are included:

[0277] Step 1) Form the node admittance matrix Y from the grid data:

[0278]

[0279] in: is the load current equivalent factor matrix, whose i-th column characterizes the load current I of node i i The impact on the voltage of each node in the network is related to the load current I i The voltage effect on each node is related to its distance from node i, and decreases with increasing distance.

[0280] Step 2) Calculate the load current virtual potential βi according to the node admittance matrix Y:

[0281] The node voltage influence characterized by the i-th column is called the load current virtual field, and the elements of each row of the i-th column are called the virtual potential of each node in the load current virtual field of node i. The load current virtual potential of node i for:

[0282]

[0283] Among them, z ii and z ij The matrix Z LL The element at row i, column i and column j.

[0284] Step 3) Determine the number of clusters and partitions, with the point with the maximum virtual potential of the node current as the cluster point and partition number:

[0285] The admittance matrix is formed from the power grid data, and the load current virtual potential of each node is calculated from the load current field.

[0286] Determine the number of partitions and the central node of the cluster. Let V i is the set of P points directly connected to node i. If for any node J∈V i ,have Then node i is the node with the local maximum potential value. The number of nodes with the local maximum potential value is the number of partitions, and each node with the local maximum potential value is the cluster center node.

[0287] Step 4) Clustering is performed based on the number of partitions to obtain preliminary partitions, and whether the constraints are met is determined. If the constraints are met, proceed to step 5). If the constraints are not met, generate the node admittance matrix after partitioning and return to step 2).

[0288] Perform cluster analysis on each node. The cluster partitioning method is as follows: calculate the electrical distance between each node and each cluster center node based on the electrical distance matrix, compare the distance between the node and each cluster center node, and based on the principle of minimum distance, classify each node and the center node with the smallest distance to it into one category to obtain the clustering result. Each cluster is the preliminary partitioning result. Calculate the power flow and short-circuit current after partitioning. If the constraints are not met, disconnect the partition tie line to form the admittance matrix after partitioning. Calculate the load current virtual potential of each node, and proceed to step 2) and step 3) until a partition that meets the conditions is obtained.

[0289] Step 5) Optimize the node boundaries after partitioning to obtain the final partitioning result.

[0290] Re-partition the boundary nodes after the preliminary partitioning. Let a and b be a pair of nodes connected by a line before partitioning. After partitioning, they belong to the adjacent c and d partitions respectively. Exchange these two boundary nodes, assign node a to partition d, calculate the power flow and short-circuit current, and retain the results if the constraints are met and the network loss is reduced. If not, return. Assign node b to partition c, calculate the power flow and short-circuit current, and retain the results if the network loss is reduced. If not, return. Continue to optimize the remaining pairs of boundary nodes using the same method until the optimization of all boundary nodes is completed, and finally obtain the partition result with the minimum network loss.

[0291] The market transaction module implements a two-stage market bidding based on a multi-agent game model to optimize the benefits of the virtual power plant; the details are as follows:

[0292] The market transaction module, combined with multi-scenario simulation technology, analyzes the uncertainty of competitors' quotations and optimizes dynamic transaction strategies.

[0293] By extracting the main components of the new power system, it can be divided into two parts: subject objects and environmental objects, which can respectively describe power market transactions and system energy flows.

[0294] The subject objects include supply subjects, demand subjects, and integrated supply and demand subjects, representing generators, users, and prosumers in the system, respectively. They also include aggregation subjects such as microgrids, distributed parks, load aggregators, and power sales companies, and management subjects such as power trading institutions and power dispatching agencies.

[0295] Environmental objects consist of two components: equipment resources and electricity contracts, which describe the physical attributes and market transaction status of entities, respectively. Equipment resources include generators, loads, energy storage devices, and transmission and distribution networks. The electricity contract serves as a medium for interaction between entities and between entities and the environment, recording transaction results and serving as the basis for electricity transfer between equipment resources.

[0296] In this study, assuming that the supply and demand entities do not fully understand the decision-making process and decision-making information of their competitors, the operation process of the electricity market is established as a Markov decision process in an uncertain environment. The Markov decision process model in this uncertain environment has various subject objects participating in the market, environmental objects, the subject's state set S and action set A, and rewards R. In transactions at a specific time scale, the subject determines its own state S at each discrete time step t by observing the information of environmental objects. t ∈S, select and execute action A according to a certain strategy t ∈A, after executing the action, the agent The state transition probability migrates to the new state S t+1 ∈S, get instant reward R t+1∈R and update the policy parameters, and finally update the environment information when necessary (generating or fulfilling the contract), such as Figure 3 shown.

[0297] In each state, the agent will adopt a behavioral strategy that maximizes the expected benefit, and the decision-making process can be represented by the Bellman optimality equation.

[0298]

[0299] Where: V * (s) is the optimal state value function, which represents the cumulative maximum expected return that can be obtained in state s; Q * (s,a) is the optimal action value function, which represents the cumulative maximum expected reward that can be obtained by performing action a in state s; s is the state of the agent; a is the optimal action in the current state s; A represents the action space; a∈A, s∈S, S represents the state space; R(s,a) represents the immediate reward after performing the action; γ is the depreciation factor; P(s'|s,a) represents the probability of transitioning to the next state s'; V * (s') represents the optimal state value function of state s; s' represents the state after state s is transferred; s'∈S; π * (s) is the optimal strategy in state s;

[0300] The intelligent agent reinforcement learning decision model uses the Q learning algorithm to estimate the optimal action value function, that is, for each possible action-state pair (s, a), the action value function is defined as the Q-Value function, which gives the expected utility of taking action a in state s, and is used to estimate Q * (s,a), thus obtaining the optimal strategy π * .

[0301] Intelligent agents are decision-making entities with learning capabilities in the power market, including generators, users, aggregators, managers, etc. The intelligent agent first initializes the Q-value lookup table through random strategies or historical learning information. In each state s, it adopts an ε-greedy strategy, performs random actions with a probability of ε, and executes a greedy strategy with a probability of 1-ε; that is, it finds an optimal strategy π based on the maximum Q-Value. * (s)∈A, after executing the strategy, the state s←s' is updated at the same time, and the Q value lookup table is updated as follows:

[0302]

[0303] In formula (22), α is the update step size; Q'(s,a) represents the updated action value function, which is the weighted result of the current Q value and the new experience; R(s,a) represents the direct benefit obtained after executing action a in state s; It represents the maximum Q value among all possible actions a' in the next state s', which represents the estimation of the optimal strategy in the future; Q(s',a') represents the maximum Q value that can be obtained by choosing the optimal action a' in the future starting from the new state s';

[0304] The decision-making process of the intelligent agent is as follows Figure 4 shown.

[0305] The multi-scenario simulation technology includes bidding revenue optimization analysis based on historical market data and simulation environment;

[0306] In order to maximize profits and measure the risks caused by electricity price uncertainty, risk factors are introduced into the objective function of power generation enterprises.

[0307] maxf=(1-ρ)B+ρV CVaR (33);

[0308] In formula (23), maxf represents the maximum value of the optimization objective; ρ is the risk factor, ρ∈[0,1], and its value is positively correlated with the conservative degree of the power generation enterprise's bidding strategy; B is the expected profit value of the power generation enterprise in participating in the day-ahead energy and ancillary service market; V CVaR is the risk indicator; f is the optimization target.

