Power distribution network power supply grid division method and system considering resource cluster clustering

Abstract

The invention provides a power distribution network power supply grid division method and system considering resource cluster clustering, and belongs to the technical field of power systems. The method comprises the steps of inputting annual load data and electric vehicle behavior data into a photovoltaic power supply output model, an energy storage charging and discharging model and an electric vehicle charging pile model, and generating time sequence data; clustering the time sequence data by adopting an improved K-means algorithm, determining an optimal clustering number through an elbow method, and determining an output curve of a typical scene according to the optimal clustering number; dynamically adjusting load imbalance degree weights and system active power loss weights in different scenes based on an output curve of a typical scene, and constructing a power distribution network power supply grid optimization model; and solving the power distribution network power supply grid optimization model to obtain an optimal grid division scheme. According to the method, the precision of power distribution network grid planning and the adaptive capability and scientificity of typical scene extraction are improved, and the complexity of power distribution network grid planning is reduced.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of power systems, and particularly relates to grid division planning of distribution network containing distributed flexible resources, and more particularly to a distribution network power supply grid division method and system considering resource cluster clustering. BACKGROUND

[0002] As one of the key components of the power system, the distribution network is responsible for receiving and distributing the power transmitted by the main network and delivering it to the users, directly affecting the life and experience of the users. Flexible resources are widely used in distribution networks due to their advantages such as easy installation, environmental friendliness, and high utilization rate. In addition, the control of the power system is gradually shifting from centralized to distributed mode, which increases the uncertainty of the operation of the power grid. However, with the continuous growth of power demand and the rapid development of emerging flexible resources such as distributed generators (DG), electric vehicles (EV), and energy storage (ES), the scale of modern medium-voltage distribution networks is becoming increasingly large, and the network structure is becoming increasingly complex, which has a huge impact on the operation of the distribution network. How to fully consider the uncertainty of DG and take into account the economy and safety of the distribution network to make reasonable allocation is a major difficulty faced by distribution network planning.

[0003] However, the existing grid division technology of the distribution network considers the application of grid division in load balancing and voltage control, but does not discuss the mutual influence between the performance of grid division and the allocation of flexible resources, making it difficult to deal with the planning problem of the distribution network under the access of flexible resources.

[0004] The existing technology is based on a specific number of clusters when performing K-means clustering, and does not dynamically measure the number of clusters k. The selection of the number of clusters has an important influence on the accuracy and efficiency of the optimization algorithm. If the value of k is too small, it may lead to insufficient clustering of samples, and the characteristics of uncertain variables cannot be accurately described. If the value of k is too large, it will increase the computational complexity and affect the solving efficiency. SUMMARY

[0005] To solve the problems existing in the prior art, the present application provides a distribution network power supply grid division method and system considering resource cluster clustering, which uses a K-means clustering algorithm improved by the elbow method to analyze the clusters of flexible resources, dynamically determines the optimal number of clusters, and generates typical operating scenarios. Based on the clustering results, an optimization model for distribution network grid division is established, with the weighted sum of system active power loss and load imbalance degree as the objective function, and the model is converted into a mixed integer second-order cone programming problem for solving. The influence of grid division strategy and grid number on the economy and operational stability of the distribution network is quantitatively analyzed.

[0006] The present invention adopts the following technical solution.

[0007] The first aspect of the present invention provides a method for dividing the power supply grid of a distribution network considering resource clustering, comprising: The annual load data and electric vehicle behavior data are input into the photovoltaic power output model, energy storage charging and discharging model, and electric vehicle charging pile model to generate time series data. An improved K-means algorithm is used to cluster time series data. The optimal number of clusters is determined by the elbow method, and the output curve for typical scenarios is determined based on the optimal number of clusters. Based on the output curves of typical scenarios, the load imbalance weight and system active power loss weight are dynamically adjusted under different scenarios to construct a power grid optimization model for the distribution network. Solve the power distribution grid optimization model to obtain the optimal grid partitioning scheme.

[0008] Preferably, the photovoltaic power output model is expressed by the following formula:

[0009] In the formula, This indicates the rated output power of a single photovoltaic array; Indicates the installed capacity of the photovoltaic array; Indicates latitude Hourly average solar irradiance at m month, d day, and h hour; Indicates the irradiation intensity under standard test conditions; This represents the power temperature coefficient of a photovoltaic array; This indicates the actual operating temperature of the photovoltaic array; Indicates the reference temperature.

[0010] Preferably, the energy storage charging and discharging model is expressed by the following formula:

[0011] In the formula, , These represent the electricity purchased and sold between the photovoltaic-storage charging station and the power grid at time t, respectively. , Let represent the charging and discharging power of the energy storage system at time t, respectively; , These represent the actual power values ​​of the conventional electrical load and the electric vehicle charging load at time t, respectively. Conventional loads include, but are not limited to, lighting and air conditioning. Let t be the actual active power generated by the photovoltaic system. The operational constraints of energy storage systems include power constraints, state of charge constraints, charge / discharge balance constraints, and charge / discharge state transition constraints.

[0012] Preferably, the electric vehicle charging pile model is represented by the following formula:

[0013] In the formula, Let represent the probability density of an electric vehicle arriving at an electric vehicle charging station at time t. This represents the average time it takes for an electric vehicle to arrive at a charging station. This represents the standard deviation of the time it takes for an electric vehicle to arrive at an electric vehicle charging station.

