Transaction scheduling optimization method and system for virtual power plant

Through hybrid deep learning and deep reinforcement learning algorithms, a multi-energy dynamic balance model is established, and the carbon trading and scheduling of virtual power plants is optimized, which solves the low-carbon economic operation and the uncertainty of virtual power plants and the 'source-load' problem, and realizes dynamic supply and demand Balanced and low-carbon economic benefits.

CN118920444APending Publication Date: 2024-11-08HUANENG ZHEJIANG ENERGY SALES CO LTD

Patent Information

Application Number
CN202410827595.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-06-25
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

In the existing technology, the carbon trading mechanism of virtual power plants is insufficiently guided, resulting in difficulty in operating the low-carbon economy, and the uncertainty of the 'source-load' and backward scheduling monitoring technology, making it difficult to effectively optimize the seasonal scheduling and multi-type distributed power generation of the power system. Scheduling of units and energy storage systems.

Method used

Adopting a hybrid deep learning model and deep reinforcement learning algorithm, a multi-energy dynamic balance model is established, and the data of the energy supply unit is obtained in real time, power balance and carbon flow balance are performed, multi-energy flow distribution scheduling is optimized, and combined with Markov distribution strategy, it is realized. Interactive regulation of low-carbon economic dispatch and source-network-load-store.

Benefits of technology

It has achieved a dynamic supply and demand balance of the integrated energy system, reduced carbon emissions, improved renewable energy utilization and economic benefits, and ensured the low-carbon economic operation and the conversion, storage and distribution between multiple energy types.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118920444A_ABST
    Figure CN118920444A_ABST
Patent Text Reader

Abstract

The invention discloses a transaction scheduling optimization method and system for a virtual power plant, and relates to the technical field of virtual power plants, and the method comprises the steps: obtaining an energy supply unit in a power side and demand side integrated energy system IES of the virtual power plant in real time, building a multi-energy dynamic balance model, and obtaining multi-energy flow distribution scheduling data; carrying out random dynamic optimization on the multi-energy flow distribution scheduling data by adopting a model based on mixed deep learning, and carrying out low-carbon optimization scheduling on the multi-energy dynamic balance model; with transaction scheduling data optimization as a target, a comprehensive energy system energy transaction scheduling model based on deep reinforcement learning is established, regional energy transaction scheduling data in different time periods are acquired, carbon flow is introduced into an energy hub according to a quota distribution model and an actual carbon emission model of a subject, and'electricity-heat-gas-carbon 'coordinated operation is realized. And low-carbon economic benefits of income maximization, carbon emission minimization and renewable energy utilization rate maximization are realized.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of virtual power plants, and in particular to a transaction scheduling optimization method and system for virtual power plants. Background Art

[0002] Virtual power plants have set off a new wave of grid transformation and energy reform, aiming to reduce the consumption of fossil fuels such as primary energy as much as possible under the requirements of low-carbon emissions, and improve the utilization efficiency of secondary energy and renewable energy. Virtual power plants can use advanced metering communication technology and automatic control systems to flexibly aggregate various types of distributed energy such as distributed generators, energy storage systems, controllable loads, electric vehicles, etc. Based on the software architecture and training algorithms of virtual power plants, the coordinated and optimized operation of "source-grid-load-storage" under multi-energy complementarity can be realized, which is more conducive to the rational optimization and utilization of resources in the overall energy system. Scientifically and rationally incorporating VPP sub-energy systems into traditional power grids can achieve the role of peak shaving and valley filling by scheduling power generation and power supply on the power supply side, power consumption and access to the Internet on the user side, and power storage plans on the energy storage side during peak power consumption periods, and appropriately alleviate the contradiction between power supply and demand when power supply is tight. In addition, combining low-carbon decarbonization technology in virtual power plants can greatly reduce carbon emissions from power grids and significantly increase renewable energy.

[0003] At present, in order to effectively alleviate the supply and demand tension of traditional power grids during peak periods of electricity consumption in my country, solve the problem of high-proportion distributed energy access, and build a new green, low-carbon, and intelligent power system under the new situation, it is necessary to optimize the scheduling of virtual power plant transactions. The current transaction scheduling optimization methods for virtual power plants still have the following problems:

[0004] (1) With the opening of the carbon trading market and the growth of carbon reduction demand, how to improve the carbon trading mechanism to guide the system to operate in a low-carbon economy and formulate a seasonal scheduling strategy for the integrated energy system has become an urgent problem to be solved;

[0005] (2) At present, the power system has problems of "source-load" uncertainty and backward dispatching and monitoring technology when it comes to energy systems containing multiple types of distributed generators, user-side demand response mechanisms, and energy storage system dispatching and control. Summary of the invention

[0006] The purpose of the present invention is to provide a trading and scheduling optimization method and system for virtual power plants, so as to solve the technical problems in the prior art on how to improve the carbon trading mechanism to guide the system to operate in a low-carbon economy, as well as the existence of "source-load" uncertainty and backward scheduling and monitoring technology.

