A virtual power plant scheduling optimization method responding to demand bidding

Abstract

The present invention belongs to the technical field of power generation and distribution, and discloses a virtual power plant scheduling optimization method that responds to demand bidding, comprising: collecting multi-source data on the day-ahead power operation status in the target power area based on a data terminal to obtain corresponding multi-source feature information, and based on the data, performing day-ahead power operation prediction at different time scales to obtain corresponding day-ahead feature prediction information; constructing a corresponding virtual power plant based on the obtained day-ahead feature prediction information: obtaining the operating constraints of the corresponding virtual power plant, and solving the objective function of the corresponding virtual power plant based on the constraints to obtain the corresponding optimal scheduling optimization scheme; generating device-level control instructions based on the optimal scheduling optimization scheme, and based on the control instructions, managing and controlling the power generation and distribution processes in the target power area, and simultaneously making rolling corrections to the corresponding optimal scheduling optimization scheme; the present invention significantly improves the economy, low carbon and reliability of the virtual power plant.

Description

Technical Field

[0001] The present invention relates to the technical field of power generation and distribution, and more specifically, to a virtual power plant scheduling optimization method that responds to demand bidding. Background Art

[0002] With the rapid development of the global economy and the continuous growth of energy demand, the power generation and distribution field is facing many severe challenges. The traditional power generation and distribution model can no longer meet current needs.

[0003] In terms of resource utilization, the country has historically relied primarily on centralized power generation and a single energy supply structure, with insufficient development and utilization of renewable energy. Due to a lack of effective scheduling and integration methods, a large amount of clean energy, such as wind and solar power, has been frequently abandoned, resulting in serious energy waste and difficulty in effectively improving power generation efficiency. In the power distribution sector, inaccurate forecasts of power demand have led to irrational power allocation, with some regions experiencing power surpluses while others face power shortages, resulting in low overall resource allocation efficiency.

[0004] In terms of cost control, the traditional power generation and distribution model faces high operating and environmental costs. In the power generation process, excessive reliance on fossil fuels has led to high fuel costs, and with increasingly stringent environmental protection requirements, the environmental costs brought by carbon emissions are also increasing. In the power distribution link, the lack of an effective demand response mechanism and the inability to reasonably guide user electricity consumption behavior have further increased the operating cost burden.

[0005] In terms of system stability and reliability, the traditional power grid structure is relatively fragile and unable to cope with increasingly complex power demands and various emergencies. Factors such as extreme weather and equipment failures can cause power supply interruptions, severely impacting social production and life. Moreover, traditional power dispatching methods make it difficult to achieve real-time monitoring and rapid adjustment of the power system, and cannot guarantee the stability and reliability of power supply.

[0006] In terms of market adaptability, with the gradual opening up of the power market and the increasing marketization, traditional power generation and distribution companies are facing fierce market competition. Due to their lack of effective analysis and forecasting capabilities of the market environment, they find it difficult to adjust their power generation and distribution strategies in a timely manner according to market changes, putting them at a disadvantage in the market competition.

[0007] In view of this, the present invention proposes a virtual power plant scheduling optimization method that responds to demand bidding to solve the above problems. Summary of the Invention

[0008] In order to overcome the above-mentioned defects of the prior art and achieve the above-mentioned objectives, the present invention provides the following technical solutions:

[0009] A virtual power plant scheduling optimization method responding to demand bidding, comprising:

[0010] Step 1: Using a data terminal, multi-source data is collected on the day-ahead power operation status in the target power area to obtain corresponding multi-source feature information. Based on this data, day-ahead power operation forecasts are performed at different time scales to obtain corresponding day-ahead feature forecast information.

[0011] Step 2: Construct a corresponding virtual power plant based on the obtained day-ahead feature prediction information. The corresponding construction process includes:

[0012] Step S1: Introduce a dynamic carbon price factor and build a corresponding carbon trading model based on the uncertainty in the carbon trading process;

[0013] Step S2: constructing a corresponding comprehensive response model based on the corresponding uncertainty of the user-side demand response;

[0014] Step S3: Modeling the virtual power plant operation scenario based on the obtained day-ahead feature prediction information to obtain a corresponding power plant operation scenario;

[0015] Step S4: constructing a corresponding virtual power plant based on the obtained carbon trading model, demand response model, and power plant operation scenario, and constructing an objective function based on the maximum net profit and minimum carbon emissions under all power plant operation scenarios;

[0016] Step 3: Obtain the operating constraints of the corresponding virtual power plant, and solve the objective function of the corresponding virtual power plant based on them to obtain the corresponding optimal scheduling optimization plan;

[0017] Step 4: Generate device-level control instructions based on the optimal scheduling optimization plan. The virtual power plant controls the power generation and distribution processes in the target power area based on the device-level control instructions. At the same time, the equipment operating status in the corresponding control process is updated in real time based on the data terminal, and the corresponding optimal scheduling optimization plan is rolled out based on it.

