Operation optimization method considering virtual power plant resource confidence interval

Abstract

The invention relates to an operation optimization method considering a virtual power plant resource confidence interval, and the method comprises the following steps: 1, constructing a distributed new energy output probability model, quantifying the output fluctuation range of resources under different confidence levels, and obtaining the comprehensive output probability interval estimation; 2, combining physical constraints of internal resources of the virtual power plant, selecting a proper calculation method from Minkowski and theory and related optimization strategies, and determining a probability feasible region of the virtual power plant; and step 3, comprehensively considering power grid scheduling requirements, resource adjustment cost and operation safety constraints, establishing a multi-target operation optimization model, solving the model through an improved genetic algorithm, outputting a real-time collaborative operation strategy of each resource of the virtual power plant, and realizing high-precision operation optimization in an uncertain environment. According to the method, the accuracy, dynamic adaptability and engineering practicability of virtual power plant operation optimization can be improved.

Description

Technical Field

[0001] This invention belongs to the field of virtual power plant operation optimization technology, and particularly relates to an operation optimization method that takes into account the resource confidence interval of virtual power plants. Background Technology

[0002] A Virtual Power Plant (VPP) is a virtual energy system that aggregates and coordinates distributed resources (such as renewable energy, energy storage systems, and demand response resources) through advanced information and communication technologies and intelligent control technologies. A VPP does not actually own a physical power plant; instead, it integrates dispersed and diverse resources into a unified whole through virtualization, enabling it to participate in grid dispatch and ancillary service provision just like a traditional power plant. VPPs play an increasingly important role in the electricity market, significantly contributing to clean energy consumption, enhancing power system flexibility, optimizing electricity market transactions, and driving the digital transformation of energy. They are a crucial tool for achieving future energy transition and sustainable development.

[0003] In the resource aggregation and operation optimization process of virtual power plants (VPPs), the uncertainty and volatility of resources are the core challenges. Existing resource aggregation methods are mainly divided into two categories: deterministic aggregation and uncertain aggregation. Deterministic aggregation methods treat the distributed resources within the virtual power plant as deterministic energy sources, assuming that the output of the resources is fixed or completely predictable. Although this method has a simple model, it fails to consider the output volatility of renewable energy sources (such as photovoltaic and wind power) and energy storage systems, resulting in a significant deviation between the aggregation results and the actual output. This leads to a disconnect between the virtual power plant's operation strategy and actual regulation capacity, making it difficult to meet the grid's technical requirements for resource reliability and flexibility.

[0004] To address resource uncertainty, scholars have proposed uncertainty aggregation methods such as scenario-based approaches and robust optimization. Scenario-based approaches describe resource output uncertainty by generating typical scenarios (e.g., high wind speed, low light intensity). However, they rely on the representativeness of historical data and suffer from insufficient coverage when dealing with multidimensional uncertainties (e.g., the coupling of multiple factors such as weather, load, and electricity prices), leading to inaccurate characterization of operational boundaries. Robust optimization methods characterize the fluctuation range of resource output by constructing uncertainty sets. While robust optimization provides some protection, its high computational complexity makes it difficult to apply rapidly in real-time markets. Furthermore, robust optimization methods typically aim for conservatism, potentially limiting the full realization of resource adjustment potential.

[0005] However, existing virtual power plant operation optimization methods have the following technical shortcomings when dealing with resource uncertainties: (1) There are shortcomings in the quantification of uncertainty. The output fluctuation range of renewable energy and other resources has not been accurately quantified by means of confidence intervals and other technical means, which makes it difficult to guarantee the reliability of the operation strategy. (2) It has poor dynamic adaptability, making it difficult to dynamically adjust and optimize strategies according to real-time resource status and grid demand, and unable to adapt to the rapid fluctuation characteristics of a high proportion of renewable energy grids; (3) There is a deficiency in the ability to coordinate multi-objective optimization. When balancing multiple technical objectives such as resource regulation accuracy, power grid stability and operation efficiency, there is a lack of a systematic optimization framework, resulting in poor overall performance of the operation strategy. (4) Furthermore, it is difficult to balance computational efficiency and accuracy. When dealing with high-dimensional uncertainty problems, existing algorithms cannot take into account both computational speed and optimization accuracy, which limits their application in practical engineering.

[0006] To address the aforementioned technical problems, this invention proposes an operation optimization method that takes into account the confidence interval of virtual power plant resources.

[0007] A search revealed no publicly available literature of the same or similar prior art as this invention. Summary of the Invention

[0008] This invention addresses the shortcomings of existing technologies by proposing an operational optimization method that takes into account the resource confidence interval of virtual power plants. By constructing a confidence interval for resource output, the uncertainty of resources is quantified and integrated into the operational strategy formulation process. This aims to improve the accuracy, dynamic adaptability, and engineering practicality of virtual power plant operational optimization, and provide technical support for the stable operation of high-proportion renewable energy power systems.

[0009] The above-mentioned objective of this invention is achieved through the following technical solution: An operation optimization method considering the resource confidence interval of a virtual power plant includes the following steps: Step 1: By combining Bootstrap resampling, ensemble learning, quantile regression and conformal quantile regression, a distributed new energy output probability model is constructed to quantify the output fluctuation range of resources at different confidence levels and obtain a comprehensive output probability interval estimate, providing accurate uncertainty boundaries for subsequent operation optimization.

[0010] Step 2: Based on the output probability interval estimate of the distributed new energy integrated obtained in Step 1, and combined with the physical constraints of the internal resources of the virtual power plant, select an appropriate calculation method from Minkowski theory and related optimization strategies to determine the probabilistic feasible region of the virtual power plant.

[0011] Step 3: Based on the probabilistic feasible region of the virtual power plant obtained in Step 2, a multi-objective operation optimization model is established by comprehensively considering grid dispatching requirements, resource regulation costs, and operational safety constraints. The model is then solved using an improved genetic algorithm to output the real-time collaborative operation strategy of each resource of the virtual power plant, thereby achieving high-precision operation optimization under uncertain conditions.

