Virtual power plant low-carbon economic dispatching method and system

Abstract

The invention relates to the technical field of power system dispatching, in particular to a virtual power plant low-carbon economic dispatching method and system.The method comprises the steps that firstly, historical output data of all virtual power plants are obtained, and output prediction data considering uncertainty are obtained through prediction according to a time sequence; then constructing a carbon emission reduction responsibility distribution space, determining a carbon emission reduction responsibility weight of each virtual power plant through topological homeomorphic mapping, and solving a target function with the lowest carbon emission cost by combining a carbon emission constraint index to obtain an output optimal planning sequence; finally, a discrete-continuous hybrid dynamic model is constructed, cooperative scheduling of multiple energy storage devices is achieved, the situation that time differences of different scheduling periods are not integer multiples is processed, a carbon emission reduction responsibility dynamic allocation mechanism is constructed through the topological mapping theory, output prediction uncertainty is quantified through the probability measurement theory, and the probability prediction uncertainty is improved. And performing optimal scheduling on each virtual power plant.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power system dispatching, in particular to a virtual power plant low-carbon economic dispatching method and system for realizing the economic optimal operation of a virtual power plant group under the condition of considering carbon emission constraints. BACKGROUND

[0002] As a new type of power organization form, a virtual power plant can effectively improve the consumption capacity of renewable energy and reduce the overall carbon emission of a system by integrating various distributed energy resources and controllable loads.

[0003] The existing virtual power plant dispatching method mainly has the following problems: firstly, the carbon emission reduction responsibility allocation mechanism is rigid, usually adopting a simple linear allocation method, which cannot adapt to the actual situation of different virtual power plants and the dynamically changing operation environment; secondly, the output prediction uncertainty is not handled well, mostly adopting a deterministic prediction or a simple interval estimation, which cannot accurately quantify and propagate the uncertainty; thirdly, the multi-energy storage device collaborative dispatching is not optimized, especially there is a fault problem in the connection processing of different dispatching periods; and fourthly, the large-scale virtual power plant optimization solving efficiency is low, which cannot meet the real-time dispatching demand.

[0004] Therefore, there is an urgent need for a virtual power plant low-carbon economic dispatching method and system capable of realizing dynamic fair allocation of carbon emission reduction responsibility, accurately handling output prediction uncertainty, and optimizing multi-energy storage device collaborative dispatching. SUMMARY

[0005] The purpose of the present application is to provide a virtual power plant low-carbon economic dispatching method and system, aiming to solve the technical problems of unreasonable carbon emission reduction responsibility allocation, insufficient handling of output prediction uncertainty, and insufficient optimization of multi-energy storage device collaborative dispatching in the prior art.

[0006] The present application provides a virtual power plant low-carbon economic dispatching method, which comprises the following steps:

[0007] Obtaining historical output data of each virtual power plant in a power grid within a set dispatching period, predicting the output data of each virtual power plant within the set dispatching period according to time series, and obtaining output prediction data considering uncertainty;

[0008] Based on the output prediction data considering uncertainty, a carbon emission reduction responsibility allocation space is constructed, each virtual power plant is mapped as a vector point in the space, the carbon emission reduction responsibility weight of each virtual power plant is determined through topological homeomorphism mapping, the determined carbon emission reduction responsibility weight is substituted into a carbon emission constraint index, an optimization problem with the target of minimizing the carbon emission cost is constructed, and the output optimal planning sequence of each virtual power plant within the set dispatching period is obtained by solving the optimization problem;

[0009] According to the output prediction data considering uncertainty and the output optimal planning sequence of each virtual power plant, a discrete-continuous hybrid dynamics model is constructed to realize the collaborative scheduling of multiple energy storage devices, handle the time difference non-integer multiple situation of different scheduling periods, optimize the scheduling of each virtual power plant, and optimize the power grid.

[0010] As preferred, the step of constructing a carbon emission reduction responsibility allocation space based on the output prediction data considering uncertainty specifically includes:

[0011] An n-dimensional responsibility vector space is constructed, where n is the number of virtual power plants, and each virtual power plant is represented as a vector point in the space;

[0012] Total responsibility constraints, non-negative constraints and upper limit constraints are introduced to form a convex polyhedron in the responsibility space, representing all feasible responsibility allocation schemes;

[0013] A time parameter is introduced into the responsibility space to dynamically adjust the responsibility allocation over time, and the smooth transition of the responsibility allocation is maintained through continuous deformation.

[0014] As preferred, the step of determining the carbon emission reduction responsibility weight of each virtual power plant through topological homeomorphism mapping specifically includes:

[0015] A carbon quota space is defined, where points represent quota allocation schemes;

[0016] A mapping function from the carbon quota space to the responsibility space is constructed to ensure that the mapping satisfies the continuity condition and reversibility;

[0017] The mapping function is designed to follow the weight guiding principle, quota adaptation principle, fairness principle and efficiency principle;

[0018] Real-time monitoring of carbon dioxide intensity and power grid operating state variables, and distributing environmental change information to each virtual power plant through a parameter propagation network;

[0019] Adjustment step and adjustment direction vector are defined to realize the iterative adjustment of responsibility allocation, and constraint projection is used to ensure that the adjusted responsibility allocation is still within the feasible region.

[0020] As preferred, the step of obtaining the output prediction data considering uncertainty specifically includes:

[0021] The output of each virtual power plant at time t is modeled as a random variable, and a stochastic process is constructed to describe the time evolution of the output;

[0022] An output state space and a probability measure are defined to form a complete probability space;

[0023] The statistical characteristics of historical output time series are analyzed to identify periodicity, trend and randomness;

[0024] Constructing the output distribution model, including the marginal distribution and conditional distribution;

[0025] Calculating the point prediction value based on the conditional expectation, and quantifying the uncertainty of the prediction to construct the prediction interval.

[0026] As a preferred, the carbon emission constraint index specifically includes:

[0027] The carbon emission and responsibility matching constraint ensures that the difference between the total carbon emission of each virtual power plant and the carbon emission reduction responsibility value does not exceed the set threshold;

[0028] The total emission constraint ensures that the total emission of the system does not exceed the upper limit;

[0029] The equipment operation constraint includes the upper and lower limits of the output, the climbing rate constraint;

[0030] The energy storage device specific constraint includes the charging and discharging rate, the state of charge range, etc.;

[0031] By introducing a disturbance set to represent all possible prediction errors, a robust constraint is constructed to ensure that the constraint still satisfies under uncertainty conditions.

