A distribution network scheduling method considering wind and light load correlation and risk characteristics

Abstract

The application discloses a distribution network scheduling method considering wind-solar load correlation and risk characteristics, and belongs to the technical field of power generation, power transformation or power distribution. First, based on wind-solar load mass daily operation history data, independent edge distribution models of each wind power, photovoltaic and load node are fitted through a non-parametric kernel density estimation method; then, wind-solar load multi-scale rank entropy correlation coefficients are calculated, core variables are selected according to the correlation between variables, a C vine Copula structure is constructed, a Pair-Copula function is selected, a wind-solar load high-dimensional joint distribution function is constructed, and the complex correlation between multi-element wind-solar loads is accurately described; finally, the joint distribution function is sampled, a hierarchical clustering method of tail risk characteristics is adopted for reduction, a typical scenario set conforming to actual operation characteristics is generated, and distribution network flexible resources are coordinated and scheduled under the typical scenario set, and the method is suitable for distribution network optimization scheduling, safety checking and risk assessment.

Description

Technical Field

[0001] This invention relates to distribution network dispatching technology, and discloses a distribution network dispatching method that takes into account the correlation and risk characteristics of wind, solar and load, belonging to the technical field of power generation, substation or distribution. Background Technology

[0002] The installed capacity and penetration rate of distributed renewable energy sources such as wind power and photovoltaics in my country's distribution networks continue to climb, and traditional distribution networks are developing towards distributed power generation. The inherent strong randomness and volatility of distributed renewable energy sources, coupled with the proactive response characteristics of load-side electricity consumption behavior, exacerbate the difficulty of power flow regulation, voltage management, and safe and stable operation of distribution networks, posing severe challenges to the optimal scheduling and risk control of distribution networks. Accurately characterizing the coupling and correlation characteristics between wind power, photovoltaics, and load variables, and generating typical operating scenarios that closely reflect the actual operating patterns of the power grid, is a core prerequisite for ensuring the accuracy and effectiveness of distribution network optimization and control models.

[0003] Currently, scholars both domestically and internationally have conducted extensive research on source-load correlation modeling and scenario generation. Traditional linear correlation analysis methods cannot meet the need to accurately characterize the nonlinear coupling correlation characteristics between multiple variables of wind, solar, and load. The Pearson correlation coefficient only quantifies the degree of linear correlation between variables and cannot capture the nonlinear, asymmetric, and tail-dependent characteristics that are prevalent between wind, solar, and load. Although rank correlation coefficients such as Kendall and Spearman can describe the nonlinear correlation between variables, they are difficult to achieve a global characterization of the spatiotemporal coupling relationship of high-dimensional variables and cannot adapt to the scenario requirements of multi-source collaborative optimization of distribution networks.

[0004] Existing technologies mostly focus on two-dimensional correlation modeling of wind and solar power, generally ignoring the high-dimensional coupling characteristics between wind power, photovoltaic power and load. This can easily lead to significant model deviations between the generated scenario and the actual operating characteristics of the distribution network. As a result, they cannot provide reliable data support for day-ahead optimization scheduling and intraday rolling control of the distribution network, and are difficult to adapt to the multi-timescale optimization and control needs of flexible interconnected distribution networks with a high proportion of new energy access.

[0005] The binary Copula function, with its excellent ability to characterize nonlinear correlations, is widely used in two-dimensional correlation modeling of wind and solar power. However, this type of method is difficult to extend to high-dimensional joint modeling scenarios involving wind power, photovoltaics, and loads, and cannot fully characterize the complex coupling relationships between multiple sources and loads. Therefore, there is an urgent need for a scenario generation method that can accurately characterize the actual output characteristics of wind and solar loads and their multi-dimensional coupling correlations to meet the needs of distribution network optimization and control. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the prior art and provide a distribution network scheduling method that takes into account the correlation and risk characteristics of wind, solar and load. By accurately capturing the high-dimensional coupling characteristics between multiple variables of wind, solar and load, a scenario that fits the actual operation law of the distribution network is generated. This achieves the invention objective of providing reliable scenario data support for the optimized control of the distribution network and thus realizing the collaborative scheduling of the distribution network. It solves the technical problem that the model bias caused by traditional modeling prevents the distribution network scheduling from obtaining the optimal flexible resource allocation based on the actual operation scenario.

[0007] To achieve the above-mentioned objectives, the present invention employs the following technical solution:

[0008] A distribution network scheduling method that takes into account the correlation between wind and solar loads and risk characteristics includes:

[0009] S1. Collect historical time-series data of wind power nodes, photovoltaic nodes and load nodes in the target distribution network, generate random variable samples of wind, solar and load for each node, and construct a high-dimensional spatiotemporal sample set representing the random variables of wind, solar and load in the target distribution network.

[0010] S2, perform nonparametric kernel density estimation on the high-dimensional spatiotemporal sample set to obtain the probability density function and marginal distribution function of the random variable samples of wind and solar load at each node, and convert the running history time series data of the random variable samples of wind and solar load at each node into pseudo-observation samples of random variable samples of wind and solar load at each node that follow a uniform distribution.

[0011] S3. Considering the synchronous rank correlation component, time-delay rank correlation component and slope change correlation component among pseudo-observation samples of wind-solar-load random variables of each node, calculate the multi-scale rank entropy correlation coefficient between pseudo-observation samples of wind-solar-load random variables of any two nodes, and construct the C-vine Copula structure.