[0014] Preferably, determining the output curve for a typical scenario based on the optimal number of clusters includes: The clustering error SSE corresponding to different numbers of clusters k is calculated by the elbow method, and the k value corresponding to the inflection point of the SSE curve is selected as the optimal number of clusters K. Based on the optimal number of clusters, the optimal number of initial cluster centers is determined by iteratively selecting the remaining time series data that has the largest sum of distances to all selected cluster centers; Calculate the distance between each time series data point and all initial cluster centers, and assign it to the cluster to which the nearest initial cluster center belongs; calculate the mean vector of all time series data points within each cluster, and set it as the new cluster center for that cluster; calculate the sum of squared distances from all time series data points to the cluster centers of their respective clusters, and set it as the clustering error SSE; if the SSE between two consecutive iterations is less than the set convergence threshold, terminate the iteration; otherwise, continue iterating to obtain the final K cluster centers; The final K cluster centers are used as output curves for K typical scenarios, and the number of time series data belonging to each cluster is counted.

[0015] Preferably, the objective function of the power distribution grid optimization model is... It can be expressed by the following formula:

[0016] In the formula, It is a set of output curves for K typical scenarios; It is the frequency of each typical scenario s throughout the year; This represents the total active power loss of the distribution network system under a typical scenario s. This is the baseline value for the active power of the distribution network system under typical scenario s; This represents the load imbalance between grids under a typical scenario s. This represents the weight of load imbalance in scenario s. This represents the active power loss weight of the system under scenario s. , Preferably, This represents the weight of load imbalance in scenario s. The active power loss weight of the system under scenario s is represented by the following formula:

[0017] In the formula, and This represents the minimum and maximum values ​​of the system's active power loss weighting coefficient. , This represents the scene category factor.

[0018] Preferably, The scene category factor is represented by the following formula:

[0019] In the formula, This represents the benchmark value for photovoltaic penetration rate. The benchmark value representing the load volatility index. Indicates the sensitivity coefficient of photovoltaic penetration. This represents the load fluctuation sensitivity coefficient. This represents the photovoltaic penetration rate index for each scenario s. This represents the load volatility index for each scenario s.

[0020] Preferably, the photovoltaic penetration rate index for each scenario s is calculated. and load fluctuation index It can be expressed by the following formula:

[0021] In the formula, This represents the photovoltaic active power output at node p' at time t in scenario s. This represents the active power of charging the electric vehicle at time t at node p in scenario s. For the scene Down The active power of the electrical load at time node p' Let represent the total active power load of the system at time t under scenario s. This represents the average total active power load of the system under scenario s, where T is the total number of time points.

[0022] The second aspect of the present invention provides a power grid partitioning system for a distribution network that considers resource clustering, and runs the power grid partitioning method for a distribution network that considers resource clustering described in the first aspect, comprising: The data acquisition module is used to input the annual load data and electric vehicle behavior data into the photovoltaic power output model, energy storage charging and discharging model, and electric vehicle charging pile model to generate time series data. The scene acquisition module is used to cluster time series data using an improved K-means algorithm, determine the optimal number of clusters using the elbow method, and determine the output curve of a typical scene based on the optimal number of clusters. The model building module is used to dynamically adjust the load imbalance weight and system active power loss weight under different scenarios based on the output curves of typical scenarios, and build a power grid optimization model for the distribution network. The scheme acquisition module is used to solve the power supply grid optimization model of the distribution network and obtain the optimal grid partitioning scheme.

[0023] Compared with the prior art, the beneficial effects of the present invention include at least the following: This invention addresses the problem that existing distribution network planning technologies fail to accurately characterize the stochastic and dynamic characteristics of flexible resources such as photovoltaics, energy storage, and electric vehicles by modeling these resources' flexibility. Traditional methods often treat these resources as fixed loads or simple power sources, failing to reflect the uncertainties caused by natural conditions (sunlight, temperature) and user behavior (electric vehicle travel). This invention improves the accuracy and realism of the model's depiction of actual distribution network operation scenarios, reduces the risk of grid planning schemes deviating from actual operating conditions due to model simplification, and enhances the accuracy of distribution network grid planning.

[0024] This invention addresses the problem in existing research where the number of clusters (k) needs to be predetermined based on experience, lacking scientific basis, through an improved K-means clustering algorithm. Traditional methods often use fixed k values ​​that are too small, leading to insufficient representativeness of the scenario, or too large, resulting in excessive computational complexity. This invention enhances the adaptability and scientific rigor of typical scenario extraction by dynamically determining the optimal k value using the elbow method. This ensures that the scenario fully encompasses the diversity of the original data, reduces the complexity of power distribution grid planning, lowers the computational scale and solution time, and improves the accuracy and efficiency of power distribution grid planning. This invention uses a dynamic weight adjustment mechanism to automatically calculate and adjust the weight coefficients of network loss and load imbalance in the objective function for different typical scenarios, such as, but not limited to, high photovoltaic penetration and high load fluctuation scenarios. This enables the grid partitioning scheme to adapt to the main contradictions of different operating scenarios. In high fluctuation scenarios, it prioritizes load balance and voltage quality, while in stable scenarios, it focuses on reducing network loss for economic reasons. This solves the shortcomings of traditional static weights that cannot adapt to the operating needs of multiple scenarios and improves the robustness and adaptability of the partitioning scheme. This invention addresses the problem of existing grid partitioning methods that separate network topology optimization from flexible resource scheduling by constructing a power distribution network grid optimization model. Traditional techniques either optimize the topology or only schedule resources, failing to systematically improve distribution network performance through the coordinated use of resource scheduling and network reconfiguration. This invention improves the overall economy of distribution network grid planning by minimizing active power losses and enhances its security by improving load balancing. It reduces the risk of system network losses and equipment overloads caused by source-load mismatch and unreasonable network structure, and coordinates grid partitioning and flexible resource optimization to improve the security and performance of distribution network grid planning. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of the power grid division process for a power distribution network that takes into account resource clustering, provided according to an embodiment of the present invention. Figure 2 These are schematic diagrams illustrating two scenarios with varying flexible resource penetration rates, provided in accordance with embodiments of the present invention. Figure 3 This is a schematic diagram comparing and analyzing the optimization results under different strategies provided in the embodiments of the present invention; Figure 4 This describes the changing trends of various indicators with the number of grids K under different scenarios provided in the embodiments of the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of this invention.