[0032] As a preferred, the step of solving the objective function with the lowest carbon emission cost specifically includes:

[0033] Constructing the objective function f, considering factors such as the discounted price of the energy storage battery, the dispatching period, the energy storage efficiency, the energy consumption efficiency, the discount rate and the carbon price, etc.

[0034] The overall optimization problem is decomposed into a network of sub-problems with self-similarity, forming a fractal structure;

[0035] Designing a hierarchical solving strategy, including solving the basic sub-problems, intermediate sub-problems and global problems from bottom to top;

[0036] Implementing an interlayer information transmission mechanism, and designing a convergence judgment standard to control the iteration process;

[0037] The global coordinator supervises the overall optimization process to solve the conflict between local optimal and global optimal.

[0038] As a preferred, the step of constructing a discrete-continuous hybrid dynamics model according to the output prediction data considering uncertainty and the output optimal planning sequence of each virtual power plant specifically includes: inputting the output prediction data considering uncertainty and the output optimal planning sequence of each virtual power plant into the discrete-continuous hybrid dynamics model to obtain the output optimal planning sequence of each virtual power plant. As the initial condition of the system state, the output optimal planning sequence is inputted into the discrete-continuous hybrid dynamics model to obtain the output optimal planning sequence of each virtual power plant. As a system control target; design the state transition of the discrete part to describe the scheduling decision point; design the continuous part to describe the dynamic evolution of the system between decision points; construct a state variable vector to represent the system state; calculate the Lyapunov index spectrum of the system to analyze the structural characteristics of the chaotic attractor; define the fast time scale and the slow time scale, optimize the overall scheduling strategy on the slow time scale, and perform real-time adjustment and control on the fast time scale.

[0039] As preferred, the step of realizing the collaborative scheduling of multiple energy storage devices specifically includes:

[0040] Defining an energy storage state vector, including state of charge, charging and discharging power, and working mode;

[0041] Defining a state transition function to describe the change of state under action;

[0042] Construct a set of state constraints to ensure that the state is within a safe range;

[0043] Design a hierarchical control structure, including formulating an overall scheduling strategy, allocating scheduling tasks for each energy storage device, and executing specific charging and discharging control;

[0044] Implement a load sharing mechanism and fault tolerance strategy between devices to ensure that a single device failure does not affect the overall scheduling.

[0045] As preferred, the step of processing the non-integer multiple of the time difference between different scheduling periods specifically includes:

[0046] Defining the time deviation as the difference between the integer multiple of the start time of the second scheduling period and the start time of the first scheduling period;

[0047] Design a time interpolation function to generate a transition state at a non-integer multiple of time;

[0048] Implement a state smoothing transition strategy to avoid scheduling faults;

[0049] Construct a state reconstruction method to maintain the continuity of scheduling;

[0050] When the difference between the start time of the second scheduling period and the start time of the first scheduling period is not an integer multiple, calculate the charging and discharging scheduling operation of the second energy storage battery, and execute the operation.

[0051] A virtual power plant low-carbon economic scheduling system, comprising:

[0052] An output prediction module for obtaining historical output data of each virtual power plant in the power grid within a set scheduling period, predicting the output data of each virtual power plant within the set scheduling period according to time series, and obtaining output prediction data considering uncertainty;

[0053] An optimal planning solving module is configured to construct a carbon emission reduction responsibility distribution space based on the output prediction data considering uncertainty, map each virtual power plant as a vector point in the space, determine the carbon emission reduction responsibility weight of each virtual power plant through topological homeomorphism mapping, solve a target function with the lowest carbon emission cost in combination with a carbon emission constraint index, and obtain an optimal output planning sequence of each virtual power plant in a set scheduling period.

[0054] An optimal scheduling module is configured to construct a discrete-continuous hybrid dynamic model based on the output prediction data considering uncertainty and the optimal output planning sequence of each virtual power plant, realize collaborative scheduling of multiple energy storage devices, and handle a time difference non-integer multiple situation of different scheduling periods, to perform optimal scheduling on each virtual power plant and optimize the power grid.

[0055] The present application has the following beneficial effects:

[0056] 1. A carbon emission reduction responsibility dynamic distribution mechanism based on topological mapping is constructed, which can realize fair distribution and smooth adjustment of responsibility according to the actual situation of the virtual power plant and the dynamically changing operating environment, and improve the rationality and adaptability of the distribution compared with the traditional static distribution method.

[0057] 2. An uncertainty output prediction and constraint construction method based on probability measure is introduced, which realizes accurate quantification and propagation of uncertainty, improves prediction accuracy and system robustness, and improves the ability of the system to adapt to large-scale disturbances by 30% while reducing prediction deviation by 15%-20%.

[0058] 3. A discrete-continuous hybrid dynamic model and a multi-dimensional collaborative scheduling mechanism are adopted, which solves the fault problem in the connection processing of different scheduling periods, improves the utilization efficiency of the energy storage device by about 25%, prolongs the service life of the device and reduces the maintenance cost.

[0059] 4. A multi-level optimization solving method with fractal structure is applied, which reduces the calculation complexity from O(n³) to O(n·logn), supports simultaneous scheduling of hundreds of virtual power plants, and reduces the operating cost by 7%-12% under the condition of meeting the same carbon emission reduction target. BRIEF DESCRIPTION OF DRAWINGS

[0060] Figure 1 A flowchart of the virtual power plant low-carbon economic scheduling method of the present application;

[0061] Figure 2 A schematic diagram of the uncertainty output prediction and constraint construction method based on probability measure of the present application;

[0062] Figure 3 A schematic diagram of the multi-level optimization solving method with fractal structure of the present application;

[0063] Figure 4The structure block diagram of the virtual power plant low-carbon economic dispatching system is shown in the figure. DETAILED DESCRIPTION

[0064] Please refer to Figure 1 - Figure 4 The application will be further described below in conjunction with the drawings and examples.