[0012] The advantages and beneficial effects of this invention are as follows: 1. This invention proposes a resource uncertainty quantification method based on "Bootstrap resampling + ensemble learning + quantile regression + conformal quantile regression". Step 1 constructs a distributed new energy output probability model to accurately characterize resource fluctuations. In step 1(1), outliers and missing data are removed and normalized through data preprocessing to provide a high-quality data foundation for subsequent modeling and avoid quantification errors caused by data bias. In step 1(2), multiple training datasets are divided using Bootstrap sampling with replacement, and the output probability intervals for each group are calculated using a base learner and quantile regression, achieving a preliminary multi-dimensional characterization of resource fluctuations. In step 1(3), multiple results are aggregated through ensemble learning, and conformal quantile regression is introduced to correct the upper and lower bounds of the intervals, ensuring the coverage accuracy of the probability intervals under different confidence levels. This solves the shortcomings of the background technology, namely, "insufficient in uncertainty quantification, failing to accurately quantify the range of resource output fluctuations" and "difficult to guarantee the reliability of the operating strategy".

[0013] 2. This invention proposes a probabilistic feasible region construction method of "resource-refined modeling + dual-method solution + confidence level quantification". Step 2 clarifies the total output adjustment boundary of the virtual power plant, taking into account both the flexibility and reliability of the operation strategy. In step 2(1), a refined model containing output characteristics and physical constraints is established for five core resources: photovoltaic, wind power, gas turbine, electric vehicle, and energy storage system, to ensure the authenticity of the constraint characterization. In step 2(2), the advantages and disadvantages of the Minkowski method and optimization method are weighed, and an appropriate solution scheme is selected through outer boundary consistency verification to accurately define the total output adjustment range under the specified confidence level. In step 2(3), the overall confidence level of the feasible region is quantified through the probability propagation algorithm to provide a quantitative basis for the reliability of the strategy. This solves the defects of the background technology, which is "difficult to balance computational efficiency and accuracy, and limited when dealing with high-dimensional uncertainty problems".

[0014] 3. This invention proposes a multi-resource collaborative optimization method of "multi-objective modeling + improved genetic algorithm + dynamic adjustment", which realizes high-precision and highly adaptable operation of virtual power plants through step 3. In step 3 (1), the grid dispatching instructions and dynamic adjustment requirements are comprehensively analyzed, and three multi-dimensional objectives are set: power tracking accuracy, resource regulation efficiency, and operation risk control, to avoid poor overall performance caused by single-objective optimization. In step 3 (2), an optimization function covering multiple objectives is constructed, integrating resource physical constraints and technical rule constraints to form a scientific optimization framework. In step 3 (3), an improved genetic algorithm is used to solve the model, and multi-resource collaboration is achieved through power allocation decision and resource dynamic adjustment. Then, through the dynamic adjustment mechanism of "1-hour data update + 6-period local optimization + threshold smooth transition", the rapid fluctuation characteristics of the power grid are adapted. This solves the defects of "poor dynamic adaptability" and "insufficient ability in multi-objective collaborative optimization" in the background technology. Attached Figure Description

[0015] Figure 1 This is a logical framework diagram of the overall technical solution of the present invention; Figure 2 The probability range for the distributed photovoltaic system of this invention; Figure 3 The probability interval for distributed wind power in this invention; Figure 4 The following is the pseudocode for the improved genetic algorithm used in this invention; Figure 5 This is a diagram of the IEEE 33-node system used in the examples of this invention; Figure 6 This is a schematic diagram of the confidence interval for the virtual power plant of the present invention. Detailed Implementation

[0016] The structure of the present invention will be further described below with reference to the accompanying drawings and embodiments. It should be noted that these embodiments are descriptive and not limiting.

[0017] An operation optimization method that takes into account the resource confidence interval of a virtual power plant, such as Figure 1 As shown, it includes the following steps: Step 1: By combining Bootstrap resampling, ensemble learning, quantile regression and conformal quantile regression, a distributed new energy output probability model is constructed to quantify the output fluctuation range of resources at different confidence levels and obtain a comprehensive output probability interval estimate, providing accurate uncertainty boundaries for subsequent operation optimization.

[0018] The specific steps of step 1 include: (1) Perform data preprocessing, collect historical output data (power generation of renewable energy) of resources within the virtual power plant, remove outliers and missing data to ensure data integrity and accuracy. Normalize the data to make it within the same scale range for easy subsequent calculations.

[0019] (2) Based on the preprocessed data obtained in step (1), use ensemble learning and quantile regression methods to construct a probability model of distributed resource output and calculate the result of the probability interval of distributed resource output. The specific steps of step 1, step (2) include: ① Samples are extracted with replacement using the Bootstrap method, and the historical data is divided into multiple training datasets; ② For each training dataset, use base learners (such as random forest, support vector machine, LSTM, etc.) and quantile regression methods to calculate the probability interval of resource output; During model training, if the input variable comes from distribution X, its quantile regression model is as shown in formula (1):

[0020] In this setting, y is denoted as the model's output value, and α represents the selected quantile. For distributed renewable energy output, the probability that its actual power falls within the corresponding confidence interval can be expressed as (1-α). The symbol F... Y (y) represents the cumulative distribution function of the random variable y. Based on the upper and lower quantiles corresponding to a given quantile α, the corresponding probability interval can be determined, specifically represented by formula (2):

[0021] Here, I record and α / 2 and 1-α / 2 represent quantile levels, respectively. They are estimates obtained through quantile regression based on historical sample data of distributed renewable energy output.

[0022] To obtain this result, it is necessary to solve an optimization problem of the form (3):

[0023] (3) Based on the distributed resource output probability interval calculated in step (2), the results of all training datasets are aggregated using an ensemble learning method to obtain the resource output probability interval of the total sample; and the upper and lower bounds of the probability interval are corrected by introducing a conformal quantile regression method to improve the accuracy of the interval estimation. The correction magnitude of the upper and lower boundaries can be determined as shown in formula (4):

[0024] in, This indicates that at the quantile level 1-α / 2, for The upper and lower bound deviation correction values ​​are calculated; φ(⋅) is the corresponding quantile calculation function. Represents the upper boundary order Column or lower boundary sequence .

[0025] The corrected probability interval can be represented by formulas (5)-(6), thus obtaining the output probability interval estimate of distributed new energy sources:

[0026]

[0027] The corrected upper and lower bounds of the interval are denoted as follows: and Correspondingly, and Indicating targeting and Error correction term.

[0028] Based on the methods mentioned above, it is possible to obtain Figure 2 The probability range of distributed photovoltaic power presented and Figure 3 The corresponding probability range for distributed wind power.

[0029] Step 2: Based on the output probability interval estimate of the distributed new energy integrated obtained in Step 1, and combined with the physical constraints of the internal resources of the virtual power plant, select an appropriate calculation method from Minkowski theory and related optimization strategies to determine the probabilistic feasible region of the virtual power plant.