[0007] In order to solve the above technical problems, the present invention specifically provides the following technical solutions:

[0008] A first aspect of the present invention provides a transaction scheduling optimization method for a virtual power plant, comprising the following steps:

[0009] Real-time acquisition of energy supply units in the integrated energy system IES on the power supply side and demand side of the virtual power plant, establishment of a multi-energy dynamic balance model for the input energy power of the energy supply unit through power balance, and acquisition of multi-energy flow allocation and scheduling data;

[0010] The multi-energy flow allocation and scheduling data is subjected to random dynamic optimization based on a hybrid deep learning model, a multi-objective optimization strategy is established for the optimized allocation and scheduling data, a low-carbon optimization scheduling is performed on the multi-energy dynamic balance model, and transaction scheduling data in the virtual power plant is obtained;

[0011] With the goal of optimizing the transaction scheduling data, an energy transaction scheduling model for an integrated energy system based on deep reinforcement learning is established. By predicting the load power of the energy supply unit and the time-of-use electricity price information, the output power of the virtual power plant is flexibly adjusted to obtain regional energy transaction scheduling data in different time periods.

[0012] As a preferred solution of the present invention, real-time acquisition of energy supply units in the integrated energy system IES on the power supply side and demand side of the virtual power plant includes:

[0013] The energy supply units on the power supply side and demand side of the virtual power plant are divided into wind turbines, photovoltaic generators, power grids and gas grids. The power of the wind turbines is solved according to the output power of the generators and the input wind speed. The power expression of the wind turbines is:

[0014]

[0015] in, represents the rated power of the wind turbine, v represents the wind speed, and v ci represents the cut-in wind speed, v w Indicates the cut-out wind speed, v o Indicates rated wind speed;

[0016] A photovoltaic power generation mathematical model is established based on the power of the wind turbine generator set to obtain the real-time input energy power of the energy supply unit, and the expression is:

[0017]

[0018] in, Indicates the input rated power of the energy supply unit, Indicates the ratio of actual power to output power, θ T ,θ S They represent the total light intensity absorbed by the photovoltaic cell and the light intensity under standard conditions, T represents the temperature of the photovoltaic cell, δv Represents the output power coefficient of photovoltaic cells.

[0019] As a preferred solution of the present invention, a multi-energy dynamic balance model is established through power balance for the input energy power of the energy supply unit to obtain multi-energy flow allocation and scheduling data, including:

[0020] The input energy power is divided into electric load demand power, thermal load demand power and gas load demand power according to the complementary effect of various types of energy. A multi-energy dynamic balance model is established by balancing the electric load demand power, thermal load demand power and gas load demand power. The expression is:

[0021]

[0022] Among them, γ gird represents the efficiency of power exchange between the virtual power plant and the grid, P tra Indicates the power exchanged between the virtual power plant and the grid, P s Indicates the electrical output power, P ev Represents photovoltaic power generation, G CHP , G P2G , H CHP They represent the electrical output power, natural gas consumption and thermal power of the virtual power plant, G NET Represents the input thermal power, G s Represents the charging and discharging power of the virtual power plant, H B represents the charging and discharging efficiency of the virtual power plant, H s Represents the grid exchange power, P l , G l , H l They correspond to the power requirements of the electric load, the thermal load and the gas load respectively;

[0023] The carbon emission quota of the virtual power plant is introduced into the multi-energy dynamic balance model in the form of carbon flow. The quotas to be traded within different time scales of the electric load demand power, thermal load demand power and gas load demand power are traded autonomously. In the multi-energy dynamic balance model, various energy flows are allocated and scheduled in a carbon flow balance mode to obtain multi-energy flow allocation and scheduling data.