[0065] Referring to Figure 1 The application provides a virtual power plant low-carbon economic dispatching method, comprising the following steps:

[0066] Firstly, the historical output data of each virtual power plant in the power grid within a set dispatching period is obtained, the output data of each virtual power plant within the set dispatching period is predicted according to the time sequence, and the output prediction data considering uncertainty is obtained.

[0067] Then, based on the output prediction data considering uncertainty, a carbon emission reduction responsibility allocation space is constructed, each virtual power plant is mapped as a vector point in the space, the carbon emission reduction responsibility weight of each virtual power plant is determined through topological homeomorphism mapping, the objective function with the lowest carbon emission cost is solved in combination with the carbon emission constraint index, and the optimal planning sequence of the output of each virtual power plant within the set dispatching period is obtained.

[0068] Finally, according to the output prediction data considering uncertainty and the optimal planning sequence of the output of each virtual power plant, a discrete-continuous hybrid dynamics model is constructed, the collaborative dispatching of multiple energy storage devices is realized, the time difference non-integer multiple situation of different dispatching periods is handled, each virtual power plant is optimally dispatched, and the power grid is optimized.

[0069] In an embodiment of the application, the step of constructing the carbon emission reduction responsibility allocation space based on the output prediction data considering uncertainty specifically comprises: constructing an n-dimensional responsibility vector space, wherein n is the number of virtual power plants, each virtual power plant is represented as a vector point in the space; introducing total responsibility constraints, non-negative constraints and upper limit constraints to form a convex polyhedron in the responsibility space, representing all feasible responsibility allocation schemes; introducing a time parameter into the responsibility space to dynamically adjust the responsibility allocation with time, and maintaining the smooth transition of the responsibility allocation through continuous deformation.

[0070] Specifically, the n-dimensional responsibility vector space can be expressed as , wherein the basis vector corresponds to the n virtual power plants. Any point represents a possible responsibility allocation scheme, wherein represents the carbon emission reduction responsibility value of the i-th virtual power plant. In the space, a metric function is defined to measure the difference between different responsibility allocation schemes, for example, the Euclidean distance can be used:

[0071] ,

[0072] wherein, is the distance between two responsibility allocation schemes and , is the carbon reduction responsibility value of the i-th virtual power plant in scheme , is the carbon reduction responsibility value of the i-th virtual power plant in scheme , n is the total number of virtual power plants, denotes the summation operation over all virtual power plants.

[0073] The constrained boundary of the responsibility space is jointly defined by the following three constraints:

[0074] 1. Total responsibility constraint: wherein denotes the summation of carbon reduction responsibility values of all virtual power plants, is the total amount of carbon reduction responsibility, this constraint ensures that the total sum of responsibilities of all virtual power plants is equal to the total amount of responsibility set by the system;

[0075] 2. Non-negative constraint: wherein is the carbon reduction responsibility value of the i-th virtual power plant, this constraint ensures that the responsibility value of each virtual power plant is non-negative;

[0076] 3. Upper limit constraint: wherein is the carbon reduction responsibility value of the i-th virtual power plant, is the maximum bearable responsibility of the i-th virtual power plant, this constraint prevents individual virtual power plants from bearing too much responsibility.

[0077] These constraints jointly constitute a convex polyhedron D in the responsibility space, representing all feasible responsibility allocation schemes. To achieve dynamic adjustment of responsibility allocation, a time parameter t is introduced, making the point P(t) in the responsibility space change over time, forming a responsibility trajectory γ(t). In practical applications, the adjustment frequency of responsibility allocation can be set according to system requirements, such as once an hour, once a day, or once a week. Preferably, the adjustment frequency can be set to once every 4 hours, which can timely respond to changes in system state, and also avoid system instability caused by too frequent adjustments.

[0078] In another embodiment of the present application, the step of determining the carbon emission reduction responsibility weight of each virtual power plant by topological homeomorphism mapping specifically comprises: defining a carbon quota space, wherein a point represents a quota allocation scheme; constructing a mapping function from the carbon quota space to the responsibility space, so that the mapping satisfies the continuity condition and the reversibility; designing the mapping function to follow the weight guiding principle, the quota adaptation principle, the fairness principle and the efficiency principle; monitoring the carbon dioxide intensity and the power grid operation state variables in real time, distributing environmental change information to each virtual power plant through a parameter propagation network; defining an adjustment step and an adjustment direction vector, realizing iterative adjustment of responsibility allocation, and ensuring that the adjusted responsibility allocation is still within the feasible region through constraint projection.

[0079] Specifically, a carbon quota space Q is defined, wherein a point represents a quota allocation scheme, is the carbon quota of the i-th virtual power plant. A mapping function is constructed, which maps the quota allocation to the responsibility allocation. The design of the mapping function needs to satisfy the following conditions:

[0080] 1. Continuity condition: a small change in the quota leads to a small change in the responsibility, that is, for any , there exists such that when , there is ; wherein represents the distance between two points and in the quota space, represents the distance between the corresponding points and in the responsibility space, and are positive real numbers;

[0081] 2. Reversibility: different quota allocations correspond to different responsibility allocations, that is, for any , there is ; wherein and are different points in the quota space, and are the corresponding points in the responsibility space.

[0082] The specific form of the mapping function can be designed as:

[0083] ,

[0084] wherein is the carbon emission reduction responsibility value of the i-th virtual power plant, is its carbon quota, is its carbon emission reduction responsibility weight, is the total amount of carbon emission reduction responsibility, and is the weight coefficient, satisfying , , is the total carbon quota of all virtual power plants, is the proportion of the carbon quota of the i-th virtual power plant in the total quota.

[0085] In practical applications, and can be flexibly adjusted according to system requirements. For example, when it is desired to give more consideration to the weight factor, the value of can be increased; when it is desired to give more consideration to the quota factor, the value of can be increased. Preferably, , so that both the weight factor and the quota factor are considered, achieving a relatively balanced responsibility allocation.

[0086] In order to realize dynamic adjustment of responsibility allocation, it is necessary to monitor the carbon dioxide intensity and the grid operation state variables (including load level, renewable energy output, etc.) in real time, and distribute environmental change information to each virtual power plant through a parameter propagation network. The iterative adjustment process of responsibility allocation can be represented as:

[0087] ,

[0088] wherein, is the responsibility allocation vector at time t, containing the responsibility values of all virtual power plants, is the responsibility allocation vector at time , is the adjustment step, which is a non-negative scalar, is the adjustment direction vector, indicating the direction of responsibility adjustment, is the time step.

