AC / DC power distribution system typical scene generation method considering source-load space-time correlation

Abstract

The invention discloses an AC / DC power distribution system typical scene generation method considering source-load space-time correlation, and relates to the technical field of scene generation, and the method comprises the steps: building a digital representation model based on the typical grid topology of an AC / DC power distribution system; according to new energy spatio-temporal evolution characteristics, constructing a time correlation model of wind and light output based on a dependency function-state transition chain, and establishing a spatial correlation model of wind and light output based on a multivariate hierarchical tree; constructing a load time sequence optimization model driven by a daily load characteristic index according to the random volatility of the daily load, and constructing a charging load generation model according to different user groups of the electric vehicle; establishing a flexibility demand and supply quantification model; and constructing an original scene according to the established model, and clustering the generated original scene to obtain a typical scene of the AC / DC power distribution system. According to the method, the source-load space-time correlation can be extracted based on historical data, the original scene is generated, and the typical scene is obtained through clustering.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of scenario generation, in particular to a method for generating typical scenarios of AC-DC power distribution system considering spatiotemporal correlation of source and load. BACKGROUND

[0002] The AC-DC power distribution system is a new type of power system that realizes flexible interconnection and closed-loop operation through flexible interconnection devices and power electronic transformers. Unlike the rigid topology of traditional distribution networks, this system forms a "meshed" topology structure with multi-modal operation capability. Through flexible interconnection devices as the core hub, different flexible interfaces of electric energy forms and voltage levels are constructed, which not only significantly improves the system's ability to accommodate new energy and suppress renewable energy fluctuations, but also has multiple functions such as continuous power flow regulation, fault current blocking, and flexible resource allocation. With the widespread access of distributed energy and electric vehicles and other diversified sources and loads, AC-DC hybrid distribution networks can realize the complementary advantages of AC and DC systems, and exhibit significant advantages in improving power supply and distribution efficiency, economy, and reliability.

[0003] Existing typical scenario generation methods are mainly based on historical data and statistical models, and extract typical operating scenarios through analysis and clustering of source and load characteristics. These methods include using probability density functions to describe the randomness of source and load output, using state transition chains to simulate time-varying characteristics, and using various clustering algorithms to identify typical scenarios. In AC distribution networks, existing research has established a load curve generation model based on daily load characteristic indicators, and an electric vehicle charging load model considering user behavior characteristics. At the same time, some research attempts to quantify flexible demand and supply to describe the impact of new energy access on the system, and uses density peak clustering methods to identify typical operating scenarios.

[0004] However, the current scenario generation method still has obvious limitations: on the one hand, existing technologies mostly focus on AC distribution networks, and the spatiotemporal correlation of source and load in AC-DC hybrid systems is not well characterized, especially the coupling characteristics of wind and light output in time and space. Traditional methods often separate the spatiotemporal correlation of wind and light, and cannot accurately represent the dynamic evolution law. On the other hand, the interaction of multiple elements in AC-DC hybrid distribution networks is complex, and existing models lack digital representation of AC-DC interconnection mode characteristics, and it is also difficult to quantify the flexibility of supply and demand in the interaction process. In addition, the spatiotemporal correlation of wind and light output is superimposed, and traditional clustering algorithms are difficult to effectively process high-dimensional and nonlinear original scenario data, resulting in insufficient extraction accuracy of typical scenarios, which cannot provide reliable support for the scheduling and planning of AC-DC power distribution systems.

[0005] In view of the problems in the related art, no effective solution has been proposed so far. SUMMARY

[0006] To address the problems in related technologies, this invention proposes a method for generating typical scenarios of AC / DC power distribution systems that considers the spatiotemporal correlation of sources and loads. This method has the advantages of characterizing the spatiotemporal correlation of sources and loads, representing AC / DC interconnection modes, and processing high-dimensional scenario data. This solves the problems of fragmented source-load correlation, insufficient representation of interconnection modes, and low scenario extraction accuracy in existing technologies.

[0007] Therefore, the specific technical solution adopted by the present invention is as follows:

[0008] A method for generating typical AC / DC power distribution system scenarios considering the spatiotemporal correlation of power sources and loads, comprising:

[0009] S1. Construct a digital representation model based on the typical network topology of AC / DC power distribution systems;

[0010] S2. Based on the spatiotemporal evolution characteristics of new energy, a time correlation model of wind and solar power output is constructed based on the dependency function-state transition chain, a spatial correlation model of wind and solar power output is established based on the multivariate hierarchical tree, and the spatial correlation model is combined with the time correlation model.

[0011] S3. Based on the random fluctuation of daily load, construct a load time series optimization model driven by daily load characteristic indicators, and based on the differences in energy consumption characteristics of different electric vehicle user groups, construct a charging load generation model based on electric vehicle type and user behavior characteristics.

[0012] S4. Based on the source-load interaction characteristics of AC / DC distribution networks, establish a flexibility demand quantification model and a flexibility supply quantification model respectively.

[0013] S5. Construct the original scene based on the established model, and cluster the generated original scene by combining the standard deviation weighted distance and fuzzy weighted K nearest neighbor clustering algorithm to obtain the typical scene of AC / DC power distribution system.

[0014] Furthermore, the construction of the digital representation model includes: establishing node type variables based on the node power supply method, and constructing node DG state variables based on the type and location of the node connected to the DG; the node type variables include AC / DC node state variables, and the node DG state variables include photovoltaic node state variables and wind power node state variables; determining the network branch type based on the node types at both ends of the branch, and establishing a branch type representation model; specifically, if both ends of the branch are AC nodes, then the branch is an AC branch; if both ends of the branch are DC nodes, then the branch is a DC branch; if both ends of the branch are AC and DC nodes respectively, then the branch is an AC / DC branch and interconnected through VSC; constructing constraints based on the node type variables and branch type variables to characterize the wind and solar scale and AC / DC interconnection mode of the distribution system, including DG quantity constraints, DG capacity constraints, and branch type quantity constraints.

[0015] Furthermore, the time-dependent model of wind and solar power output based on the dependency function-state transition chain includes: obtaining the probability density function of wind and solar power output for each time period using kernel probability density estimation based on the historical power output data of each wind and solar node in each time period; selecting a multivariate dependency function for data fitting using the ordinal correlation metric; calculating the empirical dependency function of wind and solar power output at adjacent times, selecting the dependency function with the smallest Euclidean distance as the optimal fitting function by calculating the Euclidean distance between different types of dependency functions and the empirical dependency function using the ordinal correlation metric, and deriving the conditional probability density function of the current time's power output under the condition of the previous time's power output using the marginal probability distribution and joint probability distribution; determining the degree of correlation between each time period and the current time by calculating the linear correlation coefficient between the wind and solar power output sequences at different times, so as to select the order of the higher-order time-varying state transition chain; assigning different weights to historical states according to their time proximity using the forgetting factor, and constructing a higher-order time-varying state transition chain for continuous wind and solar power output by replacing the state transition core with the dependency function.

[0016] Furthermore, the spatial correlation model of wind and solar power output based on multi-level hierarchical trees includes: using the ordinal correlation metric as weight, selecting the tree structure with the largest edge weight in the complete graph; constructing the multi-level hierarchical tree through a real-time iterative maximum spanning tree algorithm; specifically including: data preprocessing, and generating the first complete graph and the first tree with weighted edges; selecting the corresponding dependency function distribution family for the marginal variables of each edge in the first complete graph, and performing parameter estimation; generating the next complete graph through the first tree, and iteratively obtaining all tree structures; and sampling all edges in the structure by traversing the structure starting from the last tree in the multi-level hierarchical tree through a recursive algorithm.

[0017] Furthermore, combining the spatial correlation model with the temporal correlation model includes: generating high-order time-varying state transition chains for each wind and solar cluster to obtain the temporal correlation probability density matrix of the cluster; establishing a multivariate hierarchical tree model of the wind and solar cluster at each time moment, and performing multidimensional vector random sampling at each time moment; determining the output value of each wind and solar cluster at the initial time moment to obtain the conditional probability density function of each wind and solar cluster at the next time moment; using the spatial correlation sampled value as the sampled value of the conditional probability distribution, and substituting it into the inverse function of the conditional probability density function through inverse sampling to obtain the output sampled value of the wind and solar cluster at the next time moment; repeating the inverse sampling process for all wind and solar clusters and performing time recursion to obtain the output matrix of each wind and solar cluster at all times, so as to form a temporal sequence scenario that considers spatiotemporal correlation.

