Virtual power plant optimal scheduling method under dual mechanisms
By combining carbon trading and green certificates with Q-learning particle swarm optimization to optimize the scheduling of virtual power plants, the net benefit and carbon emission issues of traditional methods when renewable energy is integrated are solved, achieving efficient energy consumption and stable scheduling with low carbon emissions.
Patent Information
- Application Number
- CN202511489913.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-17
- Publication Date
- 2026-03-03
AI Technical Summary
When faced with the large-scale integration of renewable energy, existing technologies struggle to simultaneously maximize net benefits, achieve efficient renewable energy consumption, and reduce carbon emissions. Furthermore, traditional methods are prone to premature convergence and lack solution stability.
Under the dual mechanisms of carbon trading and green certificates, and combined with the particle swarm optimization algorithm of Q-learning, a virtual power plant optimization configuration model is constructed. By setting a market participation mode switch, the interaction between gas turbines, wind turbines, photovoltaic units, energy storage devices and the power grid is optimized. An objective function for maximizing net revenue is established and solved by the particle swarm optimization algorithm to dynamically balance global exploration and local convergence, thereby optimizing the scheduling scheme.
Under complex nonlinear and multi-constraint environments, higher net income, stronger renewable energy absorption and lower carbon emissions were achieved. The optimization results are stable and have good repeatability, avoiding the risk of exceeding operational limits and improving economic efficiency and low carbon emissions.
Smart Images
Figure CN121599324A_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the field of virtual power plant optimal scheduling and low-carbon operation technology, and specifically relates to a virtual power plant optimal scheduling method under a dual mechanism. Background Technology
[0002] The large-scale integration of renewable energy sources such as wind and solar power into the power system presents challenges in balancing economic efficiency, security, and low-carbon emissions due to their volatility and intermittency. On the one hand, carbon trading mechanisms use quotas and price signals to economically regulate emissions from conventional generating units, prompting power generation to reduce high-emission output. On the other hand, green certificate mechanisms quantify the environmental value of renewable energy generation with certificates and promote clean energy utilization through a "dual-drive" approach of consumption quotas and market trading. Existing technologies mostly model single market mechanisms or employ heuristic algorithms with fixed parameters. Faced with optimization problems involving high-dimensional nonlinearity and strong coupling constraints, traditional methods are prone to premature convergence and lack solution stability, making it difficult to simultaneously maximize net revenue, efficiently absorb renewable energy, and reduce carbon emissions. Summary of the Invention
[0003] This application provides a virtual power plant optimal scheduling method under a dual mechanism, which solves the problems of premature convergence and insufficient stability of solutions in traditional methods, making it difficult to simultaneously maximize net revenue, efficiently absorb renewable energy and reduce carbon emissions.
[0004] This application presents an optimal scheduling method for virtual power plants under a dual mechanism of carbon trading and green certificates, comprising: Establish a virtual power plant, including the interaction between gas turbines, wind turbines, photovoltaic units, energy storage devices and the power grid, and set up a market participation mode switch; Based on the predicted output of wind power and photovoltaic power and typical daily load data, determine the upper and lower limits and initial values of decision variables; An optimization configuration model is set up for the virtual power plant with the objective function of maximizing net revenue. The revenue includes electricity sales, carbon quota settlement and green certificate trading; the costs include gas turbine operation, wind and solar operation and maintenance, wind and solar curtailment penalties, energy storage operation and maintenance and grid power purchase and sale; constraints are set for power balance, unit output / ramp, energy storage power and energy, wind and solar forecast upper limit and tie line capacity. The objective function is solved by integrating Q-learning into the particle swarm optimization algorithm, and the optimal scheduling scheme for the virtual power plant is obtained.
[0005] Furthermore, the optimized configuration model constructs a target for maximizing the daily net revenue of the virtual power plant within the upper and lower limits of the control variables and under the initial values.
[0006] Furthermore, a market participation mode switch is set, with the switch states including no participation, participation in carbon trading only, and participation in both carbon trading and green certificate trading.
[0007] Furthermore, determining the upper and lower limits and initial values of the control variables based on the predicted output of wind power and photovoltaic power and typical daily load data includes: obtaining the typical daily load curve and wind / solar prediction curve based on the predicted output of wind power and photovoltaic power and typical daily load data, and combining the time-of-use electricity price, carbon price, and green certificate price to determine the upper and lower limits and initial values of the decision variables.
