Power distribution network flexibility resource dynamic aggregation and finite time layering optimization control method

Abstract

The invention belongs to the technical field of power system operation control, and discloses a power distribution network flexibility resource dynamic aggregation and finite time hierarchical optimization control method, which comprises the following steps: constructing a power distribution network distributed resource cluster model, clustering clusters based on different clustering indexes, analyzing the adjustable potential of each cluster, and establishing a distributed resource cluster model; calculating and quantifying scale parameters of the cluster; performing dynamic aggregation according to complementarity of different resource clusters by applying a panoramic theory; performing hierarchical optimization on the cluster according to an aggregation result, constructing an objective function and a constraint condition, and solving a hierarchical optimization model by adopting a particle swarm algorithm; a performance algorithm is preset through finite time, dynamic parameters are introduced into a particle swarm algorithm, dynamic constraint is carried out on system state errors, decision time is further shortened, and control performance is improved. According to the method, accurate clustering of different distributed resources can be realized, the cluster aggregation optimization operation efficiency of the power distribution network is effectively improved, and flexible resource planning is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of power system operation control, and specifically relates to a method for dynamic aggregation and finite-time hierarchical optimization control of distribution network flexibility resources. Background Art

[0002] The integration of a large number of distributed resources and the penetration of user-generated energy (UGE) have dramatically transformed the structure and operational characteristics of distribution networks, making them large and complex. The intermittent and fluctuating nature of these distributed resources significantly differs from traditional centralized power generation models, posing numerous challenges to the stable operation, power balance control, and power quality assurance of distribution networks. Traditional grid management models are difficult to adapt to the multi-agent management of the new power system. Simulation and deduction based on complex system principles are needed to optimize the utilization of flexible resources.

[0003] From the perspective of practical application, on the one hand, the large-scale access of distributed resources makes the distribution of distribution network power flow more complex, and traditional power grid management models find it difficult to effectively manage and coordinate the numerous distributed resources; for example, when light or wind conditions change drastically, the output fluctuations of photovoltaic power stations and wind farms will lead to unstable distribution network voltage, affecting users' electricity consumption experience and may even threaten the security of the power grid; on the other hand, the widespread application of UGE has made user-side electricity consumption behavior and energy interaction more diversified. Traditional management models are difficult to adapt to the management needs of such multiple subjects and cannot achieve efficient use of flexible resources; for example, the disorderly charging of electric vehicles will increase the load pressure on the power grid during peak hours, while its discharge potential has not been fully tapped.

[0004] In terms of optimal allocation of flexible resources in distribution networks, existing research has primarily focused on static, unilateral aggregation of distributed energy resources, lacking dynamism and compatibility with actual market scenarios. Existing methods have shortcomings in hierarchical optimization and allocation, particularly in the comprehensive consideration of dynamic aggregation and coordinated interaction of flexible resources. Furthermore, existing research has shortcomings in the timeliness of optimization algorithms. Although some studies have attempted to introduce intelligent optimization algorithms, their computational efficiency still struggles to meet real-time control requirements. The optimization process is time-consuming and converges slowly, leading to delayed decision-making and an inability to adapt to the highly dynamic and responsive operating scenarios of new distribution networks.

[0005] In summary, the existing power system operation control methods are difficult to meet the requirements of the new distribution network in terms of high reliability, high flexibility and high timeliness. Summary of the Invention

[0006] To solve the above problems, the present invention provides a method for dynamic aggregation and finite-time hierarchical optimization control of distribution network flexibility resources, so as to solve the shortcomings of the existing technology in terms of dynamic aggregation of distributed resources, real-time performance of hierarchical optimization and computing efficiency, and improve the management efficiency and control performance of distribution network flexibility resources.

[0007] The present invention provides a method for dynamic aggregation and finite-time hierarchical optimization control of distribution network flexibility resources, comprising the following steps:

[0008] Construct a multi-cluster model of a distributed resource distribution network that includes photovoltaic power stations, wind farms, energy storage, and electric vehicles. Based on the operating characteristics of photovoltaic power stations, wind farms, energy storage systems, and electric vehicles, extract key features (such as daily power curve average, charge and discharge parameters, charging behavior, etc.), and use the K-means++ clustering algorithm to accurately cluster distributed resources to form distributed photovoltaic clusters, wind power clusters, energy storage clusters, and electric vehicle clusters. By constructing an adjustable power calculation model for each cluster, quantify the cluster's scale parameters and adjustable potential. Analyze the adjustable potential of each cluster and calculate and quantify the cluster's scale parameters.