[0018] Furthermore, the process of obtaining the corresponding multi-source feature information includes:

[0019] Setting up a data terminal, which deploys a distributed message queue based on a pre-set wireless interface or API interface and builds a corresponding data channel based on it;

[0020] Based on the data channel, data is collected on the power generation side, user side and market environment involved in the corresponding target power area to obtain corresponding multi-source feature information.

[0021] Furthermore, the process of obtaining the day-ahead feature prediction information includes:

[0022] The collected multi-source feature information is input into a pre-selected and trained multi-scale hybrid prediction model to perform data prediction at different time scales to obtain corresponding day-ahead feature prediction information; the time scales include ultra-short-term prediction, medium-term prediction and long-term prediction.

[0023] Furthermore, the training process of the multi-scale hybrid prediction model includes:

[0024] The backbone framework of the multi-scale hybrid prediction model is a hierarchical federated learning framework; the infrastructure of the hierarchical federated learning framework is an edge node layer, an edge aggregation layer, and a cloud aggregation layer; the corresponding training process is as follows: the input multi-source feature information is respectively distributed to different edge nodes in the edge node layer, and the edge nodes perform time scale alignment based on pre-deployed timestamps, and slide and connect the multi-source feature information after the corresponding time scale alignment based on the pre-built time axis index to obtain the corresponding sample sequence; the corresponding sample sequence is predicted in the ultra-short term and in the medium and long term based on the parallel dual-channel input layer in the edge node to obtain the corresponding prediction results; the prediction results of each edge node are weighted based on the introduced cross-level attention mechanism and input into the edge aggregation layer; the edge aggregation layer performs edge layer aggregation and cloud aggregation every hour respectively; the corresponding training process is repeated until the function value of the multi-task loss function of the model tends to converge, the model parameters are saved, and the training is completed.

[0025] Furthermore, the process of constructing the carbon trading model in step S1 includes:

[0026] Based on the obtained day-ahead characteristic prediction information, the dynamic carbon price factor corresponding to the target power area is obtained, and the corresponding carbon trading model is constructed based on it; the carbon trading model includes the carbon emission quota model, the carbon emission model and the tiered carbon emission trading model;

[0027] The comprehensive response model in step S2 is composed of the load reduction cost and the load transfer subsidy cost under the demand response mechanism.

[0028] Furthermore, the process of constructing the power plant operation scenario in step S3 includes:

[0029] The power plant operation scenario includes several different operation scenarios. Taking the wind and solar power output scenario as an example, the construction process of the corresponding wind and solar power output scenario includes:

[0030] Based on the obtained day-ahead feature prediction information, the corresponding wind and solar output ratio data in the corresponding virtual power plant is obtained; based on this, the correlation of wind and solar output in each time period is obtained, and the corresponding wind and solar covariance matrix is ​​constructed based on this;

[0031] For each time period, construct a number of random variables based on the wind-solar covariance matrix;

[0032] Based on the wind and solar output ratios in the obtained day-ahead characteristic information, the corresponding wind power distribution probability function and photovoltaic probability distribution function are constructed respectively, and the marginal distributions are combined to obtain the corresponding joint probability distribution function;

[0033] Based on the obtained random variables, the joint probability distribution function corresponding to each time period is inversely sampled to obtain several daily wind and solar power output scenarios. Based on the pre-set number of clusters and combined with K-means clustering reduction, the corresponding wind and solar power operation scenarios are obtained.

[0034] Based on the construction process of the corresponding wind and solar operation scenarios, the power plant operation scenarios corresponding to other operation scenarios are obtained.

[0035] Furthermore, the process of defining the objective function of the virtual power plant includes:

[0036] Based on the collected multi-source characteristic information, the constructed carbon trading model, comprehensive response model and power plant operation scenario are integrated to obtain the corresponding virtual power plant, and the corresponding objective function is constructed with the expectation of maximizing the net profit of the virtual power plant and minimizing the carbon emission cost.

[0037] Furthermore, the operating constraints corresponding to the target virtual power plant are obtained, wherein the operating constraints include power balance constraints, energy storage constraints, unit ramping constraints, and wind and solar power generation operating constraints;

[0038] Furthermore, the process of obtaining the optimal scheduling optimization solution includes:

[0039] Solving and dividing the objective function based on the constraints into a cloud main problem and an edge sub-problem; the cloud main problem includes a multi-objective trade-off between maximizing net benefits and minimizing carbon emissions; the edge sub-problem includes the operation constraints that must be satisfied when local resources within the virtual power plant operate;

[0040] Separately perform optimal solutions with the goals of maximizing net benefits and minimizing carbon emissions to obtain the corresponding maximum net benefits. and the minimum carbon emissions at this time pf; and the minimum carbon emissions And the corresponding maximum net profit F0 at this time;

[0041] Based on the maximum net benefit obtained and minimum carbon emissions Set the upper and lower bounds of the virtual power plant's net income and carbon emissions; and based on these bounds, obtain the single-objective expectation coefficients of the virtual power plant's management users for net income and carbon emissions respectively;

[0042] Converting the target solution problem of the original target function into the expectation solution problem of managing the user based on the single target expectation coefficient;

[0043] The transformed desired problem and marginal subproblems are collaboratively solved based on the particle swarm optimization algorithm to obtain the corresponding optimal solution.