[0012] S4. For each edge of each tree in the C-vine Copula structure, select the Pair-Copula function for the random variables of wind, light and load on the nodes connected to it. The high-dimensional joint distribution function of wind, light and load is obtained by constructing the C-vine Copula structure with the selected Pair-Copula function.

[0013] S5, sample the high-dimensional joint distribution function of wind, solar and load to generate an initial joint operation scenario set;

[0014] S6. Based on the load power, wind power output, and photovoltaic power output of the initial joint operation scenario, calculate the net load risk sequence and tail risk index for each initial joint operation scenario.

[0015] S7. Based on the tail risk indicators, the initial joint operation scenario set is divided into a regular scenario layer and a tail risk scenario layer. The number of regular scenarios and tail risk scenarios to be retained is determined by combining the number of typical operation scenarios to be retained.

[0016] S8, cluster the retained regular scenarios and retained tail risk scenarios respectively, and select actual scenario samples in each cluster as representative scenarios, and construct a typical scenario set based on the representative scenarios;

[0017] S9 uses the constructed typical scenario set as input data for the time-series simulation of the target distribution network to obtain the flexible resource scheduling results of the target distribution network.

[0018] As a further optimization scheme for the distribution network scheduling method that considers the correlation and risk characteristics of wind, solar, and load, S2 obtains the probability density function and marginal distribution function of the wind, solar, and load random variable samples of each node by performing nonparametric kernel density estimation on the high-dimensional spatiotemporal sample set. Specifically, it constructs the first... Sample of random variables of wind and solar load at each node probability density function , for the Sample of random variables of wind and solar load at each node probability density function Integral transform to obtain the first Sample of random variables of wind and solar load at each node Marginal distribution function ,in, , For the first Sample of random variables of wind and solar load at each node The observed value at time, This represents the number of observations included in the historical time-series data of the nodal wind-solar-load random variable sample. This represents the kernel function bandwidth.

[0019] As a further optimization scheme for the distribution network scheduling method that takes into account the correlation and risk characteristics of wind, solar and load, in S2, the historical time series data of the wind, solar and load random variable samples of each node are converted into pseudo-observation samples of wind, solar and load random variables of each node that follow a uniform distribution. Specifically, the historical time series data of the wind, solar and load random variables of each node are converted into pseudo-observation samples of wind, solar and load random variables of each node that follow a uniform distribution. Sample of random variables of wind and solar load at each node Observations at time Substitute into the marginal distribution function Find the first element that follows a uniform distribution in [0,1]. Pseudo-observation samples of random variables of wind, solar and load at each node.

[0020] As a further optimization scheme of the distribution network scheduling method that takes into account the correlation and risk characteristics of wind and solar load, S3 considers the synchronous rank correlation component, time-delay rank correlation component, and ramp change correlation component among the pseudo-observation samples of wind and solar load random variables of each node, and calculates the multi-scale rank entropy correlation coefficient between any two pseudo-observation samples of wind and solar load random variables, specifically: ,in, For the first The pseudo-observation sample of the random variable wind-solar-load at node n and the first node The multi-scale rank entropy correlation coefficient among pseudo-observed samples of the random variables of wind and solar load at each node. For the first The pseudo-observation sample of the random variable wind-solar-load at node n and the first node The synchronous rank correlation component between pseudo-observed samples of wind-solar-load random variables at each node is used to describe the rank correlation strength between the two node wind-solar-load random variables at the same time. For the first The pseudo-observation sample of the random variable wind-solar-load at node n and the first node The time-delay rank correlation component between pseudo-observed samples of wind and solar load random variables at each node is used to describe the correlation strength between different node wind and solar load random variables at different time lags. For the first The pseudo-observation sample of the random variable wind-solar-load at node n and the first node The slope variation correlation component between pseudo-observation samples of the wind-solar-load random variables at each node is used to describe the consistency of the power variation trend of the wind-solar-load random variables at two nodes in adjacent time periods. , , These are the weighting coefficients for the synchronous rank correlation component, the time delay rank correlation component, and the ramp change correlation component, respectively. , , , , This is Kendall's rank correlation function. For the first The random variables of wind and solar load at each node are in Pseudo-observation samples at time points For the first The random variables of wind and solar load at each node are in Pseudo-observation samples at time points The maximum time delay order, This is the time delay order index. For time delay weights, , For the first The random variables of wind and solar load at each node are in Pseudo-observation samples at time points , , For the first The random variables of wind and solar load at each node are in Pseudo-observation samples at time points For the first The random variables of wind and solar load at each node are in Pseudo-observation samples at specific moments.

[0021] As a further optimization scheme for the distribution network scheduling method that takes into account the correlation and risk characteristics of wind, solar and load, the specific method for constructing the C-Vine Copula structure in S3 is as follows: when the sum of the multi-scale rank entropy correlation coefficients between the pseudo-observation samples of the wind, solar and load random variables of the current node and the pseudo-observation samples of the wind, solar and load random variables of other nodes reaches the maximum value, the wind, solar and load random variable of the current node corresponding to the pseudo-observation sample of the wind, solar and load random variables of the current node is selected as the core variable of the C-Vine Copula structure. The multi-scale rank entropy correlation coefficients between the pseudo-observation samples of the wind, solar and load random variables of the current node and the pseudo-observation samples of the wind, solar and load random variables of other nodes are used as the weights of the edges in the C-Vine structure. The maximum spanning tree method is used to optimize the hierarchical tree topology of the vine.