[0030] This feasible region describes the total output adjustment range that the virtual power plant can achieve at a specified confidence level. Simultaneously, based on the confidence levels corresponding to the output probability intervals of each distributed renewable energy source, the overall confidence level of the virtual power plant's probabilistic feasible region is derived and calculated using a probability propagation algorithm.

[0031] The specific steps of step 2 include: (1) Based on the output probability interval estimate of the distributed new energy obtained in step 1, various distributed resources (renewable energy, controllable distributed power sources, energy storage systems, demand response resources, etc.) in the virtual power plant are modeled to describe their output characteristics, physical constraints and regulation capabilities.

[0032] Step 2, step (1), involves modeling various distributed resources in the virtual power plant. The model includes: ① Distributed photovoltaic model, as shown below: The photovoltaic generator set was modeled using statistical methods. The probability distribution of solar radiation follows a Beta distribution function, and the photovoltaic output is positively correlated with the solar intensity, as shown in formula (7).

[0033] In the formula: —Solar radiation intensity (W / m²) 2 ); , —Shape parameters of the Beta distribution; Expected value of light radiation intensity is obtained based on historical data of light radiation intensity. and variance As shown in formulas (8) and (9):

[0034]

[0035] Solar radiation intensity The probability distribution function is given by equation (10):

[0036] In the formula: max —Upper limit of solar radiation intensity (W / m²) 2 ); min —Lower limit of solar radiation intensity (W / m²) 2 The output function of photovoltaic power generation is shown in formula (11):

[0037] In the formula: — Photovoltaic power generation conversion efficiency; —Effective area of ​​photovoltaic panels (m²) 2 ); —Solar radiation intensity at any given time (W / m²) 2 ); The constraints that photovoltaic power generation should meet are shown in formulas (12) and (13): The output constraint of the photovoltaic generator is shown in formula (12):

[0038] In the formula: —Actual output (MW) of photovoltaic unit i during time period t. —The maximum output (MW) of photovoltaic unit i during time period t; The curtailment constraint of photovoltaic power generation is shown in formula (13):

[0039] In the formula: —Maximum light rejection rate; ② Distributed wind turbine model, as shown below: Its probability density function is shown in formula (14) as follows:

[0040] Where: f(v) — wind speed probability distribution density; v — actual wind speed (m / s); k — shape parameter; c — scale parameter; When the expected value and variance of the wind speed are determined, the values ​​of k and c can be obtained by the method of moments, as shown in formulas (15) and (16):

[0041]

[0042] In the formula: —The expected value of wind speed, i.e., the average wind speed (m / s); —Standard deviation of wind speed (m / s); —Gamma function; The formula for output power is shown in (17):

[0043] In the formula: —Rated power (MW) of the wind turbine unit; — Cut-in wind speed (m / s); — Cut-off wind speed (m / s); —Rated wind speed (m / s); The constraints that wind power generation should meet are shown in formulas (18) and (19): The output constraint of the wind turbine generator set is shown in formula (18):

[0044] In the formula: —Actual output (MW) of wind turbine unit k during time period t. —The maximum output (MW) of wind turbine unit k during time period t; The wind curtailment constraint of wind turbine generators is shown in formula (19):

[0045] In the formula: —Maximum wind curtailment rate; ③ Gas turbine unit model, as shown below: Its actual output G t As shown in formula (20):

[0046] In the formula: —Power generation conversion efficiency of micro gas turbines; —The calorific value of natural gas is (34.12 MJ / m3); —Volume of natural gas used at time t (m3); D —Unit conversion constant for the length of the time period; The constraints that the gas turbine must meet during operation are shown in equations (21) to (22): The output constraint of the gas turbine is shown in formula (21):

[0047] In the formula: —Lower limit of gas turbine output (MW); —The maximum output power of the gas turbine (MW); —The start-up and shutdown status variable of the gas turbine at time t is a 0-1 variable, where 1 represents running and 0 represents shutting down; The ramp constraint of the gas turbine is shown in formula (22):

[0048] In the formula: —The output of the gas turbine during time period t —The output of the gas turbine in time period t-1 —The maximum ramp rate (MW) for gas turbines. —The lower limit of the ramp rate for gas turbines (MW); The start-stop constraints of the gas turbine are shown in formulas (23) and (24):

[0049]

[0050] In the formula: — Gas turbine operating time at time t — Gas turbine shutdown time at time t —Shortest start-up time for gas turbines —Minimum downtime of gas turbine; ④ Electric vehicle model, as shown below: The formula for the charging and discharging cost of electric vehicles in time period t is shown in (25):

[0051] In the formula: — The number of electric vehicles that the virtual power plant can dispatch (vehicles); v — the index number of the electric vehicle; —The cost of discharging an electric vehicle in time period t (RMB / MWh); —Charging cost of electric vehicles in time period t (yuan / MWh); —Discharge power (MW) of the vth electric vehicle during time period t; —Charging power (MW) of the vth electric vehicle in time period t; —Electric vehicle discharge conversion efficiency; —Electric vehicle charging conversion efficiency; The operating constraints of electric vehicles are shown in formulas (26)-(28):

[0052]

[0053]

[0054] In the formula: —The lower limit of the SOC of the vth electric vehicle. —The upper limit of the SOC of the vth electric vehicle. —Maximum permissible charging power (MW) for electric vehicles —Maximum permissible discharge power (MW) for electric vehicles; ⑤ Energy storage model, as shown below:

[0055]

[0056]

[0057] The output of the energy storage system is shown by formulas (32) and (33):

[0058]

[0059] In the formula: — The state of charge of the energy storage battery at all times; —Rated capacity of the energy storage battery (MWh); , —The energy conversion efficiency of the energy storage battery during charging and discharging; , —Energy storage batteries The charging and discharging power (MW) at any given time; The constraints for the operation of energy storage facilities should satisfy formulas (34)-(38):

[0060]

[0061]

[0062]

[0063]

[0064] In the formula: —A 0-1 variable used to represent the charge / discharge state of the energy storage battery. When =1, the energy storage system is in a charging state. When the value is 0, the energy storage system is in a discharging state.

[0065] (2) Based on the various distributed resource models in the virtual power plant constructed in step (1), obtain the physical constraints of the internal resources of the virtual power plant, combine the output probability interval estimate of the distributed new energy obtained in step 1, and solve the probabilistic feasible region of the virtual power plant by selecting a suitable optimization algorithm from the optimization methods. The probabilistic feasible region describes the range of power regulation that a virtual power plant can achieve at a certain confidence level.