[0018] Furthermore, the construction of a load time series optimization model driven by daily load characteristic indicators includes: characterizing the peak-valley steepness and flatness of the load power curve based on the daily peak-valley difference rate and daily load rate, and setting the absolute power scale according to the daily maximum load; the daily load rate is the ratio of the daily average load to the maximum load, and the daily peak-valley difference rate is the ratio of the daily peak-valley difference maximum value to the daily maximum load; with the minimum of the sum of squared errors between the generated daily load sequence and the daily load reference curve as the optimization objective, a load time series optimization model is established, and the constraints of the model are constructed by combining the daily load rate constraint, the daily peak-valley difference rate constraint, and the daily maximum load constraint; wherein, the daily load characteristic indicators include the daily load rate, the daily peak-valley difference rate, and the daily maximum load; the daily load rate constraint is achieved by ensuring that the average value of the generated load curve is consistent with the reference curve; the daily peak-valley difference rate constraint is achieved by ensuring that the load at each moment in the load curve is greater than the daily minimum load, and that the time when the daily minimum load of the generated load curve occurs is consistent with the reference curve; the daily maximum load constraint is achieved by ensuring that the load at each moment in the load curve is less than the daily maximum load, and that the time when the daily maximum load of the generated load curve occurs is consistent with the reference curve.

[0019] Furthermore, the construction of a charging load generation model based on electric vehicle types and user behavior characteristics includes: classifying electric vehicles into different types according to usage scenarios and establishing probabilistic models for travel time, mileage, and charging start time for each type; determining the corresponding charging time distribution and daily mileage distribution based on the usage behavior characteristics of each type of electric vehicle; dynamically adjusting the charging power based on grid load levels and electricity price signals, and introducing a response factor to enable the charging load to actively adapt to the grid conditions; dividing the charging process into a constant current stage and a constant voltage stage; maintaining a constant charging power during the constant current stage, and decreasing the charging power during the constant voltage stage as the state of charge (SOC) increases; setting the charging frequency according to the usage frequency of different types of electric vehicles and establishing a relationship model between the remaining battery capacity and charging duration; and using a random sampling method to simulate each electric vehicle to obtain the starting charging time and charging duration, thereby obtaining the daily basic charging load of the electric vehicle.

[0020] Furthermore, establishing a flexibility requirement quantification model includes: calculating the upward and downward flexibility requirements generated by load fluctuations based on load power demand and time scale; calculating the upward and downward flexibility requirements generated by wind power and photovoltaic power based on wind and solar power output and time scale; and summing the upward and downward flexibility requirements of load, photovoltaic power and wind power at the same time to obtain the overall system flexibility requirement.

[0021] Furthermore, establishing a quantitative model for flexibility supply includes: determining the upward and downward flexibility supply at each moment based on the power constraints of flexible interconnected devices; determining the upward and downward flexibility supply based on the loads that can be transferred in, transferred out, and shed in the demand response; determining the upward and downward flexibility supply based on the charging and discharging power, energy capacity, state of charge, and minimum and maximum state of charge of the cluster of electric vehicles; and summing the upward and downward flexibility supply of the demand response, flexible interconnected devices, and electric vehicles at the same moment to obtain the overall system flexibility supply.

[0022] Furthermore, the generated original scenes are clustered using a clustering algorithm combining standard deviation weighted distance and fuzzy weighted K nearest neighbors. This includes: treating each original scene as a data point; quantifying the flexibility supply and demand at each time step using the feature value of each data point; calculating the distance between each data point using standard deviation as a weighting factor in the distance metric; calculating the local density and center offset distance for each data point, and selecting the point with the largest product of local density and center offset distance as the cluster center; calculating the maximum neighborhood radius from each data point to its K nearest neighbors, and classifying data points with a maximum neighborhood radius greater than the average as outliers, and other points as non-outliers; clustering assignment using K nearest neighbors and breadth-first search for non-outliers; and clustering assignment using the fuzzy weighted K nearest neighbor method for outliers by calculating membership degrees and constructing a membership degree matrix to obtain typical AC / DC power distribution system scenarios.

[0023] The beneficial effects of this invention are as follows:

[0024] (1) This invention establishes a digital representation model of the AC / DC power distribution system and a source-load spatiotemporal correlation model, thereby achieving an accurate characterization of the spatiotemporal evolution characteristics of wind and solar power output. It adopts a method combining dependency function-state transition chain and multi-layered tree to solve the problem of fragmented processing of the spatiotemporal correlation of wind and solar power output. At the same time, it introduces time-series optimization driven by load characteristic index and charging load generation based on user behavior, which improves the accuracy of source-load collaborative modeling and provides reliable typical scenario support for the planning and scheduling of AC / DC power distribution systems.

[0025] (2) This invention establishes a quantitative model of flexibility demand and flexibility supply, systematically analyzes the impact of load fluctuations and changes in wind and solar power output on flexibility demand, as well as the contributions of flexible interconnection devices, demand response and electric vehicles to flexibility supply; through precise characterization of source-load interaction characteristics and quantitative evaluation of flexibility balance relationship, it can accurately capture the complex interaction of multiple elements in AC and DC power distribution system, provide optimization basis for the selection of flexible interconnection mode and flexible resource allocation of system, and improve the system's ability to cope with the uncertainty and volatility of new energy power output.

[0026] (3) This invention constructs a density peak clustering method based on standard deviation weighted distance and fuzzy weighted K nearest neighbor. By calculating the standard deviation weight of the feature dimension and the local density of the data points, it effectively distinguishes outliers and non-outliers. For non-outliers, K nearest neighbor and breadth-first search methods are used for clustering and allocation. For outliers, fuzzy weighted K nearest neighbor method is used to calculate the membership matrix for allocation. This greatly improves the processing capability of high-dimensional nonlinear original scene data and the extraction accuracy of typical scenes, providing a scientific basis for the flexibility assessment and optimization configuration of AC / DC power distribution systems. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 This is a flowchart illustrating a method for generating typical scenarios of AC / DC power distribution systems that considers the spatiotemporal correlation of source and load according to an embodiment of the present invention.

[0029] Figure 2 This is a detailed implementation diagram of a method for generating typical scenarios of AC / DC power distribution systems that considers the spatiotemporal correlation of source and load according to an embodiment of the present invention;

[0030] Figure 3 This is a network topology diagram of a typical AC / DC hybrid distribution network modified from IEEE 33 nodes, based on a method for generating typical AC / DC distribution system scenarios that considers the spatiotemporal correlation of source and load according to an embodiment of the present invention.

[0031] Figure 4 This is a quantification diagram of the flexibility requirements of the original scenario in a typical scenario generation method for AC / DC power distribution systems that considers the spatiotemporal correlation of source and load according to an embodiment of the present invention.

[0032] Figure 5This is an original scenario flexibility supply quantification diagram in a typical scenario generation method for AC / DC power distribution systems that considers the spatiotemporal correlation of source and load according to an embodiment of the present invention.

[0033] Figure 6 This is a power curve diagram of source and load elements in a typical scenario of an AC / DC power distribution system that considers the spatiotemporal correlation of source and load according to an embodiment of the present invention. Detailed Implementation

[0034] To further illustrate the various embodiments, the present invention provides accompanying drawings, which are part of the disclosure of the present invention. These drawings are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these drawings, those skilled in the art should be able to understand other possible implementation methods and the advantages of the present invention. The components in the drawings are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0035] According to an embodiment of the present invention, a method for generating typical scenarios of AC / DC power distribution systems that considers the spatiotemporal correlation of source and load is provided.

[0036] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments, such as... Figure 1 and Figure 2 As shown, according to an embodiment of the present invention, a method for generating typical scenarios of AC / DC power distribution systems considering the spatiotemporal correlation of source and load is provided. This method includes:

[0037] S1. Construct a digital representation model based on the typical network topology of AC / DC power distribution systems;

[0038] S2. Based on the spatiotemporal evolution characteristics of new energy, a time correlation model of wind and solar power output is constructed based on the dependency function-state transition chain, a spatial correlation model of wind and solar power output is established based on the multivariate hierarchical tree, and the spatial correlation model is combined with the time correlation model.

[0039] S3. Based on the random fluctuation of daily load, construct a load time series optimization model driven by daily load characteristic indicators, and based on the differences in energy consumption characteristics of different electric vehicle user groups, construct a charging load generation model based on electric vehicle type and user behavior characteristics.

[0040] S4. Based on the source-load interaction characteristics of AC / DC distribution networks, establish a flexibility demand quantification model and a flexibility supply quantification model respectively.

[0041] S5. Construct the original scene based on the established model, and cluster the generated original scene by combining the standard deviation weighted distance and fuzzy weighted K nearest neighbor clustering algorithm to obtain the typical scene of AC / DC power distribution system.

