Virtual power plant optimization scheduling and distribution method based on user excitation and carbon tax mechanism

Abstract

A virtual power plant optimization scheduling and distribution method based on user excitation and a carbon tax mechanism comprises the following steps: establishing a user output-excitation model, adding a processing scheme for carbon emission in an objective function of optimization scheduling, and establishing a penalty reward module, the excitation method of the user output-excitation model comprises a user peak regulation response model, deterministic parameter prediction and probability distribution of uncertain parameters, the user peak regulation response model is adopted when the user output-excitation model is processed, introduction of new uncertain factors is avoided, and the user output-excitation model can be processed more accurately. In the multi-time scale optimization scheduling of the virtual power plant, not only economic factors are considered, but also a carbon emission processing link is introduced, and in addition, in the internal benefit distribution of the virtual power plant, not only peak regulation contribution of each part is considered, but also risks brought by the peak regulation contribution are considered. In addition, a reward and punishment mechanism is introduced, and fair distribution and high-quality peak regulation in the virtual power plant are further promoted.

Description

Technical Field

[0001] This invention relates to the field of power system allocation optimization, and in particular to a method for optimizing scheduling and allocating related benefits based on user incentives and carbon tax mechanisms applied to virtual power plants. Background Technology

[0002] Currently, with the vigorous development of my country's power system, the proportion of large-scale new energy power such as wind power and solar power in the power system is constantly increasing, leading to fundamental changes in the structure, form, and operation control methods of the traditional power system. Due to significant changes in the randomness, intermittency, and volatility of power output, the original uncertainty on the load side has transformed into uncertainty on both the power supply and load sides.

[0003] With the rapid development of renewable energy and the continuous growth of load demand, the imbalance between energy supply and demand is becoming increasingly prominent. Furthermore, the power generation capacity of wind and solar power is directly affected by weather conditions, resulting in intermittency and fluctuations. This can lead to unstable power supply to the grid and even potential grid security and stability issues. Moreover, the power generation methods / capacities of these new energy sources are limited by natural conditions, making their dispatchability relatively poor. This increases the difficulty of grid dispatch and restricts the absorption capacity of new energy sources. Therefore, virtual power plants, as a flexible resource with energy storage characteristics, have emerged and are playing an increasingly important role in assisting the grid in peak shaving and valley filling, and in absorbing new energy sources.

[0004] Virtual power plants, by integrating dispersed energy resources, provide the power grid with greater peak-shaving capacity, effectively alleviating the energy supply-demand imbalance and providing stable power support by rapidly responding to the grid's peak-shaving needs. During peak load periods or when renewable energy output is insufficient, virtual power plants can promptly fill power gaps, preventing grid collapse and blackouts. Furthermore, by optimizing energy allocation and dispatch strategies, virtual power plants can reduce grid operating costs. They can prioritize the use of low-cost energy resources to meet load demands while reducing the use of high-cost energy sources. In addition, virtual power plants can provide greater economic benefits to energy suppliers and users through participation in electricity market transactions.

[0005] However, in practical applications of virtual power plants, it has been found that as the accuracy of forecasting new energy sources such as wind power increases with shorter time scales, simply specifying day-ahead dispatch plans is insufficient to meet actual peak-shaving demands and the required accuracy at this stage. To achieve effective equipment management by the VPP during coordinated dispatching, a hierarchical dispatching approach with short-time, rapid response has been proposed in existing technologies, including the following:

[0006] 1. The paper “Research on Multi-Time Scale Peak Shaving Dispatch Based on Wind Power Scenarios under Data-Driven Approach [J]” (Power System Protection and Control 2023, 51(16)) proposes a multi-time scale peak shaving strategy based on wind power peak shaving scenarios, which improves the grid dispatch execution efficiency under extreme wind power scenarios.

[0007] 2. The paper "Two-stage stochastic optimization scheduling model considering flexible loads [J]" (Power System Technology, 2018, 42(11)) proposes to use flexible loads to provide backup in response to the uncertainty of wind power output, promote wind power consumption and optimize the load curve;

[0008] 3. The paper "Coordinated Optimization and Scheduling of Multiple Virtual Power Plants under High Proportion of Renewable Energy Penetration [J]" (Smart Power, 2021, 49(02)) proposes a two-stage intraday optimization scheduling model for multiple virtual power plants under high proportion of renewable energy penetration, taking into account a series of uncertain factors such as renewable energy output, power load and electricity price.

[0009] 4. The paper “Rolling scheduling model considering large-scale wind power access and multi-time scale demand response resource coordination optimization [J]” (Proceedings of the CSEE, 2016, 36(17)) proposes a model based on the uncertainty of new energy sources, considering electricity price-based demand response resources, incentive-based demand response resources at three time scales, and two types of generation-side resources with different response speeds: conventional generator sets and fast-start generator sets.

[0010] However, the objective functions of the aforementioned literature mostly only consider the economics of virtual power plants and do not address carbon costs. Moreover, the treatment of user-related aspects in the aforementioned literature involves electricity prices, resulting in double uncertainty in user output and electricity prices, which in turn reduces the accuracy of subsequent decisions. In addition to these two issues, most of the aforementioned literature studies the decision-making process of VPPs in the market with economics as the objective, but it does not cover the internal profit distribution methods of VPPs much.

[0011] In addition, existing methods for profit distribution include market pricing, negotiation, cost sharing, and contribution-based allocation. The paper "Deep Peak Shaving Market Mechanism and Clearing Model Considering Virtual Power Plant Participation [J]" (Global Energy Internet, 2020, 3(05)) proposes settling accounts based on the actual deep peak shaving volume of thermal power units and virtual power plants, and the actual declared prices of each unit, divided into time periods, using the risk contribution theory to incorporate risk factors into the allocation. However, these allocation methods fail to consider the penalty and reward mechanisms during profit distribution, thus failing to effectively improve the quality of virtual power plant participation in peak shaving and promote fairness in allocation.

[0012] In summary, existing research on virtual power plants has shortcomings when considering optimal scheduling, which can be summarized into the following three points:

[0013] 1) Peak-shaving pricing is generally involved in handling user output peak-shaving, and the introduction of both user output and pricing uncertainties has a certain impact on the results of optimized dispatching.

[0014] 2) Most studies have focused on the economics of virtual power plants and their decision-making process in the market, without addressing carbon costs. However, they have also paid less attention to the methods of distributing internal benefits within virtual power plants.

[0015] 3) When distributing benefits, the method used did not take into account the penalty and reward mechanism during the distribution, which could not effectively improve the quality of virtual power plants participating in peak shaving and promote the fairness of distribution.