[0044] Based on the constructed virtual power plant, the corresponding optimal solution is simulated and run, and whether it meets the optimal solution is determined based on the operation results; if not, the solution is re-solved based on the above optimal solution process; if it meets the requirements, the corresponding optimal solution is output as the optimal scheduling optimization solution.

[0045] Furthermore, the process of rolling correction of the corresponding optimal scheduling optimization plan includes:

[0046] Based on the obtained optimal adjustment optimization plan, the control parameters required for adjustment of each local resource in the virtual power plant are obtained, and corresponding device-level control instructions are generated based on the parameters;

[0047] The virtual power plant controls the power generation or distribution process of each device terminal in the target power area based on the generated device-level control instructions;

[0048] At the same time, the data terminal is used to update the relevant data of the operation process of the virtual power plant in real time;

[0049] The real-time updated data is read at a fixed frequency, and multi-time scale predictions are performed based on the currently read data to obtain the corresponding intraday prediction results; the deviations of the intraday prediction results and the day-ahead prediction feature information of the same period are calculated to obtain the corresponding deviation calculation results;

[0050] The obtained deviation calculation results are compared with their corresponding deviation thresholds respectively; if the deviation calculation result is less than the deviation threshold, no other operations are performed; if the deviation calculation result is not less than the deviation threshold, the default penalty coefficient is introduced to correct the objective function, and the target is re-solved based on the intraday forecast results to obtain a new optimal scheduling optimization plan, and it replaces the original optimal scheduling optimization plan until the intraday scheduling is completed.

[0051] The technical effects and advantages of the virtual power plant scheduling optimization method in response to demand bidding of the present invention are as follows:

[0052] 1. Through multi-source data collection and predictions at different time scales, the power supply and demand situation in the target power area can be accurately grasped. Based on this, a virtual power plant can be built and optimized for scheduling, which can rationally plan the output of distributed resources, reduce the waste of wind and solar resources, and achieve improved power generation efficiency. It can also ensure the balance between power generation and consumption in real time, optimize the power distribution process, and improve the efficiency of power resource allocation.

[0053] 2. Introduce a dynamic carbon price factor to construct a carbon trading model, fully consider factors such as carbon emission quotas, carbon emissions and tiered carbon emission trading, and effectively control the carbon emission costs of virtual power plants; the comprehensive response model covers load reduction costs and load transfer subsidy costs, guides the user side to respond reasonably, and reduces operating costs; constructs and solves the objective function with maximum net profit and minimum carbon emissions to maximize the economic benefits of virtual power plants. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 Schematic diagram of a virtual power plant scheduling optimization method in response to demand bidding according to the present invention;

[0055] Figure 2 Schematic diagram of a virtual power plant scheduling optimization system that responds to demand bidding within the present invention. DETAILED DESCRIPTION

[0056] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0057] Example 1

[0058] See also Figure 1 As shown, the virtual power plant scheduling optimization method in response to demand bidding described in this embodiment includes:

[0059] Step 1: Using a data terminal, multi-source data is collected on the day-ahead power operation status in the target power area to obtain corresponding multi-source feature information. Based on this data, day-ahead power operation forecasts are performed at different time scales to obtain corresponding day-ahead feature forecast information.

[0060] Step 2: Construct a corresponding virtual power plant based on the obtained day-ahead feature prediction information. The corresponding construction process includes:

[0061] Step S1: Introduce a dynamic carbon price factor and build a corresponding carbon trading model based on the uncertainty in the carbon trading process;

[0062] Step S2: constructing a corresponding comprehensive response model based on the corresponding uncertainty of the user-side demand response;

[0063] Step S3: Modeling the virtual power plant operation scenario based on the obtained day-ahead feature prediction information to obtain a corresponding power plant operation scenario;

[0064] Step S4: constructing a corresponding virtual power plant based on the obtained carbon trading model, demand response model, and power plant operation scenario, and constructing an objective function based on the maximum net profit and minimum carbon emissions under all power plant operation scenarios;

[0065] Step 3: Obtain the operating constraints of the corresponding virtual power plant, and solve the objective function of the corresponding virtual power plant based on them to obtain the corresponding optimal scheduling optimization plan;

[0066] Step 4: Generate device-level control instructions based on the optimal scheduling optimization plan. The virtual power plant controls the power generation and distribution processes within the target power area based on the device-level control instructions. At the same time, the operating status of the equipment in the corresponding control process is updated in real time through the data terminal, and the corresponding optimal scheduling optimization plan is revised based on the updated status.