[0066] The main methods for solving the probabilistic feasible region of a virtual power plant fall into two categories: Minkowski summation and optimization methods. Minkowski summation methods are advantageous due to their high solution efficiency, but they struggle to incorporate network constraints. Optimization methods, on the other hand, are slower, although they can accommodate all constraints.

[0067] In solving the probabilistic feasible region of the virtual power plant, it is necessary to comprehensively weigh the advantages and disadvantages of the Minkowski method and optimization methods, and select a suitable solution.

[0068] The specific method for step (2) of step 2 is as follows: First, the Minkowski sum method and the optimization method are used simultaneously to calculate the outer boundary of the probabilistic feasible region of the virtual power plant corresponding to the upper boundary of the distributed renewable energy output, and the consistency of the two sets of calculation results is verified. If they are consistent, the inner boundary is solved using the Minkowski sum method; if the results differ, the optimization method is used to calculate the inner boundary of the probabilistic feasible region of the virtual power plant.

[0069] (3) Based on the confidence level of each distributed new energy output probability interval, the confidence level of a single resource is aggregated into the comprehensive confidence level of the entire feasible region through the probability propagation algorithm, and the confidence level of the virtual power plant in the probabilistic feasible region is quantitatively evaluated, as shown in formula (39):

[0070] The parameters in this formula are defined as follows: αi The quantile represents the probability interval of the output of the i-th distributed renewable energy source; 1-α i This indicates the confidence level for that interval. m represents the total number of distributed renewable energy sources aggregated by the virtual power plant.

[0071] Ultimately, β, as the dependent variable, measures the overall confidence level of the entire probabilistic feasible region of the virtual power plant, providing a quantitative basis for the reliability assessment of multi-resource collaborative operation strategies.

[0072] Step 3: Based on the probabilistic feasible region of the virtual power plant obtained in Step 2, a multi-objective operation optimization model is established by comprehensively considering grid dispatching requirements, resource regulation costs, and operational safety constraints. The model is then solved using an improved genetic algorithm to output the real-time collaborative operation strategy of each resource of the virtual power plant, thereby achieving high-precision operation optimization under uncertain conditions.

[0073] The specific steps of step 3 include: (1) First, conduct a power grid demand and operation target analysis, the specific steps of which include: ① Power Grid Demand Analysis: Collect power grid dispatch instructions (such as real-time power reference values, ramp rate requirements, ancillary service requirements, etc.) from the virtual power plant access points, and analyze the power grid's dynamic adjustment requirements for the virtual power plant (such as technical indicators such as load tracking, frequency response, and voltage support).

[0074] ② Operational Target Setting: Based on the technical characteristics of the virtual power plant and the grid requirements, multi-dimensional operational control targets are set. These include: power point tracking accuracy, minimizing the deviation between the actual output of the virtual power plant and the grid reference value; resource regulation efficiency, minimizing physical losses such as energy storage charging and discharging losses and gas turbine start-up and shutdown frequency; and operational risk control, controlling the output deviation risk within a preset threshold based on the confidence level of the probabilistic feasible region.

[0075] (2) Based on the analysis results of power grid demand and operation objectives in sub-step 1, a multi-objective optimization model is established in the operation optimization stage. By coordinating power tracking accuracy, resource regulation efficiency and operation risk, the safe and stable operation of the virtual power plant is achieved.

[0076] The specific steps of step 3, step (2) include: ① Establish the optimization objective function, as shown in formula (40):

[0077] Where: T—the total number of time periods within the optimization period, —The actual total output of the virtual power plant during time period t. —The grid reference power for time period t, —The weighting coefficient for time period t reflects the power grid's power tracking requirements for that time period. —Physical losses in resource adjustment during time period t. —Run risk metric (deviation risk based on probabilistic feasible region). —The risk aversion coefficient is set by the virtual power plant according to the grid stability requirements. The specific calculation of the operational risk measurement is as follows: (41)

[0078] In the formula: — is the center value of the probabilistic feasible region, representing the expected level of resource output. The specific calculation is shown in formula (42):

[0079] ② Constraint settings: These include resource constraints and technical rule constraints. For details on resource constraints, please refer to the distributed resource modeling section of the virtual power plant in step 2. The technical rule constraints are shown in formulas (43)-(45). Minimum / maximum output constraints:

[0080] Output change rate constraint (gradient constraint):

[0081] Probabilistic feasible region constraint:

[0082] In the formula: —The minimum total output (MW) allowed for the virtual power plant; —The maximum total output (MW) allowed for the virtual power plant; —The maximum ramp rate (MW / h) of output of the virtual power plant in adjacent time periods.

[0083] and Let be the lower and upper bounds of the probabilistic feasible region for the virtual power plant in time period t.

[0084] (3) Based on the multi-objective optimization model constructed in step (2), perform model solving and output the collaborative operation strategy: The specific steps of step 3 (3) include: ① Application of Optimization Algorithm: For the above-mentioned nonlinear optimization problem, this invention uses an improved genetic algorithm (GA) to solve it, and its pseudocode is as follows: Figure 4 As shown. The "improved genetic algorithm" used in this invention has been specifically improved to address the characteristics of multi-resource collaborative optimization in virtual power plants. Its main difference from the standard genetic algorithm lies in: Constrained Mutation Strategy Based on Probabilistic Feasibility Region: Standard genetic algorithms typically perform unconstrained random perturbations during mutation operations, which can easily generate "lethal individuals" that do not meet physical constraints, reducing optimization efficiency. This invention, in the mutation operation step, strictly limits the range of gene mutations of individuals to the probabilistic feasibility region of the virtual power plant solved in step 2. Between. Specifically, the implementation method is as follows: when the output gene in time period t is mutated, the mutated value... Must meet This mechanism eliminates invalid search space, ensuring that population evolution always takes place within a valid high-confidence region.

[0085] Introducing a risk-penalty fitness evaluation: Standard genetic algorithms typically only consider economic cost. This invention improves the fitness function by adding a risk metric based on the probabilistic feasible region center value, in addition to the conventional objective function (power tracking deviation + resource adjustment loss). Specifically, this is achieved by introducing a risk aversion coefficient. During the evolution process, the algorithm will automatically eliminate individuals that are low in cost but deviate from the center of the probabilistic feasible region (i.e., have higher risk), thereby finding the optimal balance between "economy" and "reliability".

[0086] The algorithm steps are as follows: Initialization: Generate an initial population, where each individual represents an operating strategy (i.e., a resource output plan for each time period), and the population size is N=100.