[0016] It is evident that the optimized scheduling of virtual power plants under existing technologies faces multiple uncertainties and uneven distribution, which seriously weakens the high quality of peak shaving by virtual power plants. Summary of the Invention

[0017] To address the problems existing in the optimal scheduling of virtual power plants under current technologies, this invention aims to resolve the following deficiencies: the introduction of two uncertain variables in handling user loads; the failure to consider carbon emission management in the objective function during optimal scheduling; and the inability to coordinate the interests of different participants, leading to unfair distribution of total revenue. This invention proposes a virtual power plant optimal scheduling and allocation method based on user incentives and a carbon tax mechanism, as follows:

[0018] The optimization scheduling and allocation method includes: establishing a user output-incentive model, adding a carbon emission treatment scheme to the objective function of optimization scheduling, and establishing a penalty and reward module.

[0019] The virtual power plant optimization scheduling and allocation method based on user incentives and carbon tax mechanisms according to the present invention is characterized in that the "establishment of a user output-incentive model" specifically refers to establishing a user output-incentive model to reduce uncertainties in optimization scheduling and to consider user peak-shaving output. This model is based on the following scheme:

[0020] 1) When the incentive does not reach the level of comfort lost by users due to peak shaving, virtual power plant users will not reduce load in response to peak shaving.

[0021] 2) When the incentive is increased, the user of the virtual power plant loses less comfort than the incentive, and the user of the virtual power plant will actively respond to peak shaving to reduce their own load.

[0022] 3) When the incentive is further increased, the response enthusiasm of virtual power plant users will also be further improved. This improvement is non-linear. That is, based on the limited load reduction of users, when the incentive reaches the threshold, the load reduction of users participating in peak shaving response will not continue to increase.

[0023] The virtual power plant optimization scheduling and allocation method based on user incentives and carbon tax mechanisms according to the present invention is characterized in that the incentive method in establishing the user output-incentive model includes establishing a user peak-shaving response model, predicting deterministic parameters, and the probability distribution of uncertain parameters, specifically as follows:

[0024] 1) Establish a user peak-shaving response model:

[0025] The impact of changes in incentives on user load shedding is expressed using elastic demand, as shown below:

[0026]

[0027] Where: E is price elasticity, I is incentive, and r is load shedding;

[0028] This method uses comfort loss cost to simulate the response behavior to stimuli. In this approach, the comfort cost has a quadratic function relationship with the load reduction amount used for peak shaving, as shown below:

[0029]

[0030] Where: α cos β cos It is a parameter of the comfort cost function, which measures the amount of load reduction and comfort sacrifice made by users participating in peak shaving.

[0031] Combining the above two equations, we can obtain the response parameters for user participation in peak shaving:

[0032]

[0033] Due to α cos β cos The value of is not fixed, therefore the peak-shaving response parameter cannot provide accurate information on the user's peak-shaving potential for refined decision-making. A piecewise function needs to be used on this basis, as follows:

[0034]

[0035] Where a0, a1, and a2 are parameters, a0 represents the degree of change of the load reduction amount of a user participating in peak shaving as the incentive changes, and is an uncertain parameter, represented by x0, while x1 to x4 are fixed values ​​of a certain user in a round of peak shaving potential assessment, which are deterministic parameters.

[0036] 2) Deterministic parameter prediction: Parameter prediction is performed using a Long Short-Term Memory (LSTM) network and a Hybrid Density Network (MDN).

[0037] 3) Probability distribution of the uncertainty parameter:

[0038] The values ​​of deterministic parameters x1 to x4 can be obtained from the above formula. Substituting the peak points of the response recorded by the user into the formula, x0 can be determined, which determines the quadratic function. Let the response data recorded by the user be (I0, r0), and the formula is as follows:

[0039]

[0040] To obtain the relationship between the uncertain parameter x0 and external conditions such as weather and date, and stimuli, a traditional neural network is combined with a hybrid density model. The traditional neural network is used to approximate the probability distribution of the target variable by approximating an arbitrary function.

[0041] The virtual power plant optimization scheduling and allocation method based on user incentives and carbon tax mechanisms according to the present invention is characterized in that the Long Short-Term Memory (LSTM) network parameter prediction in the deterministic parameter prediction specifically comprises:

[0042] 1) Input data:

[0043] The dataset is split into training and testing sets, and the output value of each neuron is computed forward. Then, the mean absolute error of the neuron is calculated, a loss function is constructed, and the network weight parameters are updated according to the gradient guidance of the loss function.

[0044] 2) In step 1) above, when the root mean square error is less than the set value, the model is validated using a test set;

[0045] 3) After verification, continue to calculate the output value of each neuron until the end.

[0046] The Long Short-Term Memory (LSTM) network used in this invention is designed based on the recursive properties of sequence data, such as language, speech, and time series. It learns the relationships and patterns between input feature values ​​and corresponding output values ​​to predict the required values. LSTM is a feedback-type neural network specifically designed for processing sequence data. It introduces a gating mechanism to solve problems such as gradient explosion and gradient vanishing that exist in RNNs. The input metrics to LSTM are load curve peak-shaving related data, such as peak / valley time percentage, hourly load rate, peak-valley difference rate, and also the response start and end times and temporal and spatial scales.

[0047] According to the virtual power plant optimization scheduling and allocation method based on user incentives and carbon tax mechanisms of the present invention, the hybrid density network (MDN) parameter prediction in the deterministic parameter prediction specifically comprises:

[0048] 1) Initialization:

[0049] Input data, set the number of hidden layer units and the number of hybrid elements, and calculate the output layer dimension;

[0050] 2) Model building:

[0051] The neural network is built using Keras, with a fully connected hidden layer using the tanh activation function, and a prediction model function is called to extract the mean, standard deviation, and scale. A custom loss function is used, which trains the model by maximizing the likelihood, i.e. minimizing the negative log-likelihood.

[0052] 3) Model compilation and training:

[0053] The model is compiled using the Adam optimizer and a custom loss function, and trained using the training data through multiple iterations by calling the model.fit function.

[0054] 4) Model Evaluation:

[0055] After the above training is completed, the model.predict function is called to predict the test data, and the get_mixture_coef function is used to extract the probability, mean and standard deviation from the prediction results.

[0056] The hybrid density network (MDN) used in this invention uses a linear combination of different probability distribution patterns to fit the probability distribution of the required parameter function. The core function of MDN is often the normal distribution function or the Laplace function. Here, the more classic normal distribution function is used.

[0057] The virtual power plant optimization scheduling and allocation method based on user incentives and carbon tax mechanisms according to the present invention is characterized in that "adding a carbon emission treatment scheme to the objective function of optimization scheduling" specifically refers to:

[0058] 1) Each user of the virtual power plant is pre-set to have a certain carbon emission quota. All these carbon emission quotas are added together to form the total carbon emission quota of the virtual power plant. Multi-timescale calculation methods are used to optimize the accuracy of the total carbon emission quota.