[0067] It should be further explained that, in the specific implementation process, the process of obtaining the corresponding multi-source feature information includes:

[0068] A data terminal is set up, and the data terminal deploys a distributed message queue based on a preset wireless interface or API interface, and builds a corresponding data channel based on it; further, based on the data channel, data is collected on the power generation side, user side and market environment involved in the corresponding target power area to obtain corresponding multi-source feature information, and the multi-source feature information includes power generation side data, user side data and market data; wherein, the power generation side data includes output data of distributed resources in the virtual power plant, real-time meteorological data, energy storage SOC status and gas turbine efficiency curve, etc.; the user side data includes user load history curve, adjustable load capacity and corresponding compensation price; the market data includes historical electricity price, carbon quota price and demand response market rules of the power spot market; wherein, the target power area can be a shopping mall building or a residential area, which is not specifically limited in the present invention; the distributed resources include renewable energy power generation resources and traditional energy power generation resources;

[0069] It should be further explained that, in the specific implementation process, the process of obtaining the day-ahead feature prediction information includes:

[0070] The collected multi-source feature information is input into a pre-selected and trained multi-scale hybrid prediction model to perform data prediction at different time scales to obtain corresponding day-ahead feature prediction information, wherein the time scales include ultra-short-term prediction, medium-term prediction, and long-term prediction; the default prediction frequencies of the ultra-short-term prediction, medium-term prediction, and long-term prediction in the present invention are 5 minutes, 4 hours, and 8 hours, respectively; wherein the day-ahead feature prediction information includes but is not limited to multiple feature data such as output share data of each distributed resource in the target power area, electricity price prediction, carbon price prediction, electricity load demand, and energy storage SOC status;

[0071] Among them, the construction process of the multi-scale hybrid prediction model includes: the backbone framework of the multi-scale hybrid prediction model is a hierarchical federated learning framework; the infrastructure of the hierarchical federated learning framework is an edge node layer, an edge aggregation layer and a cloud aggregation layer; the corresponding training process is as follows: the input multi-source feature information is respectively distributed to different edge nodes in the edge node layer, the edge nodes perform time scale alignment based on pre-deployed timestamps, and slide the multi-source feature information after the corresponding time scale alignment based on the pre-built time axis index to obtain the corresponding sample sequence; the corresponding sample sequence is predicted in the ultra-short term and medium- to long-term based on the parallel dual-channel input layer in the edge node to obtain the corresponding prediction results Based on the introduced cross-level attention mechanism, the prediction results of each edge node are weighted and input into the edge aggregation layer; the edge aggregation layer performs edge layer aggregation and cloud aggregation every hour, and the cloud aggregation layer performs edge layer aggregation and cloud aggregation every day, respectively, to achieve global knowledge sharing in the cloud. At the same time, dynamic residual correction and multi-task loss function (MSE+quantile loss+KL divergence) are introduced to optimize model robustness; the training combines the gradual expansion from single scenario to multi-scenario with the MAML initialization process, and adopts differential privacy and homomorphic encryption to ensure data security; the corresponding training process is repeated until the function value of the model's multi-task loss function converges, and the model parameters are saved, that is, the training is completed;

[0072] Among them, the edge layer aggregation refers to data aggregation of prediction results generated by local edge nodes; the cloud aggregation refers to global data aggregation of the target power area.

[0073] It should be further explained that, in the specific implementation process, the carbon trading model constructed in step S1 includes a carbon emission quota model, a carbon emission model and a tiered carbon emission trading model;

[0074] Obtain the dynamic carbon price factor corresponding to the target power area based on the obtained day-ahead characteristic forecast information Where, represents the lag term weight, which is used to measure the degree to which the carbon price at the current moment is affected by the carbon price at the previous moment; represents the dynamic carbon price factor at time t-1; Represents the market supply and demand weight, which is used to measure the carbon emission pressure in the target power area; Indicates the total amount of excess carbon emissions within the target power region; Represents the policy impact weight, which is used to measure the impact of market policies on carbon prices; represents the policy intensity index, which is determined by the reduction rate of carbon quotas, the adjustment of carbon tax and the subsidy intensity of renewable energy;

[0075] The carbon emission quota model is expressed as follows: Where, and They represent the carbon emission quota coefficients of coal-fired units and gas turbines in the virtual power plant respectively; , used to represent the amount of electricity purchased by the superior during period t; T represents the preset scheduling period. In the present invention, the scheduling period is one day; Indicates the output power of the gas turbine; Indicates the total carbon quota of the virtual power plant; in this invention, it is assumed that the main sources of carbon emissions are upstream power purchases and gas turbines within the target power area, and the source of upstream power purchases is coal-fired units;

[0076] The carbon emission model is expressed as follows: Where, Indicates actual carbon emissions; and represent the actual carbon emission coefficients of coal-fired units and gas turbines respectively;

[0077] The expression formula of the tiered carbon emissions trading model is: ; Where, represents the carbon emission step threshold of the i-th step in the target power region, where the step threshold includes an upper and lower limit; represents the benchmark carbon price corresponding to the i-th step, Indicates the dynamic carbon price factor at the previous moment; Indicates the benchmark step thresholds for each step within the power area; Represents the penalty coefficient. The higher the carbon emission level, the higher the penalty coefficient. Represents the threshold adjustment coefficient, which is used to adjust the carbon emission ladder threshold based on the real-time carbon price; represents the carbon trading cost; N1 represents the total number of steps;