[0309] The expected profit value B of a power generation enterprise is the difference between the revenue and cost in the market, where the revenue from participating in the market includes the revenue from selling electricity in the energy market. and the benefits of participating in secondary frequency modulation Cost includes unit cost and carbon emission trading costs

[0310]

[0311] In formula (24): S is the number of scenes; π s is the probability corresponding to scene s.

[0312] The revenue from electricity sales of power generation enterprises is:

[0313]

[0314] In formula (35): is the electricity price at time t in scenario s; p s,t is the winning bid power of the power generation enterprise at time t under scenario s; T is the total operating time, which is generally 24 hours.

[0315] Secondary frequency modulation benefits It includes two aspects: opportunity cost compensation and frequency regulation benefit. Among them, the opportunity cost is caused by the power generation enterprise reserving a part of the frequency regulation capacity. This part of electricity cannot benefit from the electricity market, and the market operator gives corresponding compensation for this part. Only when the system actually uses the frequency regulation capacity of the unit will the use benefit be generated.

[0316]

[0317] In formula (36): is the compensation price given by the market operator at time t under scenario s; k is the declared capacity corresponding to the opportunity cost of the unit at time t under scenario s; s,t is the actual calling coefficient of the system for the unit at time t under scenario s.

[0318] The unit cost consists of two parts: start-up and shutdown cost and operating cost. SU 、c SD They are the unit start-up and shutdown costs, and the unit operating costs, respectively, using the quadratic function express.

[0319]

[0320] In formula (37): x s,t 、y s,t They are 0-1 variables describing the start and stop status of the unit. s,t =1,y s,t =0, when the machine stops x s,t =0,y s,t =1; is the total output of the power generation enterprise, and the calculation formula is:

[0321]

[0322] In formula (38): p s,t represents the total output at time t under scenario s; k s,t The actual calling function in time and space t under scenario s;

[0323] represents the secondary frequency modulation capacity at time t in scenario s.

[0324] The carbon emission rights trading cost is the cost that a power generation enterprise must pay to purchase excess carbon emission rights when the actual carbon emissions generated by the power generation on that day exceed the carbon emission quota owned by the enterprise. The calculation formula is:

[0325]

[0326] In formula (39): is the carbon emission rights trading price; r G is the carbon emission intensity; R G is the carbon emission quota corresponding to the unit.

[0327] The constraints of the power generation enterprise joint bidding optimization model are as follows:

[0328] The unit start and stop constraints are:

[0329]

[0330] x s,t +y s,t ≤1 (42);

[0331] z s,t -z s,t-1 =x s,t -y s,t (43);

[0332] In the above formula: z s,t 、z s,i are the status of the unit at time t and time i respectively. When the value is 1, it means the unit is on, and when the value is 0, it means it is off; are the shortest start-up and shutdown time of the unit respectively. s,t-1 represents the operating status of the unit at time t-1 under scenario s; x s,t Indicates the startup status of the unit at time t in scenario s; y s,t Indicates the shutdown status of the unit at time t in scenario s; represents the minimum continuous running time constraint;

[0333] The output constraints are:

[0334] P min z s,t ≤p s,t ≤P max z s,t -P t AGC (44);

[0335] In formula (44): P max 、P min are the upper and lower limits of the unit output respectively; P t AGC Reserve frequency modulation capacity for time t; p s,t represents the actual output of the generator set at time t under scenario s;

[0336] The climbing constraint is:

[0337]

[0338] In the above formula: p on、p off are respectively the starting and stopping climbing power; p ru 、p rd are the climbing power for upward and downward operation respectively; p s,t-1 represents the actual output of the generator set at time t-1 under scenario s; k s,t k represents the actual call coefficient of the frequency modulation capacity at time t under scenario s; s,t-1 represents the actual call coefficient of the frequency modulation capacity at time t-1 under scenario s; represents the FM capacity at time t-1 under scenario s;

[0339] The auxiliary variable constraints are:

[0340] x s,t ,y s,t ,z s,t ∈{0,1}(47);

[0341] The quotation curve is required to be monotonically increasing, and the constraints are shown in equations (48) to (48). When the conventional thermal power unit reserves the frequency regulation capacity, the minimum value of the increase is set to 15% of the unit's rated capacity, as shown in equation (50);

[0342]

[0343] In the above formula: represents the market price of electricity at time t+1 under scenario s; represents the market price of electricity at time t under scenario s; p s,t+1 represents the actual output of the generator set at time t+1 under scenario s; P G is the rated capacity of the unit.

[0344] The core algorithm module uses a method that combines time series analysis with a deep learning model to optimize load forecasting accuracy and improve response speed.

[0345] The security management module builds a multi-party trust system through blockchain technology and dynamically adjusts resource transaction prices based on smart contracts to achieve transparency in resource management and credibility in resource transactions.

[0346] The present invention provides a virtual power plant optimization system that improves the management efficiency and real-time response capability of distributed resources by utilizing artificial intelligence and edge computing technologies; introduces a franchise incentive mechanism and a data-driven clustering algorithm to optimize resource aggregation planning and dynamic adjustment strategies; and constructs a two-layer random game model of virtual power plants and multi-microgrids to maximize spot market returns and controllable risks.

[0347] The virtual power plant optimization system of the present invention effectively solves the technical problems of low distributed energy integration efficiency, delayed system response and insufficient market participation capacity in the existing power system through the collaborative work of multiple modules.

[0348] The virtualization platform module integrates distributed energy resources, energy storage devices, and flexible loads to establish a system framework for resource virtualization and centralized management. The introduction of edge computing units significantly reduces the pressure of centralized management and improves resource responsiveness. The dynamic resource aggregation module uses a data-driven clustering algorithm to precisely match the responsiveness of distributed resources with the dynamic demands of the power grid, thereby optimizing resource peak and frequency regulation performance.

[0349] The core algorithm module utilizes deep learning models for load and power generation forecasting, including long short-term memory (LSTM) networks and gated recurrent units (GRUs), reducing forecast error to less than 10%. Furthermore, dynamic control optimization utilizes distributed robust optimization technology, significantly improving the system's responsiveness to emergencies.

[0350] The market trading module utilizes a two-stage bidding strategy based on the Stackelberg master-slave game model to maximize virtual power plant revenue in the upper-layer VPP game model. Simultaneously, the lower-layer VPP game model uses simulation technology to optimize market clearing costs and address the uncertainty of competitor bids. The application of blockchain technology not only enhances transparency in resource management and transaction processes but also dynamically adjusts transaction prices through smart contracts, establishing a reliable multi-party trust system.

[0351] The operating process of a virtual power plant optimization system of the present invention embodies the organic combination of resource virtualization, prediction and optimization, dynamic trading strategy and security management. Through efficient collaboration between modules, it provides a flexible, stable and economical solution for the operation of a high-proportion new energy power system.

[0352] In one possible implementation variation, the hardware component of the virtualization platform module can utilize a variety of computing devices to accommodate different operating environments and budget requirements, such as high-performance microprocessors or dedicated accelerator chips. This variation not only reduces system costs but also improves operational efficiency in specific environments.