[0024] As a preferred solution of the present invention, the multi-energy flow allocation and scheduling data is subjected to random dynamic optimization based on a hybrid deep learning model, including:

[0025] The multi-energy flow allocation and scheduling data is used as an input data set, and the factors related to the electricity price and load factor of the load sample sequence in the current virtual power plant are input as a sample sequence, and the sample sequence is inserted into the input data set in time sequence to obtain a time series matrix with original characteristics;

[0026] Detecting the time series matrix, analyzing the correlation, and obtaining a training set;

[0027] Decomposing the characteristic time series matrix of the training set into K eigenmode function subsequences containing multiple frequency sizes, and sorting the frequency subsequences in descending order from high to low;

[0028] The frequency subsequences of the same frequency level are merged to form a new frequency time series feature matrix, a convolution operation is performed on the frequency time series feature matrix to extract feature vectors, and the maximum pooling result of the convolution information is calculated after pooling the feature vectors to obtain the implicit coding information of each hidden layer;

[0029] Iteratively process the implicit coding information, obtain the optimal prediction value of each convolution layer with the minimum error as the optimization goal, and stop the training process when the frequency order of the cycle has reached the total number of frequency sub-sequences to obtain multiple frequency prediction sub-sequences;

[0030] The multi-frequency prediction sub-sequences are algebraically summed to obtain the optimized allocation and scheduling data based on multiple energy flows after random dynamic optimization.

[0031] As a preferred solution of the present invention, a multi-objective optimization strategy is established for the optimized allocation scheduling data, a low-carbon optimization scheduling is performed on the multi-energy dynamic balance model, and transaction scheduling data in the virtual power plant is obtained, including:

[0032] The optimized allocation and scheduling data is modeled with the constraints of maximizing daily income and minimizing carbon emissions, and Markov distribution is used to make sequential decisions on the optimized allocation and scheduling data to establish a multi-objective optimization strategy. The Markov distribution expression is:

[0033] F(s,t|s N ,…,s L ,t N ,…,t L )=F(s,t|s N ,t N )

[0034] Where F(s,t) is the Markov random process S tn The distribution function, s N ,…,s L represents any random state, t N ,…,t L Indicates any time;

[0035] According to the F(s, t), the sequence decision of the optimization allocation scheduling data is divided into five tuples (A, ε, θ, μ, τ), where A represents the state space, ε represents the action space, θ represents the transition probability, μ represents the reward return function, and τ represents the discount factor;

[0036] Setting the number of iterative training times, optimizing the distribution of the scheduling data according to the current moment in the Markov process in the running state, selecting an action for the running state through sequence decision making, and applying the action to the virtual power plant environment;

[0037] The action is transferred to a state according to the real-time virtual power plant environment, and the random factors of renewable energy output, electricity price changes, and load demand in the virtual power plant environment are calculated according to the optimized allocation and scheduling data to obtain transaction scheduling data in the virtual power plant.

[0038] As a preferred solution of the present invention, with the optimization of the transaction scheduling data as the goal, an energy transaction scheduling model for an integrated energy system based on deep reinforcement learning is established, including:

[0039] In the scheduling, the transaction scheduling data is optimized with the lowest operating cost and the lowest carbon emission as the optimization goals, the operating cost and photovoltaic cost in the virtual power plant are comprehensively analyzed according to the multi-energy dynamic balance model, and energy mobilization is achieved with the economic optimum as the goal through power exchange with the corresponding power grid;

[0040] By predicting the load power and time-of-use electricity price information of the energy supply unit, the output power of the virtual power plant can be flexibly adjusted to obtain regional energy trading scheduling data in different time periods.

[0041] A second aspect of the present invention provides a transaction scheduling optimization system for a virtual power plant, comprising:

[0042] Data processing module, used to process the real-time power data of energy supply units in the integrated energy system IES on the power supply side and demand side of the virtual power plant;

[0043] A data analysis module, used to analyze the influencing factors of uncertainties on the transaction scheduling of virtual power plants, and to establish an accurate physical model for the transaction scheduling data;

[0044] The data scheduling module performs hierarchical scheduling of transaction scheduling data in the operation optimization of the integrated energy system with the lowest operating cost and the lowest carbon emission as the optimization goals;

[0045] The virtual dispatching center monitors the virtual power plant status of the data dispatching module in real time, determines the power exchange efficiency between the virtual power plant and the power grid, and achieves optimal transaction dispatch.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] The present invention establishes a transaction framework for an integrated energy system and models internal entities, thereby achieving a dynamic supply and demand balance in the integrated energy system. Combined with the dynamic balance formula of the integrated energy system, an energy hub model is established to lay the foundation for the conversion, storage and distribution of multiple energy types. The carbon flow is introduced into the energy hub based on the quota allocation model of the subject and the actual carbon emission model, thereby achieving coordinated operation of "electricity-heat-gas-carbon".