[0078] It should be further explained that, in the specific implementation process, the comprehensive response model in step S2 is composed of the demand response mechanism load reduction cost and the transfer load subsidy cost;

[0079] The load reduction cost is expressed as follows:

[0080] Where, 、 and Represent the subsidy cost coefficients for load reduction during load valley, normal and peak periods respectively; C0, C1 and C2 represent load valley, normal and peak periods respectively; Indicates the load reduction at time t; represents the load reduction cost;

[0081] The formula for expressing the transfer load subsidy cost is:

[0082] Where, represents the load transfer subsidy cost, and They represent the subsidy cost coefficients corresponding to the load shifting to the off-peak period and the normal period respectively; and They represent the load during off-peak and normal periods respectively;

[0083] It should be further explained that, in the specific implementation process, the process of constructing the power plant operation scenario in step S3 includes:

[0084] Since there are many possible operating scenarios that can actually occur in a power area during actual operation, the present invention takes the construction process of a wind and solar output scenario as an example to illustrate. The construction process of the corresponding wind and solar output scenario includes:

[0085] Based on the obtained day-ahead feature prediction information, the corresponding wind and solar output ratio data in the corresponding virtual power plant is obtained; based on it, the correlation of wind and solar output in each time period is obtained, and the corresponding wind and solar covariance matrix is ​​constructed based on it. ; In the formula, each matrix element in the wind-solar covariance matrix represents the correlation of the corresponding wind and solar power output at different times; the correlation refers to the degree of mutual correlation between wind power output and photovoltaic power output at different times; Indicates the length of the corresponding scene;

[0086] Then, for each time period, several random variables X are constructed based on the wind-solar covariance matrix, and ; represents the MT-dimensional zero vector;

[0087] Based on the wind and solar output ratios obtained in the day-ahead characteristic information, the corresponding wind power distribution probability function and photovoltaic probability distribution function are constructed respectively, and the marginal distributions are combined to obtain the corresponding joint probability distribution function. ; Where e represents the Euler number, represents the Frank-Copula parameter at time t; and Indicates that regulation occurs, which is related to the marginal distribution parameters in the marginal distribution combination process; and represent the wind power distribution probability function and photovoltaic distribution probability function respectively; represents the joint probability distribution function at time t; wherein, the marginal distribution combination refers to the process of functionally merging the corresponding wind probability distribution function and photovoltaic distribution probability function based on a pre-selected binary Frank-Copula;

[0088] Then, based on the obtained random variable X, the joint probability distribution function corresponding to each time period is inversely sampled to obtain several daily wind and solar power output scenarios. Based on the pre-set number of clusters and combined with K-means clustering reduction, the corresponding wind and solar power operation scenarios are obtained.

[0089] The construction process of other power plant operation scenarios is similar to that of wind and solar operation scenarios, and will not be elaborated in detail in this invention.

[0090] It should be further explained that, in the specific implementation process, the process of defining the objective function of the virtual power plant includes:

[0091] Based on the collected multi-source characteristic information, the constructed carbon trading model, comprehensive response model and power plant operation scenario are integrated to obtain the corresponding virtual power plant. The corresponding objective function is constructed by maximizing the expected net profit of the virtual power plant and minimizing the expected carbon emission cost.

[0092] The mathematical expression formula of the corresponding objective function is:

[0093] ; where h and H represent the index and total number of power plant operation scenarios, respectively. represents the probability of occurrence of the hth power plant operation scenario; 、 、 、 and They represent the energy sales revenue of the virtual power plant under power plant operation scenario h, the operating cost of the CHP unit including P2G and CCS, the carbon trading cost, the gas turbine operating cost, and the energy storage maintenance cost; F1 and F2 represent the maximum net profit and minimum carbon emissions, respectively; represents the carbon emissions at time t.

[0094] It should be further explained that, in the specific implementation process, the process of obtaining the corresponding optimal scheduling optimization solution includes:

[0095] Obtaining operating constraints corresponding to the target virtual power plant, wherein the operating constraints include power balance constraints, energy storage constraints, unit ramping constraints, and wind and solar power generation operating constraints;

[0096] The expression formula of the wind and solar power generation operation constraint is: Where, and They represent the actual wind power generation operating power and the predicted wind power generation operating power at time t respectively;

[0097] The power balance constraint is expressed as:

[0098] Where, represents the exchange power between the virtual power plant and the upper-level large power grid at time t; represents the output of the gas turbine at time t; represents the day-ahead predicted power of the load at time t obtained based on the day-ahead characteristic information; represents the curtailed wind and solar power at time t; represents the charging or discharging power predicted at time t;

[0099] The expression formula of the energy storage constraint is: Where, Indicates the maximum charge and discharge power of the energy storage device; Indicates the amount of charge stored at time t; Indicates the electrical energy of the energy storage device at the initial moment; Indicates the charge and discharge efficiency; Indicates the real-time charging or discharging power of the energy storage device at time t; Indicates the maximum capacity of the energy storage device; and Respectively represent the state of charge at the initial moment and the end moment;

[0100] The expression formula of the unit climbing constraint is: Where, and They represent the maximum and minimum values ​​of the ramp power of the j-th unit respectively; and denote the unit power of the jth unit at time t and time t-1 respectively;

[0101] Once the constraints are met, the objective function is solved and divided based on the constraints into a cloud-based main problem and an edge sub-problem. The cloud-based main problem involves a multi-objective trade-off between maximizing net revenue and minimizing carbon emissions. The edge sub-problem involves the local resources within the virtual power plant operating under the acquired operational constraints.

