Multi-time scale optimization regulation and control method for multi-energy virtual power plant based on digital twinning

Abstract

The invention relates to a multi-time scale optimization regulation and control method for a multi-energy virtual power plant based on digital twinning, and the method comprises the following steps: obtaining real-time data of the multi-energy virtual power plant in a physical scene, and constructing a digital twinning model based on a deep neural network; constructing a dynamic stepped carbon transaction model by considering a carbon transaction reward and punishment mechanism; constructing a three-stage optimization framework of a day-ahead stage, an intra-day rolling optimization stage and a real-time adjustment stage based on the digital twinborn model and the stepped carbon transaction model; the three-stage optimization framework is solved, a regulation and control instruction is obtained, and the regulation and control process is completed. Compared with the prior art, the method has the advantages of realizing carbon emission reduction, improving the consumption rate of renewable energy sources and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent energy system optimization control, and in particular to a multi-time scale optimization regulation method for a multi-energy virtual power plant based on digital twin. Background Art

[0002] With the global emphasis on environmental protection and sustainable development, the proportion of renewable energy such as solar energy and wind energy in the power system is increasing continuously. Taking a certain country as an example, the installed capacity of wind power and photovoltaic power generation has been growing rapidly in recent years. However, renewable energy has the characteristics of intermittency, volatility and uncertainty, and its output is greatly affected by natural conditions, which brings huge challenges to the stable operation of the power system and the balance of power supply and demand. The access of a large amount of renewable energy may lead to power surplus in some local areas, resulting in the phenomena of wind curtailment and light curtailment, causing waste of energy. For example, in some areas with rich wind resources but limited grid consumption capacity, wind power may not be fully connected to the grid during the low-load period at night, resulting in an increase in the wind curtailment rate.

[0003] As a new energy management model, the virtual power plant (VPP) provides a new idea for solving the problem of renewable energy consumption. Through advanced information technology and communication technology, the virtual power plant integrates and coordinately controls dispersed resources such as distributed power sources (such as distributed photovoltaics, small-scale wind power, etc.), energy storage systems (such as battery energy storage, supercapacitors, etc.), and controllable loads (such as industrial interruptible loads, electric vehicles, etc.) to form a virtual and schedulable power generation unit. The virtual power plant can participate in power market transactions, realize the optimal allocation and efficient utilization of electric power, and improve the flexibility and reliability of the power system.

[0004] Under the background of the "dual carbon" goal, the low-carbon transformation of the power system has become an inevitable trend. The electricity-carbon coupling mechanism closely links power production and consumption with carbon emissions, and through means such as carbon emission rights trading, it encourages power enterprises and users to reduce carbon emissions. At the same time, with the popularization of electric vehicles, the vehicle-to-grid (V2G) technology has gradually become a research hotspot. Electric vehicles can not only serve as mobile energy storage units, charging during the low-load period of the grid and discharging to the grid during the high-load period to achieve power support for the grid; they can also participate in power market ancillary services such as frequency modulation and voltage regulation.

[0005] Digital twin technology is a technology that realizes real-time data interaction and two-way mapping between a physical entity and its virtual model by constructing a virtual model of the physical entity. In the power system, digital twin technology can provide more accurate models and decision-making support for the operation and management of virtual power plants. By establishing a digital twin model of a virtual power plant, the operating status of the virtual power plant can be monitored in real time, the output of renewable energy and the load demand can be predicted, the dispatching strategy of the virtual power plant can be optimized, and the operating efficiency and economic benefits of the virtual power plant can be improved.

[0006] Existing research has mainly focused on the optimal dispatching problem of multi-energy virtual power plants, considering the coordinated operation of distributed power sources, energy storage systems, and controllable loads. For example, the literature "Y. Gao and Q. Ai. Novel Optimal Dispatch Method for Multiple Energy Sources in Regional Integrated Energy Systems Considering Wind Curtailment[J]. CSEE Journal of Power and Energy Systems, 2024, 10(5): 2166-2173", "Jiming Chen, Qian Xu, Hui Gao, et al. Low-carbon Economic Optimization of a Wind Power-Carbon Capture-Power-to-Gas Virtual Power Plant Considering Carbon Emission Constraints[J]. Journal of Electrical Engineering, 2024, 19(04): 287-295", and "Hongchao Gao, Chuyi Li, Guanxiong Wang, et al. Empirical Study on the Dynamic Construction and Response of a Virtual Power Plant Aggregating Large-scale 5G Base Stations[J]. Automation of Electric Power Systems, 2024, 48(18): 47-55" proposed an optimal dispatching method for virtual power plants based on model predictive control. By predicting the output of renewable energy and the load demand, the power generation plan and energy storage charge-discharge strategy of the virtual power plant are adjusted in real time to minimize the operating cost. However, most of these studies do not fully consider the impact of the electricity-carbon coupling mechanism and vehicle-grid interaction.