[0009] Applying panoptic theory to analyze the complementarity of different resource clusters, we define matching degree and system energy models, achieve optimal dynamic aggregation by minimizing system energy, establish a loss calculation formula to measure the rationality of resource grouping, introduce complementary parameters, and distinguish the synergistic effects between different resource types. This allows highly matching resource clusters to operate in aggregate, reducing overall system energy and achieving efficient resource collaboration.

[0010] A two-layer coordinated planning model is used. The upper layer optimizes the capacity and location of flexible resources to minimize investment costs, while the lower layer optimizes resource output to minimize operating costs and voltage offsets. A particle swarm algorithm is used to achieve hierarchical solutions, design time-varying inertia weights and learning factors, balance global search and local convergence, and introduce dynamic penalty terms to accelerate the optimization process and ensure constraint satisfaction.

[0011] An asymmetric adaptive finite-time performance function is designed to constrain the system state error to converge to an allowable range within a preset time window. A time penalty term is embedded in the objective function to force the optimization process to be completed within a finite time. The algorithm parameters are dynamically adjusted to improve the convergence speed and decision timeliness to meet real-time control requirements.

[0012] The beneficial effects described in the present invention are as follows: the method described in the present invention extracts key operating characteristics based on the data characteristics of different resource entities, combines the K-means++ clustering algorithm with the silhouette coefficient evaluation, and realizes the accurate clustering and grouping of different distributed resources; the system energy model based on the panoramic theory fully considers the complementarity of different resource clusters, dynamically updates the matching parameters and complementarity weights according to the real-time operating status of the resource cluster, realizes dynamic aggregation, and improves the adaptability and stability of the aggregate through hierarchical optimization, achieving a major breakthrough in the traditional static aggregation method; introduces asymmetric adaptive performance functions and time penalty terms, dynamically constrains the system state error trajectory, combines dynamic parameter adjustment with time-varying strategies, improves the inertia weight and learning factor of the particle swarm algorithm, significantly improves the convergence speed and decision-making efficiency of the algorithm, meets the real-time operation decision-making needs of the power grid, makes up for the shortcomings of existing research in time performance, and provides a practical solution for the efficient operation of the new distribution network. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 is a flow chart of the method of the present invention;

[0014] Figure 2 It is a finite time hierarchical optimization solution flow chart;

[0015] Figure 3 This is a diagram showing the effect of a finite time preset performance algorithm. DETAILED DESCRIPTION

[0016] In order to make the contents of the present invention more clearly understood, the present invention is further described in detail below based on specific embodiments in conjunction with the accompanying drawings.

[0017] like Figure 1 As shown, the method for dynamic aggregation and finite-time hierarchical optimization control of distribution network flexibility resources described in the present invention includes the following steps:

[0018] Step 1: Build a multi-cluster model for a distributed resource distribution network that includes photovoltaic power plants, wind farms, electric vehicles, and energy storage. Cluster the clusters based on different clustering indicators, analyze the adjustable potential of each cluster, and calculate and quantify the cluster's scale parameters.

[0019] Step 2: Based on the calculation and quantification results, apply panopticon theory to dynamically aggregate different resource clusters according to their complementarity;

[0020] Step 3: Perform hierarchical optimization on the clusters based on the aggregation results, construct the objective function and constraints, and use the particle swarm algorithm to solve the hierarchical optimization model;

[0021] Step 4: Based on the solution of step 3, a finite time preset performance algorithm is used and dynamic parameters are introduced into the particle swarm algorithm to dynamically constrain the system state error, further shorten the decision time and improve the control performance.