[0042] In one embodiment, the time-dependent model obtains the probability density function of wind and solar power output at each moment through kernel density estimation, selects the dependency function to calculate the joint probability distribution, and introduces a higher-order time-varying state transition chain to realize the dynamic adjustment of the state transition probability.

[0043] The digital representation model divides nodes into DC nodes and AC nodes according to the power supply method, establishes node type variables and node DG state variables, and determines the network branch type based on the node types at both ends of the branch, thus establishing a branch type representation model.

[0044] The charging load generation model classifies electric vehicles into different types and establishes probabilistic models for travel time, driving mileage, and charging start time for each type to simulate the charging load demand of electric vehicles.

[0045] The clustering algorithm takes each original scene as a data point, uses the flexibility supply and demand quantification result at each moment as a feature value, determines the cluster center by calculating the standard deviation weight and local density, and classifies non-density peak points into outliers and non-outliers for processing.

[0046] Specifically, the specific steps of this invention include: Step A, constructing a digital representation model of the AC / DC power distribution system based on the structural characteristics of typical AC / DC power distribution system network topology and actual application scenarios; Step B, constructing a wind and solar power output time correlation model based on Copula-Markov chain and a wind and solar power output spatial correlation model based on R-Vine Copula based on the spatiotemporal evolution characteristics of new energy sources; Step C, constructing a load time series optimization model driven by daily load characteristic indicators based on the random fluctuation of daily load; Step D, constructing a charging load generation model based on electric vehicle type and user behavior characteristics based on the differences in energy consumption characteristics of different electric vehicle user groups; Step E, establishing a quantitative model of flexibility demand and flexibility supply based on the source-load interaction characteristics of AC / DC power distribution networks; Step F, clustering the above-generated original scenarios using a density peak clustering algorithm based on standard deviation weighted distance and fuzzy weighted K nearest neighbors to obtain typical AC / DC power distribution system scenarios.

[0047] In one embodiment, constructing a digital representation model includes: establishing node type variables based on the node power supply method, and constructing node DG state variables based on the type and location of the node connected to the DG; the node type variables include AC / DC node state variables, and the node DG state variables include photovoltaic node state variables and wind power node state variables; determining the network branch type based on the node types at both ends of the branch, and establishing a branch type representation model; specifically, if both ends of the branch are AC nodes, then the branch is an AC branch; if both ends of the branch are DC nodes, then the branch is a DC branch; if both ends of the branch are AC and DC nodes respectively, then the branch is an AC / DC branch and interconnected through VSC; constructing constraints based on the node type variables and branch type variables to characterize the wind and solar scale and AC / DC interconnection mode of the distribution system, including DG quantity constraints, DG capacity constraints, and branch type quantity constraints.

[0048] Specifically, node type variables include AC / DC node status variables, which take a value of 1 when the node is an AC node and a value of 0 when it is a DC node; node DG status variables include photovoltaic node status variables and wind power node status variables, where the photovoltaic node status variable takes a value of 1 when the node is connected to photovoltaics and 0 otherwise, and the wind power node status variable takes a value of 1 when the node is connected to wind power and 0 otherwise; DG quantity constraints are obtained by controlling the total number of photovoltaic nodes and the total number of wind power nodes to be between the minimum and maximum values; DG capacity constraints are obtained by setting the photovoltaic capacity and wind power capacity to be within their respective maximum and minimum value ranges, and ensuring that the wind-solar configuration ratio matches the local resource conditions and load characteristics based on the wind-solar complementarity coefficient; branch type quantity constraints are obtained by limiting the total number of AC branches, the total number of DC branches, and the total number of AC-DC branches to their respective set maximum and minimum value ranges.

[0049] Specifically, step A considers the AC / DC distribution system network topology, constructs a digital representation model, and then establishes different constraints to characterize the features of different typical interconnection modes. The distribution network topology consists of two basic elements: nodes and branches. Based on the unique structure of the AC / DC hybrid distribution network, the nodes and branches in the system can be finely classified.

[0050] Specifically, this invention classifies nodes into DC nodes and AC nodes based on their power supply method, and establishes node type variables as shown in equation (1). Based on the type and location of the node connected to the DG, node DG state variables are constructed as shown in equations (2)-(3):

[0051] (1)

[0052] (2)

[0053] (3)

[0054] In equations (1)-(3), Represents the AC / DC node state variables of node n. and These represent the sets of AC and DC nodes, respectively. This represents the photovoltaic node state variable of node n. Represents the set of photovoltaic nodes; This represents the wind power node state variable of node n. Represents the set of wind power nodes;

[0055] Specifically, the network branch type can be determined based on the types of nodes at both ends of the branch. If both nodes at both ends of the branch are AC nodes, then the branch is an AC branch; if both nodes at both ends of the branch are DC nodes, then the branch is a DC branch; if both nodes at both ends of the branch are AC and DC nodes respectively, then the branch is an AC / DC branch, interconnected through VSC; at the same time, for the same branch, there is only one possible branch type. A branch type representation model is established, as shown in equations (4)-(7):

[0056] (4)

[0057] (5)

[0058] (6)

[0059] (7)

[0060] In equations (4)-(7), Represents the AC branch state variables of branch b; Represents the set of AC branches; Represents the DC branch state variables of branch b; Represents the set of DC branches; Represents the AC-DC branch state variables of branch b; Represents the set of AC-DC branches;

[0061] Specifically, a series of constraints can be constructed based on node and branch type variables to characterize the wind and solar power scale and AC / DC interconnection mode of the power distribution system. The number of DGs connected to the AC / DC power distribution system is controlled by the node DG state variables. Based on the number of different node DG state variables, DG quantity constraints are constructed, as shown in Equation (8). At the same time, different DGs need to be set with different access capacity limits. The configuration ratio of wind power and photovoltaic power should be matched with local resource conditions and load characteristics. Therefore, DG capacity constraints based on the DG complementarity index are constructed, as shown in Equation (9). Based on the characteristics of different types of AC / DC interconnection modes, the number of different branch types is determined, and branch type quantity constraints are constructed, as shown in Equation (10).

[0062] (8)

[0063] (9)

[0064] (10)

[0065] In equations (8)-(10), and These represent the maximum and minimum values ​​of the total number of photovoltaic nodes, respectively. and These represent the maximum and minimum values ​​of the total number of wind power nodes, respectively. and These represent the maximum and minimum values ​​of the photovoltaic capacity, respectively. and These represent the maximum and minimum values ​​of wind power capacity, respectively; S n PV and S n Wind S represents the photovoltaic and wind power capacity of node n, respectively; max Indicates the theoretical maximum capacity of wind and solar power; CFI max and CFI min These represent the maximum and minimum values ​​of the wind-solar complementarity coefficient, respectively; P t PV and P t wind These represent historical data on the scenery; and These represent the maximum and minimum values ​​of the total number of AC branches, respectively; and These represent the maximum and minimum values ​​of the total number of DC branches, respectively. and These represent the maximum and minimum values ​​of the total number of AC-DC branches, respectively; N node and Nline These represent the total number of nodes and branches, respectively.

[0066] In one embodiment, constructing a time-dependent model of wind and solar power output based on a dependency function-state transition chain includes:

[0067] Based on the historical power output data of each wind and solar node in each time period, the probability density function of wind and solar power output in each time period is obtained by kernel probability density estimation.

[0068] Using the ordinal correlation metric, a multivariate dependency function is selected for data fitting; the empirical dependency function of wind and solar power output at adjacent times is calculated; by calculating the Euclidean distance between the ordinal correlation metric of different types of dependency functions and the empirical dependency function, the dependency function with the smallest Euclidean distance is selected as the optimal fitting function; and the conditional probability density function of power output at the current time under the power output condition of the previous time is derived through the marginal probability distribution and joint probability distribution.

[0069] The degree of correlation between each moment and the current moment is determined by calculating the linear correlation coefficient between the wind and solar power output sequences at different times, so as to select the order of the higher-order time-varying state transition chain. The historical states are assigned different weights according to their time proximity by using the forgetting factor, and the state transition core is replaced by the dependency function to construct a higher-order time-varying state transition chain for continuous wind and solar power output.

[0070] In one embodiment, establishing a spatial correlation model for wind and solar power output based on a multi-level hierarchical tree includes: selecting the tree structure with the largest edge weight in the complete graph using the ordinal correlation metric as weight; constructing the multi-level hierarchical tree using a real-time iterative maximum spanning tree algorithm; specifically including: data preprocessing, and generating a first complete graph and a first tree with weighted edges; selecting the corresponding dependency function distribution family for the marginal variables of each edge in the first complete graph, and performing parameter estimation; generating the next complete graph from the first tree, and iteratively obtaining all tree structures; and sampling all edges in the structure by traversing it from the last tree in the multi-level hierarchical tree using a recursive algorithm.