[0102] Then, we can solve the optimal solution separately with the goal of maximizing net benefits and minimizing carbon emissions to obtain the corresponding maximum net benefits. and the minimum carbon emissions at this time pf; and the minimum carbon emissions And the corresponding maximum net profit F0 at this time;

[0103] Based on the maximum net benefit obtained and minimum carbon emissions Set the upper and lower bounds of the net income and carbon emissions of the virtual power plant; and based on them, obtain the single-objective expectation coefficients of the virtual power plant's management users for the net income value and carbon emissions. and ;

[0104] Among them, the single target expectation coefficient ; Indicates the net revenue value that the management user of the virtual power plant is allowed to reduce; ; a1 represents the user decision coefficient, in this invention a1=0.9;

[0105] Among them, the single target expectation coefficient ; Represents the increased carbon emissions allowed by users managing virtual power plants; ; a2 represents the user decision coefficient, in this invention a2=0.9;

[0106] Then, based on the single-objective expectation coefficient, the target solution problem of the original objective function is converted into the expectation solution problem of the management user; then, based on the particle swarm optimization algorithm, the converted expectation solution problem and the marginal subproblem are collaboratively solved to obtain the corresponding optimal solution;

[0107] Based on the constructed virtual power plant, the corresponding optimal solution is simulated and run, and whether it meets the optimal solution is determined based on the operation results; if not, the solution is re-solved based on the above-mentioned optimal solution process; if it meets the requirements, the corresponding optimal solution is output as the optimal scheduling optimization solution; wherein, the optimal scheduling optimization solution includes the output proportion of distributed resources at each time of the day and the charging and discharging power of energy storage equipment.

[0108] It should be further explained that, during the specific implementation process, the process of rolling revision of the corresponding optimal scheduling optimization plan includes:

[0109] Based on the obtained optimal adjustment optimization plan, the control parameters required for adjustment of each local resource in the virtual power plant are obtained, and corresponding device-level control instructions are generated based on the parameters;

[0110] The virtual power plant controls the power generation or distribution process of each device terminal in the target power area based on the generated device-level control instructions;

[0111] At the same time, the data terminal is used to update the relevant data of the operation process of the virtual power plant in real time;

[0112] The real-time updated data is read at a fixed frequency, and then a multi-time scale forecast is performed based on the currently read data to obtain the corresponding intraday forecast results; the deviation of the intraday forecast results and the day-ahead forecast feature information of the same period is calculated to obtain the corresponding deviation calculation results;

[0113] The obtained deviation calculation results are compared with their corresponding deviation thresholds respectively; if the deviation calculation result is less than the deviation threshold, no other operations are performed; if the deviation calculation result is not less than the deviation threshold, the default penalty cost is introduced to correct the objective function, and the target is re-solved based on the intraday forecast result to obtain a new optimal scheduling optimization plan, and replace the original optimal scheduling optimization plan until the intraday scheduling is completed; wherein, the fixed frequency in the present invention is 15 minutes; the default penalty cost refers to the penalty cost generated when the actual operation deviation of a local resource exceeds the threshold; for example, the cost of curtailing wind and solar power, the default cost of demand response, etc.

[0114] The present invention can accurately grasp the power supply and demand situation in the target power area through multi-source data collection and day-ahead power operation forecasts at different time scales; based on this, a virtual power plant is constructed and scheduling optimization is performed, and the output ratio of distributed resources can be reasonably arranged. For example, an operation scenario is constructed according to wind and solar output ratio data, so that renewable energy power generation resources and traditional energy power generation resources can be used more efficiently, reducing the phenomenon of wind and solar power abandonment and improving power generation efficiency; under the power balance constraint, it can ensure the real-time balance between power generation and power consumption, optimize the power distribution process, and improve the overall power resource allocation efficiency.

[0115] Example 2

[0116] See also Figure 2 As shown, for the parts not described in detail in this embodiment, please refer to the description of Example 1. A virtual power plant scheduling optimization system that responds to demand bidding is provided, including:

[0117] Data acquisition module: collects multi-source data on the day-ahead power operation status in the target power area based on the data terminal, obtains corresponding multi-source feature information, and then makes day-ahead power operation forecasts at different time scales based on the data, obtaining corresponding day-ahead feature forecast information;

[0118] A virtual power plant construction module; constructing a corresponding virtual power plant based on the obtained day-ahead feature prediction information;

[0119] Solution generation module: used to obtain the operating constraints of the corresponding virtual power plant, and based on them, solve the objective function of the corresponding virtual power plant to obtain the corresponding optimal scheduling optimization solution;

[0120] An execution feedback module generates device-level control instructions based on the optimal scheduling optimization plan. The virtual power plant controls the power generation and distribution processes in the target power area based on the device-level control instructions. At the same time, the operating status of the equipment in the corresponding control process is updated in real time based on the data terminal, and the corresponding optimal scheduling optimization plan is revised based on the updated status.