[0059] 2) If the total carbon emission allowance exceeds the preset quota, the excess carbon emission allowance will be purchased from other power plants or users to increase the total carbon emission allowance.

[0060] 3) If the final emissions still exceed the total carbon emission allowance after purchasing carbon emission credits, taxes will be paid on the excess carbon emissions.

[0061] 4) If the virtual power plant has a carbon emission surplus, it can be sold back to other power plants or users.

[0062] Regarding carbon management, considering that the driving force of carbon emissions on the power generation side comes from consumer demand, as well as the uneven regional economic development and resource distribution, some literature points out that "consumers" should be responsible for the carbon dioxide emitted during the production process. At the same time, some literature proposes to charge users a tax on the carbon emissions generated per unit of energy consumed. Therefore, this invention makes improvements based on this.

[0063] The virtual power plant optimization scheduling and allocation method based on user incentives and carbon tax mechanisms according to the present invention is characterized in that the multi-timescale calculation method optimization includes establishing a day-ahead optimization scheduling model and an objective function for day-ahead scheduling schemes, and establishing an intraday optimization scheduling model and an objective function for intraday scheduling schemes, which are respectively:

[0064] 1) Current optimized scheduling model:

[0065] 1) Objective function:

[0066] The current optimized dispatch model is a peak-shaving model with virtual power plants, which includes wind turbines, energy storage devices, and adjustable loads. It also needs to consider users' carbon trading situation. Its objective function is to maximize the profit of the virtual power plants, as shown in the following formula:

[0067]

[0068] 1.1) Peak-shaving revenue from virtual power plants, as follows:

[0069]

[0070] Where, λ peak，t P represents the price offered by the dispatch center for peak shaving incentives at different time periods. peak，t P represents the amount of electricity that the virtual power plant participates in peak shaving. peak,t,pre For the predicted peak-shaving amount, π ω Let ω be the probability under scenario ω, and Δt be the duration;

[0071] 1.2) Costs of distributed energy units are as follows:

[0072]

[0073] Where, λ t For the unit operating cost of new energy units, F g，t F f，t The costs of curtailing solar power and wind power are respectively θ. g，t ,θ f，t The cost of curtailment of solar and wind power per unit time (t), P g,a,tω 、Pf,a,tω This refers to the curtailment power of solar and wind power.

[0074] 1.3) User load cost, as follows:

[0075] F l The costs required to mobilize demand-side loads to participate in peak shaving include the costs of incentivizing loads to participate in peak shaving and the penalty costs of load loss;

[0076]

[0077] Among them, I tω To incentivize users to participate in peak shaving, P cor,tω To incentivize corresponding users to contribute, P loss，tω Let θ represent the charge loss under scenario ω. L，tω Penalty factor for load failure

[0078] 1.4) Energy storage costs are as follows:

[0079] Energy storage costs are approximately linearly related to charging and discharging power.

[0080] F s =a x P charge +P discharge )+b x

[0081] Where a x 、b x P is the cost coefficient for energy storage devices. charge P discharge The charging and discharging power of energy storage devices;

[0082] 1.5) Carbon emission costs, as follows:

[0083]

[0084] Among them, when C em >C st When μ takes the value 1, when C em <C st When μ is 0, μ takes the value 0.

[0085] λ tax λ buy , λ sell These are the taxes required when carbon emissions are exceeded, the price required to purchase carbon emission credits, and the price when selling excess carbon emission credits;

[0086] When calculating carbon emissions, the carbon emissions from wind and solar power need to be deducted. Therefore, the relationship between carbon emissions and output power can be approximated as follows:

[0087] Nco2 =0.785*(P) paek -P g，rea -P f,rea )

[0088] 2) Constraints:

[0089] 2.1) Power balance constraints, as follows:

[0090] The amount of peak shaving contributed by a virtual power plant is equal to the sum of the peak shaving output from wind and solar power, the load reduction from user participation, and the power output from energy storage devices.

[0091]

[0092] 2.2) Network security constraints are as follows:

[0093] -Pmax≤Pl,t≤Pmax

[0094] The network throughput power must not exceed the maximum capacity of the network;

[0095] 2.3) Operational constraints for photovoltaic power generation are as follows:

[0096] 0≤P g,pre,t -P g,rea,t ≤P g,rea,t

[0097] 2.4) Operational constraints for wind and solar power generation are as follows:

[0098] 0≤P f，pre,t -P f,rea,t ≤P f,rea,t

[0099] 2.5) User constraints, as follows:

[0100] k2≤P L ≤k4

[0101] Where k2 and k4 are the upper and lower limits of the user load reduction response area;

[0102] 2.6) Energy storage constraints, as follows:

[0103] P chaege,min ≤P charge ≤P chaege,max

[0104] P dischaege,min ≤P discharge ≤P dischaege,max

[0105] Among them, P dischaege,min 、P aischaege,max The maximum and minimum charging power of energy storage devices, Pchaege,min 、P chaege,max This refers to the maximum and minimum discharge power of the energy storage device.

[0106] (ii) Intraday Optimized Scheduling Model:

[0107] 1) Objective function:

[0108] Solve the aforementioned day-ahead optimal scheduling problem, and consider the load reduction P corresponding to the incentive. cor Given this, and substituting it into the intraday optimized scheduling plan, which executes every hour and sets out the scheduling output for a four-hour timeframe each time, the accuracy of forecasting new energy sources such as wind and solar power increases as the time scale decreases. Therefore, the objective function is:

[0109]

[0110] 2) Constraints:

[0111] The constraints are the same as those for the day-ahead optimization scheduling model mentioned above.

[0112] According to the virtual power plant optimization scheduling and allocation method based on user incentives and carbon tax mechanisms of the present invention, the day-ahead scheduling formulates unit start-up and shutdown plans and output operation plans 24 hours later based on short-term forecast results. The day-ahead plan is formulated every 24 hours, and each formulation plan includes 24 hours with a resolution of 1 hour. The intraday scheduling corrects the output of the day-ahead scheduling based on ultra-short-term forecasts of 15 minutes to 4 hours. That is, the intraday scheduling is rolled every 15 minutes for 4 hours, and the output of the units in the first 15-minute period is adjusted. The real-time scheduling adjusts the scheduling plan based on the intraday plan based on ultra-short-term forecasts of 5 to 15 minutes. The real-time scheduling is rolled every 5 minutes for 5 minutes.

[0113] As the accuracy of forecasting new energy sources such as wind power increases with the shortening of time scale, it is difficult to meet the actual peak-shaving demand and the required accuracy if only day-ahead scheduling plans are specified as before. Therefore, multi-scale peak-shaving is required.