[0027] like Figure 1 As shown, Embodiment 1 of the present invention provides a method for dividing the power supply grid of a distribution network that takes into account resource clustering, including the following steps: Step 1: Input the annual load data and electric vehicle behavior data into the photovoltaic power output model, energy storage charging and discharging model, and electric vehicle charging pile model to generate time series data.

[0028] In a preferred but non-limiting embodiment of the present invention, step 1 includes: Step 1.1: Construct a photovoltaic power output model.

[0029] The power generation efficiency of photovoltaic modules is mainly constrained by sunlight conditions and ambient temperature, exhibiting a certain degree of fluctuation and regularity. Output power at time It is expressed by the following formula.

[0030] In the formula, This indicates the rated output power of a single photovoltaic array; Indicates the installed capacity of the photovoltaic array; Indicates latitude Hourly average solar irradiance at m month, d day, and h hour; Indicates the irradiation intensity under standard test conditions; This represents the power temperature coefficient of a photovoltaic array; This indicates the actual operating temperature of the photovoltaic array; Indicates the reference temperature.

[0031] Step 1.2: Construct an energy storage charging and discharging model.

[0032] The power imbalance generated by photovoltaic power generation and load consumption needs to be compensated through the cooperation of energy storage systems and the external power distribution network to meet the following power balance requirements:

[0033] In the formula, , These represent the electricity purchased and sold between the photovoltaic-storage charging station and the power grid at time t, respectively. , Let represent the charging and discharging power of the energy storage system at time t, respectively; , These represent the actual power values ​​of the conventional electrical load and the electric vehicle charging load at time t, respectively. Conventional loads include, but are not limited to, lighting and air conditioning. Let t be the actual active power generated by the photovoltaic system at time t.

[0034] The operational constraints of energy storage systems include power constraints, state of charge constraints, charge / discharge balance constraints, and charge / discharge state transition constraints.

[0035] The power constraint is expressed by the following formula:

[0036] In the formula, , These represent the maximum charging and discharging power of the energy storage system at time t, respectively. The state of charge constraint is expressed by the following formula:

[0037] In the formula, and These represent the minimum and maximum allowable states of charge of the energy storage system at time t, respectively. This represents the state of charge of the energy storage system at time t; This indicates the charging efficiency of the energy storage system. express, This indicates the discharge efficiency of the energy storage system. express.

[0038] The charge / discharge balance constraint is expressed by the following formula:

[0039] In the formula, This represents the net output power of the energy storage system at time t. The amount of charge a system receives during a charge / discharge cycle is equal to the amount of discharge.

[0040] The charge / discharge state transition constraint is expressed by the following formula:

[0041] In the formula, This represents the charging and discharging state of the energy storage system at time t, and is a 0-1 variable. =1 indicates a charging state. =0 indicates a discharge state.

[0042] Step 1.3: Construct an electric vehicle charging station (EVCS) model.

[0043] The arrival time of electric vehicles at charging stations roughly follows a normal distribution. The probability density function (PDF) of the arrival time of electric vehicles can be described as follows:

[0044] In the formula, Let represent the probability density of an electric vehicle arriving at an electric vehicle charging station at time t. This represents the average time it takes for the EV to reach the EVCS. This represents the standard deviation of the time it takes for the EV to reach the EVCS.

[0045] The daily load demand for electric vehicle charging is related to daily mileage and charging time. Generally, the daily mileage of an EV is considered to follow a normal distribution, with the following PDF:

[0046] In the formula, This indicates the daily mileage of the electric vehicle. The probability density, This indicates the daily mileage of an electric vehicle. and They represent Standard deviation and mean express The natural logarithm of .

[0047] Based on the electric vehicle's driving range and its initial state of charge (SOC), the actual SOC at the end of charging is:

[0048] In the formula, This indicates the actual state of charge of the battery at the end of charging. It is the initial state of charge of the battery when charging begins. This refers to the amount of electricity consumed by an electric vehicle to travel 100 kilometers. This indicates the total capacity of the electric vehicle's power battery.

[0049] The charging time for electric vehicles can be calculated in the following ways:

[0050] In the formula, This is the time required for an electric vehicle to complete charging; This indicates the target state of charge at the end of charging. This is the rated output power of the charging station. This indicates the average efficiency of the entire charging process.

[0051] Step 1.4: Input the annual load data and electric vehicle behavior data into the photovoltaic power output model, energy storage charging and discharging model, and electric vehicle charging pile model to generate three types of time series data: the annual output curve of the photovoltaic power source, the charging and discharging power curve of the energy storage system, and the load curve of the electric vehicle charging pile.

[0052] Step 2: The improved K-means algorithm is used to cluster the time series data. The optimal number of clusters K is determined by the elbow method. Based on the optimal number of clusters, the output curves of K typical scenarios are determined.

[0053] Step 2.1: Calculate the clustering error SSE corresponding to different cluster numbers k using the elbow method, and select the k value corresponding to the inflection point of the SSE curve as the optimal cluster number K.

[0054] Define a range of candidate cluster numbers k. For each k value, calculate the corresponding clustering error SSE using the elbow method, expressed as follows:

[0055] In the formula, Indicates the i-th cluster; For clusters The samples in; For clusters The centroid, i.e., the sample mean; For clusters The number of samples in Indicates the average distance within the class. This represents the average distance between classes.