[0121] The modules are connected via wired and / or wireless means to achieve data transmission between modules.

[0122] Example 3

[0123] This embodiment discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the operation mode of the above-mentioned virtual power plant scheduling optimization method in response to demand bidding is implemented.

[0124] Since the electronic device introduced in this embodiment is an electronic device used to implement a virtual power plant scheduling optimization method that responds to demand bidding in the embodiment of this application, based on the virtual power plant scheduling optimization method that responds to demand bidding introduced in the embodiment of this application, technical personnel in this field can understand the specific implementation of the electronic device of this embodiment and its various variations, so how the electronic device implements the method in the embodiment of this application will not be introduced in detail here. As long as technical personnel in this field implement the electronic device used in the virtual power plant scheduling optimization method that responds to demand bidding in the embodiment of this application, it falls within the scope of protection of this application.

[0125] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters and thresholds in the formulas are set by technicians in this field according to actual conditions.

[0126] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiment. All technical solutions based on the principles of the present invention are within the scope of protection of the present invention. It should be noted that for users of ordinary skill in the art, various improvements and modifications that do not depart from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A virtual power plant scheduling optimization method in response to demand bidding, characterized in that: include: Step 1: Use the data terminal to collect multi-source data on the day-ahead power operation status in the target power area to obtain the corresponding multi-source feature information. Based on this, perform day-ahead power operation forecasts at different time scales to obtain the corresponding day-ahead feature forecast information, including: The collected multi-source feature information is input into the pre-selected and trained multi-scale hybrid prediction model to perform data prediction at different time scales to obtain the corresponding day-ahead feature prediction information; The training process of the multi-scale hybrid prediction model includes: The backbone framework of the multi-scale hybrid prediction model is a hierarchical federated learning framework; the basic architecture of the hierarchical federated learning framework is an edge node layer, an edge aggregation layer, and a cloud aggregation layer; the corresponding training process is as follows: the input multi-source feature information is respectively distributed to different edge nodes in the edge node layer, the edge nodes perform time scale alignment based on pre-deployed timestamps, and slide the multi-source feature information after the corresponding time scale alignment based on the pre-built time axis index to obtain the corresponding sample sequence; the corresponding sample sequence is predicted in the ultra-short term and in the medium and long term based on the parallel dual-channel input layer in the edge node to obtain the corresponding prediction results; the prediction results of each edge node are weighted based on the introduced cross-level attention mechanism and input into the edge aggregation layer; the edge aggregation layer performs edge layer aggregation and cloud aggregation every hour; the cloud aggregation layer performs edge layer aggregation and cloud aggregation every day; the corresponding training process is repeated until the function value of the multi-task loss function of the model tends to converge, the model parameters are saved, and the training is completed; Step 2: Construct a corresponding virtual power plant based on the obtained day-ahead feature prediction information. The corresponding construction process includes: Step S1: Introduce a dynamic carbon price factor and build a corresponding carbon trading model based on the uncertainty in the carbon trading process; Step S2: constructing a corresponding comprehensive response model based on the uncertainty of user-side demand response; Step S3: Modeling the virtual power plant operation scenario based on the obtained day-ahead feature prediction information to obtain a corresponding power plant operation scenario; Step S4: constructing a corresponding virtual power plant based on the obtained carbon trading model, demand response model, and power plant operation scenario, and constructing an objective function based on the maximum net profit and minimum carbon emissions under all power plant operation scenarios; Step 3: Obtain the operating constraints of the corresponding virtual power plant, and solve the objective function of the corresponding virtual power plant based on them to obtain the corresponding optimal scheduling optimization plan, including: Based on the constraints, the objective function is divided into a cloud main problem and an edge sub-problem; Separately perform optimal solutions with the goals of maximizing net benefits and minimizing carbon emissions to obtain the corresponding maximum net benefits and minimum carbon emissions; The upper and lower bounds of the net income and carbon emissions of the virtual power plant are set based on the maximum net income and minimum carbon emissions respectively; and the single-objective expectation coefficients of the management users of the virtual power plant for the net income value and carbon emissions are obtained based on them respectively; Converting the target solution problem of the original target function into the expectation solution problem of managing the user based on the single target expectation coefficient; The transformed desired problem and marginal subproblems are collaboratively solved based on the particle swarm optimization algorithm to obtain the corresponding optimal solution. Based on the constructed virtual power plant, the corresponding optimal solution is simulated and run, and based on the operation results, it is determined whether it meets the optimal solution; if not, the solution is re-solved based on the above optimal solution process; if it meets the optimal solution, the corresponding optimal solution is output as the optimal scheduling optimization solution; Step 4: Generate device-level control instructions based on the optimal scheduling optimization plan. The virtual power plant controls the power generation and distribution processes in the target power area based on the device-level control instructions. At the same time, the equipment operating status in the corresponding control process is updated in real time based on the data terminal, and the corresponding optimal scheduling optimization plan is rolled out based on it.