[0353] The virtual power plant's system architecture can also be adjusted based on the specific application scenario. For example, in areas with densely distributed resources, the virtualization platform module can adopt a multi-layer distributed architecture, dividing management nodes into regional and global layers to optimize resource integration efficiency and system responsiveness. In areas with sparse resources, a single centralized architecture can be used to reduce system deployment and maintenance costs.

[0354] In the dynamic resource aggregation module, in addition to dynamic adjustment methods based on clustering algorithms, heuristic optimization methods such as genetic algorithms and particle swarm optimization can also be employed. These methods are suitable for large-scale distributed resource scenarios and can further improve resource responsiveness and optimization efficiency. Furthermore, the Stackelberg game model in the market transaction module can be replaced with a reinforcement learning model, enabling the system to more quickly adapt to complex market changes.

[0355] The core algorithm module offers flexible parameter settings. For example, the load forecasting model's time window can be dynamically adjusted from 1 to 48 hours to meet short-term or long-term forecasting needs. The clustering algorithm's number of clusters and threshold parameters can also be dynamically optimized based on the distribution characteristics of distributed resources and grid requirements to improve peak and frequency regulation performance.

[0356] In terms of connection and data transmission methods, edge computing units can choose from a variety of connection methods such as optical fiber, 5G network or satellite communication to adapt to different geographical and environmental conditions, while ensuring the stability and low latency of data transmission.

[0357] The virtual power plant optimization system of the present invention also offers flexibility in its installation. In actual deployment, the modular design allows for rapid deployment and adaptability to power systems of varying sizes. In experimental or testing scenarios, portable equipment can be used to reduce initial pilot costs and complexity.

[0358] The flexibility of the virtual power plant optimization system in its application scenarios also opens up the possibility for expansion. For example, virtual power plant technology can be further extended to areas such as microgrid management, smart building energy regulation, and emergency power restoration, providing technical support for a wider range of energy optimization. Regarding algorithm replacement, deep learning models such as convolutional neural networks or attention mechanism models can be used to replace existing prediction models to further improve the prediction accuracy of complex data patterns.

[0359] Furthermore, the adaptability of the virtual power plant optimization system in extreme environments can be further enhanced through industrial-grade hardware and lightweight software design, enabling it to operate normally in conditions such as high temperature, high humidity, and extreme cold. In terms of legal and regulatory compliance, the virtual power plant system can adjust trading strategies and resource integration rules based on the electricity market regulations of different countries or regions, ensuring its global applicability and compliance.

[0360] In one possible implementation, the virtualization platform module of the present invention effectively reduces the burden of centralized management by distributing edge computing units at resource access nodes, while significantly improving the response speed of distributed resources. The edge computing units can process real-time data from local resources and feed optimized control instructions back to the virtual power plant management center, thereby achieving distributed and optimized resource management.

[0361] In the above solution, the present invention achieves real-time optimization and rapid response in resource management by distributing edge computing units at resource access nodes. The primary function of the edge computing units is to collect real-time data from access nodes, such as distributed energy generation, energy storage status, and load demand, and to run optimization algorithms locally to generate immediate control instructions. These instructions act directly on access resources, enabling dynamic response and localized regulation.

[0362] The distributed deployment of edge computing units reduces the computing and communication pressure on the virtualization platform center, enabling the system to simultaneously process dynamic data from a large number of distributed resources. This approach not only improves resource integration efficiency but also significantly enhances the system's adaptability to fluctuations in renewable energy. Furthermore, the edge computing units upload aggregated and optimized data to the virtualization platform center via efficient communication interfaces to support global control and decision optimization.

[0363] Through modular design, edge computing units can be flexibly configured based on different resource characteristics to accommodate diverse resource management needs. The standardized design of communication interfaces ensures efficient data exchange and command transmission between edge computing units and the virtualization platform center, enabling efficient resource management and scheduling optimization while ensuring system operational stability.

[0364] In one possible implementation, edge computing units can choose different hardware configurations based on the specific needs of resource access points. For example, high-performance areas can use FPGAs or GPUs to enhance computing power, while standard scenarios can use low-power ARM processors to reduce energy consumption and costs. Furthermore, communication methods can be flexibly adjusted based on conditions, such as opting for fiber optic connections to ensure high data rates or using 5G or satellite communications to ensure connection stability in remote areas.

[0365] The structure of edge computing units also offers considerable flexibility. In resource-intensive areas, edge computing units can be deployed centrally to reduce hardware duplication. In scenarios where distributed resources are more dispersed, the system can deploy multiple low-power edge nodes in a distributed manner, optimizing operational efficiency and cost allocation.

[0366] For data processing steps, local optimization algorithms can be replaced with dynamic algorithms based on reinforcement learning, enabling edge computing units to adaptively adjust control strategies and quickly respond to changing resource demands. Furthermore, the frequency of data collection can be dynamically adjusted to reduce computational and communication burdens while ensuring responsiveness.

[0367] In terms of installation methods, edge computing units can adopt a combination of fixed and modular designs. For resource access points that require long-term stable operation, a fixed installation method can be selected; for temporary or experimental projects, a portable modular design can be used for rapid deployment and disassembly.

[0368] In addition, the application scenarios of edge computing units can be expanded to a wider range of energy management fields, such as supporting real-time regulation in microgrid management, optimizing load distribution in industrial energy management, or being used for post-disaster power restoration in emergency situations, providing efficient resource integration capabilities.

[0369] Regarding software and algorithms, this invention can enhance the edge computing unit's ability to model complex resource relationships by introducing advanced deep learning models (such as neural networks). Simultaneously, the use of lightweight containerization technologies such as Docker can further improve software portability and operational efficiency, enhancing its adaptability to dynamic scenarios.

[0370] This implementation also takes into account adaptability to extreme environments. In extreme heat or cold, industrial-grade hardware can be selected with enhanced heat dissipation or insulation design, while low-power operation modes and local energy storage modules can be combined to ensure reliable operation.

[0371] Finally, the system's flexibility in legal compliance is enhanced. For example, in regions with high data privacy requirements, edge computing units can meet strict data management regulations through data localization and end-to-end encryption, while ensuring system security and reliability.

[0372] In one possible implementation, the dynamic resource aggregation module of the present invention uses a data-driven clustering algorithm to dynamically match the response characteristics of distributed resources with grid demand. This algorithm clusters and groups resources based on their temporal characteristics, geographic distribution, and responsiveness, thereby optimizing resource peak and frequency regulation performance and improving overall resource utilization efficiency.

[0373] In the above scheme, the dynamic resource aggregation module of the present invention significantly improves the efficiency and flexibility of distributed resource integration by adopting a data-driven clustering algorithm. The data acquisition unit is responsible for acquiring the dynamic data of distributed resources in real time, including information such as time characteristics, geographical distribution and response capabilities. The clustering algorithm module runs a dynamic clustering algorithm based on these data, such as K-means clustering or Gaussian mixture model; resources with similar characteristics are grouped into categories to achieve efficient integration of resources. The clustering results are passed to the scheduling optimization module for formulating resource control strategies so that distributed resources can accurately respond to the dynamic needs of the power grid. In this way, the system significantly optimizes the peak-shaving and frequency regulation performance and improves resource utilization efficiency.