[0056] The elbow method is a common method for determining the number of clusters. This method calculates the average intra-cluster distance (...). ) and the average distance between classes ( The ratio of clustering error to the ratio of clustering error to clustering error is used to measure clustering error. As the number of clusters increases, the sample partitioning becomes more refined, and the intra-cluster aggregation increases, resulting in... Gradually decrease; if the number of clusters is too large, It will also decrease accordingly, leading to The rate of decline slows down. At this point, the number of clusters and... The relationships between them resemble an "elbow," with the number of clusters corresponding to the elbows being... This is the optimal number of clusters.

[0057] Plot an elbow curve with the number of clusters k on the x-axis and the clustering error SSE on the y-axis. The optimal number of clusters K is taken as the number of clusters corresponding to the elbow inflection point. As k increases, nSSE decreases, leading to a decrease in SSE; however, when k is too large, wSSE also decreases, slowing the rate of SSE decline, resulting in an "elbow" shape on the curve. The k value corresponding to the elbow inflection point is the optimal number of clusters.

[0058] Step 2.2: Based on the optimal number of clusters, determine the optimal number of initial cluster centers by iteratively selecting the remaining time series data with the largest sum of distances to all selected cluster centers.

[0059] More preferably, step 2.2 includes: Randomly select a sample from all time series data as the first initial cluster center K1; Calculate the distance between all remaining time series data and K1, and select the time series data that is farthest from K1 as the second initial cluster center K2; Calculate the sum of distances between all remaining time series data and K1 and K2 after removing K1 and K2, and set the time series data with the largest sum of distances as the third cluster center K3; When selecting the fourth and subsequent cluster centers, the sum of distances to all previously selected centers is repeatedly calculated, and the time series data with the largest sum is selected as the next center in the selection process, until K initial cluster centers are selected. That is, for selecting the l-th center (l≤K), it is necessary to calculate the distance from each remaining time series data point to all previously selected l-th centers. The sum of distances to the l-th cluster centers is used to select the time series data with the largest sum of distances as the l-th center.

[0060] Step 2.3: Calculate the distance between each time series data point and all initial cluster centers, and assign it to the cluster to which the nearest initial cluster center belongs; calculate the mean vector of all time series data points within each cluster, and set it as the new cluster center for that cluster; calculate the sum of squared distances from all time series data points to the cluster centers of their respective clusters, and set it as the clustering error SSE; if the SSE between two adjacent iterations is less than the set convergence threshold, terminate the iteration; otherwise, continue iterating to obtain the final K cluster centers.

[0061] Step 2.4: The final K cluster centers are used as the output curves of K typical scenarios, and the number of time series data belonging to each cluster is counted.

[0062] Step 3: Based on the output curves of typical scenarios, dynamically adjust the load imbalance weight and system active power loss weight under different scenarios to construct a power grid optimization model for the distribution network.

[0063] Step 3.1, define the objective function.

[0064] In the grid partitioning process of distributed generation, the optimization objectives are to minimize system active power loss and load imbalance. To comprehensively consider the impact of both during grid partitioning, the system active power loss needs to be normalized to a baseline value. Therefore, the overall objective function is to minimize the weighted sum of the normalized active power loss and the load imbalance. , can be represented as:

[0065] In the formula, It is a set of output curves for K typical scenarios; It is the frequency of each typical scenario s throughout the year; The total active power loss of the distribution network system under typical scenario s is expressed by the following formula:

[0066] In the formula, O is the number of grid cells. The set of branches within grid o Indicates a branch The resistance, This represents the branch through which the flow occurs at time t in scenario s. Current amplitude, node They are upstream and downstream nodes of the tidal current, connected by branch lines. Connecting two nodes, the system's active power loss depends on the thermal effect of the resistance on all branches in the system, where T is the total number of time points.

[0067] This is the baseline value for the active power of the distribution network system under typical scenario s; The load imbalance between meshes is defined as the load imbalance between meshes under a typical scenario s. It can be measured by the sum of the differences between the load rate of each grid and the average load rate of all grids, expressed by the following formula:

[0068] In the formula, For the total set of nodes, The set of nodes within grid o; For the scene Down The active power of the electrical load and electric vehicle charging load at time node p; This represents the total active power load of the system.

[0069] Calculate the photovoltaic penetration rate index for each scenario s. and load fluctuation index It can be expressed by the following formula:

[0070] In the formula, This represents the photovoltaic active power output at node p' at time t in scenario s. This represents the active power of charging the electric vehicle at time t at node p in scenario s. For the scene Down The active power of the electrical load at time node p' Let represent the total active power load of the system at time t under scenario s. This represents the average total active power load of the system under scenario s.

[0071] The scenario category factor is calculated based on the indicators and expressed by the following formula:

[0072] In the formula, Represents the scene category factor. This represents the baseline value for photovoltaic penetration rate, which is the average value of historical photovoltaic penetration data. The baseline value for the load volatility index is the average value of historical load volatility data. Indicates the sensitivity coefficient of photovoltaic penetration. This represents the load fluctuation sensitivity coefficient.

[0073] The loss weight is dynamically adjusted based on the scenario category factor. and imbalance weight It can be expressed by the following formula:

[0074] In the formula, and This represents the minimum and maximum values ​​of the system's active power loss weighting coefficient. , This represents the weight of load imbalance in scenario s. This represents the active power loss weight of the system under scenario s. ,when In scenarios with relatively high photovoltaic penetration and high load fluctuations, Increasing and optimizing tend to improve load balancing and voltage quality; when When smaller, Increased capacity and optimized features focus more on reducing network losses.