[0007] Some studies have begun to focus on the application of the electricity-carbon coupling mechanism in the optimization of power systems. The literature by Cai Ruitian, Yao Lijuan, and Wu Xin. Digital Twin Method for Distributed Photovoltaic Group Regulation and Control [J]. Power System Technology, 2025, 49(2): 593-603, Wang Limeng, Liu Xuemeng, Li Yang, etc. Low-Carbon Optimal Scheduling of Integrated Energy Systems Considering Demand Response under the Stepwise Carbon Trading Mechanism [J]. Electric Power Construction, 2024, 45(02): 102-114, J. Wang, et al. Joint Electricity and Carbon Sharing With PV and Energy Storage: A Low-Carbon DR-Based Game Theoretic Approach [J]. IEEE Transactions on Sustainable Energy, 2024, 15(4): 2703-2717, and R.E. Alden, E.S. Jones, et al. Smart Home HVAC Digital Twin ML Meta-Model for Electric Power Distribution Systems With Solar PV and CTA-2045 Controls [J]. IEEE Transactions on Industry Applications, 2025, 61(1): 572-582 have established an economic dispatch model for power systems considering carbon emission constraints. By introducing the carbon emission rights trading mechanism, the allocation of power generation resources is optimized, and the carbon emissions of the power system are reduced. However, these studies mainly focus on traditional power systems and do not cover scenarios of virtual power plants and vehicle-to-grid interaction. Research on vehicle-to-grid interaction mainly focuses on the charging and discharging control strategies of electric vehicles and the impact assessment on the power grid. The literature by Gao Hongchao, Li Chuyi, Wang Guanxiong, etc. Empirical Study on the Dynamic Construction and Response of a Virtual Power Plant Aggregating Large-Scale 5G Base Stations [J]. Automation of Electric Power Systems, 2024, 48(18): 47-55 proposed an optimal charging and discharging strategy for electric vehicles based on real-time electricity prices. By adjusting the charging time and power of electric vehicles, the minimization of user charging costs and the peak shaving and valley filling of the power grid load are achieved. However, most of these studies do not combine vehicle-to-grid interaction with the optimal scheduling of virtual power plants. The application of digital twin technology in power systems mainly focuses on equipment status monitoring and fault diagnosis.Literatures B. Zhou, et al. Optimal Coordination of Electric Vehicles for Virtual Power Plants With Dynamic Communication Spectrum Allocation[J]. IEEE Transactions on Industrial Informatics, 2021, 17(1): 450-462 and Wu Jiaqi, Zhang Qian, Huang Yaoyu, et al. Multi-agent Cooperative Low-carbon Economic Dispatch of Integrated Energy Systems Considering Electric Vehicles[J]. Automation of Electric Power Systems, 2024, 48(12): 36-47 use digital twin technology to establish a virtual model of a transformer, and through real-time monitoring of the transformer's operation data, realize early warning and diagnosis of transformer faults. At present, the research on applying digital twin technology to the multi-time scale optimal scheduling of virtual power plants is relatively few. Summary of the Invention

[0008] The purpose of the present invention is to provide a digital twin-based multi-time scale optimal regulation method for multi-energy virtual power plants that realizes the collaborative optimization of multi-energy devices to meet actual needs.

[0009] The purpose of the present invention can be achieved through the following technical solutions:

[0010] A digital twin-based multi-time scale optimal regulation method for multi-energy virtual power plants includes the following steps:

[0011] Obtain real-time data in the physical scenario of the multi-energy virtual power plant and construct a digital twin model based on a deep neural network;

[0012] Consider the carbon trading reward and punishment mechanism and construct a dynamic ladder carbon trading model;

[0013] Based on the digital twin model and the ladder carbon trading model, construct a three-stage optimization framework of the day-ahead stage - intraday rolling optimization stage - real-time adjustment stage;

[0014] Solve the three-stage optimization framework to obtain a regulation instruction and complete the regulation process.

[0015] Furthermore, the multi-types of energy in the multi-energy virtual power plant include distributed energy types, load types, distributed energy storage types, and multi-energy conversion types, where

[0016] The distributed energy types include renewable energy and conventional generating units,

[0017] The load types include base load, interruptible load, and shiftable load, and the shiftable load includes electric vehicles;

[0018] The distributed energy storage category includes batteries, heat storage tanks, and gas storage tanks;

[0019] The multi - energy conversion category includes gas turbines, power - to - gas, and electric boilers.