[0071] In one embodiment, combining the spatial correlation model with the temporal correlation model includes: generating a high-order time-varying state transition chain for each wind and solar cluster to obtain the temporal correlation probability density matrix of the cluster; establishing a multivariate hierarchical tree model of the wind and solar cluster at each time moment, and performing multidimensional vector random sampling at each time moment; determining the output value of each wind and solar cluster at the initial time moment to obtain the conditional probability density function of each wind and solar cluster at the next time moment; using the spatial correlation sampled value as the sampled value of the conditional probability distribution, and substituting it into the inverse function of the conditional probability density function through inverse sampling to obtain the output sampled value of the wind and solar cluster at the next time moment; repeating the inverse sampling process for all wind and solar clusters and performing time recursion to obtain the output matrix of each wind and solar cluster at all times, so as to form a temporal sequence scenario considering spatiotemporal correlation.

[0072] Specifically, step B requires obtaining a wind and solar power output conditional probability density function that can characterize the long-term output correlation based on the historical data of each wind and solar power in each time period, and then generating a dynamic time-dependent state transition chain by combining it with a high-order time-varying Markov chain. The specific steps include:

[0073] (1) Based on the historical data of each wind and solar power in each time period, the wind and solar power output in each time period is estimated by kernel density estimation, and the probability density function is obtained as shown in Equation (11):

[0074] (11)

[0075] In equation (11), C In Indicates the number of time periods; h t N represents the estimated bandwidth for the t-th time period; t P represents the number of data points in the t-th time period; t,n This represents the solar power output value of the nth data point in the t-th time period; This represents the estimated probability density; M represents the total number of samples.

[0076] (2) To select the optimal dependency function (the Copula function is used in this embodiment) for fitting, the Kendall rank correlation coefficient, used for nonlinear problems, is introduced as a measure of ordinal correlation. It represents the difference between the concordance probability and the inconsistency probability of independent random vectors (X1,Y1) and (X2,Y2) with the same distribution as (X,Y). The Kendall rank correlation coefficient is defined as:

[0077] (12)

[0078] In equation (12), τ represents the Kendall rank correlation coefficient; Indicates an event The probability of occurrence; Indicates an event The probability of occurrence.

[0079] Specifically, the empirical Copula function of wind and solar power output at adjacent times is calculated, and then the Euclidean distance of the Kendall rank correlation coefficients of the five types of Copula functions and the empirical Copula function is calculated. The Copula function with the smallest Euclidean distance is considered to have the highest goodness of fit, and the joint probability distribution function of the two is calculated using this Copula function. The conditional probability density function of power output at time t under power output at time t-1 can be derived through the marginal distribution and the joint probability distribution. As shown in equations (13)-(14):

[0080] (13)

[0081] (14)

[0082] In equations (13)-(14), and express and The probability density function (PDF); express and The joint probability distribution function (JPDF); Let represent the conditional probability density function (CCDF) of the output at time t+1, given the output at time t.

[0083] (3) In order to overcome the defect that first-order Markov chains cannot capture long-term correlations, a higher-order time-varying Markov chain is introduced to realize the dynamic adjustment of state transition probabilities over time, so as to better capture the time-varying characteristics of new energy sources. At the same time, an exponentially decaying forgetting factor is introduced to assign different weights to historical states according to their time proximity, with recent states having a greater impact.

[0084] Specifically, the order of the Markov chain is first selected to determine the adjacent time period for which the wind and solar power output is to be studied. The order L of a higher-order Markov chain indicates that the value of a variable at a certain moment is influenced by the combined influence of the values ​​at the previous L moments; that is, the power output at time t is influenced by the power outputs at times t-1, t-2, ..., tL, a total of L moments. The linear correlation coefficient (using the Pearson correlation coefficient in this embodiment) describes the correlation between wind and solar power output at different moments. The Pearson correlation coefficient between the wind and solar power output sequences at times tl and t in the historical data is denoted as ρ. t-l,t As shown in equation (15):

[0085] (15)

[0086] In equation (15), P t,n and P t-l,n This represents the nth wind and solar power output data at time tl and time t; and This represents the average wind and solar power output at time t1 and time t2. The Pearson correlation coefficient ranges from -1 to 1. A positive value indicates a positive correlation between the two sequences, while a negative value indicates a negative correlation. The absolute value of the correlation coefficient indicates the strength of the correlation. In this embodiment, a correlation greater than 0.6 is considered a strong correlation between the two sequences, while a correlation less than 0.6 is considered a weak correlation. By calculating the Pearson correlation coefficient between the wind power output sequences at different times, the degree of correlation between each time point and the current time point is determined, and the maximum step size with the strongest correlation is selected as the order L of the higher-order Markov chain.

[0087] Specifically, the state transition process of a Markov chain can be represented by a state transition kernel K. Assuming the current time is t, the state transition kernel K for the previous k time steps is shown in equation (16):

[0088] (16)

[0089] In equation (16), Let λ represent a conditional probability distribution function, denoted as the transition kernel. λ represents the forgetting factor. By replacing the core structure of the state transition kernel with the Copula function constructed in (2), a Markov state transition chain characterizing the dynamic evolution of continuous wind and solar power output can be constructed. At this time, the state transition kernel is as shown in equation (17):

[0090] (17)

[0091] In equation (17), υ i,t F(P) represents i,t DG The value of the function, i.e., υ t =F(P t ). The output at time ti is p. t-i At time t, the output force is p. t The probability of.

[0092] Specifically, step B uses a dynamic topology adaptive multivariate hierarchical tree model (in this embodiment, the R-Vine Copula model) to construct a multidimensional variable hierarchical tree diagram model for multiple wind and solar clusters at each time point based on the historical data of wind and solar at each time point. Then, the constructed multidimensional variable hierarchical tree diagram model is sampled according to a sampling algorithm.

[0093] It should be noted that traditional single copula methods are insufficient to characterize the asymmetric tail correlations between wind and solar clusters. R-Vine flexibly combines different distribution families through hierarchical conditional copulas to accurately model spatial asymmetric correlations. Unlike traditional fixed tree structures, R-Vine dynamically selects each level of the tree structure using the Maximum Spanning Tree (MST) principle. Using the Kendall rank correlation coefficient as weight, it selects the tree structure with the largest edge weight sum from the complete graph, ensuring a high degree of consistency between the tree topology and the actual geographical correlation of wind and solar clusters, thus solving the problem of mismatch between fixed tree structures and actual topology in traditional models. Simultaneously, considering spatiotemporal coupling, it dynamically adjusts the tree structure for each time period based on the differences in spatial correlation. The dynamic construction steps of the hierarchical tree structure through the real-time iterative maximum spanning tree algorithm include:

[0094] (1) Data preprocessing: The original data is converted into a uniform distribution using kernel density estimation;

[0095] (2) Generate the first graph G1: Based on the real-time geographical location and power output fluctuation characteristics of the wind-solar cluster, generate a complete graph G1 with weighted edges. Assume G1 is a complete graph with p nodes, and each edge of G1 is e=(v i v j The above is a pair of variables u vi and u vj And the Kendall τ correlation coefficient τ between these variables. e ;

[0096] (3) Generate the first tree T1: Apply the maximum spanning tree algorithm to maximize the edge weights (correlation coefficients) of the entire graph and generate the optimal tree structure in real time. This tree is T1.

[0097] (4) Selection of two-dimensional Copula model and parameter estimation: Select the corresponding two-dimensional Copula distribution family for the marginal variables of each edge of G1, and perform parameter estimation;

[0098] (5) Generate the second subgraph G2 through T1, and traverse the nodes v that have the same node in tree T1. k The two sides (v) i v k ) and (v k v j The triple {v} formed by ) i v j ;v k}, let ξ i =(v i v k ), ξ j =(v k v j ), will (ξ i ξ j The marginal variable corresponding to this edge in G2 is calculated using the following H function:

[0099] (18)

[0100] And calculate the Kendallr correlation coefficient and weight on each edge of G2;

[0101] (6) Iterate, repeat steps three, four, and five to obtain the second tree T2 and graph G3, and continue iterating until the last tree T is obtained. p-1 .

[0102] The sampling algorithm uses a recursive algorithm to sample from the last tree T in the R-vine structure. p-1 Begin by traversing all edges in the structure. When the recursive function visits edge e=(v i vj When generating marginal samples w, the first step is to generate the marginal samples w. vi and w vj The recursive function is called again, first visiting the left node v of e. i =(a m a k Generate the corresponding marginal sample w. am and w ak Next, visit the right-hand node v. j =(a k a n Similarly, corresponding marginal samples w are generated. ak and w an The steps for generating marginal samples include:

[0103] (1) If variable w vi and w vj If it already exists, return the last call to the recursive function.