[0114] According to the virtual power plant optimization scheduling and allocation method based on user incentives and carbon tax mechanisms of the present invention, the "establishment of a penalty and reward module" specifically comprises:

[0115] 1) Penalty for deviation between wind and solar power unit output and load peak shaving capacity:

[0116] To ensure the peak-shaving quality of the virtual power plant, the deviation threshold between the wind power and photovoltaic units and the load is set to 30%, based on the total deviation threshold of the virtual power plant.

[0117] 2) When the deviation exceeds 30%, this portion of the cost will not be settled, and instead, it will be transferred to the wind power and solar power units and loads with a deviation within 30%. The specific formula is as follows:

[0118] Punish1 = a i I total 70%Ppeak,i,pre>Peak,i

[0119] Set as Punish totl1 The total penalty for deviation will be allocated as a reward, with the amount of peak-shaving power and deviation value used as a reference.

[0120]

[0121] Where α j Let m be the deviation value of the j-th device, and m be the number of devices with a deviation value of 30%.

[0122] 3) Sharing the costs of carbon emissions:

[0123] In this method, wind power and photovoltaic (PV) turbines are clean energy sources that do not produce carbon emissions. Furthermore, the clean energy generated by wind power and PV turbines reduces the generation of carbon by thermal power units, thus reducing carbon emissions. Therefore, wind power and PV turbines receive rewards. The specific formula is as follows:

[0124] Formulas for calculating the carbon emission share of energy storage devices and users:

[0125]

[0126] Where Ppeak,i represents the output of energy storage or the user, and ComCO2 represents the carbon emission cost. This portion of the penalty is treated as a reward and distributed to the wind and solar power units according to the proportion of output. Let Punish total2 To allocate the total cost for this portion, the specific reward distribution is as follows:

[0127]

[0128] Where Ppeak,k represents the output of the wind power photovoltaic unit;

[0129] The benefits of wind and solar power units consist of peak-shaving output, deviation penalty rewards, and clean energy rewards:

[0130] I N =a i I tatal -μPunish1+(1-μ)award1+award2

[0131] The user's benefits consist of peak-shaving output, deviation penalty rewards, and the sharing of carbon emission treatment costs:

[0132] I u =a i I tatat -μPunish1+(1-μ)award1-Punish2

[0133] The benefits of energy storage equipment consist of peak-shaving output and the cost of carbon emission mitigation shared between the two components.

[0134] I s =a i I tatal -Punish2.

[0135] Currently, the main source of energy for users in my country is thermal power units. Therefore, users, as consumers, are responsible for this, as are energy storage equipment manufacturers. However, wind power and photovoltaic units, due to their carbon emission-free characteristics, actually share the burden of emissions from thermal power units, indirectly reducing carbon emissions.

[0136] The profit distribution mechanism of a virtual power plant needs to consider the fairness and efficiency of resource utilization. Imperfections in this mechanism often lead to potential problems such as unfair resource utilization, market distortions, and imbalances in dynamic equilibrium. To address these issues, introducing a penalty mechanism is a necessary measure in the profit distribution mechanism of a virtual power plant. The penalty-reward module established in this invention is a crucial step in this process.

[0137] According to the virtual power plant optimization scheduling and allocation method based on user incentives and carbon tax mechanism of the present invention, the total deviation threshold of the virtual power plant includes the numerical threshold of the purchased carbon emissions due to prediction deviation in "adding a carbon emission treatment scheme to the objective function of optimization scheduling".

[0138] The virtual power plant optimization scheduling and allocation method based on user incentives and carbon tax mechanisms of the present invention has achieved the following beneficial effects:

[0139] 1. The virtual power plant optimization scheduling and allocation method based on user incentive and carbon tax mechanism of the present invention adopts the user peak-shaving response model when dealing with the user output-incentive model, thus avoiding the introduction of new uncertainties;

[0140] 2. The virtual power plant optimization scheduling and allocation method based on user incentives and carbon tax mechanism of the present invention not only considers economic factors in the multi-timescale optimization scheduling of virtual power plants, but also introduces a carbon emission treatment process.

[0141] 3. The virtual power plant optimization scheduling and allocation method based on user incentives and carbon tax mechanisms of this invention considers not only the peak-shaving contribution of each part but also the risks involved in the internal benefit allocation of the virtual power plant. Furthermore, this invention introduces a reward and punishment mechanism to further promote fair allocation and high-quality peak shaving within the virtual power plant. Attached Figure Description

[0142] Figure 1 This is a schematic diagram of the load reduction incentive curve for the virtual power plant optimization scheduling and allocation method based on user incentives and carbon tax mechanisms of the present invention.

[0143] Figure 2 This is a schematic diagram of the LSTM prediction deterministic parameter results of the virtual power plant optimization scheduling and allocation method based on user incentives and carbon tax mechanisms of the present invention.

[0144] Figure 3 This is a schematic diagram illustrating the probability density of uncertainty parameters in the MDN-based virtual power plant optimization scheduling and allocation method based on user incentives and carbon tax mechanisms of the present invention.

[0145] Figure 4 This is a schematic diagram of the carbon-containing virtual power plant optimization scheduling method based on user incentives and carbon tax mechanism of the present invention.

[0146] Figure 5 This is a schematic diagram of the virtual power plant optimization scheduling method based on user incentives and carbon tax mechanism of the present invention, which includes only carbon tax. Detailed Implementation

[0147] The following description, in conjunction with the accompanying drawings and embodiments, further describes the technical means, creative features, objectives, and effects of the virtual power plant optimization scheduling and allocation method based on user incentives and carbon tax mechanisms of the present invention.

[0148] Example

[0149] The virtual power plant optimization scheduling and allocation method based on user incentives and carbon tax mechanisms includes: establishing a user output-incentive model, adding carbon emission treatment schemes to the objective function of optimization scheduling, and establishing a penalty and reward module.

[0150] like Figure 1 As shown, "Establishing a User Output-Incentive Model" specifically involves establishing a user output-incentive model to reduce uncertainties in optimized scheduling and to consider user peak-shaving output. This model is based on the following scheme:

[0151] 1) When the incentive does not reach the level of comfort lost by users due to peak shaving, virtual power plant users will not reduce load in response to peak shaving.

[0152] 2) When the incentive is increased, the user of the virtual power plant loses less comfort than the incentive, and the user of the virtual power plant will actively respond to peak shaving to reduce their own load.

[0153] 3) When the incentive is further increased, the response enthusiasm of virtual power plant users will also be further improved. This improvement is non-linear. That is, based on the limited load reduction of users, when the incentive reaches the threshold, the load reduction of users participating in peak shaving response will not continue to increase.