[0089] The adjustment step can be dynamically set according to the system state, for example, when the system state changes dramatically, the step can be appropriately reduced to ensure the stability of the adjustment; when the system state changes gently, the step can be appropriately increased to speed up the convergence. Preferably, the adjustment step can be set between 0.01 and 0.1, and the specific value can be dynamically adjusted according to actual conditions.

[0090] In order to ensure that the adjusted responsibility allocation is still within the feasible domain D, a constraint projection is needed:

[0091] ,

[0092] wherein, is the projected responsibility allocation vector, Denotes the projection operation onto the convex polyhedron D, ensuring that the projected points are within the feasible region and satisfy all constraints. By constraining the projection, it is possible to ensure that the responsibility allocation always satisfies the total responsibility constraint, the non-negativity constraint, and the upper limit constraint.

[0093] In another embodiment of the present invention, the steps of obtaining output forecast data taking into account uncertainty specifically include: modeling the output of each virtual power plant at time t as a random variable, constructing a random process to describe the time evolution of the output; defining the output state space and probability measure to form a complete probability space; analyzing the statistical characteristics of the historical output time series, identifying periodicity, trend and random components; constructing an output distribution model, including marginal distribution and conditional distribution; calculating the point prediction value based on conditional expectation, and quantifying the uncertainty of the prediction to construct a prediction interval.

[0094] Specifically, the output of the i-th virtual power plant at time t is modeled as a random variable , construct random process Describe the time evolution of output. Define the output state space S and probability measure P, and introduce -Algebra F forms a complete probability space Among them, S is the set of all possible output states, and F is the -algebra (including all measurable events), P is the probability measure function, which maps the events in F to probability values ​​on the interval [0,1].

[0095] The statistical characteristics analysis of the historical output time series includes the calculation of mean, variance, autocorrelation function, etc. Taking the autocorrelation function as an example, its calculation formula is:

[0096] ,

[0097] in, is the autocorrelation function, indicating the interval The correlation between two output values ​​in time, represents the expectation operation, is the output of the i-th virtual power plant at time t, For its The effort of every moment, is the mean output, is the variance of output, is the time delay. By analyzing the autocorrelation function, we can identify the periodic characteristics of the power. For example, for photovoltaic power generation, the autocorrelation function usually has peaks at integer multiples of 24 hours, reflecting the diurnal cycle of photovoltaic power generation.

[0098] The output distribution model can be constructed using either parametric or nonparametric methods. For parametric methods, an appropriate distribution function, such as the normal distribution or the Weibull distribution, can be selected based on the statistical characteristics of the output. For nonparametric methods, techniques such as kernel density estimation can be used to estimate the distribution directly from the data. In practical applications, different types of virtual power plants may have different output distribution characteristics, requiring the selection of an appropriate distribution model based on the specific circumstances.

[0099] The point forecast value calculated based on conditional expectation can be expressed as:

[0100] ,

[0101] in, For the i-th virtual power plant The predicted output of the moment, Indicates that in the known Under the condition of constant output, The conditional expectation of the output at a given moment, k, is the size of the observation window for historical data. In practical applications, the selection of k requires a balance between prediction accuracy and computational complexity. It is generally recommended to use a value between 12 and 24, meaning that data from 12 to 24 historical time points is considered.

[0102] The uncertainty of a forecast can be quantified using a prediction interval:

[0103] ,

[0104] in, is the lower limit of the prediction interval, It is the upper limit of the prediction interval and can be determined by setting the confidence level β. Generally, β=0.95 is taken, which means that there is a 95% probability that the actual output will fall within the prediction interval.

[0105] In another embodiment of the present invention, the carbon emission constraint indicators specifically include: carbon emission and responsibility matching constraints, ensuring that the difference between the total carbon emissions of each virtual power plant and the carbon emission reduction responsibility value does not exceed the set threshold; total emission constraints, ensuring that the total emissions of the system do not exceed the upper limit; equipment operation constraints, including upper and lower limits of output, and ramp rate constraints; energy storage equipment specific constraints, including charging and discharging rates, state of charge range, etc.; by introducing a disturbance set to represent all possible prediction errors, robust constraints are constructed to ensure that the constraints are still met under uncertainty conditions.

[0106] Specifically, the carbon emission and responsibility matching constraint can be expressed as:

[0107] ,

[0108] in, Represents the total carbon emissions of the i-th virtual power plant Its carbon emission reduction responsibility value The absolute value of the difference, To set the threshold, it indicates the maximum deviation allowed. Indicates that all virtual power plants should meet this constraint. In practical applications, The value can be set according to the system tolerance, generally , that is, a 5% deviation is allowed.

[0109] The total emission constraint can be expressed as:

[0110] ,

[0111] in, represents the sum of the total carbon emissions of all virtual power plants, The upper limit of the total carbon emissions allowed by the system.

[0112] Equipment operation constraints include upper and lower output limits and ramp rate constraints:

[0113] ,

[0114] ,

[0115] in, is the output of the i-th virtual power plant at time t, As its minimum output limit, For its maximum output limit, Indicates the absolute value of the output change between two adjacent time points. is the maximum ramp rate of the i-th virtual power plant, means that these constraints should be satisfied for all virtual power plants and at all time points.

[0116] Energy storage device-specific constraints include charge and discharge rate constraints and state of charge range constraints:

[0117] ,

[0118] ,

[0119] in, is the charge and discharge power of the i-th energy storage device at time t (positive value indicates discharge, negative value indicates charge), is the negative value of its maximum charging power, Its maximum discharge power, is the state of charge (StateofCharge) at time t, which represents the percentage of battery power, is the lower limit of its state of charge, is the upper limit of its state of charge, It is indicated that these constraints should be satisfied for all energy storage devices and all time points.