[0104] (2) If w vj If it does not exist, then: a) If edge e comes from the first-level tree T1, then w vj b) Extract directly from a uniform distribution; if edge e does not come from the first-level tree T1, consider v first. j =(a k a n The variables w corresponding to the two nodes of ) ak and w an Does it exist? If it exists, w vj Obtained directly through the H function. Otherwise, draw w from a uniform distribution. vj If w vi If it does not exist, then use H -1 Function to obtain, .

[0105] (3) If the current tree is the first tree T1, then w vi or w vj Copy to edge (v) i v j Shared node v i Or v j Go to the side.

[0106] Specifically, step B involves inverse sampling the state transition chain generated in the temporal correlation model from the sample points generated in the spatial correlation model. The specific steps include:

[0107] (1) Generate the Markov state transition chain for each wind-solar cluster to obtain the temporal correlation probability density (PDF) matrix of the cluster. ,in, Let be the CCDF at the nth scenic moment t+1.

[0108] (2) Establish the R-Vine model of the wind and light cluster at each time moment. For each time moment, randomly sample an N-dimensional vector. .

[0109] (3) Determine the output value of the nth wind-solar cluster at time t=1. Obtain the CCDF of each wind-solar cluster at time t+1. The spatial correlation sample value d n,t As sampled values ​​of the conditional probability distribution, substitute into The inverse function is used to obtain the sampled value of the nth wind-solar cluster at time t+1. ,Right now .

[0110] (4) Repeat step (3) for each wind-solar cluster to obtain the power output of each wind-solar cluster at time t+1.

[0111] (5) Let t = t + 1, and repeat steps (3) and (4) to obtain the power output matrix of each wind-solar cluster at all times. This refers to a time-series scenario that considers the spatiotemporal correlation of new energy sources.

[0112] In one embodiment, constructing a load time series optimization model driven by daily load characteristic indicators includes: characterizing the peak-valley steepness and flatness of the load power curve based on the daily peak-valley difference rate and daily load rate, and setting the absolute power scale according to the daily maximum load; the daily load rate is the ratio of the daily average load to the maximum load, and the daily peak-valley difference rate is the ratio of the daily peak-valley difference maximum value to the daily maximum load; with the minimum of the sum of squared errors between the generated daily load sequence and the daily load reference curve as the optimization objective, a load time series optimization model is established, and the constraints of the model are constructed by combining the daily load rate constraint, the daily peak-valley difference rate constraint, and the daily maximum load constraint; wherein, the daily load characteristic indicators include the daily load rate, the daily peak-valley difference rate, and the daily maximum load; the daily load rate constraint is achieved by ensuring that the average value of the generated load curve is consistent with the reference curve; the daily peak-valley difference rate constraint is achieved by ensuring that the load at each moment in the load curve is greater than the daily minimum load, and that the time when the daily minimum load of the generated load curve occurs is consistent with the reference curve; the daily maximum load constraint is achieved by ensuring that the load at each moment in the load curve is less than the daily maximum load, and that the time when the daily maximum load of the generated load curve occurs is consistent with the reference curve.

[0113] Specifically, step C requires selecting indicators that can characterize load features, and then reconstructing a load power curve that satisfies time-series characteristics based on these indicators. The daily peak-to-valley difference rate and daily load factor are selected to characterize the peak-to-valley steepness and flatness of the load power curve, respectively, thereby characterizing the shape of the load curve. Simultaneously, the absolute power scale is set using the daily maximum load to prevent subsequent optimization curves from deviating from actual capacity limits. These three indicators can accurately quantify the time-series fluctuation characteristics of the load. The specific definitions of the indicators are as follows:

[0114] The daily load factor is the ratio of the average daily load to the maximum daily load. It is used to characterize the load imbalance during the day. The calculation formula is shown in equation (19):

[0115] (19)

[0116] In equation (19), α represents the daily load factor; P t L This represents the load power at time t; This represents the maximum load value within time period T.

[0117] Specifically, the daily peak-valley difference rate refers to the ratio of the maximum daily peak-valley difference to the maximum daily load. The peak-valley difference is the difference between the maximum load and the minimum load, reflecting the peak-shaving capacity required by the power grid. The calculation formula is shown in equation (20):

[0118] (20)

[0119] In equation (20), β represents the daily peak-to-valley difference rate; This represents the minimum load value within time period T.

[0120] Specifically, the daily maximum load refers to the maximum load on a certain day, and the calculation formula is shown in equation (21):

[0121] ;(twenty one)

[0122] In equation (21), γ represents the daily maximum load.

[0123] Specifically, after obtaining the load characteristics, while satisfying the daily load characteristic index, it is also necessary to satisfy the time-series characteristics of the benchmark load. Therefore, it is necessary to use a time-series optimization model of the load to reconstruct the time-series power of the load. The optimization objective of the model is to minimize the sum of squared errors between the generated daily load sequence and the daily load benchmark curve. As shown in equation (22):

[0124] ;(twenty two)

[0125] In equation (22), P t *The daily load reference curve represents the power at time t. The curve with the smallest sum of Euclidean distances between the daily load characteristic index in historical data and the selected daily load characteristic index is taken as the daily load reference curve.

[0126] Specifically, the constraints of the optimization model include daily load factor constraints, daily peak-to-valley difference rate constraints, and daily maximum load constraints, as shown in equations (23)-(25):

[0127] ;(twenty three)

[0128] ;(twenty four)

[0129] (25)

[0130] In equations (23)-(25), equation (23) ensures that the average value of the generated load curve is consistent with the base curve; equation (24) ensures that the load at each moment in the load curve is greater than the daily minimum load. Furthermore, the time at which the daily minimum load of the generated load curve occurs is consistent with the base curve; equation (25) ensures that the load at each moment in the load curve is less than the daily maximum load. Furthermore, the time at which the daily maximum load of the generated load curve occurs is consistent with the base curve.

[0131] In one embodiment, constructing a charging load generation model based on electric vehicle type and user behavior characteristics includes: classifying electric vehicles into different types according to usage scenarios and establishing probability models for travel time, mileage, and charging start time for each type; determining the corresponding charging time distribution and daily mileage distribution based on the usage behavior characteristics of each type of electric vehicle; dynamically adjusting the charging power based on grid load level and electricity price signal, and introducing a response factor to enable the charging load to actively adapt to the grid state; dividing the charging process into a constant current stage and a constant voltage stage; maintaining a constant charging power during the constant current stage, and decreasing the charging power during the constant voltage stage as the SOC increases; setting the charging frequency according to the usage frequency of different types of electric vehicles, and establishing a relationship model between the remaining battery capacity and charging duration; and using a random sampling method to simulate each electric vehicle to obtain the starting charging time and charging duration to obtain the daily basic charging load of the electric vehicle.

[0132] Specifically, step D requires classifying electric vehicles into three categories: private cars, taxis, and buses. Probabilistic models for travel time, mileage, and charging start time are established for each category, and Monte Carlo sampling is used to simulate the charging load demand of electric vehicles.

[0133] The probability distribution of the initial charging period for private cars is shown in equation (26):

[0134] (26)

[0135] In equation (26), t p Indicates the initial charging time of a private car; μ t σ represents the expected value at the initial charging moment; t 2 This represents the variance at the initial charging time.

[0136] The probability distribution of daily mileage for private cars is shown in equation (27):

[0137] (27)

[0138] In equation (27), s represents the daily mileage of a private car; μ L σ represents the expected daily mileage. L 2 This represents the variance of daily mileage.

[0139] The charging power is dynamically adjusted based on the grid load level and electricity price signal, so that the charging load actively adapts to the grid condition. The charging power with the response factor is shown in equation (28):

[0140] (28)

[0141] In equation (28), P t base Indicates the reference charging power; P grid,t P represents the load at time t; grid,avg C represents the average load at time t; t C represents the electricity price at time t; high and C low These represent peak and off-peak electricity prices, respectively. Λ grid and Λ C Let represent the grid state response factor and the electricity price signal response factor, respectively. The charging process is divided into two stages: constant current (CC) and constant voltage (CV). In the constant current stage, the charging power remains constant until the battery SOC reaches the threshold. In the constant voltage stage, the charging power decreases as the SOC increases. The charging power in the constant voltage stage is shown in equation (29):

[0142] (29)

[0143] In equation (29), Indicates maximum charging power; SOC max Represents the maximum SOC; SOC cc This indicates the SOC threshold during the constant current phase; SOC t Let SOC be the value at time t.