[0075] Step 3.2: Define constraints, including topology constraints, distributed power output constraints, and power flow constraints.

[0076] To ensure that the distribution network retains its radial structure after reconfiguration, a set of constraints based on single-product network flow is adopted to ensure that each divided grid is a connected, acyclic radial network. The topological constraints are expressed by the following formula:

[0077] In the formula, This indicates whether node p has been assigned to grid o. This indicates that node p is assigned to grid o; otherwise, node p is not assigned to grid o. This represents the closed state of branch pq, where =1 indicates closed and =0 indicates open. L represents the set of all branches in the entire system. This represents the virtual flow through branch qp in grid m; Represents the root node. This is the set of root nodes in the spanning tree. A virtual power flow variable is introduced. This ensures that each grid is a spanning tree rooted at the substation. Equation (22) states that each load node must belong to one and only one grid. Equation (23) states that the two nodes connected by a closed branch must belong to the same grid. Equation (24) states that each node in each grid becomes the sink point of a unit virtual flow. This constraint ensures that there must be a reachable path from the root node to each node, thus guaranteeing connectivity. Equation (25) states that for any grid k, the total amount of virtual flow generated by its root node must be equal to the number of nodes contained in that grid. Equation (26) ensures that the transmission path of the virtual flow must be a subset of physically connected branches.

[0078] Distributed power generation output constraints include photovoltaic constraints, energy storage charging and discharging constraints, power flow constraints, system security constraints, and switching frequency constraints.

[0079] The photovoltaic constraint is expressed by the following formula:

[0080] In the formula, and These represent the active and reactive power outputs of the photovoltaic system at node p at time t in scenario s. and They represent The upper and lower limits, and They represent The upper and lower limits.

[0081] The charge and discharge constraints of energy storage are expressed by the following formula:

[0082] In the formula: For the scene Down The active power output of the energy stored at time node p , Scenes Down The discharge power and charging power of the stored energy at time node p; For the scene Down State of charge of energy stored at time node p , These are the charging and discharging efficiencies, respectively. and express The upper and lower limits, express The upper limit, express The upper limit, expressed by equation (29), is used to calculate the net active power output of energy storage; equation (30) is used to calculate... The state of charge of the stored energy at any time; Equations (31) and (32) are used to constrain the charging and discharging power of the stored energy; Equation (33) is used to constrain the state of charge of the stored energy; Equations (34) and (35) are complementary constraints for the charging and discharging of the stored energy to prevent the stored energy device from charging and discharging at the same time.

[0083] Power flow constraints are expressed by the following formula:

[0084] In the formula, , , , Scenes Down The magnitude of active and reactive power flowing through lines pq and hp at any given time; , , , Scenes Down Active and reactive power output of electrical load and electric vehicle charging load at time node p; , Scenes Down The active and reactive power output of the photovoltaic system at time node p; and These are the resistance and reactance of line p and q, respectively; It is a constant. Equations (36) and (37) are used to calculate the active and reactive power flowing through the lines in the distribution network, respectively; Equation (38) is used to calculate the voltage of the nodes; Equations (39) and (40) ensure that power flows only through connected lines.

[0085] System security constraints are expressed by the following formula:

[0086] In the formula, This represents the voltage at node p. and Let represent the lower and upper limits of the allowable voltage at node p, respectively. Equation (41) is the allowable voltage deviation constraint for the distribution network node; Equation (42) constrains the power flowing through the line.

[0087] The constraint on the number of switching operations is expressed by the following formula:

[0088] In the formula, This indicates the initial state of the branch switch PQ, with 1 representing a closed state and 0 representing an open state. This indicates the maximum total number of allowed switching operations.

[0089] Introducing 0-1 variables Whether a node is connected to photovoltaics, the source-load zoning constraint, is expressed by the following formula:

[0090] In the formula, This indicates the state of node p being connected to a photovoltaic system. If node p is connected to a photovoltaic system, Otherwise, it is 0. Node scaling constraints are applied to each grid. Set a minimum number of source nodes.

[0091] The source node percentage constraint is expressed by the following formula:

[0092] In the formula, This indicates the minimum percentage of photovoltaic nodes set.

[0093] Step 4: Solve the power distribution grid optimization model to obtain the optimal grid partitioning scheme.

[0094] Step 4.1: Perform convex relaxation processing on the power distribution grid optimization model from Step 3 to obtain the Mixed Integer Second-Order Cone Programming (MISOCP) model.

[0095] This invention introduces a new set of variables to replace the variables in the original model, and uses a second-order cone relaxation technique to transform the model into a convex shape. The introduced new variables are shown in the following formula:

[0096] In the formula, This represents the squared current value of branch pq in scene s and time t. This represents the squared voltage value of node p at scene s and time t.

[0097] The active power loss equation (15) containing square terms and the non-convex DistFlow power flow constraint equations (31)-(33) in the original model, after second-order cone relaxation, are respectively shown in the following formulas:

[0098] In the formula, This represents the total active power loss of grid o in scenario s. This represents the squared current value of branch pq at scene s and time t.

[0099] Inequality (53) is a standard rotational second-order cone constraint. It can be written in standard second-order cone form as shown in the following formula:

[0100] In addition, the use of absolute value functions in the switching frequency constraints also requires linearization. The linearization result is shown in the following formula:

[0101] By processing the active power loss and nonlinear terms in the power flow constraints (31)-(33) of the objective function using equations (43) and (44)-(47), and replacing the line capacity constraint (37) with equation (48), the MINP model can be transformed into a mixed-integer linear programming (MILP) model. Therefore, the distribution network power grid partitioning method considering flexible resource clustering can be expressed as follows, and this model can be solved quickly using commercial software such as Gurobi.