[0374] Furthermore, the dynamic resource aggregation module features a real-time update mechanism that dynamically adjusts clustering parameters based on changes in resource characteristics and grid demand, ensuring the system remains highly adaptable to environmental changes. Clustering results can also be used for global optimization of the virtualization platform and dynamic bidding strategies in the market trading module, providing crucial support for the stable operation and maximum profitability of the overall system.

[0375] In one possible variant implementation, the dynamic resource aggregation module can be implemented based on different algorithmic frameworks. For example, in scenarios with abundant computing resources, a high-performance distributed computing platform (such as Apache Spark) can be used to process large-scale distributed resource data. In resource-constrained scenarios, a lightweight framework (such as Python's Scikit-learn) can be selected to run clustering algorithms to reduce the system's computational overhead.

[0376] The module's structural design is also highly flexible. In resource-intensive areas, the module can adopt a centralized processing architecture to improve clustering efficiency. In areas with more widely distributed or dispersed resources, a hierarchical clustering model can be used to group resources by region before performing global optimization, achieving a balance between efficiency and accuracy.

[0377] In terms of operational methods, dynamic clustering algorithms can be replaced with other adaptable machine learning methods, such as graph-based spectral clustering algorithms, to handle complex network relationships between resources. Furthermore, hybrid approaches combining clustering with forecasting can more accurately match resource characteristics with grid demand.

[0378] For parameter adjustment, the system can dynamically optimize key parameters in the clustering algorithm, such as the number of clusters and the number of iterations. For example, when the resource scale is small, the number of clusters can be reduced to improve computing efficiency, while in complex scenarios, the parameter range can be increased to accommodate more resource characteristics.

[0379] In terms of deployment, the dynamic resource aggregation module can use containerization technologies (such as Docker) to facilitate rapid deployment and operation in different computing environments, while also supporting modular expansion. Data transmission methods can also be adjusted according to environmental requirements, such as using fiber for efficient local connections or leveraging 5G networks and low-power wide area networks (LPWANs) for long-distance transmission.

[0380] This module can also be further expanded to new areas such as new energy communities, microgrid optimization, and industrial energy regulation. For example, in smart buildings or industrial parks, dynamic aggregation can be used to optimize load distribution, thereby improving energy efficiency. In emergency response scenarios, such as grid restoration after earthquakes or floods, this module can be used to rapidly optimize resource distribution.

[0381] In terms of algorithms, more complex deep learning models, such as self-organizing maps or dynamic grouping algorithms based on attention mechanisms, can be introduced to handle more complex data patterns. In addition, through online learning technology, clustering algorithms can be continuously updated and optimized during operation, further improving real-time performance and accuracy.

[0382] The module's environmental adaptability can also be optimized through hardware design. For example, in high-temperature, low-temperature, or high-humidity environments, industrial-grade components and protective designs can be used to ensure stable operation of the module under extreme conditions.

[0383] Finally, the dynamic resource aggregation module can be customized according to the market regulations of different countries or regions, such as adjusting data processing strategies to meet privacy protection regulations, or optimizing transmission and computing methods to adapt to different electricity market rules.

[0384] In one possible implementation, the market trading module of the present invention uses multi-scenario simulation technology to analyze the uncertainty of competitor bids and dynamically optimize market trading strategies. This module employs a two-stage bidding strategy, with the upper layer maximizing virtual power plant revenue as the goal and the lower layer analyzing market clearing costs and competitor behavior to optimize revenue and control risks.

[0385] In this solution, the market trading module of the present invention dynamically optimizes the market revenue of virtual power plants through a two-stage bidding strategy and multi-scenario simulation technology. The system first uses a historical data analysis unit to extract information such as market price fluctuations, competitor behavior, and transaction settlement costs, providing basic data support for strategy optimization.

[0386] In the simulation module, the system simulates different market conditions and trading strategies based on historical data and a virtual market environment. By simulating various competitor quotes, the system can predict market liquidation outcomes and evaluate the return and risk performance of different strategies.

[0387] After analyzing the simulation results, the strategy optimization unit dynamically adjusts bidding parameters based on the virtual power plant's revenue maximization objective, while also optimizing underlying trading strategies to minimize market clearing costs. This module continuously monitors market changes during trading and updates bidding strategies in real time to maintain sensitivity and adaptability to market dynamics.

[0388] In addition, the market transaction module interacts with the virtualization platform module and the dynamic resource aggregation module through standardized interfaces, taking resource integration and optimization data as important inputs to the transaction strategy to ensure the coordination and unity of resource regulation and market returns.

[0389] In one possible variant implementation, the hardware configuration of the market trading module can be adjusted based on the specific application scenario. For example, in environments with high transaction frequency and large data processing requirements, a high-performance computing cluster based on GPUs can be used; whereas in smaller or regional trading environments, lightweight edge computing devices can be used to reduce hardware and operating costs.

[0390] The structure of the two-stage bidding strategy can be flexibly adjusted to suit different market needs. In markets with fewer trading entities, a single-tier bidding strategy can be adopted to simplify the model and improve execution efficiency. In highly competitive markets, however, the strategy can be more complex, introducing more variables for multi-dimensional optimization, such as dynamic parameters that incorporate real-time grid status or renewable energy fluctuation forecasts.

[0391] In terms of operational methods, reinforcement learning algorithms can serve as an alternative to bidding strategy optimization. By learning from market changes in real time, the system continuously optimizes bidding parameters to ensure maximum returns. Furthermore, the data processing methods in the simulation module can incorporate deep learning models based on time series analysis, such as LSTM or Transformer, to predict market price trends with higher accuracy.

[0392] The simulation module's scenario design parameters, such as the simulation time span and number of scenarios, can be adjusted based on market dynamics. In rapidly changing markets, the simulation time span and number of scenarios can be increased to more accurately assess the effectiveness of different strategies. In stable markets, the number of simulations and scenario complexity can be reduced to improve operational efficiency.

[0393] The software portion of the module can be made portable through containerization technology, supporting rapid deployment in different operating environments. At the same time, the adoption of a microservices architecture can further enhance the modularity of the system, facilitating independent updates and functional expansion.

[0394] The communication method of the transaction module can also be changed according to needs. For example, when high-speed transmission is required, a fiber optic network can be used; while in areas with long distances or poor network conditions, data transmission can be carried out through 5G networks or low-power wide area networks (LPWAN) to ensure the real-time and reliability of transaction data.

[0395] The market trading module can also be extended to other energy-related application scenarios. For example, in electric vehicle charging networks, the module can support the optimization of dynamic pricing strategies. In cross-border energy trading, it can be used to design fair and efficient cross-border trading mechanisms. Furthermore, in demand-side response scenarios, the module can help optimize load reduction and incentives to improve overall energy efficiency.

[0396] In terms of algorithms, the two-stage bidding strategy can be replaced with a mixed strategy algorithm from game theory, or meta-learning techniques can be used to improve the model's generalization capabilities across diverse market environments. Simultaneously, the simulation module can incorporate multi-agent modeling technology to enhance its ability to simulate dynamic market environments.

[0397] The trading module's hardware design can enhance its durability and stability for specific environmental conditions. For example, in high-temperature or high-humidity conditions, protective coatings and efficient heat dissipation designs can be added; in low-temperature environments, insulation layers and built-in heating modules can ensure normal operation of the device.