[0008] Furthermore, the decision variables include: gas turbine output, photovoltaic unit output, energy storage device charging and discharging power, grid interconnection line power, carbon quota trading volume, green certificate trading volume, wind and solar curtailment power, and unit ramp-up rate.
[0009] Furthermore, by solving the objective function using a particle swarm optimization algorithm incorporating Q-learning, the optimal scheduling scheme for the virtual power plant is derived, including: An initial population is randomly generated based on the power output plan of the virtual power plant over a 24-hour period. Calculate the fitness of each particle in the population; The average position of the optimal values of all individual particles is used as the global guidance center. Determine the position of the particle with the best fitness in the initial generation as the current optimal solution; Based on the Q values stored in the Q table, select the optimal action for the current state; Based on the current position and velocity of the particles and the parameters obtained from the Q table, update the particle swarm and generate a new scheduling scheme. Evaluate the fitness of each particle in the new particle swarm and update the individual optimal position of each particle; Update the global optimal solution based on the new fitness value.
[0010] Furthermore, the particle positions and velocities are updated according to the velocity update formula of the particle swarm optimization algorithm incorporating Q-learning; the fitness of each particle is calculated; and rewards are calculated based on fitness improvement and diversity changes. Rewards are then applied to the Q-table. Update the Q value in the Q table based on current rewards and expected future rewards; Execute a greedy selection strategy and select the action with the largest current Q value with a probability of 1-ε; The actions are mapped to parameters of a particle swarm algorithm that incorporates Q-learning.
[0011] Furthermore, the Q table is generated based on the state space S and the action space A, and is used to store the Q value of each state-action pair. The state space S includes a three-dimensional combination (P, D, I). Based on the current state of the particle swarm, P, D, and I are evaluated. P represents the optimization stage, which is represented by early / mid / late. D represents the fitness improvement rate of recent generations, which is represented by low / medium / high. I represents the population diversity, which is represented by low / medium / high. The three-dimensional combination of P, D, and I is located in the state space S. Action space A includes 9 pre-defined strategy combinations (ω, c1, c2), where ω represents the balance weight between exploration and development, c1 represents the intensity of local exploration, and c2 represents the intensity of local development, covering strategies of global exploration / balance / local development / convergence reinforcement.
[0012] Initialize the Q-table, initialize the particle swarm including position and velocity, set the initial state and initial ε value, where ε is a greedy selection: select the action with the largest current Q value with a probability of 1-ε, and explore randomly with a probability of ε, and ε decays exponentially with iteration; Based on the current state of the particle swarm, evaluate P, D, and I. P represents the optimization stage, which is represented by early / mid / late. D represents the fitness improvement rate of recent generations, which is represented by low / medium / high. I represents the population diversity, which is represented by low / medium / high. The three-dimensional combination of P, D, and I is located in the state space S. Action a is selected according to the ε-greedy policy, and action a is mapped to the parameters (ω, c1, c2) of the particle swarm algorithm that incorporates Q-learning. The root action space A is set, which includes 9 pre-set policy combinations (ω, c1, c2), where ω represents the balance weight between exploration and exploitation, c1 represents the intensity of local exploration, and c2 represents the intensity of local exploitation, covering global exploration / balance / local exploitation / convergence reinforcement policies.
[0013] Compared with the prior art, the advantages of this application are as follows: This application, under the dual mechanism of "carbon trading + green certificates," explicitly incorporates carbon quota settlement and green certificate incentives into the dispatch target, forming a mechanism to suppress high-emission output and positively promote wind and solar grid connection. This achieves significant economic gains and a synergistic improvement in low-carbon benefits under the same equipment and price boundaries. Combining time-of-use pricing and energy storage arbitrage signals, the optimization results exhibit a stable strategy of "charging at noon and discharging during the evening peak," effectively absorbing wind and solar power during nighttime periods of high wind and low load. This significantly reduces wind and solar curtailment, substantially improves renewable energy utilization, and simultaneously increases net revenue while reducing carbon emission intensity. This synergistic effect is particularly prominent in dual-market scenarios. Compared to baseline strategies that do not participate in or only participate in a single market, it achieves dual optimization of energy absorption and emission reduction without sacrificing economic efficiency.