[0022] As a further optimization scheme for the distribution network scheduling method that takes into account the correlation and risk characteristics of wind, solar and load, in S4, the Pair-Copula function is selected for the random variable pairs of wind, solar and load connected to each edge of each tree in the C-vine Copula structure. Specifically, the parameters of the Pair-Copula candidate functions are estimated by maximum likelihood, and the Pair-Copula function with the best fitting effect is selected by Bayesian information criterion.

[0023] As a further optimization scheme for the distribution network dispatching method that takes into account the correlation and risk characteristics of wind and solar loads, S6 calculates the net load risk sequence and tail risk index for each initial joint operation scenario based on the load power, wind power output, and solar power output of the initial joint operation scenario. The specific method is as follows: , ,in, For the first Initial joint operation scenario Net load risk at any given time For the first Initial joint operation scenario Constant load power, For the first Initial joint operation scenario Time of the first Each wind power node outputs power. For the first Initial joint operation scenario The number of wind power nodes participating in power generation at any given time. For the first Initial joint operation scenario Time of the first Each photovoltaic node contributes power. For the first Initial joint operation scenario The number of photovoltaic nodes participating in power generation at any given time. For the first Tail risk indicators for an initial joint operation scenario , For the first Initial joint operation scenario Net load risk at any given time Net load risk quantiles, , These are the weighting coefficients.

[0024] As a further optimization scheme for distribution network scheduling methods that take into account the correlation and risk characteristics of wind, solar, and loads, S7...

[0025] Based on tail risk indicators, the initial joint operation scenario set is divided into a regular scenario layer and a tail risk scenario layer. Specifically, scenarios with tail risk indicators greater than or equal to the tail risk threshold are defined as follows: The initial joint operation scenario is divided into a tail risk scenario layer, where the tail risk index is less than the tail risk threshold. The initial joint operation scenario is divided into a regular scenario layer;

[0026] The number of typical operating scenarios to be retained and the number of tail-risk scenarios to be retained are determined based on the final number of typical operating scenarios, as follows: , ,in, The number of scenarios to retain for tail-risk situations. The minimum number of items to retain for tail-risk scenarios. The number of typical operating scenarios to be retained in the end. For the first The initial probabilities of each initial joint operation scenario. For indicator functions, This is a tail risk amplification factor. The number of items to retain in a typical scenario.

[0027] As a further optimization scheme of the distribution network scheduling method that takes into account the correlation between wind, solar and load and risk characteristics, S8 specifically involves: using Euclidean distance to cluster the retained regular scenarios and the retained tail risk scenarios respectively; selecting the regular scenario sample corresponding to the cluster center of each regular scenario cluster as the representative regular scenario; and selecting the tail risk scenario sample corresponding to the cluster center of each tail risk scenario cluster as the representative tail risk scenario. The representative regular scenarios and the representative tail risk scenarios constitute a typical scenario set.

[0028] As a further optimization scheme of the distribution network scheduling method that takes into account the correlation and risk characteristics of wind, solar and load, S9 uses the constructed typical scenario set as the input data for the time series simulation of the target distribution network. With the goal of minimizing the total daily network loss and minimizing the total voltage deviation of all nodes in the network as the dual optimization objectives, the NSGA-II multi-objective genetic optimization algorithm is used to solve the flexible resource coordination and scheduling model of the distribution network and obtain the flexible resource scheduling results of the target distribution network.

[0029] The present invention, by adopting the above technical solution, has the following beneficial effects:

[0030] 1. This invention constructs a high-dimensional joint distribution model of wind power, photovoltaic power and load based on Copula, which breaks through the limitation of traditional methods that only consider the two-dimensional correlation between wind and solar power. It can more accurately characterize the nonlinear, asymmetric and tail-dependent relationships between source and load variables, and reduce the scene deviation caused by insufficient correlation modeling.

[0031] 2. This invention uses nonparametric kernel density estimation to fit the marginal distributions of various wind power, photovoltaic, and load variables. It also combines Kendall's rank correlation coefficient, Pair-Copula decomposition, and information criterion selection to construct a joint distribution model that considers the synchronous coupling strength, lag coupling strength, and fluctuation trend coupling strength among the random variables of wind, solar, and load at each node. This model globally characterizes the spatiotemporal coupling relationship of high-dimensional variables while avoiding the limitations of a single distribution assumption on actual operating data, thus improving the fitting accuracy and adaptability of the random output characteristics of wind and solar loads.

[0032] 3. In the scenario reduction stage, this invention introduces a tail risk hierarchical clustering method to distinguish between conventional scenarios and tail risk scenarios based on the net load risk index, while retaining low-probability, high-risk scenarios such as high net load, low renewable energy output, and rapid ramp-up. This makes the generated typical scenarios more in line with actual operating characteristics. Using the generated typical scenario set for distribution network optimization scheduling can achieve coordinated scheduling of flexible resources. The generated typical scenario set for distribution network safety verification and risk assessment can provide timely and accurate early warning information for preventing potential safety risks in the distribution network. Attached Figure Description

[0033] Figure 1 This is a flowchart of the distribution network scheduling method proposed in this invention, which takes into account the correlation and risk characteristics of wind, solar and load.

[0034] Figure 2 This is a structural diagram of a C-vine of a wind-blown lotus provided in one embodiment of the present invention.

[0035] Figure 3 A simulation diagram of a typical PV1 scene with scene reduction provided in an embodiment of the present invention.

[0036] Figure 4The image shows a simulated PV2 scene with scene reduction provided in one embodiment of the present invention.

[0037] Figure 5 This is a simulation diagram of a typical PV3 scene after scene reduction, provided as an embodiment of the present invention.

[0038] Figure 6 The image shows a PV4 typical scene simulation diagram with scene reduction provided in one embodiment of the present invention.