[0087] Fitness calculation: The fitness value of each individual is calculated according to the objective function, which is shown in formula (46):

[0088] Selection operation: A roulette wheel selection method is used to select individuals with high fitness to enter the next generation.

[0089] Crossover operation: A single-point crossover method is used, with a crossover probability p_c=0.8. The crossover point is selected by randomly choosing a time period as the crossover point.

[0090] Mutation operation: Mutate the individual with a mutation probability p_m=0.1. The mutation method is to randomly adjust the resource output plan within the probabilistic feasible region.

[0091] Termination condition: The maximum number of iterations is reached. Or the fitness value meets the preset threshold.

[0092] ② Optimization of multi-resource collaborative operation strategy: By using an improved genetic algorithm, combined with the probabilistic feasible region and resource physical constraints, the collaborative operation strategy of each resource within the virtual power plant is optimized; Specifically, this includes: power allocation decision-making, determining the actual absorption capacity of photovoltaic and wind power, energy storage charging and discharging power, gas turbine output, and electric vehicle cluster regulation within the probabilistic feasible region boundary; dynamic resource regulation, dynamically adjusting the output plan of each resource according to grid demand and real-time resource status to ensure that the response speed meets grid requirements; and multi-objective collaborative optimization, balancing power tracking accuracy, resource regulation efficiency, and operational risks through weighted coefficients to solve for the optimal operating strategy.

[0093] ③ To adapt to changes in grid demand, strategies will be dynamically adjusted, including: Real-time data updates: every Update the resource confidence interval and grid reference power information every hour.

[0094] Local optimization: Based on the information of the current time period, the operation strategy for subsequent time periods is locally optimized, and the optimization scope is the next k time periods (k=6).

[0095] Strategy Update: Update the optimized operating strategy to the virtual power plant energy management system. The update logic is shown in formula (47):

[0096] in To adjust the threshold and ensure a smooth transition of the strategy.

[0097] Example 1: This invention uses an IEEE 33-node example to verify the effectiveness of the technology, simultaneously configuring distributed wind power and distributed photovoltaic grid connection within an IEEE 33-node system. The IEEE 33-node distribution network topology is as follows: Figure 5As shown in the example, the internal resources aggregated in the virtual power plant include: a distributed photovoltaic (PV) power station, a wind farm, a battery energy storage system (BESS), a gas turbine system, and an electric vehicle (EV) cluster. The distributed PV has a rated capacity of 15 MW, with a historical data fitting prediction error range of ±15%; the distributed wind power has a rated capacity of 8 MW, with a historical data fitting prediction error range of ±20%; the BESS has a capacity of 10 MWh, a maximum charge / discharge power of 3 MW, a charge / discharge efficiency of 95%, and a state of charge (SOC) operating range of 20%-80%; the gas turbine system has a rated capacity of 4 MW, a minimum technical output of 1 MW, and a ramp rate of 1 MW / min, used to smooth out renewable energy fluctuations; the EV cluster contains 500 electric vehicles, and the aggregated total charging power demand curve is derived based on typical user behavior predictions, with a maximum adjustable charging power of 1.5 MW and a response delay ≤5 minutes. The optimization cycle is 24 hours, with each time period lasting 1 hour, simulating the daily operation optimization process of a virtual power plant. In the control objective weights, power tracking accuracy weight ω1=0.6, resource regulation loss weight ω2=0.2, and operational risk weight ω3=0.2. Risk aversion coefficient... Reflecting the requirements for power grid stability, the risk aversion coefficient in the method of this invention... The value is set to 0.5. Evaluation metrics include output deviation rate, resource utilization rate, and computational efficiency.

[0098] To highlight the technical advantages of this invention, the following three operational optimization strategies were designed for comparison: Strategy 1 (Deterministic Model): Use single-point predicted values ​​of photovoltaic and wind power as deterministic available power, and operate gas turbines and EV clusters according to a predetermined plan, ignoring their uncertainties, and allocate the output of each resource based on linear programming.

[0099] Strategy 2 (Fixed Margin Model): Considering uncertainty, a fixed margin is added only to the output of photovoltaic / wind power. During the bidding process, only the predicted power point is used, and the confidence interval is not included in the bidding strategy optimization model.

[0100] Strategy 3 (Method of this invention): Using the method proposed in this invention, firstly, confidence intervals for uncertain resources such as photovoltaic and wind power are constructed. Then, combining the physical constraints of the resources and market rules, the overall probabilistic feasible region of the virtual power plant is calculated. The probabilistic feasible region of the virtual power plant is combined with the improved genetic algorithm to optimize the multi-resource collaborative operation strategy.

[0101] This invention first utilizes ensemble learning and quantile regression methods to construct output confidence intervals for photovoltaic and wind power at a 95% confidence level. Based on these confidence intervals, and considering the output constraints of gas turbines, the charging and discharging states of energy storage, and the regulation potential of electric vehicle clusters, the overall power regulation probability feasible region of the virtual power plant is calculated, as follows: Figure 6 As shown, according to formula (39), the confidence interval of the probabilistic feasible region of the virtual power plant can be calculated to be 90.25%, meaning that the probabilistic feasible region of the virtual power plant has a 90.25% probability of being calculated accurately. The area between the solid line segment and the dashed line segment in the figure represents the range of uncertainty in resource output. This feasible region defines the upper and lower limits of power, providing a reliable technical boundary for multi-resource collaborative operation strategies.

[0102] Regarding the bid deviation rate, Strategy 1, due to its failure to consider uncertainty, exhibits a bid deviation rate of 18%-25% during periods of sudden changes in sunlight / wind speed (e.g., 10:00-12:00, 16:00-18:00), exceeding the allowable deviation threshold of the power grid. Strategy 2 reduces the deviation rate to 8%-12%, but its excessive margin leads to underutilization of resource regulation potential. Strategy 3 dynamically adjusts the power output plan through confidence intervals, controlling the deviation rate at 2.3%-4.8% throughout the entire period, meeting technical specifications. In terms of resource utilization, the method of this invention coordinates multiple resources through probabilistic feasible domains, achieving a total regulation of 92.3% of the theoretical maximum for energy storage, gas turbines, and EV clusters, a 10%-15% improvement over Strategy 1 (78.5%) and Strategy 2 (81.2%), validating the technical advantages of resource coordination and optimization. Regarding computational efficiency, the method of this invention takes 28.6 seconds for a single optimization, compared to 45.3 seconds for Strategy 2 and 15.2 seconds for Strategy 1 (the deterministic model is simple but has low accuracy). This invention, while ensuring accuracy, meets the technical requirements for real-time power grid operation optimization (≤5 minutes), proving its engineering applicability.