[0022] Specifically, in step 1, a multi-cluster model of the distribution network including distributed resources such as photovoltaic power stations, wind farms, electric vehicles, and energy storage is constructed, and clusters are clustered based on different clustering indicators:

[0023] The K-means++ clustering algorithm is used to cluster wind farms, photovoltaic power stations, energy storage systems, and electric vehicles based on wind power-related data, photovoltaic-related data, energy storage-related data, and electric vehicle-related data, resulting in distributed wind power clusters, distributed photovoltaic power clusters, distributed electric energy storage systems, and electric vehicle clusters. Specifically:

[0024] (1) Extract key features from different resources and construct a standardized feature matrix:

[0025] For wind farms and photovoltaic power stations, the mean, standard deviation, peak-valley period ratio and other characteristics of the daily power curve are selected as the criteria for cluster division; for energy storage, the rated capacity, maximum charge and discharge power, charge and discharge power and other characteristics are selected as the criteria for cluster division; for electric vehicles, the average daily number of charging times, single charging time, and charging amount are selected as the criteria for cluster division.

[0026] (2) Standardize features of different dimensions (Z-score normalization):

[0027]

[0028] Among them, z i is the standardized eigenvalue, x i is the original eigenvalue, μ is the characteristic mean, and σ is the standard deviation.

[0029] (3) Perform K-means++ clustering, and the objective function is to minimize the sum of squared Euclidean distances from samples within the cluster to the centroid;

[0030]

[0031] Among them, K is the number of clusters, x is the sample value, C k is the kth cluster, μ k is the centroid of the kth cluster;

[0032] The K-means++ clustering steps are:

[0033] Step 1: Initialize the centroid: Randomly select the first centroid μ1, and the subsequent centroids with probability D(x) 2 / ∑D(x)2 Select, where D(x) is the shortest distance from sample x to the selected centroid, avoiding the initial centroid being too close;

[0034] Step 2: Assign samples: Assign each sample to the cluster corresponding to the nearest centroid:

[0035]

[0036] Among them, x is the sample value, C k is the kth cluster, μ k is the centroid of the kth cluster, μ j is the centroid of the jth cluster;

[0037] Step 3: Update the centroid: Calculate the new centroid of each cluster:

[0038]

[0039] If the centroid change is less than the threshold or the maximum number of iterations is reached, the process ends; otherwise, return to Step 2.

[0040] (4) The degree of separation between a sample and its own cluster and other clusters is measured according to the silhouette coefficient. The calculation formula of the silhouette coefficient s(i) is:

[0041]

[0042] Among them, a(i) is the average distance between sample i and other samples in the same cluster, and b(i) is the average distance between sample i and the nearest other cluster samples. The closer the silhouette coefficient s(i) is to 1, the better the clustering effect.

[0043] Furthermore, in step 1, the adjustable potential of each cluster is analyzed, and the scale parameters of the cluster are calculated and quantified:

[0044] Based on the operating mechanisms and characteristics of distributed power clusters, distributed electric energy storage systems, and electric vehicle clusters, we construct the adjustable power calculation model for distributed power clusters, the adjustable power calculation model for distributed electric energy storage systems, and the adjustable power calculation model for electric vehicle clusters. The distributed power clusters include distributed photovoltaic power clusters and distributed wind power clusters.

[0045] The calculation model of the adjustable power of distributed photovoltaic power cluster is:

[0046]

[0047] Where ΔP PV Indicates the adjustable power of the photovoltaic cluster, n PV represents the number of photovoltaic units, Pp,i,max(t+1) represents the predicted maximum power of photovoltaic unit i at the next moment, P p,i(t) represents the output power of photovoltaic unit i under the current conditions, P N,p,i Indicates the rated power of photovoltaic unit i;

[0048] The calculation model of the adjustable power of distributed wind power cluster is:

[0049]

[0050] Where ΔP WT Indicates the adjustable power of the photovoltaic cluster, n WT represents the number of wind turbines, Pp,i,max(t+1) represents the predicted maximum power of photovoltaic unit i at the next moment, and P p,i (t) represents the output power of photovoltaic unit i under the current conditions, P N,p,i Indicates the rated power of photovoltaic unit i;

[0051] The calculation model of the adjustable power of the distributed electric energy storage system is:

[0052]

[0053] Where, Indicates the power of the energy storage load group before regulation, Indicates the power of the energy storage load group after regulation; the power of the N energy storage charging and discharging group is:

[0054]

[0055] Where k j (t) represents the operating status of energy storage system j, represents the charging and discharging rated power of energy storage system j;

[0056] The calculation model of the adjustable power of electric vehicle cluster is:

[0057]

[0058] Where n EV represents the number of electric vehicles, P EV,c,i (t) represents the charging response potential of user i at time t, P EV,d,i (t) represents the discharge response potential of user i at time t.