[0398] Finally, the module's design can adapt its strategies and architecture to the energy market regulations of different countries or regions. For example, in regions with high privacy protection requirements, data encryption storage and transmission technology can be used to meet compliance requirements. In cases where market rules limit the range of bids, the bidding algorithm can be adjusted to ensure optimal performance within the legal and regulatory framework.

[0399] In one possible implementation, the security management module of the present invention leverages blockchain technology to establish a multi-party trust system and uses smart contracts to dynamically adjust resource transaction prices. This module ensures transparency in the virtual power plant's resource management and market transactions, while also improving the system's security and reliability.

[0400] In this solution, the security management module utilizes blockchain technology to establish a multi-party trust system, providing robust security for the virtual power plant's resource management and market transactions. The distributed ledger records all resource management and transaction data, ensuring transparency and immutability, while also providing a fully traceable management mechanism.

[0401] The smart contract engine automatically adjusts resource transaction prices based on pre-set trading rules. Specifically, the smart contract calculates the optimal transaction price based on real-time resource supply and demand and market conditions, and automatically triggers transaction settlement when these conditions are met. This dynamic price adjustment mechanism significantly improves transaction fairness and market efficiency while reducing the risk of human intervention.

[0402] The encryption module protects data through public key encryption technology, ensuring the security of transaction information during transmission and storage. In addition, a hash algorithm is used to generate a unique identifier for transaction data, further preventing the possibility of data tampering.

[0403] The security management module is seamlessly integrated with the virtualization platform module and the market transaction module through standardized interfaces. The distributed ledger provides trusted data support for resource integration and transactions, while the smart contract optimizes the dynamic allocation and price adjustment of resources at the transaction execution level.

[0404] Through the innovative application of blockchain technology, this module not only ensures the high security and transparency of the system, but also improves the operating efficiency and reliability of virtual power plants in complex trading environments involving multiple entities.

[0405] In one possible implementation variant, the security management module can choose from different blockchain technology frameworks, such as Ethereum to support the complex logic of smart contracts or the Fabric framework for higher transaction throughput and privacy protection. For the distributed ledger's storage nodes, the system can use enterprise-grade hardware-based servers or high-performance SSD devices to meet the data storage and processing requirements of different transaction scenarios.

[0406] The network structure of a distributed ledger can also be adjusted based on the application scenario. For example, where resource access is small or market participants are limited, an efficient consortium chain architecture can be chosen to speed up transaction confirmation. In an environment with dispersed participants and high data transparency requirements, a public chain model can be adopted to ensure the openness and fairness of the system.

[0407] Smart contract logic can be replaced through a rules engine. This allows users to dynamically adjust transaction rule templates based on business needs and supports real-time parameter updates to optimize price adjustment flexibility. Furthermore, to further improve data query efficiency, a hybrid storage solution combining blockchain and traditional databases can be used, storing high-frequency data in a relational database while maintaining long-term transaction records on the blockchain for security.

[0408] In terms of technical parameters, the choice of blockchain consensus algorithm can be flexibly adjusted based on system requirements. For example, for scenarios requiring high security, PoW (Proof of Work) can be used to ensure the system's tamper resistance; while for scenarios requiring higher efficiency, PoS (Proof of Stake) or PBFT (Practical Byzantine Fault Tolerance) can be used to increase transaction confirmation speed and reduce resource consumption.

[0409] In addition, the module's software design can achieve cross-platform compatibility and rapid deployment capabilities through containerized deployment (such as Docker), while using the microservice architecture to independently develop and update each functional unit, thereby improving development efficiency and system stability.

[0410] In terms of connection methods, modules can communicate securely through a dedicated virtual private network (VPN), or use quantum encryption technology in sensitive data transmission to further enhance data security and transmission reliability.

[0411] The security management module can be installed flexibly. For example, the module can be centrally deployed at the core node of the virtual power plant for unified management, or edge computing units can be installed at each access point in a distributed manner to support local data processing and improve regional transaction efficiency.

[0412] In terms of expanding application scenarios, the security management module is not only suitable for resource management and market transactions of virtual power plants, but can also be extended to scenarios such as regional microgrids, cross-border power transactions and new energy communities, and can even be applied to multi-regional energy scheduling and resource integration in the industrial field.

[0413] Smart contract design can also incorporate machine learning technology to analyze historical transaction data and market trends to predict future price trends and optimize trading strategies. Furthermore, the consensus algorithm can be replaced with a DAG (directed acyclic graph) architecture to support higher-frequency transactions.

[0414] Under extreme working conditions (such as high temperature, low temperature or high humidity environment), the hardware part of the module can adopt industrial-grade materials and design to ensure the stability and durability of the system; in low power supply scenarios, continuous operation can be guaranteed by introducing low-power hardware and local energy storage solutions.

[0415] Finally, the security management module can also be adjusted according to the legal and regulatory requirements of different markets, such as meeting the needs of privacy protection regulations through localized data storage and distributed encryption technology, and optimizing transaction rules to ensure compliance with regional market regulations and legal frameworks.

Claims

1. A virtual power plant optimization system, characterized by The system includes: A virtualization platform module is used to integrate distributed energy, energy storage equipment, electric vehicles, and load resources to achieve virtualization and centralized management of resources; Dynamic resource aggregation module, used to dynamically match resource characteristics with power system requirements; Market trading module, used to optimize the benefits of virtual power plants; Core algorithm module, used to optimize load forecasting accuracy.

2. The virtual power plant optimization system according to claim 1, characterized in that: In the virtualization platform module, in order to achieve the coordinated optimization of the distributed resources in the virtual power plant (VPP), the output characteristics and operating characteristics of each distributed resource are established, and the corresponding mathematical model is analyzed and established: 1) Photovoltaic power generation modeling is as follows: In formula (1), P PV Represents photovoltaic output power; f PV is the loss factor of the photovoltaic panel; P PV,cap is the output power of the photovoltaic array under standard test conditions; G T,STC is the light intensity under standard test conditions; G T is the actual light intensity; α P is the power temperature coefficient; T cell Indicates the actual operating temperature of the photovoltaic cell; T cell,STC Indicates the temperature under standard test conditions; 2) The wind power generation WT model is as follows: In formula (2), P WT Indicates the actual output power of the wind turbine; v i 、v v Respectively represent the cut-in and cut-out wind speeds; v represents the current wind speed; v r Indicates the cut-out wind speed; v o Indicates rated wind speed; 3) The energy storage system charging and discharging model is as follows: In formula (3), SOC t Indicates the state of charge of the energy storage system at time t; SOC t-1 represents the state of charge of the energy storage system at time t-1; λ b Indicates the self-discharge rate of the energy storage system; Indicates the charging efficiency of the energy storage system; Indicates the discharge efficiency of the energy storage system; represents the charging power of the energy storage system at t-1; represents the discharge power of the energy storage system at t-1; △t represents the time interval between two adjacent time points; SOC min Indicates the lower limit of the state of charge; SOC max Represents the upper limit of the state of charge.