[0020] Furthermore, the steps of constructing the digital twin model based on the deep neural network include:

[0021] Based on the real - time data, a certain proportion of sample data sets are obtained by using the random sampling method, and a mechanism model of the virtual power plant is constructed according to the actual physical property expression;

[0022] Initialize the deep neural network, use the mechanism model as input for feature extraction, and obtain a data - driven model;

[0023] Superimpose the data - driven model and the mechanism model to obtain a digital twin model based on the deep neural network.

[0024] Furthermore, the deep neural network includes a CNN layer, a GRU layer, a fully - connected layer, and an output layer. The steps of obtaining the data - driven model include:

[0025] Use the mechanism model as input to perform feature extraction and dimensionality reduction in the CNN layer;

[0026] Take the difference between the measured value of the actual output of the virtual power plant and the corresponding mechanism model as the output of the GRU layer;

[0027] Use the GRU layer to learn the feature vectors of the dimensionality - reduced features to learn the mapping relationship between the input and the output, and then process and output through the fully - connected layer and the output layer;

[0028] Repeat the above steps for iterative processing to obtain a data - driven model.

[0029] Furthermore, it also includes optimizing the digital twin model based on the deep neural network. The specific steps include:

[0030] Use the standardized error evaluation index to evaluate the accuracy of the digital twin model based on the deep neural network, and adjust and optimize the connection weights of the deep neural network. Among them, the standardized error evaluation index of the digital twin model based on the deep neural network is approximately generalized through the loss function C(w, b), and the expression of the loss function C(w, b) is:

[0031]

[0032] (w * ,b * )=argminC(w,b)

[0033] Wherein, w and b are the weight connection parameter and the bias parameter matrix, and (w * , b * ) are the global optimal parameters, Y DT is the digital twin model based on the deep neural network, I is the sample dimension, Y re is the measured value of the actual output power, and n is the number of samples.

[0034] Furthermore, the steps of constructing the dynamic stepped carbon trading model include:

[0035] Determine the initial carbon emission quota of the multi-energy virtual power plant according to the baseline method, and the expression is:

[0036]

[0037] Wherein, are respectively the initial carbon quotas of the multi-energy virtual power plant, the conventional thermal power unit, the gas turbine, the power purchase from the external power grid, and the carbon capture power plant, and δ fp , δ mt,e , δ mt,h , δ g , δ cg are respectively the corresponding carbon quota coefficients, nfp is the number of conventional thermal power units, T is the scheduling time scale, is the power of the i-th conventional thermal power unit, nmt is the number of gas turbines, is the power of the i-th gas turbine, is the heat emission of the i-th gas turbine at time t, P t g is the power of the power purchase from the external power grid at time t, P t cg is the power of the carbon capture power plant at time t;

[0038] Determine the carbon emission sources of the multi-energy virtual power plant, and combine the emission factor method to obtain the actual carbon emissions of the multi-energy virtual power plant. The expression is:

[0039]

[0040] Wherein, D VPP , D fp , D mt , D g , D cg are respectively the actual carbon emissions of the multi-energy virtual power plant, the conventional thermal power unit, the gas turbine, the power purchase from the external power grid, and the carbon capture power plant, and λ fp , λ mt,e , λ mt,h , λ gThey are the corresponding carbon emission coefficients respectively;

[0041] According to the initial carbon emission quota and the actual carbon emissions, calculate the carbon trading volume of the multi - energy virtual power plant. The calculation formula is:

[0042]

[0043] In the formula, ΔD VPP is the carbon trading volume of the multi - energy virtual power plant;

[0044] Based on the carbon trading volume, considering the carbon trading reward and punishment mechanism, construct a dynamic stepped carbon trading model. The expression is:

[0045]

[0046] In the formula, is the carbon trading price, ε is the carbon trading benchmark price; α is the growth rate of the stepped carbon trading price; τ is the step size of the carbon emission interval.

[0047] Furthermore, the regulation process in the day - ahead stage of the three - stage optimization framework includes:

[0048] Based on the wind - solar load prediction data, with the total day - ahead economic cost as the optimization goal, and superimposing the electric vehicle charging and discharging rewards and the stepped carbon trading model, determine the decision variables to obtain the day - ahead regulation instructions. Among them, the decision variables include the charge - discharge state of energy storage, the output of distributed power sources, and the tie - line power. The expression of the optimization goal in the day - ahead stage is:

[0049]

[0050] In the formula, C da represents the total day - ahead economic cost; C op,t , C pl,t represent the system operation cost and the pollutant treatment cost at time t respectively; C g,t , C gs,t are the system - grid interaction cost and the gas purchase cost at time t respectively; C c,tr,t is the carbon trading cost at time t.