[0120] To handle the uncertainty of the output prediction, a set of disturbances U is introduced to represent all possible prediction errors:

[0121]

[0122] where U is the set of disturbances, which is an ellipsoidal set, u is an n-dimensional prediction error vector, each component corresponds to the prediction error of a virtual power plant, denotes the transpose of u, is the covariance matrix of n x n-dimensional prediction error, which describes the correlation between the prediction errors of different virtual power plants, is the inverse matrix of is the inverse matrix of is a robustness level parameter that controls the size of the ellipsoidal set, is its square value. By adjusting the value of , the degree of conservatism of the robust constraint can be controlled. Generally speaking, the value of can be in the range of 1 to 3, with smaller values corresponding to lower conservatism and larger values corresponding to higher conservatism. Preferably, , the robust constraint can cover about 95% of the prediction error cases, achieving a good balance between conservatism and economy.

[0123] Based on the set of disturbances, a robust constraint is constructed:

[0124]

[0125] where is the constraint function, x is the decision variable vector containing all variables to be optimized, and u is the prediction error vector, indicates that the constraint should be satisfied for any prediction error u in the set of disturbances U. This robust constraint ensures that the constraint condition is satisfied under any possible prediction error, thereby enhancing the robustness of the system.

[0126] In still another embodiment of the present application, the step of solving the objective function with the lowest carbon emission cost specifically comprises: constructing an objective function f, considering factors such as the discounted price of energy storage batteries, scheduling time period, energy storage efficiency, energy consumption efficiency, discount rate and carbon price; decomposing the overall optimization problem into a network of sub-problems with self-similarity to form a fractal structure; designing a hierarchical solving strategy, including solving the basic sub-problems, intermediate sub-problems and global problems from bottom to top; implementing an inter-layer information transmission mechanism and designing a convergence judgment standard to control the iteration process; supervising the overall optimization process through a global coordinator to solve the conflict between local optimum and global optimum.

[0127] ​Specifically, the carbon emission reduction responsibility weight of each virtual power plant is determined by a topological homeomorphism mapping will be substituted into the carbon emission and responsibility matching constraint In combination with the total emission constraint , the equipment operation constraint and the energy storage device specific constraint, an optimization problem is constructed to minimize the carbon emission cost.

[0128] Specifically, the objective function f can be expressed as:

[0129] ,

[0130] where f is the carbon emission cost objective function, represents the summation of all n virtual power plants, represents the summation of all T scheduling periods, is the discounted electricity price obtained by the energy storage battery in the i-th virtual power plant (yuan / kWh), is the charging and discharging power of the energy storage battery in the i-th virtual power plant at period t (kW), is the length of a scheduling period (hours), is the energy consumption efficiency of the energy storage battery (dimensionless, taking a value between 0 and 1), is the energy storage efficiency of the energy storage battery in the i-th virtual power plant at period t (dimensionless, taking a value between 0 and 1), is the total energy consumption of the energy storage battery in the i-th virtual power plant at period t (kWh), and r is the discount rate (dimensionless, representing the time value of money), is the discount factor at period t, and p is the carbon price (yuan / ton), is the total carbon emission of the i-th virtual power plant (tons).

[0131] In practical applications, the values of each parameter need to be determined according to specific circumstances. For example, the discounted electricity price can be set according to the electricity market price and the energy storage characteristics, generally between 0.3 yuan / kWh and 0.8 yuan / kWh; the scheduling period length can be set according to system demand, generally 15 minutes, 30 minutes or 1 hour; the energy storage efficiency is influenced by factors such as battery type and operating state, generally between 0.85 and 0.95; the discount rate can be set according to the time value of money, generally between 0.05 and 0.1; the carbon price is influenced by market factors, generally between 50 yuan / ton and 200 yuan / ton.

[0132] The overall optimization problem is decomposed into a network of self-similar sub-problems, forming a fractal structure, which can greatly improve the solving efficiency. The specific decomposition method is as follows:

[0133] 1. Define problem decomposition mapping : , decompose global problem into sub-problems, where is decomposition mapping function, is global optimization problem, is decomposed sub-problem;

[0134] 2. Ensure sub-problems maintain core structural properties of original problem, i.e. each sub-problem can also be further decomposed in similar manner, forming self-similar nested structure;

[0135] 3. Construct dependency graph G between problems, representing dependency relationships between sub-problems, G is a directed graph, nodes represent sub-problems, edges represent dependency relationships.

[0136] Hierarchical solving strategy includes the following steps:

[0137] 1. First layer: solve basic sub-problems, which are small in size and low in complexity, can be directly solved;

[0138] 2. Second layer: solve intermediate sub-problems based on first layer results, use first layer solutions as input;

[0139] 3. Third layer: integrate lower layer results to solve global problem, combine solutions of each sub-problem into global solution.

[0140] Inter-layer information transfer mechanism ensures coordination and consistency between problems at different levels. Convergence criteria can be set as relative change rate of objective function value less than given threshold (such as 0.1%) or iteration number reaching upper limit (such as 100 times).

[0141] Global coordinator adjusts parameters of each sub-problem to promote local optimization towards global optimal direction. When local optimal conflicts with global optimal, appropriate relaxation of part of constraints or adjustment of objective function weight can be made to balance relationship between different objectives.

[0142] In another embodiment of the present application, according to the output prediction data considering uncertainty and the output optimal planning sequence of each virtual power plant, a discrete-continuous hybrid dynamics model is constructed. Specifically, the output prediction data is taken as the initial condition of the system state, and the output optimal planning sequence is taken as the control target of the system, and the following hybrid dynamics model is constructed:

[0143] ,

[0144] in, is the system state vector, describing the state of the system at time t; for The system state vector at time t; for The system state vector at time t; is the control input vector, which represents the control applied at time t; for The control input vector at time t; is a discrete mapping function describing the scheduling decision point State transfer; is a continuous flow function that describes the dynamic evolution of the system between decision points; is the output prediction data vector; is the output optimal planning sequence vector; represents the time derivative of the state vector; represents the kth scheduling decision time point; represents the k+1th scheduling decision time point; Indicates that time t is within the interval between two adjacent decision points, is the output forecast data vector, which contains the predicted output value of each virtual power plant at time t, for The optimal output planning sequence vector at time.

[0145] The state variable vector can contain the following components:

[0146]

[0147] in, is the system state vector, the superscript T represents the vector transpose, is the output of the i-th virtual power plant at time t (kW), is the state of charge of the j-th energy storage device at time t (dimensionless, indicating the proportion of the remaining battery capacity to the total capacity), n is the number of virtual power plants, m is the number of energy storage devices, and the ellipsis indicates that other state variables such as grid frequency and voltage can also be included.