[0039] Figure 7 The following is a simulation diagram of a typical WP1 scene after scene reduction, provided as an embodiment of the present invention.

[0040] Figure 8 The following is a simulation diagram of a typical WP2 scene after scene reduction, provided as an embodiment of the present invention.

[0041] Figure 9 The simulation diagram of a typical Load1 scenario after scene reduction is provided for one embodiment of the present invention.

[0042] Figure 10 The Pareto front simulation diagram for a typical scenario generated by the method proposed in this invention is provided as an embodiment of this invention.

[0043] Figure 11 The Pareto front simulation diagram for a scenario generated by a conventional method according to an embodiment of the present invention. Detailed Implementation

[0044] The present invention will be further explained in detail below with reference to the accompanying drawings and specific embodiments, but it should be understood that the scope of protection of the present invention is not limited to the specific embodiments.

[0045] To verify the effectiveness of the high-dimensional source-load correlation modeling and scene generation method based on Copula, measured data of five-element wind-solar-load were analyzed. Monte Carlo sampling was performed on the joint distribution model of wind-solar-load Copula to generate large-scale joint wind-solar-load power output scenes. After clustering and reduction, a typical scene set was obtained, and then the typical scene set was optimized and scheduled under multiple objectives.

[0046] In one embodiment of the present invention, as follows is provided Figure 1 The distribution network scheduling method shown includes S1 to S9, which takes into account the correlation and risk characteristics of wind, solar and load.

[0047] S1: Collect multi-day historical time-series data of wind power nodes, photovoltaic nodes, and load nodes within the target distribution network, generate random variable samples of wind, solar, and load for each node, and construct a high-dimensional spatiotemporal sample set representing the random variables of wind, solar, and load in the target distribution network. , No. Sample of random variables of wind and solar load at each node By the Historical time series data of wind, solar and load random variables of individual nodes Composition. For example... Figure 2 As shown in T1: At that time, the first Nodal wind-solar-load random variable samples Corresponding to the historical time-series data of PV1 at node 1; At that time, the first Nodal wind-solar-load random variable samples Corresponding to the historical time-series data of PV2 on node 2; At that time, the first Nodal wind-solar-load random variable samples Corresponding to node 3, WP1; At that time, the first Nodal wind-solar-load random variable samples Corresponding to node 4, WP2; At that time, the first Nodal wind-solar-load random variable samples This corresponds to node 5, Load.

[0048] S2, nonparametric kernel density estimation is performed on the high-dimensional spatiotemporal sample set to obtain the random variable samples of wind-solar-load at each node. probability density function and marginal distribution function The historical time series data of the random variable samples of wind and solar load at each node are converted into pseudo-observation samples of random variable samples of wind and solar load at each node that follow a uniform distribution.

[0049] Using the Gaussian kernel function as the kernel function, the random variable samples of wind and solar loads at each node are constructed using the kernel density estimation formula shown in equation (1). probability density function ,

[0050] In equation (1), For the first Sample of random variables of wind and solar load at each node The observed value at time, This represents the number of observations included in the historical time-series data of the nodal wind-solar-load random variable sample. This represents the kernel function bandwidth.

[0051] Wind-solar-load random variable samples at each node Marginal distribution function By analyzing the probability density function Integral transform to obtain, Substitution Find the first one that follows a uniform distribution in [0,1]. Pseudo-observation samples of random variables of wind, solar and load at each node.

[0052] S3, Calculate the multi-scale rank entropy correlation coefficient between pseudo-observed samples of wind-solar-load random variables at any two nodes. The current node wind-solar load random variable corresponding to the pseudo-observed sample that maximizes the sum of the multi-scale rank entropy correlation coefficients between the current node's wind-solar load random variable pseudo-observed sample and those of other nodes' wind-solar load random variable pseudo-observed samples is selected as the core variable of the C-vine Copula structure. The multi-scale rank entropy correlation coefficients between the current node's wind-solar load random variable pseudo-observed sample and those of other nodes' wind-solar load random variable pseudo-observed samples are used as the weights of the edges in the C-Vine structure. The maximum spanning tree method is employed, that is, the hierarchical tree topology of the vine is optimized by satisfying the maximum weight connection rule of the C-vine adjacency constraint so that the hierarchical tree topology of the vine achieves the best representation effect on the correlation between high-dimensional random variables. The final constructed C-vine Copula structure is as follows: Figure 2 As shown.

[0053] The wind-solar-load multi-scale rank entropy correlation coefficient is used to simultaneously characterize the synchronous coupling, lag coupling, and fluctuation trend coupling strength between wind, solar and load. It is composed of a weighted average of synchronous rank correlation components, time-lag rank correlation components, and slope change correlation components.

[0054] No. The pseudo-observation sample of the random variable wind-solar-load at node n and the first node Multiscale rank entropy correlation coefficient among pseudo-observed samples of wind, solar, and load random variables at each node Calculated using equation (2):

[0055]

[0056] In equation (2), For the first The pseudo-observation sample of the random variable wind-solar-load at node n and the first node The synchronous rank correlation components among pseudo-observed samples of the random variables of wind and solar load at each node For the first The pseudo-observation sample of the random variable wind-solar-load at node n and the first node The time-delay rank correlation components between pseudo-observed samples of the wind-solar-load random variables at each node For the first The pseudo-observation sample of the random variable wind-solar-load at node n and the first node The correlation components of slope change among pseudo-observed samples of the random variables of wind and solar load at each node. , , Let be the weighting coefficients of the synchronous rank correlation component, the time delay rank correlation component, and the ramp change correlation component, and satisfy . .