3. The virtual power plant optimization system according to claim 2, characterized in that: To achieve safe and stable operation of the energy storage system, it is divided into five charging and discharging action intervals: Interval 1: When 0≤SOC t <0.2SOC max When the charge power range is P d =0,P d Indicates the discharge power of the energy storage system in interval 1; Interval 2: When 0.2SOC max ≤SOC t <0.35SOC max When the charging power range is: 0.2SOC max ≤SOC t <0.35SOC max , the discharge power range is Interval 3: When 0.35SOC max ≤SOC t <0.65SOC max When the charging power range is P C , Interval 4: When 0.65SOC max ≤SOC t <0.8SOC max When the charging power range is Discharge power range is Interval 5: When 0.8SOC max ≤SOC t <SOC max When the charging power is P C =0, non-rechargeable state; discharge power At the same time, ensure that the state of charge of the energy storage system is not less than the minimum state of charge and not greater than the maximum state of charge. Then, the energy storage system is divided into free trading energy storage and backup energy storage.

4. The virtual power plant optimization system according to claim 2, characterized in that: For multiple virtual power plants (VPPs), a two-layer VPP game model is established. The upper-layer VPP game model takes maximizing the VPP revenue as its objective function, while the lower-layer VPP game model takes minimizing the total power generation cost as its objective function, thereby modeling the internal game of the VPP. The upper-level VPP game model maximizes the benefits of the virtual power plant VPP and establishes an objective function, which is expressed as follows: In formula (4), represents the revenue obtained by the i-th VPP from selling electricity to the load at time t; The gains of the interaction of the i-th VPP with a wider grid are shown; represents the revenue obtained from selling electricity from the i-th VPP to other VPPs; For any virtual power plant (VPP), it is necessary to ensure that the system maintains power balance at any point during operation. The sum of the power of all power sources in the VPP interacting with other VPPs and the sum of the power interacting with the main grid equals the load demand. The expression is as follows: In formula (5), Indicates the load demand power; Indicates photovoltaic output power; Indicates the output power of the wind turbine; Indicates the output power of standby energy storage; Indicates the free energy storage output power; It is the power input from the large power grid to the VPP; With the objective function of reducing the overall operating costs of VPP, a lower-level VPP game model is constructed: In formula (9), represents the total power generation cost of the i-th VPP at time t; represents the photovoltaic power generation cost in the i-th VPP; represents the cost of electricity generated by wind turbines in the i-th VPP; represents the operating cost of the energy storage system in the i-th VPP; N PV represents the number of photovoltaic cells in the i-th VPP; N W N represents the number of wind turbines in the i-th VPP; b represents the amount of energy storage in the i-th VPP; represents the minimum total power generation cost of VPP at time t; represents the photovoltaic power generation cost of the i-th virtual power plant; represents the wind power generation cost of the i-th virtual power plant; represents the cost of the energy storage system in the i-th VPP.

5. The virtual power plant optimization system according to claim 4, characterized in that: The upper-level VPP game model includes constraints: 1) Wind power and photovoltaic output constraints: The production capacity of wind and photovoltaic installations is limited by their maximum output ratings: In formula (6), represents the actual photovoltaic output power of the i-th virtual power plant at time t; Represents the maximum output power of the PV; represents the actual output power of the wind turbine of the i-th VPP at time t; is the maximum power output of the fan; 2) The electricity price traded between VPPs should not be higher than the price of electricity purchased from the grid, nor lower than the price of electricity sold from the grid: l da,t ≤λ i,t ≤λ' da,t (7); In formula (7), λ i,t is the transaction price of VPP; da,t is the price of electricity sold by VPP to the grid; da,t The price of electricity purchased by the VPP from the grid; 3) Power trading exists between VPPs and the main grid, and between multiple VPPs with tie-line power constraints: In formula (8), represents the energy of interaction between the i-th VPP and the large grid; is the maximum interaction power allowed by the contact line between the first VPP and the grid; P ij,t is the energy of interaction between the i-th VPP and the j-th VPP; is the maximum interaction power allowed by the contact line between the i-th VPP and the j-th VPP.

6. The virtual power plant optimization system according to claim 4, characterized in that: The lower-level VPP game model satisfies power balance constraints, energy storage charging and discharging power constraints, and wind and solar curtailment constraints. The energy storage system constrains that the energy storage charging and discharging power must be less than the maximum output power and maximum discharge power of the energy storage system. In formula (10), Indicates the maximum charging power; represents the net charge and discharge power of the energy storage system in the i-th VPP at time t; Indicates the maximum discharge power; Abandoning wind power generation WT and photovoltaic power generation output constraints: In formula (11), ρ pv,t and ρ w,t They represent the constraint coefficients of photovoltaic power generation and wind power generation WT reduction at time t; △P w,t represents the wind curtailment of the ith VPP at time t; △P pv,t represents the amount of abandoned light of the i-th VPP at time t.

7. The virtual power plant optimization system according to claim 4, characterized in that: The multiple VPPs in the two-layer VPP game model need to consider both the upper-layer VPP game model and the lower-layer VPP game model, and the two influence each other. The two-layer particle swarm optimization method is used to solve the two-layer VPP game model containing multiple VPPs. In the two-layer particle swarm optimization method, the upper-layer VPP game model is solved using the outer-layer particle swarm method, where the particle position represents the direct transaction power of a single VPP, and the fitness value represents the sum of the economic benefits of all VPPs. The lower-layer VPP game model is solved using the inner-layer particle swarm method, where the particle position represents the output of each unit and the user load power demand, and the fitness value represents the VPP operating cost. The two-layer particle swarm optimization method finally converges to the game equilibrium solution through the iterative coupling of the outer-layer PSO and the inner-layer PSO. In the structure and coupling relationship of the two-tier VPP game model, the goal of the upper-tier VPP game model is to maximize the benefits of each VPP: in: Represents the revenue from electricity sales; Indicates the benefits of interacting with the main network; It indicates that the transaction revenue between VPPs depends on the power generation cost optimized at the lower level; Decision variable: transaction price λ between VPPs i,t and interactive power P g,t ; The cost minimization objective of the lower-level VPP game model is to minimize the internal power generation cost of the VPP; Decision variables: output P of each unit pv,t ,P w,t ,P b,t and the amount of wind and solar power curtailment △P w,t ,△P pv,t ; Coupling mechanism: The upper-layer VPP game model's revenue calculation uses the lower-layer VPP game model's cost as input, while the lower-layer VPP game model optimization accepts the transaction price and power constraints assigned by the upper-layer VPP game model; the two achieve dynamic equilibrium through iterative feedback of the two-layer PSO.