[0014] At the solution level, the particle swarm optimization algorithm integrating Q-learning dynamically balances global exploration and local convergence through an adaptive parameter tuning mechanism of "state-action-reward-Q-table update," effectively suppressing the defects of traditional particle swarm optimization algorithms such as premature convergence and large fluctuations in solution quality. Simultaneously, by combining diverse rewards and feasible region projection / penalty processing, the solution achieves stable convergence under hard constraints such as power balance, gas turbine ramping, energy storage power and energy conservation, wind and solar power caps, and tie-line capacity, avoiding the risk of exceeding operational limits. Multiple runs under different stochastic initial values and different wind, solar, and electricity price scenarios demonstrate that this method converges faster, yields more robust results, has a smaller optimal value variance, and exhibits good repeatability in its strategy structure (such as energy storage charging and discharging periods and gas turbine load patterns).
[0015] The model in this application is constructed using a "benefit / cost breakdown + constraint modularization" approach, making it easy to extend to more refined costs and constraints, or incorporate more flexible resources without changing the core framework. The output is a 24-hour hourly power and energy sequence, which can be directly interfaced with the scheduling master station's planning layer. It is simple to implement in engineering and has moderate computational resource requirements. If insufficient state information is encountered or low-risk operation is required, Q-learning can be disabled or the exploration intensity reduced; the algorithm smoothly degrades to the standard particle swarm optimization algorithm, still providing a feasible solution. When the market is closed or only a single mechanism is used, the model can automatically simplify the corresponding terms and constraints without requiring a change of solver.
[0016] In summary, this application achieves simultaneous improvements in "higher net income, stronger renewable energy absorption, and lower carbon emissions" under complex nonlinear and multi-constraint environments through the coupled innovation of "unified modeling of dual market incentives + Q-learning adaptive parameter tuning". It has obvious non-obviousness and engineering promotion value. Attached Figure Description
[0017] Figure 1 This is a diagram of the virtual power plant structure under the dual mechanisms of carbon trading and green certificates in this application; Figure 2 This application uses a particle swarm optimization algorithm incorporating Q-learning to solve the objective function, resulting in a flowchart of the optimal scheduling scheme for the virtual power plant. Figure 3 Typical daily wind power and photovoltaic power output and electricity load curves are provided for embodiments of this application; Figure 4 The time-of-use electricity price curve provided in the embodiments of this application; Figure 5 A diagram illustrating a 24-hour virtual power plant dispatching scheme for carbon-free trading and green certificate trading provided in this application embodiment; Figure 6 A diagram illustrating a 24-hour dispatch scheme for a virtual power plant based solely on carbon trading, provided as an embodiment of this application. Figure 7A diagram illustrating a 24-hour dispatch scheme for a virtual power plant under a dual mechanism of carbon trading and green certificates, as provided in this application embodiment. Detailed Implementation
[0018] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.
[0019] The carbon trading discussed in this application refers to a market-based environmental policy tool designed to reduce greenhouse gas emissions, particularly carbon dioxide (CO2), through economic incentives. The core idea of carbon trading is to control total carbon emissions by setting an overall carbon emission cap (Cap) and allocating emission rights in the form of allowances to various enterprises or industries. Enterprises can buy and sell these emission allowances through market transactions to meet their emission needs. Setting an emission cap (Cap): An overall carbon emission cap is set, which is typically reduced annually to gradually reduce greenhouse gas emissions. The emission cap is broken down into specific emission allowances and allocated to various enterprises or industries. Allowances can be allocated free of charge or through auctions or other paid methods. If an enterprise's actual emissions are lower than its allowances, it can sell excess allowances to other enterprises; if its actual emissions exceed its allowances, it needs to purchase additional allowances. Through market trading, the price of carbon emission rights is determined by market supply and demand.
[0020] Green certificate trading refers to the process by which green electricity generated by renewable energy power generation projects is sold on a dedicated trading platform through the issuance of green electricity certificates.