[0103] This example demonstrates the effectiveness of the invention through three core technical indicators: output deviation rate, resource utilization rate, and computational efficiency. The analysis results show that, in complex virtual power plant scenarios involving multiple heterogeneous resources, the proposed "operation optimization method considering the resource confidence interval of the virtual power plant" can still effectively quantify and manage resource uncertainty, providing a reliable boundary for operation optimization. Compared with traditional deterministic methods and methods that do not fully utilize confidence interval information, the method of this invention can achieve collaborative optimization among multiple energy sources. While effectively controlling risks, it significantly improves the multi-resource collaborative optimization capability of virtual power plants participating in grid dispatch, verifying the effectiveness and engineering applicability of the proposed invention.

[0104] The innovation of this invention lies in: When virtual power plants participate in grid dispatch and ancillary services, their internal resources (such as renewable energy, energy storage systems, and demand response resources) exhibit significant volatility and uncertainty. Existing operation optimization methods are typically based on deterministic assumptions and fail to adequately consider the impact of resource uncertainty on operational strategies. This leads to a disconnect between operational plans and actual regulation capabilities, potentially causing problems such as the inability to execute resource dispatch instructions and untimely responses to grid ancillary services.

[0105] Therefore, the present invention aims to solve the following technical problems: (1) How to quantify the uncertainty of internal resources of a virtual power plant and integrate it into the process of formulating an operation strategy; (2) How to balance the accuracy of resource regulation with the speed of grid response in operation optimization, and improve the flexibility and reliability of operation strategy; (3) How to optimize the multi-resource collaborative control strategy to improve the accuracy and dynamic adaptability of the overall operation optimization of the virtual power plant.

[0106] To solve the above-mentioned technical problems, the technical solution provided by the present invention is as follows: An operation optimization method considering the resource confidence interval of a virtual power plant includes the following steps: Step 1: By combining Bootstrap resampling, ensemble learning, quantile regression and conformal quantile regression, a distributed new energy output probability model is constructed to quantify the output fluctuation range of resources at different confidence levels and obtain a comprehensive output probability interval estimate, providing accurate uncertainty boundaries for subsequent operation optimization.

[0107] Step 2: Based on the distributed renewable energy output probability intervals obtained in Step 1, and considering the physical constraints of the virtual power plant's internal resources, a suitable calculation method is selected from Minkowski theory and related optimization strategies to determine the probabilistic feasible region of the virtual power plant. This feasible region describes the total output adjustment range that the virtual power plant can achieve at a specified confidence level. Simultaneously, based on the confidence levels corresponding to each distributed renewable energy output probability interval, the overall confidence level of the virtual power plant's probabilistic feasible region is derived and calculated using a probability propagation algorithm.

[0108] Step 3: Based on the probabilistic feasible region of the virtual power plant obtained in Step 2, a multi-objective operation optimization model is established, comprehensively considering grid dispatching requirements, resource regulation costs, and operational safety constraints. An improved genetic algorithm is used to solve the model, outputting real-time collaborative operation strategies for each resource of the virtual power plant, achieving high-precision operation optimization under uncertain environments.

[0109] The present invention aims to solve the above-mentioned technical problems through the following technical means: (1) How to accurately quantify the uncertainty of resources within a virtual power plant and integrate it into the operation strategy formulation process: Addressing the problem that existing technologies cannot accurately characterize the volatility of resources such as photovoltaic and wind power, this invention employs "Bootstrap resampling + ensemble learning + quantile regression" to construct a basic probability model, and further introduces "conformal quantile regression" technology to correct the deviation of the upper and lower bounds of the probability interval. This technique can accurately obtain the probability interval of the comprehensive output of distributed new energy under different confidence levels, providing a high-precision uncertainty boundary input for the operation strategy.

[0110] (2) How to balance resource regulation accuracy and grid response speed in operation optimization, and improve the flexibility and reliability of operation strategies: Addressing the problems of complex and conservative traditional robust optimization calculations and the neglect of constraints by deterministic models, this invention adopts a probabilistic feasible region construction method that combines Minkowski theory with optimization. By performing refined physical modeling of resources such as photovoltaics, wind power, and gas turbines, and using an outer boundary consistency verification mechanism to adaptively select the solution algorithm, while simultaneously using a probability propagation algorithm to quantify the comprehensive confidence level of the feasible region. This approach ensures both an accurate description of the physical constraints of resources and effectively improves the computational efficiency of the feasible region boundary.

[0111] (3) How to optimize multi-resource collaborative control strategies to improve the accuracy and dynamic adaptability of overall virtual power plant operation optimization: Addressing the issues that single-objective optimization cannot simultaneously meet the needs of multiple parties and that static strategies are difficult to adapt to real-time fluctuations, this invention establishes a multi-objective optimization model that includes power tracking, regulation costs, and operational risks, and uses an "improved genetic algorithm (GA)" for solving it. Simultaneously, combined with a dynamic adjustment mechanism of "real-time data update + local rolling optimization + threshold smooth transition," multi-timescale collaborative control of photovoltaic, wind power, energy storage, electric vehicles, and gas turbines is achieved, ensuring that the operation strategy can adapt to rapid changes in the power grid in real time.

[0112] Although embodiments and drawings of the present invention have been disclosed for illustrative purposes, those skilled in the art will understand that various substitutions, variations and modifications are possible without departing from the spirit and scope of the present invention and the appended claims. Therefore, the scope of the present invention is not limited to the contents disclosed in the embodiments and drawings.

Claims

1. An operation optimization method considering the resource confidence interval of a virtual power plant, characterized in that: Includes the following steps: Step 1: By combining Bootstrap resampling, ensemble learning, quantile regression and conformal quantile regression, a distributed new energy output probability model is constructed to quantify the output fluctuation range of resources at different confidence levels and obtain a comprehensive output probability interval estimate, providing accurate uncertainty boundaries for subsequent operation optimization. Step 2: Based on the output probability interval estimate of the distributed new energy integrated obtained in Step 1, and combined with the physical constraints of the internal resources of the virtual power plant, select an appropriate calculation method from Minkowski theory and related optimization strategies to determine the probabilistic feasible region of the virtual power plant. Step 3: Based on the probabilistic feasible region of the virtual power plant obtained in Step 2, a multi-objective operation optimization model is established by comprehensively considering grid dispatching requirements, resource regulation costs, and operational safety constraints. The model is then solved using an improved genetic algorithm to output the real-time collaborative operation strategy of each resource of the virtual power plant, thereby achieving high-precision operation optimization under uncertain conditions.