[0057] The synchronous rank correlation component is used to describe the rank correlation strength between two nodal wind-solar-load random variables at the same time. The pseudo-observation sample of the random variable wind-solar-load at node n and the first node Synchronous rank correlation components among pseudo-observed samples of the wind-solar-load random variable at each node The expression is:

[0058]

[0059] In equation (3), This represents the Kendall rank correlation function. and For respectively the first The random variable of wind-solar load at each node, the first node The random variables of wind and solar load at each node are in Pseudo-observation samples at specific moments.

[0060] The time-delay rank correlation component is used to describe the correlation strength of wind-solar-load random variables at different nodes under different time lags. The pseudo-observation sample of the random variable wind-solar-load at node n and the first node Expression of the time-delay rank correlation component between pseudo-observed samples of the wind-solar-load random variable at each node for:

[0061]

[0062] In equation (4), The maximum time delay order, This is the time delay order index. It is a time-delay weight, and satisfies .

[0063] The slope variation correlation component is used to describe the consistency of the power variation trend of the wind-solar-load random variables of two nodes in adjacent time periods. The pseudo-observation sample of the random variable wind-solar-load at node n and the first node The correlation components of slope change among pseudo-observed samples of wind-solar-load random variables at each node The expression is:

[0064]

[0065] In equation (5), , .

[0066] The core variable of Copula is selected from the pseudo-observed sample of the current node wind-solar-load random variable that maximizes the sum of the multi-scale rank entropy correlation coefficients between the pseudo-observed sample of the current node wind-solar-load random variable and the pseudo-observed samples of the wind-solar-load random variables of other nodes. Essentially, it selects the node wind-solar-load random variable with the strongest overall correlation with the random variables of the wind-solar-load random variables of other nodes as the core variable. The first tree constructed by directly connecting the core variable to the random variables of the wind-solar-load random variables of all other nodes can capture the most global dependency information.

[0067] For the The multi-scale rank entropy correlation coefficient between the pseudo-observed samples of wind and solar load random variables at each node and the pseudo-observed samples of wind and solar load random variables at the other nodes is calculated using equation (6). Sum of absolute values :

[0068]

[0069] The node wind-solar-load random variable with the highest overall correlation is selected as the core variable of the first tree of C-vine, and its index is... satisfy:

[0070]

[0071] Figure 2 The figure shows the C-vine Copula hierarchical tree topology of the five-element wind-solar-load variables. According to the figure caption, the variable correspondence is: 1-PV1, 2-PV2, 3-WP1, 4-WP2, 5-Load.

[0072] The first tree T1 consists of five circular nodes representing the original wind-solar-load random variables of the five nodes. Circular node 1 represents the wind-solar-load random variable of node PV1, circular node 2 represents the wind-solar-load random variable of node PV2, circular node 3 represents the wind-solar-load random variable of node WP1, circular node 4 represents the wind-solar-load random variable of node WP2, and circular node 5 represents the wind-solar-load random variable of node Load. Node 1 is the core variable, forming a star-shaped network connecting nodes 2, 3, 4, and 5. The four edges are labeled 1,2, 1,3, 1,4, and 1,5, corresponding to four binary pair-copulas.

[0073] The second tree T2 consists of four edges of the ellipse node T1. Connecting outwards from edge (1,2) as the core, the three edges are labeled 2,3|1, 2,4|1, and 2,5|1, representing the pair-copula conditions of PV2 with WP1, WP2, and Load, respectively, given PV1.

[0074] The third tree T3 consists of three edges with node T2. Two edges are labeled 4,3|21 and 3,5|12, representing the conditional pair-copula of WP2 and WP1, and WP1 and Load, given {PV1,PV2}.

[0075] The fourth tree T4 consists of two edges connected to node T3. One edge is labeled 5,4|123, representing the conditional pair-copula of Load and WP2 given {PV1,PV2,WP1}.

[0076] S4. For each pair of wind-solar-load random variables connected by each edge of each tree in the C-vine Copula structure, a family of Pair-Copula functions is selected. After maximum likelihood estimation of the parameters of each candidate Pair-Copula function, the Bayesian Information Criterion (BIC) is used to select the Pair-Copula function with the best fitting effect, and finally a high-dimensional joint distribution function of wind-solar-load is constructed. The Pair-Copula alternative function set includes Gaussian Copula, Gumbel Copula, Clayton Copula, Frank Copula, Joe Copula, and the 90°, 180°, and 270° rotation variants of the above Copula functions.

[0077] S5. Sample the high-dimensional joint distribution function of wind, solar, and load to generate an initial joint operation scenario set. Perform Monte Carlo random sampling on the high-dimensional joint distribution function of wind, solar, and load. The sample size of the Monte Carlo random sampling is 500-2000 sets, and each sample set is a sequence of wind power, photovoltaic, and load output corresponding to a 24-hour time scale and hourly time resolution, which constitutes the initial joint operation scenario set.

[0078] S6. Based on the load power, wind power output, and photovoltaic power output of the initial joint operation scenario, calculate the net load risk sequence and tail risk index for each initial joint operation scenario. The tail risk index is used to characterize the degree of high net load, rapid net load ramp-up, and the persistence of exceeding the risk threshold.