[0051] Furthermore, the regulation process in the intra - day rolling optimization stage of the three - stage optimization framework includes:

[0052] Aiming at the uncertainty of the source - load power prediction, with the day - ahead regulation as the benchmark, input the real - time wind - solar load data, and with the goal of optimizing the total intra - day economic cost and minimizing the deviation between the power of each distributed device and the day - ahead value, determine the decision variables to obtain the intra - day regulation instructions. Among them, the decision variables include the source - load power adjustment amount. The expression of the optimization goal in the intra - day rolling optimization stage is:

[0053]

[0054] In the formula, C din represents the total economic cost of each rolling period Δt within a day; t0 is the starting moment of the rolling period; d is the number of rolling periods; respectively represent the operating cost of the system within a day and the pollutant treatment cost; are respectively the cost of interacting with the external power grid within a day and the gas purchase cost; is the carbon trading cost within a day; is the penalty cost for the adjustment amount of each distributed device within a day.

[0055] Furthermore, in the three-stage optimization framework, the real-time adjustment stage uses a distributed model predictive control algorithm to solve the real-time optimization objective, so as to further adjust the optimization strategy of the rolling optimization stage within a day and obtain real-time control instructions. The expression of the real-time optimization objective in the real-time adjustment stage is:

[0056]

[0057] In the formula, X(t) is the state variable, which is composed of the power of various distributed resource aggregations and the power P of the external power grid tie line t g to form a column vector; U(t) is the control variable, which is composed of to form a column vector; W(t) is the disturbance variable, which is composed of the power of the distributed wind and solar aggregation and the load disturbance ΔP L,t to form a column vector; Y(t) is the state variable.

[0058] Furthermore, the steps of using the distributed model predictive control algorithm to solve the real-time optimization objective include:

[0059] Regarding various distributed resource aggregations and their corresponding controllers in the multi-energy virtual power plant as the i-th subsystem, and introducing a coordination mechanism to construct the local optimization objective of each subsystem. The expression of the local optimization objective is:

[0060]

[0061] In the formula, F i is the optimization objective of the i-th subsystem, G i , J i , P i , Q i are weight coefficients; y i (t + Δt|t) represents the predicted value of the output of subsystem i at time t for the rolling time t + Δt; y i,ref(t + Δt|t) represents the local reference trajectory of the tracking subsystem i; u i (t + Δt) represents the control variable of the subsystem i at the moment of t + Δt; y ref (t + Δt|t) is the global reference trajectory;

[0062] Solve the local optimization objectives of each subsystem to obtain the control instructions of each subsystem, and complete the regulation of the multi - energy virtual power plant in the real - time adjustment stage.

[0063] Compared with the prior art, the present invention has the following beneficial effects:

[0064] (1) The present invention uses a deep neural network to construct a digital twin model of a multi - energy virtual power plant, which can learn the correlation relationship between data to obtain a more accurate digital twin model, and constructs a three - stage optimization framework of the day - ahead stage - intraday rolling optimization stage - real - time adjustment stage in combination with the carbon trading reward and punishment mechanism. At three time scales of day - ahead, intraday and real - time, optimize the scheduling strategies respectively to achieve the multi - time - scale collaborative optimization of multi - energy devices that meets the actual requirements.

[0065] (2) The deep neural network of the present invention includes a CNN layer and a GRU layer. The CNN layer performs feature extraction and dimensionality reduction, and the GRU layer fully learns the feature vectors to mine the mapping relationship between the input and the output, which helps to improve the accuracy of the digital twin model. In addition, a loss function is constructed to adjust and optimize the connection weights of the deep neural network, making the digital twin model infinitely approach the measured values of the real model.

[0066] (3) The three - stage optimization framework of the present invention constructs multiple optimization objectives and adopts a distributed model predictive control algorithm to decompose the global optimization problem into a subsystem collaborative optimization problem, reducing the computational complexity and improving the real - time response ability.