[0021] The present application will now be described in further detail with reference to the accompanying drawings. Figure 1 As shown, a virtual power plant optimal scheduling method under a dual mechanism is characterized by the following: Establish a virtual power plant, including the interaction between gas turbines, wind turbines, photovoltaic units, energy storage devices and the power grid, and set up a market participation mode switch; Based on the predicted output of wind power and photovoltaic power and typical daily load data, determine the upper and lower limits and initial values of decision variables; An optimization configuration model is set up for the virtual power plant with the objective function of maximizing net revenue. The revenue includes electricity sales, carbon quota settlement and green certificate trading; the costs include gas turbine operation, wind and solar operation and maintenance, wind and solar curtailment penalties, energy storage operation and maintenance and grid power purchase and sale; constraints are set for power balance, unit output / ramp, energy storage power and energy, wind and solar forecast upper limit and tie line capacity. The objective function is solved by integrating Q-learning into the particle swarm optimization algorithm, and the optimal scheduling scheme for the virtual power plant is obtained.
[0022] The virtual power plant in this embodiment consists of a gas turbine, a wind turbine, a photovoltaic unit, and an energy storage system, and exchanges power with the external power grid. A market participation mode switch SW1 is set to distinguish three operating scenarios: C1 (no participation), C2 (participation in carbon trading only), and C3 (participation in both carbon trading and green certificate trading).
[0023] This application's embodiment uses 24 time periods within a day for scheduling. Typical daily load curves and upper limits for wind / solar forecasts are shown below. Figure 3 As shown; Time-of-use electricity pricing is as follows: Figure 4 As shown; carbon price and green certificate price are treated as constants. Equipment capacity, efficiency, and limitations are configured according to the engineering parameter table: upper and lower limits of gas turbine rating and ramping limits, upper and lower limits of energy storage capacity and efficiency tie-line capacity, etc.
[0024] Without compromising system security, the embodiments of this application can operate in all three modes; preferably, the C3 scenario can achieve a better overall effect in terms of economy and low carbon emissions.
[0025] Establish a virtual power plant: Construct a virtual power plant structure including gas turbines, wind power, photovoltaics, energy storage, and grid interaction, and set market participation mode switches SW1∈{non-participation (C1), carbon only (C2), carbon + green certificates (C3)}; Determine the upper and lower limits and initial values of decision variables based on wind and photovoltaic power forecast output and typical daily load data: Based on typical daily load curves and wind / solar forecast curves, and combined with time-of-use electricity prices, carbon prices, and green certificate prices, determine the upper and lower limits and initial values of decision variables; With the goal of maximizing net revenue, revenue includes electricity sales, carbon quota settlement, and green certificate trading; costs include gas turbine operation, wind and solar operation and maintenance, wind and solar curtailment penalties, energy storage operation and maintenance, and grid power purchase and sale; Set constraints such as power balance, unit output / ramp-upgrade, energy storage power and energy, wind and solar forecast upper limits, and tie-line capacity; Adopt a Q-learning enhanced particle swarm optimization algorithm, adaptively adjust inertia weights and learning factors under the state-action-reward-Q table update framework, and output a 24-hour optimized scheduling scheme and economic-low-carbon indicators.
[0026] Under the constraints of each device, the objective function of the optimization configuration model is solved by the particle swarm optimization algorithm that integrates Q-learning, and the optimal scheduling scheme of the virtual power plant for carbon trading and green certificate trading is obtained.
[0027] Specifically, the data includes wind power output data and typical daily demand data, by determining the upper and lower limits of control variables and the initial values of decision variables. The optimization configuration model for each piece of equipment includes: establishing an optimization model with the objective of maximizing net profit, and constructing a virtual power plant's intraday (24-hour) net profit maximization objective within the upper and lower limits and initial values of the decision variables. In the formula, For the revenue of VPP (Virtual Power Plant); Cost of VPP; Decision variables include: gas turbine output, photovoltaic unit output, energy storage device charging and discharging power, grid interconnection line power, carbon quota trading volume, green certificate trading volume, wind and solar curtailment power, and unit ramp-up rate.
[0028] The mathematical model for calculating the cost of the virtual power plant is as follows: 1) Gas turbine operating costs: In the formula, , , For gas turbine operating cost factors, Let i be the rated output power of the i-th gas turbine in period t. 2) Wind turbine operating costs: In the formula, This is the wind power operating cost coefficient. Let be the planned output power of the j-th wind turbine in period t.