[0079] For the The initial joint operation scenario, in the first Net load risk at any given time for:

[0080]

[0081] In equation (8), For the first Initial joint operation scenario Constant load power, For the first Initial joint operation scenario Time of the first Each wind power node outputs power. For the first Initial joint operation scenario The number of wind power nodes participating in power generation at any given time. For the first Initial joint operation scenario Time of the first Each photovoltaic node contributes power. For the first Initial joint operation scenario The number of photovoltaic nodes participating in power generation at any given time.

[0082] No. Tail risk indicators for initial joint operation scenarios for:

[0083]

[0084] In equation (9), , Net load risk quantiles, , These are the weighting coefficients.

[0085] S7. Based on the tail risk index, divide the initial joint operation scenario set into a regular scenario layer and a tail risk scenario layer: define scenarios where the tail risk index is greater than or equal to the tail risk threshold. The initial joint operation scenario is divided into a tail risk scenario layer, where the tail risk index is less than the tail risk threshold. The initial joint operation scenario is divided into a regular scenario layer. The number of typical operating scenarios retained was determined based on the empirical quantiles of the tail risk indicators for all initial joint operation scenarios. Number of tail-risk scenarios retained for:

[0086]

[0087] In equation (10), The minimum number of items to retain for tail-risk scenarios. For the first The initial probabilities of each initial joint operation scenario. For indicator functions, This is the tail risk amplification factor.

[0088] Number of items retained in typical scenarios for:

[0089]

[0090] S8. Euclidean distance is used to cluster the retained regular scenarios and tail-risk scenarios respectively. Real-world scenario samples are selected from each cluster as representative scenarios, and a typical scenario set is constructed based on these representative scenarios. For example... Figures 3 to 9 As shown, scenarios 1 and 2 are representative tail-risk scenarios. The process of performing Euclidean distance clustering on the retained regular scenarios or tail-risk scenarios is as follows:

[0091] S81. Randomly divide the retained regular scenarios or retained tail risk scenarios into k groups of scenarios, and randomly designate one scenario sample in each group as the initial cluster center.

[0092] S82. Calculate the Euclidean distance between each scene sample and the cluster center of each group of scenes, and assign each scene sample to the group to which the cluster center with the smallest Euclidean distance belongs.

[0093] S83. Calculate the new cluster centers in the groups after partitioning using the following formula:

[0094]

[0095] In equation (12), For the first The number of scene samples within the group. For the first The Euclidean distance between the m-th scene sample and the m-th scene sample;

[0096] S84. Repeat steps S82-S83 until the cluster centers do not change. The resulting cluster centers are the representative scenarios.

[0097] S9. Use the constructed typical scenario set as input data for the time-series simulation of the target distribution network to obtain the flexible resource scheduling results of the target distribution network.

[0098] After completing the modeling, sampling, and hierarchical clustering reduction processes S1 to S8, six typical wind-solar-load operation scenarios with probability weights were obtained. To quantitatively verify the fidelity and engineering practical value of the scenarios generated by this invention, comparative tests were conducted from two dimensions: statistical distribution fitting accuracy and actual operational benefits of multi-objective optimization scheduling of the distribution network. The control group used existing traditional scenario generation methods. The test results are as follows:

[0099] I. Quantitative Validation of Scene Distribution Fidelity Based on KL Divergence and JS Divergence

[0100] KL divergence and JS divergence were used as quantitative evaluation indicators. The smaller the values ​​of these two indicators, the smaller the difference between the probability distribution of the generated scenario and the actual time-series sample distribution of wind and solar load in the distribution network, the lower the degree of scenario distortion, and the closer it is to the actual output and load variation patterns on site. The KL divergence and JS divergence between the scenario sets generated by the method of this invention and the traditional method and the measured historical samples were calculated respectively, and the statistical results are shown in Table 1.

[0101] Table 1. The degree of similarity between the generated scenarios and the real-world scenarios from different solutions.

[0102]

[0103] The quantitative data in Table 1 clearly shows that, compared with traditional modeling methods, the KL divergence and JS divergence values ​​of the generated scenarios are significantly reduced. This proves that the present invention fully characterizes the three-dimensional synchronization, time delay, and ramp-up coupling relationship of wind, solar and load through multi-scale rank entropy correlation coefficients, and effectively alleviates the scene distortion problem of traditional methods by using tail risk hierarchical clustering to retain extreme operating conditions. The generated wind, solar and load time series scenarios are highly consistent with the real operation data of the distribution network in terms of probability distribution, fluctuation characteristics, and coverage of extreme operating conditions, and can provide a high-precision input data source for subsequent power grid simulation and scheduling calculations.

[0104] II. Performance Verification of Distribution Network Scheduling Based on NSGA-II Multi-Objective Optimization Algorithm

[0105] The typical scenario set generated by this invention and the scenario set generated by traditional methods are used as inputs for distribution network time-series simulation. With the dual optimization objectives of minimizing the total daily network loss and minimizing the total voltage deviation of all network nodes, the NSGA-II multi-objective genetic optimization algorithm is used to solve the flexible resource coordination and scheduling model of the distribution network. The comparison results of the Pareto optimal frontier distributions corresponding to the two scenario inputs are as follows: Figure 10 , Figure 11 As shown.

[0106] The scheduling performance differences between the two scenarios can be quantitatively distinguished from the results of multi-objective optimization:

[0107] When optimizing scheduling using the typical scenarios of this invention, the resulting Pareto fronts are all concentrated in the low-loss range of 1.28~1.36MWh, corresponding to a total voltage deviation of 27.8~30.6pu for all nodes in the network over 24 hours, resulting in lower active power loss and better voltage quality in the system.