[0154] In this embodiment, the incentive method in establishing the user output-incentive model includes establishing a user peak-shaving response model, predicting deterministic parameters, and determining the probability distribution of uncertain parameters, specifically as follows:

[0155] 1) Establish a user peak-shaving response model:

[0156] The impact of changes in incentives on user load shedding is expressed using elastic demand, as shown below:

[0157]

[0158] Where: E is price elasticity, I is incentive, and r is load shedding;

[0159] This method uses comfort loss cost to simulate the response behavior to stimuli. In this approach, the comfort cost has a quadratic function relationship with the load reduction amount used for peak shaving, as shown below:

[0160]

[0161] Where: α cos β cos It is a parameter of the comfort cost function, which measures the amount of load reduction and comfort sacrifice made by users participating in peak shaving.

[0162] Combining the above two equations, we can obtain the response parameters for user participation in peak shaving:

[0163]

[0164] Due to α cos β cos The value of is not fixed, therefore the peak-shaving response parameter cannot provide accurate information on the user's peak-shaving potential for refined decision-making. A piecewise function needs to be used on this basis, as follows:

[0165]

[0166] a0, a1, a2

[0167] Where x0 is a parameter, a0 represents the degree of change of the load reduction amount of a user participating in peak shaving as the incentive changes, and is an uncertain parameter, represented by x0, while x1 to x4 are fixed values ​​of a certain user in a round of peak shaving potential assessment, which are deterministic parameters.

[0168] 2) Deterministic parameter prediction: Parameter prediction is performed using a Long Short-Term Memory (LSTM) network and a Hybrid Density Network (MDN).

[0169] 3) Probability distribution of the uncertainty parameter:

[0170] The values ​​of deterministic parameters x1 to x4 can be obtained from the above formula. Substituting the peak points of the response recorded by the user into the formula, x0 can be determined, which determines the quadratic function. Let the response data recorded by the user be (I0, r0), and the formula is as follows:

[0171]

[0172] To obtain the relationship between the uncertain parameter x0 and external conditions such as weather and date, and stimuli, a traditional neural network is combined with a hybrid density model. The traditional neural network is used to approximate the probability distribution of the target variable by approximating an arbitrary function.

[0173] like Figure 2 As shown, the parameter prediction of the Long Short-Term Memory (LSTM) network in deterministic parameter prediction is specifically as follows:

[0174] 1) Input data:

[0175] The dataset is split into training and testing sets, and the output value of each neuron is computed forward. Then, the mean absolute error of the neuron is calculated, a loss function is constructed, and the network weight parameters are updated according to the gradient guidance of the loss function.

[0176] 2) In step 1) above, when the root mean square error is less than the set value, the model is validated using a test set;

[0177] 3) After verification, continue to calculate the output value of each neuron until the end.

[0178] like Figure 3 As shown, the parameter prediction for hybrid density network (MDN) in deterministic parameter prediction is as follows:

[0179] 1) Initialization:

[0180] Input data, set the number of hidden layer units and the number of hybrid elements, and calculate the output layer dimension;

[0181] 2) Model building:

[0182] The neural network is built using Keras, with a fully connected hidden layer using the tanh activation function, and a prediction model function is called to extract the mean, standard deviation, and scale. A custom loss function is used, which trains the model by maximizing the likelihood, i.e. minimizing the negative log-likelihood.

[0183] 3) Model compilation and training:

[0184] The model is compiled using the Adam optimizer and a custom loss function, and trained using the training data through multiple iterations by calling the model.fit function.

[0185] 4) Model Evaluation:

[0186] After the above training is completed, the model.predict function is called to predict the test data, and the get_mixture_coef function is used to extract the probability, mean and standard deviation from the prediction results.

[0187] "Add a carbon emission treatment scheme to the objective function of the optimized scheduling," which specifically means:

[0188] 1) Each user of the virtual power plant is pre-set to have a certain carbon emission quota. All these carbon emission quotas are added together to form the total carbon emission quota of the virtual power plant. Multi-timescale calculation methods are used to optimize the accuracy of the total carbon emission quota.

[0189] 2) If the total carbon emission allowance exceeds the preset quota, the excess carbon emission allowance will be purchased from other power plants or users to increase the total carbon emission allowance.

[0190] 3) If the final emissions still exceed the total carbon emission allowance after purchasing carbon emission credits, taxes will be paid on the excess carbon emissions.

[0191] 4) If the virtual power plant has a carbon emission surplus, it can be sold back to other power plants or users.

[0192] refer to Figure 4 Carbon-containing virtual power plant optimization scheduling and Figure 5 The optimization of virtual power plant scheduling with only carbon tax is achieved through multi-timescale computational methods, including establishing day-ahead optimization scheduling models and objective functions for day-ahead scheduling schemes, and establishing intraday optimization scheduling models and objective functions for intraday scheduling schemes, as follows:

[0193] 1) Current optimized scheduling model:

[0194] 1) Objective function:

[0195] The current optimized dispatch model is a peak-shaving model with virtual power plants, which includes wind turbines, energy storage devices, and adjustable loads. It also needs to consider users' carbon trading situation. Its objective function is to maximize the profit of the virtual power plants, as shown in the following formula:

[0196]

[0197] 1.1) Peak-shaving revenue from virtual power plants, as follows:

[0198]

[0199] P peak,t,pre P peak,t ≥P peak,t,pre

[0200] F peak,t =P peak,t 0.7P peak,t,pre ≤P peak,t <P peak,t,pre

[0201] 0 0.7P peak,t,pre >P peak,t

[0202] Where, λ peak，t P represents the price offered by the dispatch center for peak shaving incentives at different time periods. peak，t P represents the amount of electricity that the virtual power plant participates in peak shaving. peak，t，pre For the predicted peak-shaving amount, π ω Let ω be the probability under scenario ω, and Δt be the duration;

[0203] 1.2) Costs of distributed energy units are as follows:

[0204]

[0205] Where, λ t For the unit operating cost of new energy units, F g，t F f,t The costs of curtailing solar power and wind power are respectively θ. g，t ,θ f，t The cost of curtailment of solar and wind power per unit time (t), P g,a,tω 、P f,a,tω This refers to the curtailment power of solar and wind power.

[0206] 1.3) User load cost, as follows:

[0207] F l The costs required to mobilize demand-side loads to participate in peak shaving include the costs of incentivizing loads to participate in peak shaving and the penalty costs of load loss;

[0208]

[0209] Among them, I tω To incentivize users to participate in peak shaving, P cor，tω To incentivize corresponding users to contribute, P loss，tω Let θ represent the charge loss under scenario ω. L,tω Penalty factor for load failure

[0210] 1.4) Energy storage costs are as follows:

[0211] Energy storage costs are approximately linearly related to charging and discharging power.

[0212] F s =a x (P charge +P discharge )+b x

[0213] Where a x 、b x P is the cost coefficient for energy storage devices. charge 、P discharge The charging and discharging power of energy storage devices;

[0214] 1.5) Carbon emission costs, as follows:

[0215]

[0216] Among them, when C em <C st When μ takes the value 1, when C em <C st When μ is 0, μ takes the value 0.