[0148] Lyapunov exponent spectrum of the system It can be calculated by the following steps:

[0149] 1. Construct the Jacobian matrix J(x,u) of the system. J is an n×n dimensional matrix, where n is the dimension of the state vector and the matrix elements are Represents state variables State derivative The partial derivative of

[0150] 2. Solve the variational equation where is a small perturbation of the state, is the derivative of the perturbation, denotes the multiplication of a matrix and a vector;

[0151] 3. Calculate the expansion rate in each direction to obtain the Lyapunov exponent where denotes the size of the perturbation in the i-th direction at time t, denotes the initial perturbation size, denotes the natural logarithm, denotes the limit as time tends to infinity.

[0152] The sign of the Lyapunov exponent indicates the stability of the system in the corresponding direction: a positive value indicates instability, a negative value indicates stability, and a zero value indicates neutrality. In practical applications, it is generally desirable for the Lyapunov exponent spectrum of the system to have only a small number of positive or zero values, and most of them to be negative, so that the system has good stability.

[0153] Define the fast time scale and the slow time scale , which satisfy , i.e. the fast time scale is much smaller than the slow time scale. The overall scheduling strategy is optimized on the slow time scale, such as global optimization every hour or every day; real-time adjustment and control are performed on the fast time scale, such as local adjustment every minute or every second. Preferably, minutes, hours, so as to ensure the effect of global optimization and timely response to rapid changes in the system.

[0154] In another embodiment of the present application, the step of realizing the coordinated scheduling of multiple energy storage devices specifically includes: defining an energy storage state vector, including state of charge, charging and discharging power, and working mode; defining a state transition function to describe the change of state under action; constructing a state constraint set to ensure that the state is within a safe range; designing a hierarchical control structure, including formulating an overall scheduling strategy, allocating scheduling tasks for each energy storage device, and executing specific charging and discharging control; implementing a load sharing mechanism and fault tolerance strategy between devices to ensure that a single device failure does not affect the overall scheduling.

[0155] Specifically, the energy storage state vector can be represented as:

[0156] ,

[0157] Where s is the energy storage state vector, SoC is the state of charge, which represents the energy storage level (dimensionless, generally a real number between 0 and 1), P is the charge and discharge power (kW, positive values ​​indicate discharge, negative values ​​indicate charge), and mode is the operating mode (an integer, such as 1 for charging, 2 for discharging, and 3 for standby).

[0158] State transfer function T(s, ) described in the action The following status changes:

[0159] ,

[0160] in, is the new state after transfer, T is the state transfer function, s is the current state, is the action applied (such as changing the charge or discharge power).

[0161] For example, the state of charge transition can be expressed as:

[0162] ,

[0163] in, is the state of charge after transfer, SoC is the current state of charge, P is the charge and discharge power (kW), Δt is the time step (hours), η is the charge and discharge efficiency (ηch for charging and 1 / ηdis for discharging, both are dimensionless and range from 0 to 1), and Ecap is the energy storage capacity (kWh).

[0164] The state constraint set Scon ensures that the state is within a safe range:

[0165] ,

[0166] in, is a state constraint set, which contains all states that satisfy the constraints. is the lower limit of the state of charge (dimensionless), is the upper limit of the state of charge (dimensionless), is the maximum charging power (kW), is the maximum discharge power (kW), and the mode value set {1, 2, 3} indicates charging mode, 2 indicates discharging mode, and 3 indicates standby mode.

[0167] In practical applications, different types of energy storage devices may have different parameter values. For example, for lithium-ion batteries, Generally, it is 0.1 to 0.2. Generally, it is set between 0.9 and 0.95 to protect the battery and prolong its life; the charge and discharge efficiency and Typically between 0.9 and 0.95. For flywheel energy storage, It is desirable to approach a value of 0, with higher charge and discharge efficiency, up to 0.95 or more.

[0168] The hierarchical control structure includes three levels:

[0169] 1. Top level: Formulate overall scheduling strategy, determine the operation target of each energy storage device;

[0170] 2. Middle level: Assign scheduling tasks to each energy storage device, task allocation based on device characteristics and operating state;

[0171] 3. Bottom level: Execute specific charge and discharge control, achieve the set power curve.

[0172] The load sharing mechanism between devices can be designed based on the capacity, efficiency and life characteristics of each device. For example, a weighted allocation method can be used:

[0173]

[0174] where, Pi is the allocated power of the i-th energy storage device (kW), wi is its weight (dimensionless, representing the relative importance of the device), Σwi is the sum of the weights of all m energy storage devices, Ptotal is the total demand power (kW), wi / Σwi represents the weight proportion of the i-th device. The weight wi can be determined according to the capacity, current state of charge and life characteristics of the device, for example:

[0175]

[0176] where, wi is the weight of the i-th energy storage device, Ci is the device capacity (kWh), f(SOCi) is a function related to the state of charge, g(Li) is a function related to the device life. The design of functions f and g should ensure that when the state of charge approaches the boundary or the remaining cycle times are few, the weight is correspondingly reduced, thereby reducing the burden on the device.

[0177] Fault tolerance strategy ensures that a single device failure does not affect the overall scheduling. When a device failure is detected, the system will immediately isolate the faulty device and redistribute tasks to other normally operating devices. At the same time, the system will assess the impact of the failure on the overall scheduling and adjust the scheduling plan if necessary.

[0178] ​​In another embodiment of the present application, the step of processing the non-integer multiple of time difference of different scheduling periods specifically includes: defining the time deviation as the difference between the integer multiple of the starting time of the second scheduling period and the starting time of the first scheduling period; designing a time interpolation function to generate a transition state at a non-integer multiple of time; implementing a state smoothing transition strategy to avoid scheduling faults; constructing a state reconstruction method to maintain the continuity of scheduling; when the difference between the starting time of the second scheduling period and the starting time of the first scheduling period is not an integer multiple, calculating the charging and discharging scheduling operation of the second energy storage battery and executing the operation.