[0067] (4) The present invention constructs a dynamic stepped carbon trading model to more accurately measure the carbon emissions of different devices, realizes the electricity - carbon collaborative optimization, encourages low - carbon scheduling, and achieves the carbon emission reduction target. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 It is a schematic diagram of the method flow of the present invention;

[0069] Figure 2 It is the overall architecture of the MEVPP digital twin system of the present invention;

[0070] Figure 3 It is the digital twin modeling based on the deep neural network of the present invention;

[0071] Figure 4 It is the flow chart of the optimization algorithm of the present invention. DETAILED DESCRIPTION

[0072] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0073] This embodiment provides a multi-time scale optimization and control method for a multi-energy virtual power plant based on digital twins, such as Figure 2 As shown in the figure, first, based on the real-time data of the physical system, a digital twin model of the virtual power plant is constructed, including multiple energy resources such as wind energy, photovoltaics, energy storage, and electric vehicles; secondly, combined with the carbon trading mechanism, the electricity-carbon coupling optimization objective function of the virtual power plant is established, taking into account both economy and low carbon, and optimizing scheduling strategies are formulated on three time scales: day-ahead, intraday, and real-time, to achieve multi-time scale collaborative optimization; the charging and discharging behavior of electric vehicles is included in the optimization model to give full play to its role as a flexible resource. Finally, according to the fluctuation of carbon prices and the carbon emissions of the virtual power plant, the optimization strategy is dynamically adjusted to achieve the carbon reduction target.

[0074] Specifically, Figure 1 As shown, the method comprises the following steps:

[0075] Step 1: Build a digital twin model based on deep neural network:

[0076] The multi-energy virtual power plant is classified and modeled according to "source-load-storage-conversion". Among them, the distributed energy category includes renewable energy such as wind and solar power and conventional power generating units, the load category includes base load, interruptible load and transferable load, the distributed energy storage category includes batteries, heat storage tanks and gas storage tanks, and the multi-energy conversion category includes gas turbines, power-to-gas, electric boilers, etc.

[0077] like Figure 3 As shown in the figure, a certain proportion of sample data sets are obtained by random sampling. According to the known physical property expression, the mechanism model f(x i ); secondly, the deep neural network is initialized, the mechanism model is used as input, feature extraction and dimension reduction are performed in the convolutional neural network CNN layer, and the difference between the actual output measurement of wind / light resources and the corresponding mechanism model is used as the output YNN of the recurrent neural network GRU layer; thirdly, the recurrent neural network GRU layer fully learns the extracted feature vectors and explores the mapping relationship between input and output; finally, the trained data-driven model and the mechanism model are superimposed to obtain the digital twin model Y DT . The standardized error evaluation index is used to evaluate Y DT Carry out accuracy assessment, adjust and optimize the connection weights of the deep neural network, so that the digital twin model can infinitely approach the real model measurement value.

[0078] Among them, the loss function C(w, b) can be used to approximate the generalization of Y DT The accuracy evaluation index is expressed as follows:

[0079]

[0080] (w * , b * ) = argmin C(w, b) (2)

[0081] Among them, the weight connection parameter and the bias parameter matrix are w and b respectively. (w * , b * ) are their global optimal parameters, used to adjust the connection weights of the deep neural network; I is the sample dimension.

[0082] Step 2: Construct a dynamic stepped carbon trading model:

[0083] Determine the initial carbon emission quota of the multi - energy virtual power plant according to the baseline method, including conventional thermal power units, gas turbines, purchasing electricity from the external power grid, user load, and carbon capture power plants:

[0084]

[0085] Among them, are the initial carbon quotas of the multi - energy virtual power plant, conventional thermal power units, gas turbines, purchasing electricity from the external power grid, and carbon capture power plants respectively; δ fp , δ mt,e , δ mt,h , δ g , δ cg are the corresponding carbon quota coefficients respectively.

[0086] The carbon emission sources mainly include purchased electricity, gas turbines, and conventional thermal power units. According to the emission factor method, assuming that the actual carbon emissions of the unit are proportional to the unit output, the actual carbon emissions are expressed as follows:

[0087]

[0088] Among them, D VPP , D fp , D mt , D g , D cg are the actual carbon emissions of the multi - energy virtual power plant, conventional thermal power units, gas turbines, purchasing electricity from the external power grid, and carbon capture power plants respectively; λ fp , λ mt,e , λ mt,h , λ g are the corresponding carbon emission coefficients respectively.

[0089] To promote power generation enterprises to formulate reasonable carbon emission plans, a reasonable carbon trading mechanism needs to be designed. First, calculate the carbon trading volume of the multi - energy virtual power plant, which is expressed as follows:

[0090]

[0091] Design a dynamic stepped carbon trading model considering reward and punishment mechanisms, which is expressed as follows:

[0092]

[0093] Among them, is the carbon trading price; ε is the carbon trading benchmark price; α is the growth rate of the stepped carbon trading price; τ is the step size of the carbon emission interval.

[0094] Step 3: Construct a three - stage optimization framework of the day - ahead stage - intra - day rolling optimization stage - real - time adjustment stage and solve it.