[0048] Construct a low-carbon economic dispatch optimization model based on deep reinforcement learning algorithm, and realize interactive regulation of source-grid-load-storage integration under random changes in the external environment of the virtual power plant, creating low-carbon economic benefits with maximization of profitable income, minimization of carbon emissions and maximization of renewable energy utilization. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the implementation methods of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the implementation methods or the description of the prior art. Obviously, the drawings in the following description are only exemplary, and for ordinary technicians in this field, other implementation drawings can be derived from the provided drawings without creative work.

[0050] Figure 1 A flow chart of a transaction scheduling optimization method for a virtual power plant provided by an embodiment of the present invention;

[0051] Figure 2 A block diagram of a trading scheduling optimization system for a virtual power plant provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0052] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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 creative work are within the scope of protection of the present invention.

[0053] like Figure 1 As shown, the present invention provides a transaction scheduling optimization method for a virtual power plant, comprising the following steps:

[0054] Real-time acquisition of energy supply units in the integrated energy system IES on the power supply side and demand side of the virtual power plant, establishment of a multi-energy dynamic balance model for the input energy power of the energy supply unit through power balance, and acquisition of multi-energy flow allocation and scheduling data;

[0055] In this embodiment, the integrated energy system IES with coordinated planning and operation of electricity, heat and gas is taken as the main body, the energy supply units are divided into wind turbines, photovoltaic power generation arrays, power grids and gas grids, the IES electricity-heat-gas energy hub model is built, and a carbon trading mechanism based on a step-by-step carbon price is established. Secondly, it is proposed to regard carbon emission quotas as carbon flows, establish a carbon flow power balance formula, and introduce it into the energy hub for unified planning and scheduling.

[0056] The hybrid deep learning model is used to perform random dynamic optimization on the multi-energy flow allocation and scheduling data, a multi-objective optimization strategy is established for the optimized allocation and scheduling data, a low-carbon optimization scheduling is performed on the multi-energy dynamic balance model, and the transaction scheduling data in the virtual power plant is obtained;

[0057] In this embodiment, a hybrid deep learning model is used to perform numerical decomposition on the characteristic time series matrix of multi-energy flow allocation scheduling data to achieve information feature extraction at different frequency scales, improve the prediction accuracy of the prediction module in the hybrid deep learning model, and greatly reduce the residual noise in the frequency subsequence components, thereby avoiding the problem of modal aliasing and making up for the shortcomings of other decomposition algorithms.

[0058] With the goal of optimizing transaction scheduling data, an energy transaction scheduling model for an integrated energy system based on deep reinforcement learning is established. By predicting the load power of the energy supply unit and the time-of-use electricity price information, the output power of the virtual power plant is flexibly adjusted to obtain regional energy transaction scheduling data in different time periods.

[0059] In this embodiment, the virtual power plant is treated as an independent interest entity, and the economic dispatch of the virtual power plant is achieved with the optimization of transaction dispatch data as the goal.

[0060] Real-time acquisition of energy supply units in the integrated energy system IES on the power supply side and demand side of the virtual power plant, including:

[0061] The energy supply units on the power supply side and demand side of the virtual power plant are divided into wind turbines, photovoltaic generators, power grids and gas grids. The power of the wind turbine is solved according to the output power of the generator and the input wind speed. The power expression of the wind turbine is:

[0062]

[0063] in, represents the rated power of the wind turbine, v represents the wind speed, and v ci represents the cut-in wind speed, v w Indicates the cut-out wind speed, v o Indicates rated wind speed;

[0064] In this embodiment, a mathematical model is established for the wind turbine generator system, and the photovoltaic effect is used to generate an electromotive force from the directionally moving free charges, thereby providing current to the load and providing input power for the power supply side and the demand side of the virtual power plant.

[0065] The mathematical model of photovoltaic power generation is established based on the power of wind turbines to obtain the real-time input energy power of the energy supply unit. The expression is:

[0066]

[0067] in, Indicates the input rated power of the energy supply unit, Indicates the ratio of actual power to output power, θ T ,θ S They represent the total light intensity absorbed by the photovoltaic cell and the light intensity under standard conditions, T represents the temperature of the photovoltaic cell, δ v Represents the output power coefficient of photovoltaic cells.

[0068] In this embodiment, a mathematical model of an energy hub for coordinated operation of multiple energy types is established based on the power balance formula of wind turbines.