[0029] 3) Operating costs of photovoltaic units: In the formula, This is the photovoltaic operating cost coefficient. Let k be the planned output power of the k-th photovoltaic power generation in period t. 4) Cost of wind curtailment penalties: In the formula, As a cost factor for wind curtailment penalties, Let j be the predicted power of the j-th wind power in period t. 5) Cost of abandoning light penalty: In the formula, As a cost factor for light abandonment penalties, Let j be the predicted power of the j-th photovoltaic power at time t; 6) Energy storage operating costs: In the formula, This is the operating cost coefficient for energy storage devices. The charging power of the energy storage system at time t. The discharge power of the energy storage system at time t 7) Grid interaction costs: In the formula, This refers to the price the VPP pays to the grid. The electricity purchased by the VPP from the grid. This refers to the price at which the VPP sells electricity to the grid. The amount of electricity sold by the VPP to the grid; The specific calculation model for the revenue of the established virtual power plant is as follows: 1) Revenue from electricity sales: , In the formula, The revenue generated by the VPP during period t; 2) Carbon trading revenue: , , In the formula, The revenue that VPP earns from participating in the carbon certificate exchange during period t. The price of carbon emission rights in carbon trading. This represents the carbon emission limit for VPP in period t. This represents the carbon emissions of VPP in period t. Carbon emission rights allocation coefficient per unit of electricity; 3) Benefits of Green Certificates: In the formula, The unit price of a green certificate. The renewable energy generation of the VPP in period t. The renewable energy consumption quota for VPP in period t The constraints of the optimized configuration model should include at least the following: 1) Power balance constraints: In the formula, Let t be the load demand during period t.
[0030] 2) Gas turbine output constraints: In the formula, This represents the lower limit of the output power of the i-th gas turbine. The upper limit of the output power of the i-th gas turbine And satisfy the climbing constraint: In the formula, This represents the lower limit of the ramp power for the i-th device. This represents the upper limit of the climb power for the i-th device; 3) Energy storage constraints: In the formula, This is the minimum discharge power of the energy storage device. This represents the maximum discharge power of the energy storage device. This refers to the minimum charging power of the energy storage device. This refers to the maximum charging power of the energy storage device. The amount of electricity stored in the energy storage device during period t. The discharge efficiency of energy storage devices. Improve the charging efficiency of energy storage devices. This is the minimum power required for the energy storage device during the dispatch period. The maximum power of the energy storage device during the dispatch period 4) Wind and solar power output constraints: , ; 5) Power constraints of grid-connected tie lines: , In the formula, This is the maximum power limit for the power grid contact wire. This is the minimum power limit for the power grid contact wire.
[0031] In one embodiment, the objective function is further solved by incorporating Q-learning particle swarm optimization to obtain the optimal scheduling scheme for the virtual power plant, including: Input raw data and basic parameters of the QPSO algorithm: including the operating data of the input VPP (virtual power plant), including the power demand and the predicted output of renewable energy within 24 hours; and the parameters of the input QPSO algorithm, such as the number of particles, the maximum number of iterations, the inertia weight, and the learning factor.
[0032] An initial population is randomly generated based on the power output plan of the virtual power plant within 24 hours: Based on the power output plan of the virtual power plant, a set of initial solutions is randomly generated, each solution representing a possible scheduling scheme; Calculate the fitness of each particle in the population, evaluate the performance of each particle (scheduling scheme), and calculate its fitness value. Fitness is usually related to cost, benefit, or environmental impact. The average position of the optimal values of all individual particles is used as the global guidance center: find the optimal solution of each particle in the current population, and calculate the average position of these optimal solutions as the guidance center for subsequent iterations. Obtaining the optimal position in the first generation: Determine the position of the particle with the best fitness in the initial generation as the current optimal solution; Obtain the inertia parameters and learning factor from the Q table: Based on the Q value stored in the Q table, select the optimal action for the current state, which corresponds to the setting of the inertia parameters and learning factor.
[0033] Obtain a new population: Update the particle swarm based on the current particle position, velocity, and parameters obtained from the Q table, and generate a new scheduling scheme; Calculate the fitness of the new population to find the optimal position of new particles: evaluate the fitness of each particle in the new particle population and update the individual optimal position of each particle.