2. A virtual power plant scheduling optimization method in response to demand bidding according to claim 1, characterized in that: The process of obtaining the corresponding multi-source feature information includes: Setting up a data terminal, which deploys a distributed message queue based on a pre-set wireless interface or API interface and builds a corresponding data channel based on it; Based on the data channel, data is collected on the power generation side, user side and market environment involved in the corresponding target power area to obtain corresponding multi-source feature information.

3. The virtual power plant scheduling optimization method in response to demand bidding according to claim 1 is characterized in that: The time scales include very short-term forecasts, medium-term forecasts and long-term forecasts.

4. The method for optimizing virtual power plant scheduling in response to demand bidding according to claim 1, characterized in that: The process of constructing the carbon trading model in step S1 includes: Based on the obtained day-ahead characteristic prediction information, the dynamic carbon price factor corresponding to the target power area is obtained, and the corresponding carbon trading model is constructed based on it; the carbon trading model includes the carbon emission quota model, the carbon emission model and the tiered carbon emission trading model; The comprehensive response model in step S2 is composed of the load reduction cost and the load transfer subsidy cost under the demand response mechanism.

5. The virtual power plant scheduling optimization method in response to demand bidding according to claim 1 is characterized in that: The process of constructing the power plant operation scenario in step S3 includes: The power plant operation scenario includes several different operation scenarios. Taking the wind and solar power output scenario as an example, the construction process of the corresponding wind and solar power output scenario includes: Based on the obtained day-ahead feature prediction information, the corresponding wind and solar output ratio data in the corresponding virtual power plant is obtained; based on this, the correlation of wind and solar output in each time period is obtained, and the corresponding wind and solar covariance matrix is ​​constructed based on this; For each time period, construct a number of random variables based on the wind-solar covariance matrix; Based on the wind and solar output ratios in the obtained day-ahead characteristic information, the corresponding wind power distribution probability function and photovoltaic probability distribution function are constructed respectively, and the marginal distributions are combined to obtain the corresponding joint probability distribution function; Based on the obtained random variables, the joint probability distribution function corresponding to each time period is inversely sampled to obtain several daily wind and solar power output scenarios. Based on the pre-set number of clusters and combined with K-means clustering reduction, the corresponding wind and solar power operation scenarios are obtained. Based on the construction process of the corresponding wind and solar operation scenarios, the power plant operation scenarios corresponding to other operation scenarios are obtained.

6. The method for optimizing virtual power plant scheduling in response to demand bidding according to claim 1, characterized in that: The objective function construction process includes: Based on the collected multi-source characteristic information, the constructed carbon trading model, comprehensive response model and power plant operation scenario are integrated to obtain the corresponding virtual power plant, and the corresponding objective function is constructed with the expectation of maximizing the net profit of the virtual power plant and minimizing the carbon emission cost.

7. The method for optimizing virtual power plant scheduling in response to demand bidding according to claim 1, characterized in that: The operating constraints corresponding to the target virtual power plant are obtained, where the operating constraints include power balance constraints, energy storage constraints, unit ramp constraints, and wind and solar power generation operating constraints.

8. The method for optimizing virtual power plant scheduling in response to demand bidding according to claim 1, characterized in that: The cloud main problem includes the multi-objective trade-off between maximizing net benefits and minimizing carbon emissions; the edge sub-problem includes the operation constraints that must be met when the local resources within the virtual power plant are running.

9. The method for optimizing virtual power plant scheduling in response to demand bidding according to claim 1, characterized in that: The process of rolling revision of the corresponding optimal scheduling optimization plan includes: Based on the obtained optimal adjustment optimization plan, the control parameters required for adjustment of each local resource in the virtual power plant are obtained, and corresponding device-level control instructions are generated based on the parameters; The virtual power plant controls the power generation or distribution process of each device terminal in the target power area based on the generated device-level control instructions; At the same time, the data terminal is used to update the relevant data of the operation process of the virtual power plant in real time; The real-time updated data is read at a fixed frequency, and multi-time scale predictions are performed based on the currently read data to obtain the corresponding intraday prediction results; the deviations of the intraday prediction results and the day-ahead prediction feature information of the same period are calculated to obtain the corresponding deviation calculation results; The obtained deviation calculation results are compared with their corresponding deviation thresholds respectively; if the deviation calculation result is less than the deviation threshold, no other operations are performed; if the deviation calculation result is not less than the deviation threshold, the default penalty coefficient is introduced to correct the objective function, and the target is re-solved based on the intraday forecast results to obtain a new optimal scheduling optimization plan, and it replaces the original optimal scheduling optimization plan until the intraday scheduling is completed.

Images