[0095] Day - ahead stage:

[0096] In the day - ahead stage, with the total day - ahead economic cost as the optimization objective, the dispatching time scale is 24h and the time interval is 1h, which is expressed as follows:

[0097]

[0098] Among them, C da represents the total day - ahead economic cost; C op,t , C pl,t represent the system operation cost and pollutant treatment cost at time t respectively; C g,t , C gs,t are the system - grid interaction cost and gas purchase cost at time t respectively; C c,tr,t is the carbon trading cost at time t.

[0099] Intra - day rolling optimization stage:

[0100] In the intra - day rolling optimization stage, with the optimal total economic cost and the minimum deviation of the power of each distributed device from the day - ahead as the objectives, the dispatching time scale is 24h and the time interval is 15min, which is expressed as follows:

[0101]

[0102] Among them, C din represents the total economic cost of each rolling cycle Δt within the day; t0 is the starting time of the rolling cycle; d is the number of rolling cycles; represent the system operation cost and pollutant treatment cost within the day respectively; are the external grid interaction cost and gas purchase cost within the day respectively; is the carbon trading cost within the day; The penalty cost for the adjustment amount of each distributed device within a day.

[0103] Real-time adjustment stage:

[0104] In the real-time adjustment stage, the distributed model predictive control (DMPC) method is used to further adjust the intraday rolling optimization strategy, with a time interval of 5 minutes, expressed as follows:

[0105]

[0106] Among them, X(t) is the state variable, which consists of the power of various distributed resource aggregates and the power of the external grid connection line P t g to form a column vector; U(t) is the control variable, which consists of to form a column vector; W(t) is the disturbance variable, which consists of the power of the distributed wind-solar aggregate and the load disturbance amount ΔP L,t to form a column vector; Y(t) is the state variable.

[0107] To reduce the difficulty of large-scale cluster rolling optimization of the system, the global optimization problem is decomposed into local optimization problems of multiple subsystems, and at the same time, a coordination mechanism is introduced into the objective function to ensure that the local optimization is consistent with the global goal. Regarding each distributed resource aggregate and its corresponding controller as the i-th subsystem, the objective function is expressed as follows:

[0108]

[0109] This formula consists of three parts: the local tracking error term, the control variable change penalty term, and the global coordination term. Among them, G i , J i , P i , Q i are weight coefficients; y i (t + Δt|t) represents the predicted value of the i-th subsystem's output at time t for the rolling time t + Δt; y i,ref (t + Δt|t) represents the local reference trajectory for tracking the i-th subsystem; u i (t + Δt) represents the control variable of the i-th subsystem at time t + Δt; y ref (t + Δt|t) is the global reference trajectory.

[0110] Combining the above process and Figure 4 , the steps to achieve three-stage collaborative optimization control include:

[0111] Construct a system that includes electrical, thermal, and gas energy as well as multi - energy conversion devices. Analyze the constraint relationships between distributed devices, energy flow losses, and establish algorithmic constraint conditions. With the lowest total operating cost as the objective function, integrate the economy of the power grid, and at the same time convert environmental and stability factors into economic costs. Adopt "three - stage coordinated scheduling":

[0112] Day - ahead plan (1 hour): Based on the predicted data of wind and solar power loads, with the total economic cost as the optimization goal, determine decision variables such as the charge - discharge state of energy storage, the output of distributed generation (DG), and the power of tie - lines. This includes day - ahead operating costs, pollutant treatment costs, power and gas purchase costs, and carbon trading costs, and superimpose charge - discharge rewards and carbon trading reward - punishment mechanisms.

[0113] Intra - day rolling optimization (15 minutes): Aiming at the uncertainty of source - load power prediction, with the day - ahead plan as the benchmark, input the real - time data of wind and solar power loads, optimize the total economic cost, and determine the adjustment amount of source - load power. Consider intra - day operating costs, pollutant treatment costs, energy purchase costs, carbon trading costs, and equipment adjustment penalty costs, and fine - tune the day - ahead plan through the feedback of measurement information to balance equipment constraints with economic, environmental, and stability goals.

[0114] DMPC real - time adjustment (5 minutes): Relying on the local controllers of flexible resources such as photovoltaic, wind power, batteries, and electric vehicles, set and adjust control commands in real - time to respond to the real - time changes of the system and ensure efficient operation.

[0115] If the above functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer - readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or part of this technical solution can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. And the aforementioned storage medium includes: USB flash drives, mobile hard disks, read - only memory (ROM), random access memory (RAM), magnetic disks, or optical discs and other various media that can store program codes.

[0116] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memory, CD-ROM, optical memory, etc.) containing computer-usable program code. The solutions in the embodiments of the present invention can be implemented in various computer languages. For example, object-oriented programming languages such as Java and interpreted scripting languages such as JavaScript, etc.