2. The operation optimization method considering the resource confidence interval of a virtual power plant according to claim 1, characterized in that: The specific steps of step 1 include: (1) Perform data preprocessing, collect historical output data of resources inside the virtual power plant, remove outliers and missing data, and normalize the data; (2) Based on the preprocessed data obtained in step (1), use ensemble learning and quantile regression methods to construct a probability model of distributed resource output and calculate the result of the probability interval of distributed resource output. (3) Based on the distributed resource output probability interval calculated in step (2), the results of all training datasets are aggregated using an ensemble learning method to obtain the resource output probability interval of the total sample; and the upper and lower bounds of the probability interval are corrected by introducing a conformal quantile regression method, and the correction magnitude of the upper and lower boundaries can be determined as shown in formula (4): ； in, This indicates that at the quantile level 1-α / 2, for The upper and lower bound deviation correction values ​​are calculated; φ(⋅) is the corresponding quantile calculation function. Represents the upper boundary order Column or lower boundary sequence ; The corrected probability interval can be represented by formulas (5)-(6), thus obtaining the output probability interval estimate of distributed new energy sources: ； ； The corrected upper and lower bounds of the interval are denoted as follows: and Correspondingly, and Indicating targeting and Error correction term.

3. The operation optimization method considering the resource confidence interval of a virtual power plant according to claim 2, characterized in that: The specific steps of step 1, step (2) include: ① Samples are extracted with replacement using the Bootstrap method, and the historical data is divided into multiple training datasets; ② For each training dataset, use the base learner and quantile regression method to calculate the probability interval of resource output; During model training, if the input variable comes from distribution X, its quantile regression model is as shown in formula (1): ； In this setting, y is denoted as the model's output value, and α represents the selected quantile; for distributed renewable energy output, the probability that its actual power falls within the corresponding confidence interval can be expressed as (1-α); symbol F Y (y) is used to represent the cumulative distribution function of the random variable y; based on the upper and lower quantiles corresponding to the given quantile α, the corresponding probability interval can be determined, specifically represented by formula (2): ； Here, I record and They represent quantile levels α / 2 and 1-α / 2, respectively, which are estimates obtained through quantile regression based on historical sample data of distributed renewable energy output. To obtain this result, it is necessary to solve an optimization problem of the form (3): 。 4. The operation optimization method considering the resource confidence interval of a virtual power plant according to claim 1, characterized in that: The specific steps of step 2 include: (1) Based on the output probability interval estimate of the distributed new energy obtained in step 1, various distributed resources in the virtual power plant are modeled to describe their output characteristics, physical constraints and regulation capabilities. (2) Based on the various distributed resource models in the virtual power plant constructed in step (1), obtain the physical constraints of the internal resources of the virtual power plant, combine the output probability interval estimate of the distributed new energy obtained in step 1, and solve the probabilistic feasible region of the virtual power plant by selecting a suitable optimization algorithm from the optimization methods.

5. The operation optimization method considering the resource confidence interval of a virtual power plant according to claim 4, characterized in that: Step 2, step (1), involves modeling various distributed resources in the virtual power plant. The model includes: ① Distributed photovoltaic model, as shown below: ； In the formula: —Solar radiation intensity (W / m²) 2 ); , —Shape parameters of the Beta distribution; Expected value of light radiation intensity is obtained based on historical data of light radiation intensity. and variance As shown in formulas (8) and (9): ； ； Solar radiation intensity The probability distribution function is given by equation (10): ； In the formula: max —Upper limit of solar radiation intensity (W / m²) 2 ); min —Lower limit of solar radiation intensity (W / m²) 2 The output function of photovoltaic power generation is shown in formula (11): ； In the formula: — Photovoltaic power generation conversion efficiency; —Effective area of ​​photovoltaic panels (m²) 2 ); —Solar radiation intensity at any given time (W / m²) 2 ); The constraints that photovoltaic power generation should meet are shown in formulas (12) and (13): The output constraint of the photovoltaic generator is shown in formula (12): ； In the formula: —Actual output (MW) of photovoltaic unit i during time period t. —The maximum output (MW) of photovoltaic unit i during time period t; The curtailment constraint of photovoltaic power generation is shown in formula (13): ； In the formula: —Maximum light rejection rate; ② Distributed wind turbine model, as shown below: Its probability density function is shown in formula (14) as follows: ； Where: f(v) — wind speed probability distribution density; v — actual wind speed (m / s); k — shape parameter; c — scale parameter; When the expected value and variance of the wind speed are determined, the values ​​of k and c can be obtained by the method of moments, as shown in formulas (15) and (16): ； ； In the formula: —The expected value of wind speed, i.e., the average wind speed (m / s); —Standard deviation of wind speed (m / s); —Gamma function; The formula for output power is shown in (17): ； In the formula: —Rated power (MW) of the wind turbine unit; — Cut-in wind speed (m / s); — Cut-off wind speed (m / s); —Rated wind speed (m / s); The constraints that wind power generation should meet are shown in formulas (18) and (19): The output constraint of the wind turbine generator set is shown in formula (18): ； In the formula: —Actual output (MW) of wind turbine unit k during time period t. —The maximum output (MW) of wind turbine unit k during time period t; The wind curtailment constraint of wind turbine generators is shown in formula (19): ； In the formula: —Maximum wind curtailment rate; ③ Gas turbine unit model, as shown below: Its actual output G t As shown in formula (20): ； In the formula: —Power generation conversion efficiency of micro gas turbines; —The calorific value of natural gas is (34.12 MJ / m3); —Volume of natural gas used at time t (m3); D —Unit conversion constant for the length of the time period; The constraints that the gas turbine must meet during operation are shown in equations (21) to (22): The output constraint of the gas turbine is shown in formula (21): ； In the formula: —Lower limit of gas turbine output (MW); —The maximum output power of the gas turbine (MW); —The start-up and shutdown status variable of the gas turbine at time t is a 0-1 variable, where 1 represents running and 0 represents shutting down; The ramp constraint of the gas turbine is shown in formula (22): ； In the formula: —The output of the gas turbine during time period t —The output of the gas turbine in time period t-1 —The maximum ramp rate (MW) for gas turbines. —The lower limit of the ramp rate for gas turbines (MW); The start-stop constraints of the gas turbine are shown in formulas (23) and (24): ； ； In the formula: — Gas turbine operating time at time t — Gas turbine shutdown time at time t —Shortest start-up time for gas turbines —Minimum downtime of gas turbine; ④ Electric vehicle model, as shown below: The formula for the charging and discharging cost of electric vehicles in time period t is shown in (25): ； In the formula: — The number of electric vehicles that the virtual power plant can dispatch (vehicles); v — the index number of the electric vehicle; —The cost of discharging an electric vehicle in time period t (RMB / MWh); —Charging cost of electric vehicles in time period t (yuan / MWh); —Discharge power (MW) of the vth electric vehicle during time period t; —Charging power (MW) of the vth electric vehicle in time period t; —Electric vehicle discharge conversion efficiency; —Electric vehicle charging conversion efficiency; The operating constraints of electric vehicles are shown in formulas (26)-(28): ； ； ； In the formula: —The lower limit of the SOC of the vth electric vehicle. —The upper limit of the SOC of the vth electric vehicle. —Maximum permissible charging power (MW) for electric vehicles —Maximum permissible discharge power (MW) for electric vehicles; ⑤ Energy storage model, as shown below: ； ； ； The output of the energy storage system is shown by formulas (32) and (33): ； ； In the formula: — The state of charge of the energy storage battery at all times; —Rated capacity of the energy storage battery (MWh); , —Energy conversion efficiency of energy storage batteries during charging and discharging; , —Energy storage batteries The charging and discharging power (MW) at any given time; The constraints for the operation of energy storage facilities should satisfy formulas (34)-(38): ； ； ； ； ； In the formula: —A 0-1 variable used to represent the charge / discharge state of the energy storage battery. When =1, the energy storage system is in a charging state. When the value is 0, the energy storage system is in a discharging state.