[0102] Step 4.2: Solve the transformed convex optimization model to obtain the optimal mesh partitioning scheme.

[0103] Embodiment 2 of the present invention provides a power grid partitioning system for a distribution network that considers resource clustering, and implements the power grid partitioning method for a distribution network that considers resource clustering described in Embodiment 1, including: The data acquisition module is used to input the annual load data and electric vehicle behavior data into the photovoltaic power output model, energy storage charging and discharging model, and electric vehicle charging pile model to generate time series data. The scene acquisition module is used to cluster time series data using an improved K-means algorithm, determine the optimal number of clusters using the elbow method, and determine the output curve of a typical scene based on the optimal number of clusters. The model building module is used to dynamically adjust the load imbalance weight and system active power loss weight under different scenarios based on the output curves of typical scenarios, and build a power grid optimization model for the distribution network. The scheme acquisition module is used to solve the power supply grid optimization model of the distribution network and obtain the optimal grid partitioning scheme.

[0104] This invention, in its third embodiment, uses a 26-node distribution network in a region of East my country as an example, combining raw data on photovoltaic (PV), electricity load, and electric vehicle charging load over a 365-day, 24-hour period in the region for computational verification. The voltage level is 10kV, and the allowable fluctuation range of node voltage is 0.95 to 1.05 pu. The total active load is 4.316MW, the total reactive load is 3.710MVAr, the branch power limit is 8MW, and the maximum allowable transmission capacity of the main transformer is 50MVA. Simulation verification is performed on the scenario. To simulate the impact of flexible resources on grid partitioning, two configuration scenarios for PV, ES, and EVCS are set: Scenario A represents medium flexible resource penetration, and Scenario B represents high flexible resource penetration. The specific nodes accessed by flexible resources are shown in Table 1, and the system topology is as follows. Figure 2As shown in the table. Scenario A has 4 photovoltaic nodes, 2 energy storage nodes and 1 electric vehicle charging station, while Scenario B has 6 photovoltaic nodes, 4 energy storage nodes and 2 electric vehicle charging stations. The relevant parameters for photovoltaic, energy storage and electric vehicle charging stations are shown in Tables 2-4.

[0105] Table 1 Flexible Resource Access Nodes

[0106] Table 2. Parameters related to photovoltaic power supply

[0107] Table 2 (Continued) - Parameters Related to Photovoltaic Power Sources

[0108]

[0109] Table 3 Energy Storage Related Parameters

[0110] Table 4 Parameters of Electric Vehicle Charging Stations

[0111] To analyze the advantages of flexible resource clustering and grid partitioning methods in regional distribution network segmentation optimization, this invention designs and compares four optimization strategies under scenario A, with other parameters remaining the same. Strategy 1: No flexible resource access, no power grid partitioning; Strategy 2: Flexible resource access, no power grid partitioning; Strategy 3: No flexible resource access, power grid partitioning; Strategy 4: The method proposed in this invention. The objective function, total active power loss, and total load imbalance in the four sets of comparative examples are shown in Table 5. The comparative analysis of optimization results under different strategies is as follows: Figure 3 As shown.

[0112] Table 5 Comparison of Objective Functions under Different Strategies

[0113] Comparing Strategy 4 and Strategy 1, compared to traditional distribution network grid partitioning methods that do not consider flexible resource access or grid division, the distribution network grid partitioning method that considers flexible resource clustering in this example achieves significant optimization results. The comprehensive objective function value decreases from 0.8588 to 0.5529, a reduction of 35.6%. This is mainly due to the sharp decrease in total active power loss, from 94.1kW to 51.87kW, a reduction of 44.9%; the load imbalance is also significantly improved, decreasing from 0.7766 to 0.5871, a reduction of 24.4%. This result demonstrates that by optimizing the partitioning of the distribution network and coordinating the scheduling of flexible resources, node affiliation can be optimized, structural load balancing can be achieved, the power flow distribution of the system can be fundamentally changed, long-distance, high-load power transmission can be avoided, thereby greatly reducing network losses and improving the overall economic efficiency of the system.

[0114] Comparing strategies 2 and 1, without changing the network topology, introducing only flexible resources resulted in almost no change in total active power loss, while the total load imbalance improved by 13.5%. This demonstrates that the main role of flexible resources is to smooth load fluctuations and balance supply and demand, but under a fixed, suboptimal network structure, their ability to reduce line losses is limited.

[0115] Comparing strategy 3 and strategy 1, without accessing flexible resources, simply performing grid partitioning reduced total active power loss by 17.8% and total load imbalance by a significant 45.0%. This demonstrates that optimized network topology is a crucial means of improving distribution network performance, and a reasonable grid partitioning scheme can significantly reduce network losses and enhance load balancing benefits.

[0116] Comparing the total load imbalance of strategies 3 and 4 reveals that strategy 4 shows an increase in this metric compared to strategy 3, highlighting the complex trade-offs in multi-objective optimization. While the method of this invention, in pursuing the minimization of total active power loss, transfers some load from the main line to previously less loaded branches, resulting in a more rational load distribution between grids and significantly reducing losses, it may also lead to greater differences in line load rates within some grids. However, from the perspective of the comprehensive objective function value, strategy 4 remains the optimal solution among the three, indicating that the economic benefits of the significant reduction in total active power loss far outweigh the impact of the slight increase in load imbalance. This also verifies that the method proposed in this invention can find a comprehensive optimal solution that takes into account different optimization objectives.