[0117] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be realized by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for realizing the functions specified in one Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0118] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including instruction means, and the instruction means realizes the functions specified in one Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0119] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process. Thus, the instructions executed on the computer or other programmable device provide steps for realizing the functions specified in one Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0120] Although the preferred embodiments of the present invention have been described, those skilled in the art can make additional changes and modifications once they know the basic creative concepts. Therefore, the appended claims are intended to be construed to include the preferred embodiments and all changes and modifications falling within the scope of the present invention.

[0121] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these modifications and variations.

Claims

1. A multi-time-scale optimal regulation method for a multi-energy virtual power plant based on digital twin, characterized in that, It includes the following steps: Obtain real-time data in the physical scenario of a multi-energy virtual power plant and construct a digital twin model based on a deep neural network; Consider the carbon trading reward and punishment mechanism and construct a dynamic stepped carbon trading model; Based on the digital twin model and the stepped carbon trading model, construct a three-stage optimization framework of the day-ahead stage - intraday rolling optimization stage - real-time adjustment stage; Solve the three-stage optimization framework to obtain a control instruction and complete the control process.

2. The multi-time scale optimal regulation method of a multi-energy virtual power plant based on digital twin according to claim 1, wherein The multiple types of energy in the multi-energy virtual power plant include distributed energy types, load types, distributed energy storage types, and multi-energy conversion types, where The distributed energy types include renewable energy and conventional generating units; The load types include base load, interruptible load, and shiftable load, and the shiftable load includes electric vehicles; The distributed energy storage types include batteries, thermal storage tanks, and gas storage tanks; The multi-energy conversion types include gas turbines, power-to-gas, and electric boilers.

3. A multi-time scale optimal regulation method for a multi-energy virtual power plant based on digital twin according to claim 1, characterized in that The steps of constructing the digital twin model based on a deep neural network include: Based on the real-time data, use the random sampling method to obtain a certain proportion of sample data sets, and construct a mechanism model of the virtual power plant according to the actual physical characteristic expressions; Initialize the deep neural network, use the mechanism model as the input for feature extraction, and obtain a data-driven model; Overlay the data-driven model and the mechanism model to obtain a digital twin model based on a deep neural network.

4. A multi-time scale optimal regulation method for a multi-energy virtual power plant based on digital twin according to claim 3, characterized in that, The deep neural network includes a CNN layer, a GRU layer, a fully connected layer, and an output layer. The steps of obtaining the data-driven model include: Use the mechanism model as the input to perform feature extraction and dimensionality reduction in the CNN layer; Use the difference between the actual output measurement value of the virtual power plant and the corresponding mechanism model as the output of the GRU layer; Use the GRU layer to learn the feature vectors of the dimensionality-reduced features to learn the mapping relationship between the input and the output, and then process and output through the fully connected layer and the output layer; Repeat the above steps for iterative processing to obtain a data-driven model.

5. A multi-time-scale optimal regulation method for a multi-energy virtual power plant based on digital twin according to claim 1, characterized in that, It also includes optimizing the digital twin model based on a deep neural network. The specific steps include: Use the standardized error evaluation index to evaluate the accuracy of the digital twin model based on a deep neural network, and adjust and optimize the connection weights of the deep neural network. Among them, the standardized error evaluation index of the digital twin model based on a deep neural network is approximately generalized through the loss function C(w,b), and the expression of the loss function C(w,b) is: (w * ,b * ) = argmin C(w,b) where \(w\) and \(b\) are the weight connection parameter and the bias parameter matrix, \((w * ,b * )\) are the global optimal parameters, \(Y DT \) is the digital twin model based on the deep neural network, \(I\) is the sample dimension, \(Y re \) is the measured value of the actual output, and \(n\) is the number of samples.