[0108] When optimizing scheduling using traditional distortion scenarios, the Pareto front shifts to a higher loss range, with total network loss reaching 1.64~1.76MWh throughout the day, and the voltage deviation across the entire network increases simultaneously.

[0109] The simulation results of this project fully demonstrate that traditional methods suffer from severe distortion in the generated scenarios due to the lack of high-dimensional temporal coupling characteristics of wind, solar and load in the modeling and the omission of extreme risk conditions at the tail end. The scheduling strategies formulated based on the distorted scenarios cannot match the actual source-load fluctuations of the power grid, which can easily lead to high active power loss in the distribution network and an increased risk of node voltage exceeding limits. In contrast, this invention eliminates the systematic distortion defects of the scenarios. Based on the scheduling scheme obtained by solving typical scenarios that are close to reality, it can simultaneously realize the economic operation of the distribution network and voltage safety control, and has clear application value in power system engineering. It solves the technical shortcomings of existing scenario generation technologies that only stay at the data level and cannot support the safe and economical regulation of the distribution network.

[0110] One embodiment of the present invention provides a computer system including a memory and a processor. The memory stores a computer program that runs on the processor. When the processor runs the computer program, it executes the steps of the above-described distribution network scheduling method that takes into account the correlation between wind, solar and load and risk characteristics.

[0111] One embodiment of the present invention is a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the above-described distribution network scheduling method that takes into account the correlation and risk characteristics of wind, solar and load.

[0112] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the specific embodiments described above. The specific embodiments and descriptions in the specification are merely for further illustrating the principles of the invention. Various changes and modifications can be made to the present invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of protection claimed by the present invention. The scope of protection of the present invention is defined by the claims and their equivalents.

Claims

1. A distribution network scheduling method considering the correlation and risk characteristics of wind, solar, and load, characterized in that, include: S1. Collect historical time-series data of wind power nodes, photovoltaic nodes and load nodes in the target distribution network, generate random variable samples of wind, solar and load for each node, and construct a high-dimensional spatiotemporal sample set representing the random variables of wind, solar and load in the target distribution network. S2, perform nonparametric kernel density estimation on the high-dimensional spatiotemporal sample set to obtain the probability density function and marginal distribution function of the random variable samples of wind and solar load at each node, and convert the running history time series data of the random variable samples of wind and solar load at each node into pseudo-observation samples of random variable samples of wind and solar load at each node that follow a uniform distribution. S3, considering the synchronous rank correlation component, time-delay rank correlation component and slope change correlation component between pseudo-observation samples of wind and solar load random variables of each node, calculate the multi-scale rank entropy correlation coefficient between pseudo-observation samples of wind and solar load random variables of any two nodes, and construct the C-vine Copula structure. S4, select a Pair-Copula function for the random variable pairs of wind and solar loads connected to each edge of each tree in the C-vine Copula structure, and construct a high-dimensional joint distribution function of wind and solar loads from the C-vine Copula structure with the selected Pair-Copula function; S5, sample the high-dimensional joint distribution function of wind, solar and load to generate an initial joint operation scenario set; S6. Based on the load power, wind power output, and photovoltaic power output of the initial joint operation scenario, calculate the net load risk sequence and tail risk index for each initial joint operation scenario. S7. Based on the tail risk index, the initial joint operation scenario set is divided into a regular scenario layer and a tail risk scenario layer. The number of regular scenarios and tail risk scenarios to be retained is determined by combining the number of typical operation scenarios to be retained. S8, cluster the retained regular scenarios and retained tail risk scenarios respectively, and select actual scenario samples in each cluster as representative scenarios, and construct a typical scenario set based on the representative scenarios; S9 uses the constructed typical scenario set as input data for the time-series simulation of the target distribution network to obtain the flexible resource scheduling results of the target distribution network.

2. The distribution network scheduling method considering wind-solar-load correlation and risk characteristics according to claim 1, characterized in that, In step S2, nonparametric kernel density estimation is performed on the high-dimensional spatiotemporal sample set to obtain the probability density function and marginal distribution function of the wind-solar-load random variable samples at each node. Specifically, the kernel density estimation is used to construct the first... Sample of random variables of wind and solar load at each node probability density function , for the Sample of random variables of wind and solar load at each node probability density function Integral transform to obtain the first Sample of random variables of wind and solar load at each node Marginal distribution function ,in, , For the first Sample of random variables of wind and solar load at each node The observed value at time, This represents the number of observations included in the historical time-series data of the nodal wind-solar-load random variable sample. This represents the kernel function bandwidth.

3. The distribution network scheduling method considering wind-solar-load correlation and risk characteristics according to claim 2, characterized in that, In step S2, the historical time-series data of the wind-solar load random variable samples at each node are converted into pseudo-observation samples of the wind-solar load random variables at each node that follow a uniform distribution. Specifically, this involves converting the first... Sample of random variables of wind and solar load at each node Observations at time Substitute into the marginal distribution function Find the first element that follows a uniform distribution in [0,1]. Pseudo-observation samples of random variables of wind, solar and load at each node.