[0217] λ tax , λ buy , λ sell These are the taxes required when carbon emissions are exceeded, the price required to purchase carbon emission credits, and the price when selling excess carbon emission credits;

[0218] When calculating carbon emissions, the carbon emissions from wind and solar power need to be deducted. Therefore, the relationship between carbon emissions and output power can be approximated as follows:

[0219] N co2 =0.785*(P) paek -P g,rea -P f,rea )

[0220] 2) Constraints:

[0221] 2.1) Power balance constraints, as follows:

[0222] The amount of peak shaving contributed by a virtual power plant is equal to the sum of the peak shaving output from wind and solar power, the load reduction from user participation, and the power output from energy storage devices.

[0223]

[0224] 2.2) Network security constraints are as follows:

[0225] -Pmax≤Pl,t≤Pmax

[0226] The network throughput power must not exceed the maximum capacity of the network;

[0227] 2.3) Operational constraints for photovoltaic power generation are as follows:

[0228] 0≤P g,pre,t -P g,rea,t ≤P g,rea,t

[0229] 2.4) Operational constraints for wind and solar power generation are as follows:

[0230] 0≤P f,pre,t -P f,rea,t ≤P f,rea,t

[0231] 2.5) User constraints, as follows:

[0232] k2≤P L ≤k4

[0233] Where k2 and k4 are the upper and lower limits of the user load reduction response area;

[0234] 2.6) Energy storage constraints, as follows:

[0235] P chaege,min ≤P charge ≤P chaege,max

[0236] P dischaege,min ≤P discharge ≤P dischaege,max

[0237] Among them, P chaege,min 、P chaege,max The maximum and minimum charging power of energy storage devices, P dischaege,min 、P dischaege,max This refers to the maximum and minimum discharge power of the energy storage device.

[0238] (ii) Intraday Optimized Scheduling Model:

[0239] 1) Objective function:

[0240] Solve the aforementioned day-ahead optimal scheduling problem, and consider the load reduction P corresponding to the incentive.cor Given this, and substituting it into the intraday optimized scheduling plan, which executes every hour and sets out the scheduling output for a four-hour timeframe each time, the accuracy of forecasting new energy sources such as wind and solar power increases as the time scale decreases. Therefore, the objective function is:

[0241]

[0242] 2) Constraints:

[0243] The constraints are the same as those for the day-ahead optimization scheduling model mentioned above.

[0244] The day-ahead dispatcher formulates the unit start-up and shutdown plan and output operation plan 24 hours later based on short-term forecast results. This day-ahead plan is formulated every 24 hours, and each formulation plan includes 24 hours with a resolution of 1 hour. The intraday dispatcher, on the other hand, corrects the output of the day-ahead dispatcher based on ultra-short-term forecasts of 15 minutes to 4 hours. That is, the intraday dispatcher rolls the plan every 15 minutes, with each roll lasting 4 hours, and adjusts the output of the units in the first 15-minute period. The real-time dispatcher adjusts the dispatch plan based on the intraday plan using ultra-short-term forecasts of 5 to 15 minutes. The real-time dispatcher rolls the plan every 5 minutes, with each roll lasting 5 minutes.

[0245] "Establish a punishment and reward module," which specifically includes:

[0246] 1) Penalty for deviation between wind and solar power unit output and load peak shaving capacity:

[0247] To ensure the peak-shaving quality of the virtual power plant, the deviation threshold between the wind power and photovoltaic units and the load is set to 30%, based on the total deviation threshold of the virtual power plant.

[0248] 2) When the deviation exceeds 30%, this portion of the cost will not be settled, and instead, it will be transferred to the wind power and solar power units and loads with a deviation within 30%. The specific formula is as follows:

[0249] Punish1 = a i I total 70%Ppeak,i,pre>Peak,i

[0250] Set as Punish total1 The total penalty for deviation will be allocated as a reward, with the amount of peak-shaving power and deviation value used as a reference.

[0251]

[0252] Where α j Let m be the deviation value of the j-th device, and m be the number of devices with a deviation value of 30%.

[0253] 3) Sharing the costs of carbon emissions:

[0254] In this method, wind power and photovoltaic (PV) turbines are clean energy sources that do not produce carbon emissions. Furthermore, the clean energy generated by wind power and PV turbines reduces the generation of carbon by thermal power units, thus reducing carbon emissions. Therefore, wind power and PV turbines receive rewards. The specific formula is as follows:

[0255] Formulas for calculating the carbon emission share of energy storage devices and users:

[0256]

[0257] Where Ppeak,i represents the output of energy storage or the user, and ComCO2 represents the carbon emission cost. This portion of the penalty is treated as a reward and distributed to the wind and solar power units according to the proportion of output. Let Punish total2 To allocate the total cost for this portion, the specific reward distribution is as follows:

[0258]

[0259] Where Ppeak,k represents the output of the wind power photovoltaic unit;

[0260] The benefits of wind and solar power units consist of peak-shaving output, deviation penalty rewards, and clean energy rewards:

[0261] I N =a i I tatal -μPunish1+(1-μ)award1+award2

[0262] The user's benefits consist of peak-shaving output, deviation penalty rewards, and the sharing of carbon emission treatment costs:

[0263] I u =a i I tatal -μPunish1+(1-μ)award1-Punish2

[0264] The benefits of energy storage equipment consist of peak-shaving output and the cost of carbon emission mitigation shared between the two components.

[0265] I s =a i I tatal -Punish2.

[0266] In this embodiment, the "establishment of a penalty and reward module" can be referenced from the examples in Tables 1 and 2 below.

[0267]

[0268] Table 1. Internal Benefit Distribution Mechanism of Virtual Power Plant Based on Reward and Punishment Mechanism

[0269]

[0270] Table 2 Comparison of Different Allocation Mechanisms

[0271] It should be noted that the total deviation threshold for the virtual power plant includes the numerical threshold for the carbon emissions purchased due to prediction bias in "Adding a carbon emission treatment scheme to the objective function of optimized scheduling".

[0272] The virtual power plant optimization scheduling and allocation method based on user incentives and carbon tax mechanisms of this invention adopts a user peak-shaving response model when handling the user output-incentive model, avoiding the introduction of new uncertainties. Furthermore, in the multi-timescale optimization scheduling of virtual power plants, this invention considers not only economic factors but also carbon emission treatment. In addition, the internal benefit allocation of this invention considers not only the peak-shaving contribution of each component but also the associated risks. Moreover, this invention introduces a reward and penalty mechanism to further promote fair allocation and high-quality peak-shaving within the virtual power plant.