[0144] The charging needs of private cars are related to their daily mileage, therefore their charging time can be expressed as:

[0145] (30)

[0146] In equation (30), ψ 100 This indicates the electricity consumption per 100 kilometers for a private car; P p Indicates the charging power of a private car; θ p Indicates charging efficiency; T p This indicates the charging time for a private car.

[0147] Taxis and buses have relatively long operating hours and mileages, and their electricity consumption behavior is relatively regular. Therefore, taxis are set to be charged twice a day, and public buses are set to be charged once a day. The initial SOC and the user's charging start time for each period follow a normal distribution. The charging time for these two types of electric vehicles can be derived from the relationship model between the remaining battery power and the charging time, as shown in equation (31):

[0148] (31)

[0149] In equation (31), SOC tb C represents the percentage of remaining battery power in a taxi or bus. tb θ represents the battery capacity of a taxi or bus. tb P represents the charging efficiency of three types of electric vehicles. tb This indicates the charging power of a taxi or bus.

[0150] The starting charging time and charging duration are obtained by sampling each electric vehicle using a random sampling method (the Monte Carlo method is used in this embodiment), thereby obtaining the daily basic charging load W for each electric vehicle. EV (t), as shown in equation (32):

[0151] (32)

[0152] In equation (32), W EV (t) represents the electric vehicle in t The charging load at any given moment.

[0153] In one embodiment, establishing a flexibility demand quantification model includes: calculating the upward and downward flexibility demands generated by load fluctuations based on load power demand and time scale; calculating the upward and downward flexibility demands generated by wind power and photovoltaic power based on wind and solar power output and time scale; and summing the upward and downward flexibility demands of load, photovoltaic power, and wind power at the same time to obtain the overall system flexibility demand. Establishing a flexibility supply quantification model includes: determining the upward and downward flexibility supply provided at each time based on the power constraints of flexible interconnection devices; determining the provided upward and downward flexibility supply based on the loads that can be transferred in, transferred out, and shelved in the demand response; determining the provided upward and downward flexibility supply based on the charging and discharging power, energy capacity, state of charge, and minimum and maximum state of charge of the cluster electric vehicles; and summing the upward and downward flexibility supplies of demand response, flexible interconnection devices, and electric vehicles at the same time to obtain the overall system flexibility supply.

[0154] Specifically, in step E, quantitative models of flexibility demand and flexibility supply need to be constructed based on the source and load characteristics of the AC / DC power distribution system.

[0155] E1. Flexibility Demand Quantification Model: Flexibility demand refers to the power regulation capability required by a system to maintain real-time supply-demand balance within a certain time scale, in response to uncertainties and fluctuations in renewable energy output and load uncertainty. Specifically, it includes:

[0156] (1) The flexibility requirements arising from load fluctuations are quantified as shown in equation (33):

[0157] (33)

[0158] In equation (33), Let Δt represent the power demand of the i-th load at time t, where Δt represents the time scale. and Let represent the upward and downward flexibility requirements of the i-th load at time t, respectively.

[0159] (2) The flexibility requirements generated by wind power and photovoltaic power are quantified as shown in equation (34):

[0160] (34)

[0161] In equation (34), This represents the output of the i-th scenic spot at time t; and Let represent the upward and downward flexibility requirements of the i-th landscape at time t, respectively.

[0162] (3) The overall system flexibility requirements are quantified as shown in equation (35):

[0163] (35)

[0164] In equation (35), F t de,u and F t de,d These represent the system's upward and downward flexibility requirements at time t, respectively. and These represent the upward and downward flexibility requirements of the load at time t, respectively. and These represent the upward and downward flexibility requirements of photovoltaics at time t, respectively. and These represent the upward and downward flexibility requirements of wind power at time t, respectively.

[0165] E2. Quantitative Model of Flexibility Supply: Flexibility supply refers to the capabilities or resources that the system can provide to meet the aforementioned flexibility requirements. Specifically, this includes:

[0166] (1) The flexibility provided by the Flexible Interconnect Device (FID) is quantified as shown in Equation (36):

[0167] (36)

[0168] In equation (36), This represents the transmission power of the i-th FID at time t; and Let represent the upward and downward flexibility supply of the i-th FID at time t, respectively.

[0169] (2) The supply of flexibility provided by demand response (DR) is quantified as shown in equation (37):

[0170] (37)

[0171] In equation (37), This represents the load that can be transferred to the i-th load at time t; This represents the load that can be transferred out at time t for the i-th load; This represents the load that can be cut off at time t for the i-th load. and Let represent the upward and downward flexibility supply of the i-th DR at time t, respectively.

[0172] (3) The flexibility provided by the cluster of electric vehicles is quantified as shown in equation (38):

[0173] (38)

[0174] In equation (38), where, and Let represent the charging and discharging power of the i-th electric vehicle at time t; and E represents the maximum charging and discharging power of the electric vehicle; i EV This represents the electrical energy capacity of the i-th electric vehicle; This represents the State of Charge (SOC) of the i-th electric vehicle at time t. and Let S and SOC represent the minimum and maximum SOC of the i-th electric vehicle, respectively.

[0175] (4) The overall system flexibility supply quantification is shown in equation (39):

[0176] (39)

[0177] In equation (39), F t su,u and F t su,d These represent the upward and downward flexibility supply of the system at time t, respectively; and These represent the upward and downward flexibility of supply in the demand response at time t, respectively. and These represent the upward and downward flexibility supply of FID at time t, respectively; and These represent the upward and downward flexibility supply of the electric vehicle at time t, respectively.

[0178] In one embodiment, clustering the generated original scenes using a clustering algorithm combining standard deviation weighted distance and fuzzy weighted K-nearest neighbors includes: treating each original scene as a data point; the feature value of each data point is the flexibility supply and demand quantification result at each time step; calculating the distance between each data point using standard deviation as a weighting factor in the distance metric; calculating the local density and center offset distance of each data point, and taking the point with the largest product of local density and center offset distance as the cluster center; calculating the maximum neighborhood radius of each data point to its K-nearest neighbors, and classifying data points with a maximum neighborhood radius greater than the average as outliers, and classifying other points as non-outliers; using K-nearest neighbors and breadth-first search methods for cluster assignment of non-outliers; and using the fuzzy weighted K-nearest neighbors method for outliers, calculating membership degrees and constructing a membership degree matrix for cluster assignment to obtain typical AC / DC power distribution system scenarios.

[0179] Specifically, step F requires clustering the original scenarios constructed in step AE, using a density peak clustering method based on standard deviation weighted distance and fuzzy weighted K-nearest neighbors to obtain typical AC / DC power distribution system scenarios. Each original scenario is treated as a data point x, and the flexibility supply and demand quantification result at each time step is used as the feature value of data point x. The entire clustering process is mainly divided into three parts, with the specific steps as follows:

[0180] F1. All non-density peak points are divided into outliers and non-outliers, which mainly include:

[0181] (1) Calculate the standard deviation s of each feature dimension according to equation (40). k And calculate the standard deviation weight w for each feature dimension according to equation (41). k ;

[0182] (40)

[0183] (41)

[0184] In equations (40)-(41), x lk This represents the k-th feature value of the l-th data point; represents the average of the k-th feature value across all data points; n represents the total number of data points; m represents the total dimension of the feature.

[0185] (2) Using the standard deviation as the weighting factor in the distance metric, the distance sd between each data point is calculated using formula (42). ij ;

[0186] (42)

[0187] In equation (42), x ik and x jk This represents the k-th feature value of the i-th and j-th data points.

[0188] (3) Calculate the local density ρ of each data point using equations (43) and (44). i and center offset distance δ i Based on the number of clusters, the points with the largest product of local density and center offset distance are selected as cluster centers;

[0189] (43)

[0190] (44)

[0191] In equations (43)-(44), KNN i Let K represent the set of K nearest neighbors of point i.

[0192] (4) Calculate the maximum neighborhood radius r from each data point to its K nearest neighbor using equation (45). i K :

[0193] (45)

[0194] (5) Classify all non-cluster centers. Compare the maximum neighborhood radius of each data point with the average value, classify data points with a maximum neighborhood radius greater than the average value as outliers, and the remaining points as non-outliers. The specific formulas are shown in (46) and (47):

[0195] (46)

[0196] (47)

[0197] In equations (46)-(47), τ represents the average of the maximum neighborhood radii of all data points; i Let i represent the i-th outlier.

[0198] F2. For non-outlier points, K-nearest neighbor and breadth-first search methods are used. This mainly includes:

[0199] (1) Initialization: Mark all density peaks (cluster centers) as unvisited.

[0200] (2) Main loop: When there are unvisited cluster centers: select an unvisited cluster center c i Set as the center of the new cluster and mark it as visited. Initialize queue Q. (The last part, "c", appears to be a typo and can be left as is.) i The K nearest neighbors are assigned to the cluster and added to queue Q.