6. The operation optimization method considering the resource confidence interval of a virtual power plant according to claim 4, characterized in that: The specific method for step (2) of step 2 is as follows: First, the outer boundary of the probabilistic feasible domain of the virtual power plant corresponding to the upper boundary of the distributed renewable energy output is calculated using both the Minkowski sum method and the optimization method, and the consistency of the two sets of calculation results is verified. If they are consistent, the inner boundary is solved using the Minkowski sum method. If there is a difference in the results, the optimization method is used to calculate the inner boundary of the probabilistic feasible domain of the virtual power plant.

7. The operation optimization method considering the resource confidence interval of a virtual power plant according to claim 4, characterized in that: The following steps are included after step (2) of step 2: (3) Based on the confidence level of each distributed new energy output probability interval, the confidence level of a single resource is aggregated into the comprehensive confidence level of the entire feasible region through the probability propagation algorithm, and the confidence level of the virtual power plant in the probabilistic feasible region is quantitatively evaluated, as shown in formula (39): ； The parameters in this formula are defined as follows: α i The quantile represents the probability interval of the output of the i-th distributed renewable energy source; 1-α i This indicates the confidence level of the interval; m represents the total number of distributed renewable energy sources aggregated by the virtual power plant. Ultimately, β, as the dependent variable, measures the overall confidence level of the entire probabilistic feasible region of the virtual power plant, providing a quantitative basis for the reliability assessment of multi-resource collaborative operation strategies.

8. The operation optimization method considering the resource confidence interval of a virtual power plant according to claim 1, characterized in that: The specific steps of step 3 include: (1) First, conduct a power grid demand and operation target analysis; (2) Based on the analysis results of power grid demand and operation objectives in step (1), a multi-objective optimization model is established in the operation optimization stage; (3) Based on the multi-objective optimization model constructed in step (2), perform model solving and output collaborative operation strategy.

9. The operation optimization method considering the resource confidence interval of a virtual power plant according to claim 8, characterized in that: The specific steps of step 3, step (2) include: ① Establish the optimization objective function, as shown in formula (40): ； Where: T—the total number of time periods within the optimization period, —The actual total output of the virtual power plant during time period t. —The grid reference power for time period t, —The weighting coefficient for time period t reflects the power grid's power tracking requirements for that time period. —Physical losses in resource adjustment during time period t. —Run risk metric (deviation risk based on probabilistic feasible region). —The risk aversion coefficient is set by the virtual power plant according to the grid stability requirements; the specific calculation of the operation risk measurement is as follows: (41) ； In the formula: — is the center value of the probabilistic feasible region, representing the expected level of resource output; the specific calculation is shown in formula (42): ； ② Constraint settings: These include resource constraints and technical rule constraints. For details on resource constraints, please refer to the distributed resource modeling section of the virtual power plant in step 2. The technical rule constraints are shown in formulas (43)-(45). Minimum / maximum output constraints: ； Output change rate constraint (gradient constraint): ； Probabilistic feasible region constraint: ； In the formula: —The minimum total output (MW) allowed for a virtual power plant; —The maximum total output (MW) allowed by the virtual power plant; —The maximum ramp rate (MW / h) of output of the virtual power plant in adjacent time periods.

10. The operation optimization method considering the resource confidence interval of a virtual power plant according to claim 8, characterized in that: The specific steps of step 3 (3) include: ① An improved genetic algorithm (GA) is used to solve the problem. The specific steps are as follows: Initialization: Generate an initial population, where each individual represents a running strategy, and the population size is N=100; Fitness calculation: The fitness value of each individual is calculated according to the objective function, which is shown in formula (46): ； Selection operation: A roulette wheel selection method is used to select individuals with high fitness to enter the next generation; Crossover operation: A single-point crossover method is used, with a crossover probability p_c=0.

8. The crossover point is selected by randomly choosing a time period as the crossover point. Mutation operation: Mutate the individual with a mutation probability p_m=0.

1. The mutation method is to randomly adjust the resource output plan within the probability feasible region. Termination condition: The maximum number of iterations is reached. Or the fitness value meets the preset threshold; ② Optimization of multi-resource collaborative operation strategy: By using an improved genetic algorithm, combined with the probabilistic feasible region and resource physical constraints, the collaborative operation strategy of each resource within the virtual power plant is optimized; ③ To adapt to changes in grid demand, strategies will be dynamically adjusted, including: Real-time data updates: every Update the resource confidence interval and grid reference power information every hour. Local optimization: Based on the information of the current time period, the operation strategy for subsequent time periods is locally optimized, and the optimization scope is the next k time periods (k=6). Strategy Update: Update the optimized operating strategy to the virtual power plant energy management system. The update logic is shown in formula (47): ； in To adjust the threshold and ensure a smooth transition of the strategy.

Images