[0179] Specifically, the time deviation can be represented as:

[0180]

[0181] wherein, is the time deviation (in hours or minutes), is the starting time of the first scheduling period, is the starting time of the second scheduling period, and k is an integer representing the number of times the first scheduling period can be completely placed before the starting time of the second scheduling period. When , the non-integer multiple of time difference needs to be processed.

[0182] Time interpolation function is used to generate a transition state at a non-integer multiple of time. For example, linear interpolation can be used:

[0183]

[0184] wherein, is the system state at the starting time of the second scheduling period, is the system state at the end of the kth first scheduling period, is the system state at the end of the k+1th first scheduling period, represents the interpolation ratio (dimensionless, taking a value between 0 and 1), and are the weights of the two adjacent states, respectively.

[0185] The state smoothing transition strategy ensures that the system state changes continuously in time, avoiding scheduling faults caused by non-integer multiple of time difference. The specific implementation can use the sliding window method, i.e., at the beginning of each scheduling period, based on the current system state and historical scheduling plan, a smooth scheduling curve for a future period of time is generated.

[0186] The state reconstruction method is used to maintain the continuity of scheduling. When a non-integer multiple of time difference is detected, the system re-evaluates the current state and generates a new scheduling plan based on the latest state. This dynamic adjustment mechanism ensures the continuity and adaptability of the scheduling.​​

[0187] When the difference between the second scheduling period start time and the first scheduling period start time is not an integer multiple of zero, the charge-discharge scheduling operation of the second energy storage battery needs to be calculated. The specific steps are as follows:

[0188] 1. Obtain the charge-discharge scheduling plan of the first energy storage battery;

[0189] 2. Calculate the time deviation ;

[0190] 3. Generate the transition state of the second scheduling period start time based on the time interpolation function;

[0191] 4. According to the transition state and the output optimal planning sequence of the second scheduling period, calculate the charge-discharge scheduling operation of the second energy storage battery;

[0192] 5. Perform the calculated charge-discharge scheduling operation.

[0193] This processing mechanism solves the scheduling fault problem caused by the non-integer multiple of the time difference in the traditional scheduling method, and improves the continuity and stability of the system.

[0194] With reference Figure 4 , the application also provides a virtual power plant low-carbon economic scheduling system, comprising an output prediction module 1, an optimal planning solution module 2 and an optimization scheduling module 3.

[0195] The output prediction module 1 is used to obtain the historical output data of each virtual power plant in the power grid within a set scheduling period, predict the output data of each virtual power plant within the set scheduling period according to the time sequence, and obtain the output prediction data considering uncertainty.

[0196] The optimal planning solution module 2 is used to construct a carbon emission reduction responsibility allocation space based on the output prediction data considering uncertainty, map each virtual power plant as a vector point in the space, determine the carbon emission reduction responsibility weight of each virtual power plant through topological homeomorphism mapping, solve the target function with the lowest carbon emission cost combined with the carbon emission constraint index, and obtain the output optimal planning sequence of each virtual power plant within the set scheduling period.

[0197] The optimization scheduling module 3 is used to construct a discrete-continuous hybrid dynamics model according to the output prediction data considering uncertainty and the output optimal planning sequence of each virtual power plant, realize the collaborative scheduling of multiple energy storage devices, handle the non-integer multiple of the time difference of different scheduling periods, optimize the scheduling of each virtual power plant, and optimize the power grid.

[0198] Preferably, the system further comprises a data collection and preprocessing module 4 and a system control center 5. The data collection and preprocessing module 4 is used to collect virtual power plant real-time operation data, historical output data and external environment data, perform data cleaning, anomaly detection and missing value processing, and distribute the processed data to the output prediction module 1 and the system control center 5. The system control center 5 is used to coordinate the operation of each functional module, allocate carbon emission reduction responsibility weights based on topology mapping, monitor the overall operation state of the system, and handle abnormal situations and emergency responses.

[0199] In the system, the data interaction process between modules is as follows:

[0200] First, the data collection and preprocessing module 4 collects various data and performs preprocessing, and then distributes the processed data to the output prediction module 1 and the system control center 5.

[0201] Second, the output prediction module 1 executes a prediction algorithm based on historical data and external environment data, generates output prediction data considering uncertainty, and sends the prediction results to the optimal planning solution module 2.

[0202] Then, the optimal planning solution module 2 combines the output prediction data and the carbon emission reduction responsibility weights allocated by the system control center 5, constructs and solves the optimization problem, obtains the output optimal planning sequence of each virtual power plant, and sends the optimization results to the optimal scheduling module 3.

[0203] Finally, the optimal scheduling module 3 generates specific device scheduling instructions according to the output optimal planning sequence, handles the case of non-integer multiples of time difference values, coordinates the charging and discharging scheduling of multiple energy storage devices, and issues the scheduling instructions to the control systems of each virtual power plant.

[0204] Through the collaborative work of the above modules, the system can achieve the economic optimal operation of the virtual power plant group under low-carbon constraints, and provide strong support for the low-carbon transformation of the power system.