8. The virtual power plant optimization system according to claim 4, characterized in that: The dynamic resource aggregation module combines the responsiveness of distributed resources with market supply and demand dynamics, and uses a data-driven clustering algorithm to dynamically match resource characteristics with power system requirements to optimize peak and frequency regulation performance. The details are as follows: Distributed resources are divided into energy storage and power sources. Energy storage includes distributed energy storage and electric vehicles, which have energy accumulation capabilities. Power sources include fully controlled power sources (FCPs) such as micro gas turbines, and semi-controlled power sources (HCPs) such as distributed wind power and distributed photovoltaics. The constraints of the fully controlled power supply (FCP) include power, capacity, and ramp constraints, namely: In formula (14): and are the active and reactive outputs of FCPi at time t respectively; and are the minimum and maximum active output of FCPi respectively; is the rated capacity of FCPi; and They are the upper and lower limits of FCPi’s climbing ability, respectively; represents the active and reactive output of FCPi at time t+1; A semi-controlled power supply HCP resource output model considering the uncertainty of wind and solar power output is constructed, and the chance constraints containing random variables are converted into deterministic constraints through the Gaussian mixture model (GMM) to achieve the solution of the virtual power plant feasible region (VFR). The resource output model of the semi-controlled power supply HCP is: In formula (13): and are the active and reactive outputs of HCPi at time t respectively; and are the minimum and maximum power factor angles of HCPi respectively; Indicates the maximum active power output limit of FCPi at time t.

9. The virtual power plant optimization system according to claim 8, characterized in that: The objective function of the economic optimization model of multiple VPPs is to optimize the total cost of each aggregated VPP. The total cost of VPPi includes the active power cost. and reactive power costs Right now: In formula (16): P i and Q i are the active and reactive power inputs of VPPs, respectively; I is the number of VPPs included in the distribution network; represents the minimum total cost of all VPPs; C represents the total cost of all VPPs; Among them, active cost The piecewise quadratic function aggregation cost model is adopted, which is the superposition of the costs of various resources within the VPP. The cost functions of various internal resources are shown in formula (17); Reactive power cost Including fixed costs and opportunity costs, the specific expressions of the active power cost model and reactive power cost model of VPP are: In the above formula: are the costs of FCP, HCP and energy storage resources respectively; and is the cost coefficient of FCPm; and is the cost coefficient of HCPn; and is the cost coefficient of energy storage resource l; α i,n , β i,n and γ i,n are the secondary, primary and constant cost coefficients of VPPi output active power in segment n, usually n∈[2,5], which is determined by the internal resource type of VPP; and are the active powers of FCPm, HCPn and energy storage resource l respectively; D i,n is the boundary value of VPPi output active power; Q,i is the discounted fixed cost; is the cost coefficient; P i,max and S i,max are the maximum active power and apparent power of VPPi respectively; α i represents the secondary cost coefficient of VPPi output active power; β i represents the primary cost coefficient of VPPi output active power; γ i represents the constant cost coefficient of VPPi output active power; P i Indicates the active power of VPPi; Q i Indicates the reactive power of VPPi; D i1 Indicates VPP i Minimum technical output; D i2 Indicates the output limit or operation mode switching point of the first type of resources; D i3 Indicates the output upper limit of the second type of resources or the maximum output allowed by the system.

10. The virtual power plant optimization system according to claim 8, characterized in that: The cluster analysis combines the load current field and electrical distance. The cluster center is determined by the load current field, and then clustering is completed based on the electrical distance, avoiding the repeated iteration problem in traditional clustering algorithms. (1). Calculation of load current field and virtual potential: The virtual potential φ of the load current at node i is defined as i : Among them, Z ij is the matrix Z u Elements, I j is the load current of node j; this formula reflects the distribution characteristics of node voltage affected by load current; Describe the relationship between the voltage drop at node i and the load current: △U i =Z ij I j (24); Among them, △U i represents the voltage drop at node i; Equation (24) is used to equivalently represent the voltage impact between nodes; (2) Definition of electrical distance: The electrical distance matrix D between nodes is defined based on the inverse matrix of the Jacobian matrix J, namely the susceptance matrix B; D ij =|(J -1 ) ii +(J -1 ) jj -2(J -1 ) ij | (25); Formula (25) is used to quantify the degree of electrical connection between nodes and is the basis of cluster analysis; (3) Determination of cluster center nodes The cluster center node is determined by the maximum point of the local potential value of the load current field: For node i, if its potential value φ i If the potential value of i is greater than that of all directly connected nodes, then i is the point with the maximum local potential value; the number of points with the maximum local potential value is the number of partitions, and these points serve as cluster center nodes; (4) Cluster analysis process: Clustering method: Calculate the electrical distance D between each node and the cluster center node ij , divide the nodes into the partitions to which the cluster centers with the smallest distance belong; Partition optimization: By exchanging boundary nodes, optimization is performed with the minimum active network loss as the objective function: in, and are the active output and active load of the node respectively.

11. The virtual power plant optimization system according to claim 1, characterized in that: The market transaction module implements a two-stage market bidding based on a multi-agent game model to optimize the benefits of the virtual power plant; the details are as follows: Assuming that the supply and demand entities do not fully understand the decision-making process and decision-making information of their competitors, the operation process of the electricity market is established as a Markov decision process in an uncertain environment. The Markov decision process model in this uncertain environment has various subject objects participating in the market, environmental objects, the subject's state set S and action set A, and reward R; in transactions at a specific time scale, the subject determines its own state S by observing the environmental object information at each discrete time step t. t ∈S, select and execute action A according to a certain strategy t ∈A, after executing the action, the agent The state transition probability migrates to the new state S t+1 ∈S, get instant reward R t+1 ∈R and update the policy parameters, and finally generate or perform the contract to update the environment information.

12. The virtual power plant optimization system according to claim 11, characterized in that: In each state, the agent will adopt a behavioral strategy that maximizes expected benefits. This strategy process is represented by the Bellman optimality equation. Where: V * (s) is the optimal state value function, which represents the cumulative maximum expected return that can be obtained in state s; Q * (s,a) is the optimal action value function, which represents the cumulative maximum expected reward that can be obtained by performing action a in state s; s is the state of the agent; a is the optimal action for the current state s; A represents the action space; a∈A, s∈S, S represents the state space; R(s,a) represents the immediate reward after executing the action; γ is the depreciation factor; P(s'|s,a) represents the probability of transitioning to the next state s'; V * (s') represents the optimal state value function of state s; s' represents the state after state s is transferred; s'∈S; π * (s) is the optimal strategy in state s; The intelligent agent reinforcement learning decision model uses the Q learning algorithm to estimate the optimal action value function, that is, for each possible action-state pair (s, a), the action value function is defined as the Q-Value function, which gives the expected utility of taking action a in state s, and is used to estimate Q * (s,a), thus obtaining the optimal strategy π * ; The intelligent agent first initializes the Q-value lookup table through random strategies or historical learning information. In each state s, it adopts the ε-greedy strategy, performs random actions with a probability of ε, and performs the greedy strategy with a probability of 1-ε; that is, it finds an optimal strategy π based on the maximum Q-Value. * (s)∈A, after executing the strategy, the state s←s' is updated at the same time, and the Q value lookup table is updated as follows: In formula (22), α is the update step size; Q'(s,a) represents the updated action value function, which is the weighted result of the current Q value and the new experience; R(s,a) represents the direct benefit obtained after executing action a in state s; It represents the maximum Q value among all possible actions a' in the next state s', which represents the estimation of the optimal strategy in the future; Q(s',a') represents the maximum Q value that can be obtained by choosing the optimal action a' in the future starting from the new state s'.