[0273] Those skilled in the art should recognize that the above embodiments are merely illustrative of this application and are not intended to limit this application. Any variations or modifications to the above embodiments that fall within the scope of the essential spirit of this application will fall within the scope of the claims of this application.

Claims

1. A virtual power plant optimization scheduling and allocation method based on user incentives and carbon tax mechanisms is as follows: The optimized scheduling and allocation method includes: Establish a user output-incentive model, add carbon emission treatment schemes to the objective function of optimization scheduling, and establish a penalty and reward module.

2. The virtual power plant optimization scheduling and allocation method based on user incentives and carbon tax mechanisms as described in claim 1, characterized in that, The aforementioned "establishment of a user output-incentive model" specifically refers to the establishment of a user output-incentive model to reduce uncertainties in optimized scheduling and to consider user peak-shaving output. This model is based on the following scheme: 1) When the incentive does not reach the level of comfort lost by users due to peak shaving, virtual power plant users will not reduce load in response to peak shaving. 2) When the incentive is increased, the user of the virtual power plant loses less comfort than the incentive, and the user of the virtual power plant will actively respond to peak shaving to reduce their own load. 3) When the incentive is further increased, the response enthusiasm of virtual power plant users will also be further improved. This improvement is non-linear. That is, based on the limited load reduction of users, when the incentive reaches the threshold, the load reduction of users participating in peak shaving response will not continue to increase.

3. The virtual power plant optimization scheduling and allocation method based on user incentives and carbon tax mechanisms as described in claim 2, characterized in that, The incentive method in establishing the user output-incentive model includes establishing a user peak-shaving response model, predicting deterministic parameters, and determining the probability distribution of uncertain parameters, specifically as follows: 1) Establish a user peak-shaving response model: The impact of changes in incentives on user load shedding is expressed using elastic demand, as shown below: Where: E is price elasticity, I is incentive, and r is load shedding; This method uses comfort loss cost to simulate the response behavior to stimuli. In this approach, the comfort cost has a quadratic function relationship with the load reduction amount used for peak shaving, as shown below: Where: α cos β cos It is a parameter of the comfort cost function, which measures the amount of load reduction and comfort sacrifice made by users participating in peak shaving. Combining the above two equations, we can obtain the response parameters for user participation in peak shaving: Due to α cos β cos The value of is not fixed, therefore the peak-shaving response parameter cannot provide accurate information on the user's peak-shaving potential for refined decision-making. A piecewise function needs to be used on this basis, as follows: Where a0, a1, and a2 are parameters, a0 represents the degree of change of the load reduction amount of a user participating in peak shaving as the incentive changes, and is an uncertain parameter, represented by x0, while x1 to x4 are fixed values ​​of a certain user in a round of peak shaving potential assessment, which are deterministic parameters. 2) Deterministic parameter prediction: Parameter prediction is performed using a Long Short-Term Memory (LSTM) network and a Hybrid Density Network (MDN). 3) Probability distribution of the uncertainty parameter: The values ​​of deterministic parameters x1 to x4 can be obtained from the above formula. Substituting the peak points of the response recorded by the user into the formula, x0 can be determined, which determines the quadratic function. Let the response data recorded by the user be (I0, r0), and the formula is as follows: To obtain the relationship between the uncertain parameter x0 and external conditions such as weather and date, and stimuli, a traditional neural network is combined with a hybrid density model. The traditional neural network is used to approximate the probability distribution of the target variable by approximating an arbitrary function.

4. The virtual power plant optimization scheduling and allocation method based on user incentives and carbon tax mechanisms as described in claim 3, characterized in that, The deterministic parameter prediction in the Long Short-Term Memory (LSTM) network specifically includes: 1) Input data: The dataset is split into training and testing sets, and the output value of each neuron is computed forward. Then, the mean absolute error of the neuron is calculated, a loss function is constructed, and the network weight parameters are updated according to the gradient guidance of the loss function. 2) In step 1) above, when the root mean square error is less than the set value, the model is validated using a test set; 3) After verification, continue to calculate the output value of each neuron until the end.

5. The virtual power plant optimization scheduling and allocation method based on user incentives and carbon tax mechanisms as described in claim 3, characterized in that, The deterministic parameter prediction of the hybrid density network (MDN) specifically includes: 1) Initialization: Input data, set the number of hidden layer units and the number of hybrid elements, and calculate the output layer dimension; 2) Model building: The neural network is built using Keras, with a fully connected hidden layer using the tanh activation function, and a prediction model function is called to extract the mean, standard deviation, and scale. A custom loss function is used, which trains the model by maximizing the likelihood, i.e. minimizing the negative log-likelihood. 3) Model compilation and training: The model is compiled using the Adam optimizer and a custom loss function, and trained using the training data through multiple iterations by calling the model.fit function. 4) Model Evaluation: After the above training is completed, the model.predict function is called to predict the test data, and the get_mixture_coef function is used to extract the probability, mean and standard deviation from the prediction results.

6. The virtual power plant optimization scheduling and allocation method based on user incentives and carbon tax mechanisms as described in claim 1, characterized in that, The aforementioned "adding a carbon emission treatment scheme to the objective function of optimized scheduling" specifically refers to: 1) Each user of the virtual power plant is pre-set to have a certain carbon emission quota. All these carbon emission quotas are added together to form the total carbon emission quota of the virtual power plant. Multi-timescale calculation methods are used to optimize the accuracy of the total carbon emission quota. 2) If the total carbon emission allowance exceeds the preset quota, the excess carbon emission allowance will be purchased from other power plants or users to increase the total carbon emission allowance. 3) If the final emissions still exceed the total carbon emission allowance after purchasing carbon emission credits, taxes will be paid on the excess carbon emissions. 4) If the virtual power plant has a carbon emission surplus, it can be sold back to other power plants or users.