[0201] (3) Queue processing loop: When queue Q is not empty, take out the first element q. Iterate through each point p in the K nearest neighbors of q. If p is not assigned and is not an outlier, calculate the mean θ of the K nearest neighbor distances of p. If the distance from p to q is less than θ, assign p to the same cluster as q and add p to queue Q.

[0202] 4) Termination condition: All cluster centers have been visited and queue Q is empty.

[0203] F3. For outliers, a fuzzy weighted K-nearest neighbor method is used, which mainly includes:

[0204] (1) Calculate membership degree: First, for each unassigned point i, calculate its membership degree to each cluster c according to the formula (48).

[0205] (48)

[0206] In equation (28), Let represent the normalized weights representing the probability that data point i belongs to cluster c. This represents the similarity between data points i and j.

[0207] (2) Construct a membership matrix S: where each row represents the membership degree of a point to be assigned to each cluster, and each column represents the membership degree of all points to be assigned to each cluster.

[0208] (49)

[0209] (3) Allocation main loop: Find the maximum value in the membership matrix S and assign the corresponding point to the corresponding cluster. Then, set the membership degree of the point to all clusters to 0, indicating that the point has been assigned. Subsequently, for each point q in the K nearest neighbors of the point, update its membership degree to each cluster, as shown in Equation (50).

[0210] (50)

[0211] (4) Update evaluation: Repeat the above assignment process until the maximum value in the membership matrix is ​​0, that is, all points have been assigned or the membership is 0.

[0212] To facilitate understanding of the above technical solution of the present invention, the following is a detailed explanation using a typical scenario of AC / DC power distribution system in a city and rural area as an example:

[0213] First, according to S1 of the present invention, a digital representation model is constructed based on the typical network topology of AC / DC power distribution systems. In this embodiment, based on the IEEE 33 system, four typical network topologies are designed, such as... Figure 3 As shown, based on the node power supply method and the type and location of the node connected to the DG, the node type variable and the node DG state variable are clearly defined, thereby determining the network branch type and establishing a complete digital representation model.

[0214] Secondly, following the S2 technical solution, a spatiotemporal correlation model for wind and solar power output is constructed based on the spatiotemporal evolution characteristics of new energy sources. For the historical power output data of each wind and solar node in each time period, kernel probability density estimation is used to obtain the probability density function of wind and solar power output for each time period, and the optimal dependency function is selected for fitting using the ordinal correlation metric. Simultaneously, using the ordinal correlation metric as weights, a multivariate hierarchical tree is constructed through a real-time iterative maximum spanning tree algorithm, achieving accurate modeling of the spatial correlation of wind and solar power output. Regarding the integration of the spatiotemporal correlation model, by using spatial correlation sample values ​​as sample values ​​for the conditional probability distribution, inverse sampling techniques are employed to successfully construct a time-series scenario for wind and solar power output that considers spatiotemporal correlation.

[0215] Next, following S3, a load time-series optimization model and a charging load generation model are constructed. Based on characteristic indicators such as daily load factor, daily peak-valley difference rate, and daily maximum load, a load time-series optimization model is established, and a charging load generation model is constructed according to the differences in energy consumption characteristics among different user groups of electric vehicles. In this embodiment, electric vehicles are divided into three types according to usage scenarios: private cars, taxis, and buses. Probabilistic models for travel time, mileage, and charging start time are established for each type. The charging process is divided into two stages: constant current and constant voltage, thus achieving accurate generation of electric vehicle charging load.

[0216] Based on S4, and considering the source-load interaction characteristics of AC / DC distribution networks, a flexibility demand quantification model and a flexibility supply quantification model were established respectively. The flexibility demand quantification model considers the upward and downward flexibility demands generated by load power demand, wind power, and photovoltaic output at different time scales; the flexibility supply quantification model comprehensively considers the upward and downward flexibility supply provided by flexible interconnection devices, demand response, and clustered electric vehicles.

[0217] Finally, following S5, the generated original scenes are clustered using a clustering algorithm combining standard deviation weighted distance and fuzzy weighted K nearest neighbors. Each original scene is treated as a data point, and the flexibility supply and demand quantification result at each time step is used as a feature value. The local density and center offset distance are calculated, and outliers and non-outliers are identified based on the maximum neighborhood radius of the K nearest neighbors. Different clustering assignment methods are applied to obtain typical AC / DC power distribution system scenes.

[0218] Based on the characteristics of four typical network topologies and practical application scenarios, this embodiment sets up four scenario sets. Using the IEEE 33 system as a foundation, four typical network topologies are designed, such as... Figure 3 As shown, the parameter settings for the scene set are shown in Table 1-5.

[0219] Table 1 Interconnection Modes of Scenario Sets and Their Applicable Scenarios

[0220] Table 2 Load parameter settings for the scenario set

[0221] Table 3 Parameter settings for landscape and electric vehicles in the scene set

[0222] Table 4. User Behavioral Characteristics of Private Cars

[0223] Table 5. User Behavior Characteristics of Taxi and Bus Drivers

[0224] Following the aforementioned methods and parameter settings, 1000 original scenarios are generated for each scenario set, and the flexibility requirements and supply for each scenario are calculated separately. In the flexibility requirement calculation, the upward and downward flexibility requirements arising from load fluctuations are calculated based on load power demand and time scale. Furthermore, the upward and downward flexibility requirements arising from wind and solar power are calculated based on wind and solar power output and time scale. Finally, the upward and downward flexibility requirements of load, solar, and wind power at the same moment are summed to obtain the overall system flexibility requirement. In the flexibility supply calculation, the upward and downward flexibility supply provided at each moment is determined based on the power limitations of flexible interconnection devices, the transferable load characteristics of demand response, and the charging and discharging characteristics of clustered electric vehicles.

[0225] like Figure 4 and Figure 5 As shown, Figure 4 and Figure 5 In the graph, blue represents downward flexibility demand and supply, while red represents upward flexibility demand and supply. Darker colors indicate smaller flexibility demand and supply, and vice versa. The graph shows that the distribution of flexibility demand in scenarios 1 and 2 across different time periods approximates the load curve, while scenarios 3 and 4 approximate the photovoltaic output curve. The quantitative definition of flexibility demand shows that it is primarily determined by the net load curve. Scenario 1 and 2 represent urban scenarios, so their net load curves are mainly affected by load fluctuations. Scenario 3 and 4 represent rural scenarios, with larger wind and solar scales and much higher wind and solar penetration rates than urban scenarios, so their net load curves are mainly affected by wind and solar output fluctuations. Flexibility supply is higher in urban scenarios than in rural scenarios because demand response is involved, and the scale of electric vehicles is much larger in urban scenarios. Within urban scenarios, the flexibility supply in scenario 2 is much greater than the other three scenarios because scenario 2 represents residential areas and data centers, areas with a large amount of transferable load, and the largest scale of electric vehicles among the three scenarios. The grid structure in Scenario 2 is an expanded AC-DC interconnected type, which allows more power to flow between the AC and DC regions, thus providing greater flexibility.

[0226] Finally, a clustering algorithm combining standard deviation weighted distance and fuzzy weighted K-nearest neighbor was used to cluster the generated original scenes. After cluster analysis, six typical daily scenes were finally obtained, such as... Figure 6 As shown in the figure, the time-series curves of the source and load elements for each scenario are specifically displayed. These typical scenarios can accurately reflect the source and load characteristics and flexible supply and demand relationship of the AC / DC power distribution system under different operating conditions, thereby providing effective support for system planning and scheduling.

[0227] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for generating typical scenarios of AC / DC power distribution systems considering the spatiotemporal correlation of source and load, characterized in that, include: S1. Construct a digital representation model based on the typical network topology of AC / DC power distribution systems; S2. Based on the spatiotemporal evolution characteristics of new energy, a time correlation model of wind and solar power output is constructed based on the dependency function-state transition chain, a spatial correlation model of wind and solar power output is established based on the multivariate hierarchical tree, and the spatial correlation model is combined with the time correlation model. S3. Based on the random fluctuation of daily load, construct a load time series optimization model driven by daily load characteristic indicators, and based on the differences in energy consumption characteristics of different electric vehicle user groups, construct a charging load generation model based on electric vehicle type and user behavior characteristics. S4. Based on the source-load interaction characteristics of AC / DC distribution networks, establish a flexibility demand quantification model and a flexibility supply quantification model respectively. S5. Construct the original scene based on the established model, and cluster the generated original scene by combining the standard deviation weighted distance and fuzzy weighted K nearest neighbor clustering algorithm to obtain the typical scene of AC / DC power distribution system.