[0034] New fitness calculation and optimal solution update: Update the global optimal solution based on the new fitness value; Among them, the implementation reward is obtained based on the change in fitness after the selected action is implemented. The reward function is as follows: ,in In order to improve adaptability, Rewards for changes in particle diversity; Q-table update: Update the Q value in the Q-table based on the current reward and expected future reward. ; Greedy selection (GR-greedy): Executes a greedy selection strategy, selecting the action with the largest current Q value with a certain probability and exploring randomly with a certain probability; Using ε-greedy selection: Selecting the action with the largest current Q value with a probability of 1-ε and exploring randomly with a probability of ε, where ε decays exponentially with iteration; Choose action a according to the ε-greedy strategy and map action a to the parameters (ω, c1, c2) of the particle swarm algorithm that incorporates Q-learning; Specifically, the Q table is generated based on the state space S and the action space A.
[0035] The Q-table is initialized to store the Q-value of each state-action pair. The state space S includes a three-dimensional combination (P, D, I). Based on the current state of the particle swarm, P, D, and I are evaluated. P represents the optimization stage, which is represented by early / mid / late. D represents the fitness improvement rate of recent generations, which is represented by low / medium / high. I represents the population diversity, which is represented by low / medium / high. The three-dimensional combination of P, D, and I is located in the state space S. The action space A includes nine pre-set strategy combinations (ω, c1, c2). ω represents the balance weight between exploration and exploitation. c1 represents the intensity of local exploration. c2 represents the intensity of local exploitation. The strategies cover global exploration / balance / local exploitation / convergence reinforcement, including: global exploration strategy, balanced exploration and exploitation strategy, local exploitation strategy, convergence reinforcement strategy, four different mixed strategies, and a random strategy. The actions are mapped to ω, c1, and c2 in the QPSO algorithm's velocity update, and the particles are updated accordingly until convergence, outputting a 24-hour optimized scheduling scheme.
[0036] Comparison and effects of three market scenarios: Figures 5-7 The diagram shows an overlay of three 24-hour dispatching schemes: (1) C1 (not participating): During the peak photovoltaic period at noon and the period of high wind and low load at night, power curtailment is likely to occur, and the gas turbine maintains a high output; the participation of the energy storage system is limited.
[0037] (2) C2 (carbon only): Carbon price suppresses high emission output, gas turbine output is reduced from 10:00 to 15:00, and photovoltaic grid connection is improved; energy storage shows "noon charging - evening discharging", and peak discharge is shaving during the evening peak (18:00-21:00).
[0038] (3) C3 (carbon + green certificate): Photovoltaic power is basically fully absorbed from 9:00 to 16:00, and wind power is fully absorbed from 0:00 to 4:00; energy storage is stably charged at noon and discharged at night; power purchase and sale are strategically switched during peak / valley periods, resulting in the highest net income, the lowest carbon intensity, and the highest renewability.
[0039] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the protection scope of this application.
Claims
1. A method for optimal scheduling of virtual power plants under a dual mechanism of carbon trading and green certificates, characterized in that, The method includes: Establish a virtual power plant, including the interaction between gas turbines, wind turbines, photovoltaic units, energy storage devices and the power grid, and set up a market participation mode switch; Based on the predicted output of wind power and photovoltaic power and typical daily load data, determine the upper and lower limits and initial values of decision variables; An optimization configuration model is set up for the virtual power plant with the objective function of maximizing net revenue. The revenue includes electricity sales, carbon quota settlement and green certificate trading; the costs include gas turbine operation, wind and solar operation and maintenance, wind and solar curtailment penalties, energy storage operation and maintenance and grid power purchase and sale; constraints are set for power balance, unit output / ramp, energy storage power and energy, wind and solar forecast upper limit and tie line capacity. The objective function is solved by integrating Q-learning into the particle swarm optimization algorithm, and the optimal scheduling scheme for the virtual power plant is obtained.
2. The optimal scheduling method for virtual power plants under the dual mechanism of carbon trading and green certificates as described in claim 1, characterized in that, The optimized configuration model constructs a target for maximizing the daily net revenue of a virtual power plant within the upper and lower limits of the control variables and under the initial values.
3. The optimal scheduling method for virtual power plants under the dual mechanism of carbon trading and green certificates as described in claim 1, characterized in that, Set a market participation mode switch, with the switch states including no participation, participation in carbon trading only, and participation in both carbon trading and green certificate trading.