[0069] The input energy power of the energy supply unit is used to establish a multi-energy dynamic balance model through power balance to obtain multi-energy flow allocation and scheduling data, including:

[0070] The input energy power is divided into the power demand of electric load, the power demand of thermal load and the power demand of gas load according to the complementary effect of various types of energy. The multi-energy dynamic balance model is established by balancing the power demand of electric load, the power demand of thermal load and the power demand of gas load. The expression is:

[0071]

[0072] Among them, γ gird represents the efficiency of power exchange between the virtual power plant and the grid, P tra Indicates the power exchanged between the virtual power plant and the grid, P s Indicates the electrical output power, P ev Represents photovoltaic power generation, G CHP , G P2G , H CHP They represent the electrical output power, natural gas consumption and thermal power of the virtual power plant, G NET Represents the input thermal power, G s Represents the charging and discharging power of the virtual power plant, H B represents the charging and discharging efficiency of the virtual power plant, H s Represents the grid exchange power, P l , G l , H lThey correspond to the power requirements of the electric load, the thermal load and the gas load respectively;

[0073] The carbon emission quota of the virtual power plant is introduced into the multi-energy dynamic balance model in the form of carbon flow. The quotas to be traded within different time scales of electric load demand power, thermal load demand power and gas load demand power are traded autonomously. In the multi-energy dynamic balance model, various energy flows are allocated and scheduled in the carbon flow balance mode to obtain multi-energy flow allocation and scheduling data.

[0074] The multi-energy flow allocation and scheduling data is stochastically dynamically optimized based on a hybrid deep learning model, including:

[0075] The multi-energy flow allocation and scheduling data is used as the input data set, and the factors related to the electricity price and load factor of the load sample sequence in the current virtual power plant are input as the sample sequence. The sample sequence is inserted into the input data set in time sequence to obtain the time series matrix with original characteristics;

[0076] In this embodiment, the factors related to the electricity price and load factor of the load sample sequence in the current virtual power plant are input, mainly by combining the historical electricity price sample time series, historical load sample time series and 17 rows of influencing factor sample time series on the predicted day side by side to form a time series feature matrix for the predicted day.

[0077] Detect the time series matrix, analyze the correlation, and obtain the training set;

[0078] Decompose the characteristic time series matrix of the training set into K eigenmode function subsequences with multiple frequency sizes, and sort each frequency subsequence in descending order from high to low;

[0079] In this embodiment, in order to aggregate the sample characteristic information between electricity price / load and different influencing factors, the IMF sub-sequence components at the same frequency band level are merged to reduce the negative impact of multi-frequency differences on the prediction effect.

[0080] The frequency subsequences of the same frequency level are merged to form a new frequency time series feature matrix. The frequency time series feature matrix is ​​convolved to extract feature vectors. After pooling the feature vectors, the maximum pooling result of the convolution information is calculated to obtain the implicit coding information of each hidden layer.

[0081] Iterate the implicit coding information, and obtain the optimal prediction value of each convolution layer with the minimum error as the optimization goal. When the frequency order of the cycle has reached the total number of frequency sub-sequences, stop the training process and obtain the multi-frequency prediction sub-sequences;

[0082] Perform algebraic summation on the multi-frequency prediction sub-sequences to obtain the optimal allocation and scheduling data based on multiple energy flows after random dynamic optimization.

[0083] In this embodiment, for time series matrices with high randomness, high nonlinearity and high volatility, the ICEEMDAN decomposition algorithm is used to numerically decompose the input feature time series matrix to be trained to achieve information feature extraction at different frequency scales. In order to improve the prediction accuracy of the prediction module in the hybrid prediction model, the feature time series matrix of the input layer is subjected to multiple EMD decomposition processes to obtain several multi-band IMF subsequence components.

[0084] In this embodiment, in the ICEEMDAN decomposition algorithm, adaptive Gaussian white noise is added in each round of EEMD decomposition stage, and each modal component needs to be obtained by calculating a unique residual signal. Compared with other time series decomposition methods, the ICEEMDAN decomposition algorithm can greatly reduce the residual noise in the IMF subsequence components, thereby avoiding the problem of "modal aliasing" to make up for the shortcomings of other decomposition algorithms.

[0085] Establish a multi-objective optimization strategy for the optimized allocation and dispatching data, perform low-carbon optimization dispatching on the multi-energy dynamic balance model, and obtain transaction dispatching data in the virtual power plant, including:

[0086] The optimal allocation and scheduling data is modeled with the constraints of maximizing daily income and minimizing carbon emissions. The Markov distribution is used to make sequential decisions on the optimal allocation and scheduling data, and a multi-objective optimization strategy is established. The Markov distribution expression is:

[0087] F(s,t|s N ,…,s L ,t N ,…,t L )=F(s,t|s N ,t N )

[0088] Where F(s,t) is the Markov random process S tn The distribution function, s N ,…,s L represents any random state, t N ,…,t L Indicates any time;

[0089] According to F(s, t), the sequential decisions of optimizing the allocation scheduling data are divided into five tuples (A, ε, θ, μ, τ), where A represents the state space, ε represents the action space, θ represents the transition probability, μ represents the reward return function, and τ represents the discount factor;

[0090] In this embodiment, on the basis of satisfying the electricity / heat / cooling load and other constraints, a set of optimal "source-grid-load-storage" low-carbon economic dispatch strategies is determined to maximize the daily income profit and minimize carbon emissions of the integrated energy system IES.