[0117] To verify the effectiveness of the proposed collaborative optimization method considering both flexible resources and grid partitioning under different resource penetration rates, and to analyze the impact of grid partitioning granularity on the optimization results, this section designs two flexible resource access scenarios, Scenario A and Scenario B. In both scenarios, optimization solutions are obtained for strategies with no grid partitioning (K= / ), partitioning into 2 grids (K=2), and partitioning into 3 grids (K=3), respectively.

[0118] It should be noted that, considering the power distribution system in this example consists of three independent substation feeders, to ensure that each independently operating grid contains at least one main power source to achieve basic source-load balance, the maximum number of grids is set to 3. Further dividing the grid into 4 or more grids will inevitably create isolated grids without a main power source, failing to meet the constraints of independent operation; therefore, these are not considered effective strategies for comparison. The optimization results under each strategy are shown in Table 6, and the trends of each indicator with the number of grids K under different scenarios are as follows: Figure 4 As shown.

[0119] Table 6 Comparison of results under different resource penetration and grid partitioning strategies

[0120] In scenarios A and B, the three optimized meshing schemes achieved the best overall benefits compared to not meshing, consistent with the conclusions drawn from the method effectiveness analysis above. However, it is worth noting that an unreasonable meshing granularity (K=2) can actually worsen system performance. In both scenarios, the total active power loss and total load imbalance of the K=2 strategy are significantly worse than the baseline case without meshing. This indicates that blindly partitioning the network, if it fails to create an efficient and balanced power supply topology, can lead to performance degradation due to issues such as power flow detours and mesh load imbalances, further validating the necessity of optimizing the meshing strategy.

[0121] By comparing the results of scenario A and scenario B under the same K value, the impact of flexible resource penetration rate on system performance can be analyzed.

[0122] First, without grid partitioning, the total active power loss in scenario B is higher than in scenario A. This is because, under a fixed, suboptimal network topology, the addition of photovoltaic, energy storage, and charging loads alters the original power flow distribution. Especially during peak photovoltaic power generation periods, localized power may be fed back to upstream substations; and the added charging load may also exacerbate transmission pressure on certain branches. These new and complex power exchanges lead to an increase in overall system network losses. However, scenario B has a better overall load imbalance than scenario A. This is mainly due to the greater availability of flexible resources such as energy storage in scenario B. These resources effectively smooth out net load fluctuations and improve the system's balancing capability through peak shaving and valley filling scheduling behaviors.

[0123] Secondly, after adopting the optimal 3-grid partitioning, the optimization potential of high-penetration scenario B was fully realized. Its objective function and total load imbalance are significantly better than the corresponding values ​​in scenario A. Simulation results also show that the loss value in scenario B is not much lower than in scenario A. This may mean that under this network structure, the system's loss reduction capability has reached a bottleneck. With the weight settings, the optimization model tends to utilize the newly added flexible resources in scenario B to significantly improve load balance, rather than further reducing the already low loss level.

[0124] Compared with the prior art, the beneficial effects of the present invention include at least the following: This invention addresses the problem that existing distribution network planning technologies fail to accurately characterize the stochastic and dynamic characteristics of flexible resources such as photovoltaics, energy storage, and electric vehicles by modeling these resources' flexibility. Traditional methods often treat these resources as fixed loads or simple power sources, failing to reflect the uncertainties caused by natural conditions (sunlight, temperature) and user behavior (electric vehicle travel). This invention improves the accuracy and realism of the model's depiction of actual distribution network operation scenarios, reduces the risk of grid planning schemes deviating from actual operating conditions due to model simplification, and enhances the accuracy of distribution network grid planning.

[0125] This invention addresses the problem in existing research where the number of clusters (k) needs to be predetermined based on experience, lacking scientific basis, through an improved K-means clustering algorithm. Traditional methods often use fixed k values ​​that are too small, leading to insufficient representativeness of the scenario, or too large, resulting in excessive computational complexity. This invention enhances the adaptability and scientific rigor of typical scenario extraction by dynamically determining the optimal k value using the elbow method. This ensures that the scenario fully encompasses the diversity of the original data, reduces the complexity of power distribution grid planning, lowers the computational scale and solution time, and improves the accuracy and efficiency of power distribution grid planning. This invention uses a dynamic weight adjustment mechanism to automatically calculate and adjust the weight coefficients of network loss and load imbalance in the objective function for different typical scenarios, such as, but not limited to, high photovoltaic penetration and high load fluctuation scenarios. This enables the grid partitioning scheme to adapt to the main contradictions of different operating scenarios. In high fluctuation scenarios, it prioritizes load balance and voltage quality, while in stable scenarios, it focuses on reducing network loss for economic reasons. This solves the shortcomings of traditional static weights that cannot adapt to the operating needs of multiple scenarios and improves the robustness and adaptability of the partitioning scheme. This invention addresses the problem of existing grid partitioning methods that separate network topology optimization from flexible resource scheduling by constructing a power distribution network grid optimization model. Traditional techniques either optimize the topology or only schedule resources, failing to systematically improve distribution network performance through the coordinated use of resource scheduling and network reconfiguration. This invention improves the overall economy of distribution network grid planning by minimizing active power losses and enhances its security by improving load balancing. It reduces the risk of system network losses and equipment overloads caused by source-load mismatch and unreasonable network structure, and coordinates grid partitioning and flexible resource optimization to improve the security and performance of distribution network grid planning.

[0126] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.