[0205] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any skilled person in the art can easily think of changes or replacements within the technical scope disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A virtual power plant low-carbon economic dispatching method, characterized in that, The method comprises the following steps: obtaining historical output data of each virtual power plant in the power grid within a set scheduling period, predicting the output data of each virtual power plant within the set scheduling period according to the time sequence, and obtaining output prediction data considering uncertainty; based on the output prediction data considering uncertainty, constructing a carbon emission reduction responsibility allocation space, mapping each virtual power plant as a vector point in the space, determining the carbon emission reduction responsibility weight of each virtual power plant through topological homeomorphism mapping, substituting the determined carbon emission reduction responsibility weight into the carbon emission constraint index, constructing an optimization problem with the minimum carbon emission cost as the target, and solving the optimization problem to obtain the optimal output planning sequence of each virtual power plant within the set scheduling period; based on the output prediction data considering uncertainty and the optimal output planning sequence of each virtual power plant, constructing a discrete-continuous hybrid dynamics model, realizing the collaborative scheduling of multiple energy storage devices, and processing the time difference non-integer multiple situation of different scheduling periods, optimizing the scheduling of each virtual power plant, and optimizing the power grid. 2.The low-carbon economic dispatching method of a virtual power plant according to claim 1, wherein, The step of constructing a carbon emission reduction responsibility allocation space based on the output prediction data considering uncertainty comprises the following steps: constructing an n-dimensional responsibility vector space, where n is the number of virtual power plants, and each virtual power plant is represented as a vector point in the space; introducing total responsibility constraints, non-negative constraints and upper limit constraints to form a convex polyhedron in the responsibility space, representing all feasible responsibility allocation schemes; introducing a time parameter into the responsibility space to dynamically adjust the responsibility allocation with time, and maintaining the smooth transition of the responsibility allocation through continuous deformation. 3.The low-carbon economic dispatching method of a virtual power plant according to claim 2, characterized in that, The step of determining the carbon emission reduction responsibility weight of each virtual power plant through topological homeomorphism mapping comprises the following steps: defining a carbon quota space, where points represent quota allocation schemes; constructing a mapping function from the carbon quota space to the responsibility space, so that the mapping satisfies the continuity condition and reversibility; designing the mapping function to follow the weight guiding principle, the quota adaptation principle, the fairness principle and the efficiency principle; monitoring carbon dioxide intensity and power grid operating state variables in real time, and distributing environmental change information to each virtual power plant through a parameter propagation network; defining an adjustment step and an adjustment direction vector to realize the iterative adjustment of the responsibility allocation, and ensuring that the adjusted responsibility allocation is still within the feasible domain through constraint projection. 4.The low-carbon economic dispatch method of a virtual power plant according to claim 1, wherein, The step of obtaining output prediction data considering uncertainty comprises the following steps: modeling the output of each virtual power plant at time t as a random variable, and constructing a stochastic process to describe the time evolution of the output; defining an output state space and a probability measure to form a complete probability space; analyzing the statistical characteristics of the historical output time series, identifying periodicity, trend and randomness; constructing an output distribution model, including marginal distribution and conditional distribution; calculating the point prediction value based on the conditional expectation, quantifying the uncertainty of the prediction, and constructing the prediction interval.

5. The low-carbon economic dispatch method of a virtual power plant according to claim 1, characterized in that, The carbon emission constraint index comprises: carbon emission and responsibility matching constraint, ensuring that the difference between the total carbon emission of each virtual power plant and the carbon emission reduction responsibility value does not exceed a set threshold; total emission constraint, ensuring that the total emission of the system does not exceed the upper limit; device operation constraints, including output upper and lower limits, ramp rate constraints; specific constraints of energy storage devices, including charge and discharge rate, state of charge range, etc. Robust constraints are constructed by introducing a set of disturbances representing all possible prediction errors, ensuring that the constraints are still satisfied under uncertainty.

6. A low-carbon economic dispatch method for a virtual power plant according to claim 1, characterized in that: The steps of solving the objective function of the lowest carbon emission cost include: The objective function f is constructed by considering factors such as the discounted price of energy storage batteries, scheduling period, energy storage efficiency, energy consumption efficiency, discount rate, and carbon price. The overall optimization problem is decomposed into a network of self-similar subproblems, forming a fractal structure. A hierarchical solving strategy is designed, including solving the basic subproblem, the intermediate subproblem, and the global problem from bottom to top. An interlayer information transmission mechanism is implemented, and a convergence judgment standard is designed to control the iteration process. A global coordinator is implemented to supervise the overall optimization process and solve the conflict between local optimization and global optimization.

7. The low-carbon economic dispatch method of a virtual power plant according to claim 1, characterized in that, The steps of constructing a discrete-continuous hybrid dynamics model based on the output forecast data considering uncertainty and the output optimal planning sequence of each virtual power plant include: As the initial condition of the system state, the output optimal planning sequence As the system control target; design a discrete part to describe the state transition of the scheduling decision point; design a continuous part to describe the dynamic evolution of the system between decision points; construct a state variable vector to represent the system state; calculate the Lyapunov exponent spectrum of the system to analyze the structural characteristics of the chaotic attractor; define fast time scales and slow time scales, optimize the overall scheduling strategy on the slow time scale, and perform real-time adjustment and control on the fast time scale. 8.The low-carbon economic dispatch method of a virtual power plant according to claim 1, wherein, The steps of realizing the collaborative scheduling of multiple energy storage devices include: Define the state vector of energy storage, including state of charge, charging and discharging power, and working mode. Define the state transition function to describe the change of state under action. Construct a set of state constraints to ensure that the state is within a safe range. Design a hierarchical control structure, including formulating an overall scheduling strategy, allocating scheduling tasks for each energy storage device, and executing specific charging and discharging control. Implement a load sharing mechanism and fault tolerance strategy between devices to ensure that the failure of a single device does not affect the overall scheduling. 9.The low-carbon economic dispatch method of a virtual power plant according to claim 1, wherein, The steps of handling the case where the time difference between different scheduling periods is not an integer multiple include: Define the time offset as the difference between the integer multiple of the start time of the second scheduling period and the start time of the first scheduling period. Design a time interpolation function to generate a transition state at a non-integer multiple time point. Implement a state smoothing transition strategy to avoid scheduling faults. Construct a state reconstruction method to maintain the continuity of scheduling. When the difference between the start time of the second scheduling period and the start time of the first scheduling period is not an integer multiple, calculate the charging and discharging scheduling operation of the second energy storage battery and execute the operation.

10. A virtual power plant low-carbon economic dispatching system, characterized in that, It includes: An output prediction module is used to obtain historical output data of each virtual power plant in the power grid within a set scheduling period, predict the output data of each virtual power plant within the set scheduling period based on time series, and obtain output prediction data considering uncertainty. An optimal planning solving module is used to construct a carbon emission reduction responsibility allocation space based on the output prediction data considering uncertainty, map each virtual power plant to a vector point in the space, determine the carbon emission reduction responsibility weight of each virtual power plant through topological homeomorphism mapping, solve the objective function of the lowest carbon emission cost combined with the carbon emission constraint index, and obtain the optimal planning sequence of the output of each virtual power plant within the set scheduling period. An optimal scheduling module is used to construct a discrete-continuous hybrid dynamics model based on the output prediction data considering uncertainty and the optimal planning sequence of the output of each virtual power plant, realize the collaborative scheduling of multiple energy storage devices, handle the case where the time difference between different scheduling periods is not an integer multiple, and perform optimal scheduling for each virtual power plant to optimize the power grid.

Images