[0091] In this embodiment, Markov random process stochastic dynamic optimization is used to achieve low-carbon economic scheduling.

[0092] Set the number of iterative training times, optimize the distribution of the operating state of the scheduling data in the Markov process according to the current moment, select an action for the operating state through sequence decision making, and apply the action to the virtual power plant environment;

[0093] The action is transferred to a certain state according to the real-time virtual power plant environment. The random factors of renewable energy output, electricity price changes, and load demand in the virtual power plant environment are calculated based on the optimized allocation and scheduling data to obtain the transaction scheduling data in the virtual power plant.

[0094] In this embodiment, the state transfer strategy is executed according to the virtual power plant environment, and the environmental state is transferred according to the Markov process, and then transferred to the environmental state at the next moment. The mathematical operation scenario at the next moment is calculated and generated in combination with the random factors such as the renewable energy output, electricity price changes, load demand, etc. of the virtual power plant environment at the current moment, and the output of the electric-heat-cold energy supply equipment and other units is calculated. The reward return function at the current moment is further calculated and fed back to the intelligent agent. The intelligent agent saves the learning record of this iteration in the experience replay pool of the deep reinforcement learning model, and then samples the action strategy and environmental state of the small batch actor and inputs them into the critic. The critic will output a set of action-value functions to judge the pros and cons of the action strategy.

[0095] With the goal of optimizing transaction scheduling data, an energy transaction scheduling model for an integrated energy system based on deep reinforcement learning is established, including:

[0096] In the dispatching process, the transaction dispatching data is optimized with the lowest operating cost and the lowest carbon emission as the optimization goals. The operating cost and photovoltaic cost of the virtual power plant are comprehensively analyzed according to the multi-energy dynamic balance model. Through power exchange with the corresponding power grid, energy mobilization is achieved with the goal of economic optimization.

[0097] By predicting the load power of the energy supply unit and the time-of-use electricity price information, the output power of the virtual power plant can be flexibly adjusted, and regional energy trading scheduling data in different time periods can be obtained.

[0098] In this embodiment, the deep reinforcement learning model adopts the learning framework of actor-critic, which is a type of intelligent algorithm in which the agent continuously tries and errors and adaptively updates the learning. Deep reinforcement learning includes two types of objects: the agent and the environment. The algorithm relies on the agent to continuously interact with the environment. First, the agent applies an action to the environment and obtains the corresponding state and reward or punishment feedback from the environment. The agent will guide itself to update the action strategy based on the reward or punishment obtained from the feedback, and continuously adopt the optimized action strategy to change the state of the environment, and finally achieve the highest cumulative reward return in the process of continuous iterative training.

[0099] like Figure 2 As shown, the second embodiment of the present invention provides a transaction scheduling optimization system for a virtual power plant, including:

[0100] Data processing module, used to process the real-time power data of energy supply units in the integrated energy system IES on the power supply side and demand side of the virtual power plant;

[0101] Data analysis module, used to analyze the impact of uncertainties on virtual power plant transaction scheduling and establish an accurate physical model for transaction scheduling data;

[0102] The data scheduling module performs hierarchical scheduling of transaction scheduling data in the operation optimization of the integrated energy system with the lowest operating cost and the lowest carbon emission as the optimization goals;

[0103] The virtual dispatching center monitors the virtual power plant status of the data dispatching module in real time, determines the power exchange efficiency between the virtual power plant and the power grid, and achieves optimal transaction dispatch.

[0104] In this embodiment, the virtual power plant's intelligent "source-grid-load-storage" flexible scheduling is realized through the virtual dispatching center. Based on the principle of maintaining internal supply and demand balance, the integrated energy system IES is divided into energy layer and carbon flow layer, and is committed to ensuring the low-carbon economic operation of the system.

[0105] The present invention establishes a transaction framework for an integrated energy system and models internal entities, thereby achieving a dynamic supply and demand balance in the integrated energy system. Combined with the dynamic balance formula of the integrated energy system, an energy hub model is established to lay the foundation for the conversion, storage and distribution of multiple energy types. The carbon flow is introduced into the energy hub based on the quota allocation model of the subject and the actual carbon emission model, thereby achieving coordinated operation of "electricity-heat-gas-carbon".