[0127] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. A power distribution network power grid partitioning method considering resource cluster clustering, characterized in that: input the annual load data and electric vehicle behavior data into the photovoltaic power output model, energy storage charging and discharging model and electric vehicle charging pile model to generate time series data; use an improved K-means algorithm to cluster the time series data, determine the optimal number of clusters by the elbow method, and determine the output curve of the typical scene according to the optimal number of clusters; based on the output curve of the typical scene, dynamically adjust the load imbalance degree weight and system active power loss weight under different scenes, and construct a power distribution network power grid optimization model; solve the power distribution network power grid optimization model to obtain the optimal grid partitioning scheme.

2. The power distribution network power grid partitioning method considering resource cluster clustering according to claim 1, characterized in that: the photovoltaic power output model is represented by the following formula: In the formula, represents the rated output power of a single photovoltaic array; represents the installed capacity of a photovoltaic array; represents the solar radiation hourly average value at the latitude , m month, d day, h time; represents the irradiance intensity under standard test conditions; represents the power temperature coefficient of a photovoltaic array; represents the actual operating temperature of a photovoltaic array; represents the reference temperature.

3. The power distribution network power grid partitioning method considering resource cluster clustering according to claim 2, characterized in that: the energy storage charging and discharging model is represented by the following formula: In the formula, , These represent the electricity purchased and sold between the photovoltaic-storage charging station and the power grid at time t, respectively. , Let represent the charging and discharging power of the energy storage system at time t, respectively; , These represent the actual power values ​​of the conventional electrical load and the electric vehicle charging load at time t, respectively. Conventional loads include, but are not limited to, lighting and air conditioning. Let t be the actual active power generated by the photovoltaic system. The energy storage system operation constraints include power constraints, state of charge constraints, charging and discharging capacity balance constraints and charging and discharging state transition constraints.

4. The power distribution network power grid partitioning method considering resource cluster clustering according to claim 3, characterized in that: the electric vehicle charging pile model is represented by the following formula: wherein denotes the probability density of the electric vehicle arriving at the electric vehicle charging column at time t, denotes the mean value of the time of the electric vehicle arriving at the electric vehicle charging column, denotes the standard deviation of the time of the electric vehicle arriving at the electric vehicle charging column.

5. The power distribution network power grid partitioning method considering resource cluster clustering according to claim 1, characterized in that: determining the output curve of the typical scene according to the optimal number of clusters includes: calculating the clustering error SSE corresponding to different clustering numbers k by the elbow method, and selecting the k value corresponding to the inflection point of the SSE curve as the optimal number of clusters K; based on the optimal number of clusters, the initial clustering centers of the optimal number of clusters are determined by iteratively selecting the remaining time series data with the maximum distance sum from all selected clustering centers; calculate the distance between each time series data and all initial clustering centers, and assign it to the cluster to which the nearest initial clustering center belongs; calculate the mean vector of all time series data in each cluster and set it as the new clustering center of the cluster; calculate the distance sum of squares of all time series data to the clustering center of the cluster to which it belongs and set it as the clustering error SSE; if the SSE between the adjacent two iterations is less than the set convergence threshold, terminate the iteration; otherwise, continue the iteration to obtain the final K clustering centers; the final K clustering centers are taken as the output curves of the K typical scenes, and the number of time series data belonging to each cluster is counted.

6. The power distribution network power grid partitioning method considering resource cluster clustering according to claim 1, characterized in that: Objective function of a power distribution grid optimization model is expressed by the following equation: wherein is a set of output curves of K typical scenarios; is the frequency of each typical scenario s in a year; is the total active power loss of the distribution grid system under the typical scenario s; is the reference value of the active power of the distribution grid system under the typical scenario s; is the load imbalance degree between grids under the typical scenario s, denotes the load imbalance degree weight under the scenario s, denotes the system active power loss weight under the scenario s, .

7. The power distribution network power grid partitioning method considering resource cluster clustering according to claim 6, characterized in that: represents the load imbalance degree weight under the scenario s, represents the system active power loss weight under the scenario s, and is expressed in the following formula: In the formula, and represent the minimum and maximum values of the system active power loss weight coefficient, , represents the scene category factor.

8. The power distribution network power grid partitioning method considering resource cluster clustering according to claim 7, characterized in that: representing a scene category factor, expressed as follows: wherein, represents a reference value for the photovoltaic penetration index, represents a reference value for the load fluctuation index, represents a photovoltaic penetration sensitivity coefficient, represents a load fluctuation sensitivity coefficient, represents the photovoltaic penetration index for each scenario s, represents the load fluctuation index for each scenario s.

9. The power distribution network power grid partitioning method considering resource cluster clustering according to claim 8, characterized in that: The photovoltaic penetration index is calculated for each scenario s and the load volatility index is expressed as follows: wherein, represents the photovoltaic active power output at time t at node p' for scenario s, represents the charging active power of the electric vehicle at time t at node p for scenario s, is the scenario under the active power of the electricity consumption load at time t at node p' for scenario s, represents the total active power load of the system at time t for scenario s, represents the average total active power load of the system for scenario s, T being the total number of times.

10. A power distribution network power supply grid partitioning system considering resource cluster clustering, a power distribution network power supply grid partitioning method according to any one of claims 1-9 is run, characterized in that: a data acquisition module is used to input the annual load data and electric vehicle behavior data into the photovoltaic power output model, the energy storage charging and discharging model and the electric vehicle charging pile model to generate time series data; a scene acquisition module is used to cluster the time series data by using an improved K-means algorithm, determine the optimal clustering number by the elbow method, and determine the output curve of the typical scene according to the optimal clustering number; a model construction module is used to dynamically adjust the load imbalance degree weight and the system active power loss weight under different scenes based on the output curve of the typical scene, and construct a power distribution network power supply grid optimization model; a scheme acquisition module is used to solve the power distribution network power supply grid optimization model to obtain an optimal grid partitioning scheme.

Images