4. The distribution network scheduling method considering wind-solar-load correlation and risk characteristics according to claim 3, characterized in that, In S3, considering the synchronous rank correlation component, time-delay rank correlation component, and slope change correlation component among the pseudo-observation samples of wind-solar-load random variables at each node, the multi-scale rank entropy correlation coefficient between any two pseudo-observation samples of wind-solar-load random variables is calculated as follows: ,in, For the first The pseudo-observation sample of the random variable wind-solar-load at node n and the first node The multi-scale rank entropy correlation coefficient among pseudo-observed samples of the random variables of wind and solar load at each node. For the first The pseudo-observation sample of the random variable wind-solar-load at node n and the first node The synchronous rank correlation component between pseudo-observed samples of wind-solar-load random variables at each node is used to describe the rank correlation strength between the two node wind-solar-load random variables at the same time. For the first The pseudo-observation sample of the random variable wind-solar-load at node n and the first node The time-delay rank correlation components among pseudo-observed samples of wind and solar load random variables at each node are used to describe the correlation strength of wind and solar load random variables at different time lags. For the first The pseudo-observation sample of the random variable wind-solar-load at node n and the first node The slope change correlation component between pseudo-observation samples of wind-solar-load random variables at each node is used to describe the consistency of the power change trend of the two node wind-solar-load random variables in adjacent time periods. , , These are the weighting coefficients for the synchronous rank correlation component, the time delay rank correlation component, and the ramp change correlation component, respectively. , , , , This is Kendall's rank correlation function. For the first The random variables of wind and solar load at each node are in Pseudo-observation samples at time points For the first The random variables of wind and solar load at each node are in Pseudo-observation samples at time points The maximum time delay order, This is the time delay order index. For time delay weights, , For the first The random variables of wind and solar load at each node are in Pseudo-observation samples at time points , , For the first The random variables of wind and solar load at each node are in Pseudo-observation samples at time points For the first The random variables of wind and solar load at each node are in Pseudo-observation samples at specific times.

5. The distribution network scheduling method considering wind-solar-load correlation and risk characteristics according to claim 4, characterized in that, In S3, the specific method for constructing the C-vine Copula structure is as follows: when the sum of the multi-scale rank entropy correlation coefficients between the pseudo-observed samples of the wind-solar-load random variable of the current node and the pseudo-observed samples of the wind-solar-load random variable of the other nodes reaches the maximum value, the wind-solar-load random variable of the current node corresponding to the pseudo-observed sample of the wind-solar-load random variable of the current node is selected as the core variable of the C-vine Copula structure, and the multi-scale rank entropy correlation coefficients between the pseudo-observed samples of the wind-solar-load random variable of the current node and the pseudo-observed samples of the wind-solar-load random variable of the other nodes are used as the weights of the edges in the C-Vine structure. The maximum spanning tree method is used to optimize the hierarchical tree topology of the vine.

6. The distribution network scheduling method considering wind-solar-load correlation and risk characteristics according to claim 5, characterized in that, In step S4, a Pair-Copula function is selected for the random variable pairs of wind, light, and load connected to each edge of each tree in the C-vine Copula structure. Specifically, the parameters of the candidate Pair-Copula functions are estimated by maximum likelihood, and the Pair-Copula function with the best fitting effect is selected by using the Bayesian information criterion.

7. The distribution network scheduling method considering wind-solar-load correlation and risk characteristics according to claim 6, characterized in that, In step S6, based on the load power, wind power output, and photovoltaic output of the initial joint operation scenario, the net load risk sequence and tail risk index for each initial joint operation scenario are calculated. The specific method is as follows: , ,in, For the first Initial joint operation scenario Net load risk at any given time For the first Initial joint operation scenario Constant load power, For the first Initial joint operation scenario Time of the first Each wind power node outputs power. For the first Initial joint operation scenario The number of wind power nodes participating in power generation at any given time. For the first Initial joint operation scenario Time of the first Each photovoltaic node contributes power. For the first Initial joint operation scenario The number of photovoltaic nodes participating in power generation at any given time. For the first Tail risk indicators for an initial joint operation scenario , For the first Initial joint operation scenario Net load risk at any given time Net load risk quantiles, , These are the weighting coefficients.

8. The distribution network scheduling method considering wind-solar-load correlation and risk characteristics according to claim 7, characterized in that, In S7 Based on the tail risk index, the initial joint operation scenario set is divided into a regular scenario layer and a tail risk scenario layer. Specifically, the tail risk index is set to be greater than or equal to the tail risk threshold. The initial joint operation scenario is divided into a tail risk scenario layer, where the tail risk index is less than the tail risk threshold. The initial joint operation scenario is divided into a regular scenario layer; The number of typical operating scenarios to be retained and the number of tail-risk scenarios to be retained are determined based on the final number of typical operating scenarios, as follows: , ,in, The number of scenarios to retain for tail-risk situations. The minimum number of items to retain for tail-risk scenarios. The number of typical operating scenarios to be retained in the end. For the first The initial probabilities of each initial joint operation scenario. For indicator functions, This is a tail risk amplification factor. The number of items to retain in a typical scenario.

9. The distribution network scheduling method considering wind-solar-load correlation and risk characteristics according to claim 8, characterized in that, Specifically, S8 involves using Euclidean distance to cluster the retained regular scenarios and retained tail risk scenarios, selecting the regular scenario sample corresponding to the cluster center of each regular scenario cluster as the representative regular scenario, and selecting the tail risk scenario sample corresponding to the cluster center of each tail risk scenario cluster as the representative tail risk scenario. The representative regular scenarios and the representative tail risk scenarios together form a typical scenario set.

10. A distribution network scheduling method considering wind-solar-load correlation and risk characteristics according to claim 9, characterized in that, In S9, the constructed typical scenario set is used as the input data for the time-series simulation of the target distribution network. The dual optimization objectives are to minimize the total daily network loss and the total voltage deviation of all network nodes. The NSGA-II multi-objective genetic optimization algorithm is used to solve the flexible resource coordination and scheduling model of the distribution network and obtain the flexible resource scheduling results of the target distribution network.

Images