[0106] Construct a low-carbon economic dispatch optimization model based on deep reinforcement learning algorithm, and realize interactive regulation of source-grid-load-storage integration under random changes in the external environment of the virtual power plant, creating low-carbon economic benefits with maximization of profitable income, minimization of carbon emissions and maximization of renewable energy utilization.

[0107] The above embodiments are only exemplary embodiments of the present application and are not intended to limit the present application. The protection scope of the present application is defined by the claims. Those skilled in the art may make various modifications or equivalent substitutions to the present application within the essence and protection scope of the present application, and such modifications or equivalent substitutions shall also be deemed to fall within the protection scope of the present application.

Claims

1. A transaction scheduling optimization method for a virtual power plant, characterized in that: The following steps are involved: Real-time acquisition of energy supply units in the integrated energy system IES on the power supply side and demand side of the virtual power plant, establishment of a multi-energy dynamic balance model for the input energy power of the energy supply unit through power balance, and acquisition of multi-energy flow allocation and scheduling data; The multi-energy flow allocation and scheduling data is subjected to random dynamic optimization based on a hybrid deep learning model, a multi-objective optimization strategy is established for the optimized allocation and scheduling data, a low-carbon optimization scheduling is performed on the multi-energy dynamic balance model, and transaction scheduling data in the virtual power plant is obtained; With the goal of optimizing the transaction scheduling data, an energy transaction scheduling model for an integrated energy system based on deep reinforcement learning is established. By predicting the load power of the energy supply unit and the time-of-use electricity price information, the output power of the virtual power plant is flexibly adjusted to obtain regional energy transaction scheduling data in different time periods.

2. A transaction scheduling optimization method for a virtual power plant according to claim 1, characterized in that: Real-time acquisition of energy supply units in the integrated energy system IES on the power supply side and demand side of the virtual power plant, including: The energy supply units on the power supply side and demand side of the virtual power plant are divided into wind turbines, photovoltaic generators, power grids and gas grids. The power of the wind turbines is solved according to the output power of the generators and the input wind speed. The power expression of the wind turbines is: in, represents the rated power of the wind turbine, v represents the wind speed, and v ci represents the cut-in wind speed, v w Indicates the cut-out wind speed, v o Indicates rated wind speed; A photovoltaic power generation mathematical model is established based on the power of the wind turbine generator set to obtain the real-time input energy power of the energy supply unit, and the expression is: in, Indicates the input rated power of the energy supply unit, Indicates the ratio of actual power to output power, θ T ,θ S They represent the total light intensity absorbed by the photovoltaic cell and the light intensity under standard conditions, T represents the temperature of the photovoltaic cell, δ v Represents the output power coefficient of photovoltaic cells.

3. A transaction scheduling optimization method for a virtual power plant according to claim 2, characterized in that: The input energy power of the energy supply unit is used to establish a multi-energy dynamic balance model through power balance to obtain multi-energy flow allocation and scheduling data, including: The input energy power is divided into electric load demand power, thermal load demand power and gas load demand power according to the complementary effect of various types of energy. A multi-energy dynamic balance model is established by balancing the electric load demand power, thermal load demand power and gas load demand power. The expression is: Among them, γ gird represents the efficiency of power exchange between the virtual power plant and the grid, P tra Indicates the power exchanged between the virtual power plant and the grid, P s Indicates the electrical output power, P ev Represents photovoltaic power generation, G CHP , G P2G , H CHP They represent the electrical output power, natural gas consumption and thermal power of the virtual power plant, G NET Represents the input thermal power, G s Represents the charging and discharging power of the virtual power plant, H B represents the charging and discharging efficiency of the virtual power plant, H s Represents the grid exchange power, P l , G l , H l They correspond to the power requirements of the electric load, the thermal load and the gas load respectively; The carbon emission quota of the virtual power plant is introduced into the multi-energy dynamic balance model in the form of carbon flow. The quotas to be traded within different time scales of the electric load demand power, thermal load demand power and gas load demand power are traded autonomously. In the multi-energy dynamic balance model, various energy flows are allocated and scheduled in a carbon flow balance mode to obtain multi-energy flow allocation and scheduling data.