6. The multi-time scale optimal regulation method of a multi-energy virtual power plant based on digital twin according to claim 1, wherein, The steps of constructing the dynamic stepped carbon trading model include: Determine the initial carbon emission quota of the multi-energy virtual power plant according to the baseline method. The expression is: In the formula, are the initial carbon quotas of the multi - energy virtual power plant, conventional thermal power unit, gas turbine, power purchase from the external power grid, and carbon capture power plant respectively, and δ fp , δ mt,e , δ mt,h , δ g , δ cg are the corresponding carbon quota coefficients respectively, nfp is the number of conventional thermal power units, T is the scheduling time scale, is the power of the i - th conventional thermal power unit, nmt is the number of gas turbines, is the power of the i - th gas turbine, is the heat emission of the i - th gas turbine at time t, is the power of power purchase from the external power grid at time t, is the power of the carbon capture power plant at time t; Determine the carbon emission sources of the multi-energy virtual power plant, and combine the emission factor method to obtain the actual carbon emissions of the multi-energy virtual power plant. The expression is: Where D VPP , D fp , D mt , D g , D cg are the actual carbon emissions of the multi - energy virtual power plant, conventional thermal power unit, gas turbine, power purchase from the external power grid, and carbon capture power plant respectively, and λ fp , λ mt,e , λ mt,h , λ g are the corresponding carbon emission coefficients respectively; According to the initial carbon emission quota and the actual carbon emissions, calculate the carbon trading volume of the multi-energy virtual power plant. The calculation expression is: where ΔD VPP is the carbon trading volume of the multi - energy virtual power plant; Based on the carbon trading volume, consider the carbon trading reward and punishment mechanism and construct a dynamic stepped carbon trading model. The expression is: In the formula, is the carbon trading price, ε is the carbon trading benchmark price; α is the growth rate of the stepped carbon trading price; τ is the step size of the carbon emission interval.

7. A multi-time scale optimal regulation method for a multi-energy virtual power plant based on digital twin according to claim 1, characterized in that, The control process in the day-ahead stage of the three-stage optimization framework includes: Based on the predicted data of wind and light loads, with the total economic cost of the day-ahead as the optimization objective, and superimposing the rewards for electric vehicle charging and discharging and the stepped carbon trading model, decision variables are determined to obtain the day-ahead regulation instructions. Among them, the decision variables include the charge and discharge states of energy storage, the output of distributed power sources, and the tie-line power. The expression of the optimization objective in the day-ahead stage is: Where, C da represents the total economic cost before the day; C op,t , C pl,t respectively represent the system operation cost and pollutant treatment cost at time t; C g,t , C gs,t are respectively the interaction cost between the system and the power grid and the gas purchase cost at time t; C c,tr,t is the carbon trading cost at time t.

8. A multi-time-scale optimal regulation method for a multi-energy virtual power plant based on digital twin according to claim 1, characterized in that, The regulation process in the intraday rolling optimization stage of the three-stage optimization framework includes: In response to the uncertainty of source-load power prediction, taking the day-ahead regulation as the benchmark, inputting the real-time data of wind and light loads, and with the optimal total economic cost of the intraday and the minimum deviation between the power of each distributed device and the day-ahead as the optimization objectives, decision variables are determined to obtain the intraday regulation instructions. Among them, the decision variables include the adjustment amount of source-load power. The expression of the optimization objective in the intraday rolling optimization stage is: where C din represents the total economic cost of each rolling period Δt within a day; t0 is the starting time of the rolling period; d is the number of rolling periods; respectively represent the operating cost of the system within a day and the cost of pollutant treatment; are respectively the cost of interaction with the external power grid within a day and the cost of gas purchase; is the carbon trading cost within a day; is the penalty cost for the adjustment amount of each distributed device within a day.

9. A multi-time scale optimal regulation method for a multi-energy virtual power plant based on digital twin according to claim 1, characterized in that, In the real-time adjustment stage of the three-stage optimization framework, the distributed model predictive control algorithm is used to solve the real-time optimization objective to further adjust the optimization strategy in the intraday rolling optimization stage to obtain the real-time regulation instructions. The expression of the real-time optimization objective in the real-time adjustment stage is: where X(t) is the state variable, which is composed of the power of various distributed resource aggregates and the power of the external power grid connection line to form a column vector; U(t) is the control variable, which is composed of to form a column vector; W(t) is the disturbance variable, which is composed of the power of the distributed wind and solar aggregates and the load disturbance ΔP L,t to form a column vector; Y(t) is the state variable.

10. A multi-time scale optimal regulation method for a multi-energy virtual power plant based on digital twin according to claim 9, characterized in that, The steps of using the distributed model predictive control algorithm to solve the real-time optimization objective include: Regarding the various distributed resource aggregations and their corresponding controllers in the multi-energy virtual power plant as the i-th subsystem, and introducing a coordination mechanism to construct the local optimization objective of each subsystem. The expression of the local optimization objective is: where F i is the optimization objective of the i-th subsystem, G i , J i , P i , Q i are weight coefficients; y i (t + Δt|t) represents the predicted value of the output of subsystem i at time t for the rolling time t + Δt; y i,ref (t + Δt|t) represents the local reference trajectory of tracking subsystem i; u i (t + Δt) represents the control variable of subsystem i at time t + Δt; y ref (t + Δt|t) is the global reference trajectory; Solve the local optimization objective of each subsystem to obtain the control instructions of each subsystem, and complete the regulation of the multi-energy virtual power plant in the real-time adjustment stage.

Images