2. The method for generating typical scenarios of AC / DC power distribution systems considering the spatiotemporal correlation of source and load as described in claim 1, characterized in that, The construction of the digital representation model includes: Node type variables are established based on the node power supply method, and node DG state variables are constructed based on the type and location of the node connected to the DG; the node type variables include AC / DC node state variables, and the node DG state variables include photovoltaic node state variables and wind power node state variables; The network branch type is determined based on the types of nodes at both ends of the branch, and a branch type representation model is established. Specifically, if both nodes at both ends of the branch are AC nodes, the branch is an AC branch; if both nodes at both ends of the branch are DC nodes, the branch is a DC branch; if both nodes at both ends of the branch are AC and DC nodes respectively, the branch is an AC / DC branch and is interconnected through VSC. Constraints are constructed based on node type variables and branch type variables to characterize the wind and solar power scale and AC / DC interconnection mode of the power distribution system, including DG quantity constraints, DG capacity constraints, and branch type quantity constraints.

3. The method for generating typical scenarios of AC / DC power distribution systems considering the spatiotemporal correlation of source and load as described in claim 1, characterized in that, The time-dependent model for wind and solar power output constructed based on the dependency function-state transition chain includes: Based on the historical power output data of each wind and solar node in each time period, the probability density function of wind and solar power output in each time period is obtained by kernel probability density estimation. Using the ordinal correlation metric, a multivariate dependency function is selected for data fitting; the empirical dependency function of wind and solar power output at adjacent times is calculated; by calculating the Euclidean distance between the ordinal correlation metric of different types of dependency functions and the empirical dependency function, the dependency function with the smallest Euclidean distance is selected as the optimal fitting function; and the conditional probability density function of power output at the current time under the power output condition of the previous time is derived through the marginal probability distribution and joint probability distribution. The degree of correlation between each moment and the current moment is determined by calculating the linear correlation coefficient between the wind and solar power output sequences at different times, so as to select the order of the higher-order time-varying state transition chain. The historical states are assigned different weights according to their time proximity by using the forgetting factor, and the state transition core is replaced by the dependency function to construct a higher-order time-varying state transition chain for continuous wind and solar power output.

4. The method for generating typical scenarios of AC / DC power distribution systems considering the spatiotemporal correlation of source and load as described in claim 1, characterized in that, The spatial correlation model for wind and solar power output based on a multi-layered hierarchical tree includes: Using the ordinal correlation metric as the weight, select the tree structure with the largest edge weight in the complete graph; The construction of a multi-level hierarchical tree is carried out through a real-time iterative maximum spanning tree algorithm. Specifically, it includes: data preprocessing, generating the first complete graph with weighted edges and the first tree; selecting the corresponding dependency function distribution family for the marginal variables of each edge in the first complete graph and performing parameter estimation; generating the next complete graph from the first tree and iteratively obtaining all tree structures. The algorithm samples all edges in the multi-layered tree by traversing the structure, starting from the last tree in the multi-layered tree, using a recursive algorithm.

5. The method for generating typical scenarios of AC / DC power distribution systems considering the spatiotemporal correlation of source and load as described in claim 1, characterized in that, The combination of spatial correlation model and temporal correlation model includes: Generate high-order time-varying state transition chains for each wind and solar cluster to obtain the temporal correlation probability density matrix of the cluster. A multi-dimensional hierarchical tree model of the landscape cluster at each moment is established, and multi-dimensional vector random sampling is performed at each moment; By determining the power output value of each wind and solar cluster at the initial moment, the conditional probability density function of each wind and solar cluster at the next moment is obtained. The spatial correlation sample value is used as the sample value of the conditional probability distribution. By substituting the inverse function of the conditional probability density function into the sample value of the wind and solar cluster output at the next moment, the power output sample value of the wind and solar cluster at the next moment is obtained. The reverse sampling process is repeated for all wind and solar clusters and time recursion is performed to obtain the output matrix of each wind and solar cluster at all times, so as to form a time-series scene that considers the spatiotemporal correlation.

6. The method for generating typical scenarios of AC / DC power distribution systems considering the spatiotemporal correlation of source and load as described in claim 1, characterized in that, The construction of a load time-series optimization model driven by daily load characteristic indicators includes: Based on the daily peak-valley difference rate and the daily load rate, the peak-valley steepness and flatness of the load power curve are respectively characterized, and the absolute power scale is set according to the daily maximum load; the daily load rate is the ratio of the daily average load to the maximum load, and the daily peak-valley difference rate is the ratio of the daily peak-valley difference maximum value to the daily maximum load. With the goal of minimizing the sum of squared errors between the generated daily load sequence and the daily load baseline curve, a load time series optimization model is established, and the constraints of the model are constructed by combining the daily load rate constraint, the daily peak-valley difference rate constraint, and the daily maximum load constraint. The daily load characteristic indicators include daily load factor, daily peak-to-valley difference rate, and daily maximum load. The daily load factor constraint is achieved by ensuring that the average value of the generated load curve remains consistent with the baseline curve; The daily peak-valley difference rate constraint is achieved by ensuring that the load at each moment in the load curve is greater than the daily minimum load, and that the time when the daily minimum load occurs in the generated load curve is consistent with the reference curve. The daily maximum load constraint is achieved by ensuring that the load at each moment in the load curve is less than the daily maximum load, and that the moment when the daily maximum load occurs in the generated load curve is consistent with the reference curve.

7. The method for generating typical scenarios of AC / DC power distribution systems considering the spatiotemporal correlation of source and load as described in claim 6, characterized in that, The construction of the charging load generation model based on electric vehicle type and user behavior characteristics includes: Electric vehicles are classified into different types according to usage scenarios, and probabilistic models for travel time, driving mileage, and charging start time are established for each type. Based on the usage behavior characteristics of different types of electric vehicles, the corresponding charging time period distribution and daily driving mileage distribution are determined; The charging power is dynamically adjusted based on the grid load level and electricity price signal, and a response factor is introduced to enable the charging load to actively adapt to the grid condition. The charging process is divided into a constant current stage and a constant voltage stage; the charging power remains constant during the constant current stage, while the charging power decreases as the state of charge (SOC) increases during the constant voltage stage. The charging frequency is set according to the usage frequency of different types of electric vehicles, and a model is established to show the relationship between the remaining battery power and the charging time. A random sampling method was used to simulate each electric vehicle to obtain the starting charging time and charging duration, so as to obtain the daily basic charging load of the electric vehicle.

8. The method for generating typical scenarios of AC / DC power distribution systems considering the spatiotemporal correlation of source and load as described in claim 1, characterized in that, The establishment of a flexibility requirement quantification model includes: Calculate the upward and downward flexibility requirements caused by load fluctuations based on load power demand and time scale; Calculate the upside and downside flexibility requirements of wind and solar power based on wind and solar output and time scale; The overall system flexibility requirement is obtained by summing the upward and downward flexibility requirements of load, photovoltaic, and wind power at the same moment.

9. A method for generating typical scenarios of AC / DC power distribution systems considering the spatiotemporal correlation of source and load, as described in claim 8, is characterized in that... The establishment of a flexible supply quantification model includes: Based on the power limitations of flexible interconnected devices, determine the upward and downward flexibility supply provided at each time point; The supply of upward and downward flexibility is determined based on the load that can be transferred in, the load that can be transferred out, and the load that can be cut off in the demand response. The supply of upward and downward flexibility is determined based on the charging and discharging power, energy capacity, state of charge, and minimum and maximum state of charge of the cluster of electric vehicles. The overall system flexibility supply is obtained by summing the upward and downward flexibility supply of demand response, flexible connected devices, and electric vehicles at the same moment.

10. The method for generating typical scenarios of AC / DC power distribution systems considering the spatiotemporal correlation of source and load as described in claim 1, characterized in that, The clustering of the generated original scene by combining the standard deviation weighted distance and fuzzy weighted K nearest neighbor clustering algorithms includes: Each original scene is treated as a data point; the feature value of each data point is the result of quantifying the supply and demand of flexibility at each moment. The distance between each data point is calculated using the standard deviation as a weighting factor in the distance metric. Calculate the local density and center offset distance for each data point, and take the point with the largest product of local density and center offset distance as the cluster center; Calculate the maximum neighborhood radius of each data point to its K nearest neighbors, and classify data points with a maximum neighborhood radius greater than the average as outliers, and other points as non-outliers; For non-outliers, K-nearest neighbors and breadth-first search methods are used for clustering and assignment. For outliers, a fuzzy weighted K-nearest neighbor method is used to perform clustering and assignment by calculating membership degree and constructing membership degree matrix to obtain typical scenarios of AC / DC power distribution system.

Images