7. The virtual power plant optimization scheduling and allocation method based on user incentives and carbon tax mechanisms as described in claim 6, characterized in that, The optimization of the multi-timescale calculation method includes establishing a day-ahead optimized scheduling model and an objective function for the day-ahead scheduling scheme, and establishing an intraday optimized scheduling model and an objective function for the intraday scheduling scheme, which are respectively: 1) Current optimized scheduling model: 1) Objective function: The current optimized dispatch model is a peak-shaving model with virtual power plants, which includes wind turbines, energy storage devices, and adjustable loads. It also needs to consider users' carbon trading situation. Its objective function is to maximize the profit of the virtual power plants, as shown in the following formula: 1.1) Peak-shaving revenue from virtual power plants, as follows: P peak,t,pre P peak,t ≥P peak,t，pre F peak,t =P peak,t ,0.7P p,eak,t,pre ≤P p,eak,t <P peak,t,pre 0 0.7P p,eak,t,pre >P peak,t Where, λ peak，t P represents the price offered by the dispatch center for peak shaving incentives at different time periods. peak，t P represents the amount of electricity that the virtual power plant participates in peak shaving. peak，t，pre For the predicted peak-shaving amount, π ω Let ω be the probability under scenario ω, and Δt be the duration; 1.2) Costs of distributed energy units are as follows: Where, λ t For the unit operating cost of new energy units, F g，t F f，t The costs of curtailing solar power and wind power are respectively θ. g,t ,θ f,t The cost of curtailment of solar and wind power per unit time (t), P g,a,tω 、P f,a,tω This refers to the curtailment power of solar and wind power. 1.3) User load cost, as follows: F l The costs required to mobilize demand-side loads to participate in peak shaving include the costs of incentivizing loads to participate in peak shaving and the penalty costs of load loss; Among them, I tω To incentivize users to participate in peak shaving, P cor，tω To incentivize corresponding users to contribute, P loss，tω Let θ represent the charge loss under scenario ω. L，tω Penalty factor for load failure 1.4) Energy storage costs are as follows: Energy storage costs are approximately linearly related to charging and discharging power. F s =a x (P charge +P discharge )+b x Where a x 、b x P is the cost coefficient for energy storage devices. charge P discharge The charging and discharging power of energy storage devices; 1.5) Carbon emission costs, as follows: Among them, when C em >C st When μ takes the value 1, when C em <C st When μ is 0, μ takes the value 0. λ tax , λ buy , λ sell These are the taxes required when carbon emissions are exceeded, the price required to purchase carbon emission credits, and the price when selling excess carbon emission credits; When calculating carbon emissions, the carbon emissions from wind and solar power need to be deducted. Therefore, the relationship between carbon emissions and output power can be approximated as follows: N co2 =0.785*(P paek -P g,rea -P f,rea ) 2) Constraints: 2.1) Power balance constraints, as follows: The amount of peak shaving contributed by a virtual power plant is equal to the sum of the peak shaving output from wind and solar power, the load reduction from user participation, and the power output from energy storage devices. 2.2) Network security constraints are as follows: -Pmax≤Pl,t≤Pmax The network throughput power must not exceed the maximum capacity of the network; 2.3) Operational constraints for photovoltaic power generation are as follows: 0≤P g,pre,t -P g,rea,t P g,rea,t 2.4) Operational constraints for wind and solar power generation are as follows: 0≤P f,pre,t -P f，rea,t ≤P f,rea,t 2.5) User constraints, as follows: k2≤P L ≤k4 Where k2 and k4 are the upper and lower limits of the user load reduction response area; 2.6) Energy storage constraints, as follows: P chaege,min ≤P charge ≤P chaege,max P dischaege,min ≤P discharge ≤P dischaege,max Among them, P dischaege，min 、P dischaege，max The maximum and minimum charging power of energy storage devices, P chaege，min 、P chaege，max This refers to the maximum and minimum discharge power of the energy storage device. (ii) Intraday Optimized Scheduling Model: 1) Objective function: Solve the aforementioned day-ahead optimal scheduling problem, and consider the load reduction P corresponding to the incentive. cor Given this, and substituting it into the intraday optimized scheduling plan, which executes every hour and sets out the scheduling output for a four-hour timeframe each time, the accuracy of forecasting new energy sources such as wind and solar power increases as the time scale decreases. Therefore, the objective function is: 2) Constraints: The constraints are the same as those for the day-ahead optimization scheduling model mentioned above.

8. The virtual power plant optimization scheduling and allocation method based on user incentives and carbon tax mechanisms as described in claim 7, characterized in that, The day-ahead scheduling is based on short-term forecasts to formulate unit start-up and shutdown plans and power output operation plans 24 hours later. The day-ahead plan is formulated every 24 hours, and each formulation plan includes 24 hours with a resolution of 1 hour. The intraday scheduling is based on ultra-short-term forecasts of 15 minutes to 4 hours to correct the output of the day-ahead scheduling. That is, the intraday scheduling is rolled every 15 minutes, and each roll lasts for 4 hours, adjusting the unit output for the first 15-minute period. The real-time scheduling is based on ultra-short-term forecasts of 5 to 15 minutes to adjust the scheduling plan on the basis of the intraday plan. The real-time scheduling is rolled every 5 minutes, and each roll lasts for 5 minutes.

9. The virtual power plant optimization scheduling and allocation method based on user incentives and carbon tax mechanisms as described in claim 1, characterized in that, The "establishment of a punishment and reward module" specifically refers to: 1) Penalty for deviation between wind and solar power unit output and load peak shaving capacity: To ensure the peak-shaving quality of the virtual power plant, the deviation threshold between the wind power and photovoltaic units and the load is set to 30%, based on the total deviation threshold of the virtual power plant. 2) When the deviation exceeds 30%, this portion of the cost will not be settled, and instead, it will be transferred to the wind power and solar power units and loads with a deviation within 30%. The specific formula is as follows: Punish1=a i I total 70%Ppeak,i,pre>Peak,i Set as Punish total1 The total penalty for deviation will be allocated as a reward, with the amount of peak-shaving power and deviation value used as a reference. Where α j Let m be the deviation value of the j-th device, and m be the number of devices with a deviation value of 30%. 3) Sharing the costs of carbon emissions: In this method, wind power and photovoltaic (PV) turbines are clean energy sources that do not produce carbon emissions. Furthermore, the clean energy generated by wind power and PV turbines reduces the generation of carbon by thermal power units, thus reducing carbon emissions. Therefore, wind power and PV turbines receive rewards. The specific formula is as follows: Formulas for calculating the carbon emission share of energy storage devices and users: Where P peak，i ComCO2 represents the carbon emission cost for energy storage or user output. This portion of the penalty is treated as a reward and distributed to wind and solar power units proportional to their output. Let Punish... total2 To allocate the total cost for this portion, the specific reward distribution is as follows: Among them, P peak，k For the output of wind power and solar power units; The benefits of wind and solar power units consist of peak-shaving output, deviation penalty rewards, and clean energy rewards: I N =a i I tatal -μPunish1+(1-μ)award1+award2 The user's benefits consist of peak-shaving output, deviation penalty rewards, and the sharing of carbon emission treatment costs: I u =a i I tatal -μPunish1+(1-μ)award1-Punish2 The benefits of energy storage equipment consist of peak-shaving output and the cost of carbon emission mitigation shared between the two components. I s =a i I tatal -Punish2。 10. The virtual power plant optimization scheduling and allocation method based on user incentives and carbon tax mechanisms as described in claim 9, characterized in that, The total deviation threshold of the virtual power plant includes the numerical threshold for the amount of carbon emissions purchased due to prediction bias in "adding a carbon emission treatment scheme to the objective function of optimized scheduling".

Images