13. The virtual power plant optimization system according to claim 12, characterized in that: Including bidding revenue optimization analysis based on historical market data and simulation environment; In order to maximize profits and measure the risks caused by electricity price uncertainty, risk factors are introduced into the objective function of power generation enterprises. maxf=(1-ρ)B+ρV CVaR (33); In formula (23), maxf represents the maximum value of the optimization objective; ρ is the risk factor, ρ∈[0,1], and its value is positively correlated with the conservative degree of the power generation enterprise's bidding strategy; B is the expected profit value of the power generation enterprise in participating in the day-ahead energy and ancillary service market; V CVaR is the risk indicator; f is the optimization target; The expected profit value B of a power generation enterprise is the difference between the revenue and cost in the market, where the revenue from participating in the market includes the revenue from selling electricity in the energy market. and the benefits of participating in secondary frequency modulation Cost includes unit cost and carbon emission trading costs In formula (24): S is the number of scenes; π s is the corresponding probability of scene s; The revenue from electricity sales of power generation enterprises is: In formula (35): is the electricity price at time t in scenario s; p s,t is the winning bid power of the power generation enterprise at time t under scenario s; T is the total running time; Secondary frequency modulation benefits It includes two aspects: opportunity cost compensation and frequency modulation call benefits; In formula (36): is the compensation price given by the market operator at time t under scenario s; k is the declared capacity corresponding to the opportunity cost of the unit at time t under scenario s; s,t is the actual calling coefficient of the system for the unit at time t under scenario s; The unit cost consists of two parts: start-up and shutdown cost and operating cost. SU 、c SD They are the unit start-up and shutdown costs, and the unit operating costs, respectively, using the quadratic function express; In formula (37): x s,t 、y s,t They are 0-1 variables describing the start and stop status of the unit. s,t =1,y s,t =0, when the machine stops x s,t =0,y s,t =1; is the total output of the power generation enterprise, and the calculation formula is: In formula (38): p s,t represents the total output at time t under scenario s; k s,t The actual calling function in time and space t under scenario s; represents the secondary frequency modulation capacity at time t in scenario s; The calculation formula for carbon emission rights trading costs is: In formula (39): is the carbon emission rights trading price; r G is the carbon emission intensity; R G is the carbon emission quota corresponding to the unit.

14. The virtual power plant optimization system according to claim 12, characterized in that: The constraints of the power generation enterprise joint bidding optimization model are as follows: The unit start and stop constraints are: z s,t -z s,t-1 =x s,t -y s,t (43); In the above formula: z s,t 、z s,i are the status of the unit at time t and time i respectively. When the value is 1, it means the unit is on, and when the value is 0, it means it is off; are the shortest start and stop time of the unit respectively; s,t-1 represents the operating status of the unit at time t-1 under scenario s; x s,t Indicates the startup status of the unit at time t in scenario s; y s,t Indicates the shutdown status of the unit at time t in scenario s; Represents the minimum continuous running time constraint; The output constraints are: P min z s,t ≤p s,t ≤P max z s,t -P t AGC (44); In formula (44): P max 、P min are the upper and lower limits of the unit output respectively; P t AGC Reserve frequency modulation capacity for time t; p s,t represents the actual output of the generator set at time t under scenario s; The climbing constraint is: In the above formula: p on 、p off are respectively the starting and stopping climbing power; p ru 、p rd are the climbing power for upward and downward operation respectively; p s,t-1 represents the actual output of the generator set at time t-1 under scenario s; k s,t k represents the actual call coefficient of the frequency modulation capacity at time t under scenario s; s,t-1 represents the actual call coefficient of the frequency modulation capacity at time t-1 under scenario s; represents the FM capacity at time t-1 under scenario s; The auxiliary variable constraints are: x s,t ,y s,t ,z s,t ∈{0,1}(47); The quotation curve is required to be monotonically increasing, and the constraints are shown in equations (48) to (48); when the conventional thermal power unit reserves the frequency regulation capacity, the minimum value of the increase is set to 15% of the unit's rated capacity, as shown in equation (50); In the above formula: represents the market price of electricity at time t+1 under scenario s; represents the market price of electricity at time t under scenario s; p s,t+1 represents the actual output of the generator set at time t+1 under scenario s; P G is the rated capacity of the unit.

15. The virtual power plant optimization system according to claim 1, characterized in that: The security management module builds a multi-party trust system through blockchain technology and dynamically adjusts resource transaction prices based on smart contracts to achieve transparency in resource management and credibility in resource transactions.

16. The virtual power plant optimization system according to claim 1, characterized in that: It also includes a security management module for dynamically adjusting resource transaction prices.

17. A method for optimizing power grid partitioning by combining load current field and cluster analysis, characterized by: The number of partitions and cluster centers are determined by the load current field, and clustering is performed using electrical distance. Finally, the boundary node division is optimized with active network loss as the optimization target to achieve the optimized grid partitioning, including the following steps: Step 1) Form the node admittance matrix Y from the grid data: in: is the load current equivalent factor matrix, whose i-th column characterizes the load current I of node i i The impact on the voltage of each node in the network is related to the load current I i It is approximately in direct proportion. The voltage influence on each node is related to the distance from node i, and decreases with the increase of distance. Step 2) Calculate the load current virtual potential βi according to the node admittance matrix Y: The node voltage influence characterized by the i-th column is called the load current virtual field, and the elements of each row of the i-th column are called the virtual potential of each node in the load current virtual field of node i. The load current virtual potential of node i for: Among them, z ii and z ij The matrix Z LL The element at row i, column i and column j; Step 3) Determine the number of clusters and partitions, with the point with the maximum virtual potential of the node current as the cluster point and partition number: The admittance matrix is formed by the power grid data, and the load current virtual potential of each node is calculated by the load current field; the number of partitions and the central node of the cluster are determined; let V i is the set of P points directly connected to node i. If for any node J∈V i ,have Then node i is the node with the local maximum potential value; the number of nodes with the local maximum potential value is the number of partitions, and each node with the local maximum potential value is the cluster center node; Step 4) Clustering is performed based on the number of partitions to obtain preliminary partitions, and whether the constraints are met is determined. If the constraints are met, step 5 is performed. If the constraints are not met, the node admittance matrix after partitioning is generated and the process returns to step 2). Calculate the electrical distance between each node and each cluster center node according to the electrical distance matrix, compare the distance between the node and each cluster center node, and divide each node and the center node with the smallest distance into one category based on the principle of minimum distance, and obtain the clustering result. Each cluster is the preliminary partition result. Calculate the power flow and short-circuit current of the power grid after partitioning. If the constraint conditions are not met, disconnect the partition tie line to form the admittance matrix after partitioning, calculate the load current virtual potential of each node, and go to step 2) and step 3) until the partition that meets the conditions is obtained; Step 5) Optimize the node boundaries after partitioning to obtain the final partitioning result: Re-partition the boundary nodes after the preliminary partitioning. Let a and b be a pair of nodes connected by a line before partitioning. After partitioning, they belong to the adjacent c and d partitions respectively. Exchange these two boundary nodes, assign node a to partition d, calculate the power flow and short-circuit current, and retain the results if the constraints are met and the network loss is reduced. If not, return. Assign node b to partition c, calculate the power flow and short-circuit current, and retain the results if the network loss is reduced. If not, return. Continue to optimize the remaining pairs of boundary nodes using the same method until the optimization of all boundary nodes is completed, and finally obtain the partition result with the minimum network loss.

Images