4. The optimal scheduling method for virtual power plants under the dual mechanism of carbon trading and green certificates as described in claim 1, characterized in that, The step of determining the upper and lower limits and initial values of control variables based on wind power and solar power forecast output and typical daily load data includes: obtaining typical daily load curves and wind / solar forecast curves based on wind power and solar power forecast output and typical daily load data, and combining time-of-use electricity prices, carbon prices, and green certificate prices to determine the upper and lower limits and initial values of decision variables.
5. The optimal scheduling method for virtual power plants under the dual mechanism of carbon trading and green certificates as described in claim 1, characterized in that, Decision variables include: gas turbine output, photovoltaic unit output, energy storage device charging and discharging power, grid interconnection line power, carbon quota trading volume, green certificate trading volume, wind and solar curtailment power, and unit ramp-up rate.
6. The optimal scheduling method for virtual power plants under the dual mechanism of carbon trading and green certificates as described in claim 1, characterized in that, The objective function is solved using a particle swarm optimization algorithm incorporating Q-learning, resulting in the optimal scheduling scheme for the virtual power plant, including: An initial population is randomly generated based on the power output plan of the virtual power plant over a 24-hour period. Calculate the fitness of each particle in the population; The average position of the optimal values of all individual particles is used as the global guidance center. Determine the position of the particle with the best fitness in the initial generation as the current optimal solution; Based on the Q values stored in the Q table, select the optimal action for the current state; Based on the current position and velocity of the particles and the parameters obtained from the Q table, update the particle swarm and generate a new scheduling scheme. Evaluate the fitness of each particle in the new particle swarm and update the individual optimal position of each particle; Update the global optimal solution based on the new fitness value.
7. The optimal scheduling method for virtual power plants under the dual mechanism of carbon trading and green certificates as described in claim 6, characterized in that, The particle position and velocity are updated according to the velocity update formula of the particle swarm optimization algorithm that incorporates Q-learning; the fitness of each particle is calculated; and rewards are calculated based on fitness improvement and diversity changes. Rewards are then applied to the Q-table. Update the Q value in the Q table based on current rewards and expected future rewards; Execute a greedy selection strategy and select the action with the largest current Q value with a probability of 1-ε; The actions are mapped to parameters of a particle swarm algorithm that incorporates Q-learning.
8. The optimal scheduling method for virtual power plants under the dual mechanism of carbon trading and green certificates as described in claim 6, characterized in that, The Q table is generated based on the state space S and the action space A, and is used to store the Q value of each state-action pair. The state space S includes a three-dimensional combination (P, D, I). Based on the current state of the particle swarm, P, D, and I are evaluated. P represents the optimization stage, which is represented by early / medium / late. D represents the fitness improvement rate of recent generations, which is represented by low / medium / high. I represents the population diversity, which is represented by low / medium / high. The three-dimensional combination of P, D, and I is located in the state space S. Action space A includes 9 pre-defined strategy combinations (ω, c1, c2), where ω represents the balance weight between exploration and development, c1 represents the intensity of local exploration, and c2 represents the intensity of local development, covering strategies of global exploration / balance / local development / convergence reinforcement.
9. Initialize the Q-table, initialize the particle swarm, including position and velocity, set the initial state and initial ε value, where ε is a greedy selection: select the action with the largest current Q value with a probability of 1-ε, and explore randomly with a probability of ε, and ε decays exponentially with iteration; Based on the current state of the particle swarm, evaluate P, D, and I. P represents the optimization stage, which is represented by early / mid / late. D represents the fitness improvement rate of recent generations, which is represented by low / medium / high. I represents the population diversity, which is represented by low / medium / high. The three-dimensional combination of P, D, and I is located in the state space S. Action a is selected according to the ε-greedy policy, and action a is mapped to the parameters (ω, c1, c2) of the particle swarm algorithm that incorporates Q-learning. The root action space A is set, which includes 9 pre-set policy combinations (ω, c1, c2), where ω represents the balance weight between exploration and exploitation, c1 represents the intensity of local exploration, and c2 represents the intensity of local exploitation, covering global exploration / balance / local exploitation / convergence reinforcement policies.