4. A transaction scheduling optimization method for a virtual power plant according to claim 3, characterized in that: The multi-energy flow allocation and scheduling data is subjected to random dynamic optimization based on a hybrid deep learning model, including: The multi-energy flow allocation and scheduling data is used as an input data set, and the factors related to the electricity price and load factor of the load sample sequence in the current virtual power plant are input as a sample sequence, and the sample sequence is inserted into the input data set in time sequence to obtain a time series matrix with original characteristics; Detecting the time series matrix, analyzing the correlation, and obtaining a training set; Decomposing the characteristic time series matrix of the training set into K eigenmode function subsequences containing multiple frequency sizes, and sorting the frequency subsequences in descending order from high to low; The frequency subsequences of the same frequency level are merged to form a new frequency time series feature matrix, a convolution operation is performed on the frequency time series feature matrix to extract feature vectors, and the maximum pooling result of the convolution information is calculated after pooling the feature vectors to obtain the implicit coding information of each hidden layer; Iteratively process the implicit coding information, obtain the optimal prediction value of each convolution layer with the minimum error as the optimization goal, and stop the training process when the frequency order of the cycle has reached the total number of frequency sub-sequences to obtain multiple frequency prediction sub-sequences; The multi-frequency prediction sub-sequences are algebraically summed to obtain the optimized allocation and scheduling data based on multiple energy flows after random dynamic optimization.

5. A transaction scheduling optimization method for a virtual power plant according to claim 4, characterized in that: Establishing a multi-objective optimization strategy for the optimized allocation scheduling data, performing low-carbon optimization scheduling on the multi-energy dynamic balance model, and obtaining transaction scheduling data in the virtual power plant, including: The optimized allocation and scheduling data is modeled with the constraints of maximizing daily income and minimizing carbon emissions, and Markov distribution is used to make sequential decisions on the optimized allocation and scheduling data to establish a multi-objective optimization strategy. The Markov distribution expression is: F(s,t|s N ,…,s L ,t N ,…,t L )=F(s,t|s N ,t N ) Where F(s,t) is the Markov random process S tn The distribution function, s N ,…,s L represents any random state, t N ,…,t L Indicates any time; According to the F(s, t), the sequence decision of the optimization allocation scheduling data is divided into five tuples (A, ε, θ, μ, τ), where A represents the state space, ε represents the action space, θ represents the transition probability, μ represents the reward return function, and τ represents the discount factor; Setting the number of iterative training times, optimizing the distribution of the scheduling data according to the current moment in the Markov process in the running state, selecting an action for the running state through sequence decision making, and applying the action to the virtual power plant environment; The action is transferred to a state according to the real-time virtual power plant environment, and the random factors of renewable energy output, electricity price changes, and load demand in the virtual power plant environment are calculated according to the optimized allocation and scheduling data to obtain transaction scheduling data in the virtual power plant.

6. A transaction scheduling optimization method for a virtual power plant according to claim 5, characterized in that: With the goal of optimizing the transaction scheduling data, an energy transaction scheduling model for an integrated energy system based on deep reinforcement learning is established, including: In the scheduling, the transaction scheduling data is optimized with the lowest operating cost and the lowest carbon emission as the optimization goals, the operating cost and photovoltaic cost in the virtual power plant are comprehensively analyzed according to the multi-energy dynamic balance model, and energy mobilization is achieved with the economic optimum as the goal through power exchange with the corresponding power grid; By predicting the load power and time-of-use electricity price information of the energy supply unit, the output power of the virtual power plant can be flexibly adjusted to obtain regional energy trading scheduling data in different time periods.

7. A trading scheduling optimization system for a virtual power plant applied to claims 1-6, characterized in that: include: Data processing module, used to process the real-time power data of energy supply units in the integrated energy system IES on the power supply side and demand side of the virtual power plant; A data analysis module, used to analyze the influencing factors of uncertainties on the transaction scheduling of virtual power plants, and to establish an accurate physical model for the transaction scheduling data; The data scheduling module performs hierarchical scheduling of transaction scheduling data in the operation optimization of the integrated energy system with the lowest operating cost and the lowest carbon emission as the optimization goals; The virtual dispatching center monitors the virtual power plant status of the data dispatching module in real time, determines the power exchange efficiency between the virtual power plant and the power grid, and achieves optimal transaction dispatch.

Citation Information

Patent Citations

  • Integrated energy system energy hub management and control method considering carbon flow

    CN114050571A

  • Carbon transaction method based on intelligent park virtual power plant

    CN116523683A

  • Multi-virtual power plant electricity-carbon joint optimization operation method and device and storage medium

    CN116720690A

Cited By

  • Question answering method, device, equipment, system and program product

    CN119808922A

  • Virtual power plant coordinated operation method, device and system considering multi-energy complementation

    CN121150037A