[0059] Furthermore, in step 2, based on the operating status of the distributed resource system and considering the optimal aggregation mode of scheduling resources to participate in demand response, the following distributed resource cluster optimization scheduling strategy based on panoptic theory is proposed:

[0060] Establish the loss calculation formula:

[0061]

[0062] In the formula, X represents the grouping situation; s j represents the scale parameter of the jth individual; p ij represents the matching degree between individuals i and j; d ij Represents the distance between individuals i and j. The physical meaning of the above model is: when two individuals with a large matching degree are not in the same group, the loss of the group will increase; otherwise, the loss of the group will decrease.

[0063] From this, the system energy model can be defined:

[0064]

[0065] Combining the above two equations, we can get the system energy in the case of X groups:

[0066]

[0067] The above equation shows that system energy is determined by the individual scale parameters, the matching degree of the two resource clusters operating together, and the distance between individual groups. Panoramic theory groups members based on the actions taken by each member to reduce their own loss. Only one member is allowed to move to another group at a time, and this action reduces its own loss. If one member's loss decreases, the loss of other members will also decrease, thereby reducing the overall energy; conversely, if the overall energy decreases, the loss of each member will also decrease.

[0068] When two individuals with high matching degrees are in the same group and two individuals with low matching degrees are not in the same group, the energy of the system is lower. When the system energy reaches the minimum value, it is the best grouping situation.

[0069] In a specific embodiment, it is assumed that the set F a ={a1,a2,…,α,…,a m} and set F b ={b1,b2,…,β,…,b n} represents two groups of aggregations, aggregation a and aggregation b, where there are m clusters in aggregation a and n clusters in aggregation b. m and b1,b2,…,b n A value of α, β is 1, 2, 3 or 4. A value of 1 indicates a photovoltaic cluster, a value of 2 indicates a wind power cluster, a value of 3 indicates an energy storage cluster, and a value of 4 indicates an electric vehicle cluster. For example, F a ={1,1,2,3,4,4}, indicating that there are two photovoltaic clusters, one wind power cluster, one energy storage cluster, and two electric vehicle clusters in group a.

[0070] For β∈F b , define sa (β) is the attraction of cluster β in cluster b to cluster a. Considering the complementarity between different resource clusters, s a (β) is calculated as follows:

[0071]

[0072] Specifically, the parameters s are adjusted according to the complementarity of different distributed power sources. α Setting of (β):

[0073]

[0074] The meaning of the numerical representation of the above formula is: when β≠α, different cluster types have complementary properties, β is positive to α, that is, β is willing to work with α, so s α (β)>0; when β=3, β is the energy storage cluster. Since the energy storage output constraint is less and the scheduling is convenient, other clusters have greater expectations for it and the complementarity is more obvious, so s α (β) = +2. When β = α, the clusters are identical and do not have complementarity. When they are all photovoltaic, wind power, and electric vehicles, their randomness increases, making the aggregate output volatility increase. At this time, setting s α (β) = -3, indicating that β is extremely unwilling to work with α; when they are both energy storage, their desire for each other is in a balanced state, and whether they work together or not has little impact on the overall situation, so setting s α (β)=0. Therefore, s α The setting of (β) reflects the strength of complementarity between different resources.

[0075] The complementarity of each cluster determines the matching degree between aggregations, so the matching degree between aggregations a and b can be obtained:

[0076]

[0077] Through the definition of the above parameters, the panoramic theory energy function is used to solve the optimal grouping, that is, the multi-cluster aggregation operation status.

[0078] The present invention obtains the operating status of the resource cluster in real time, periodically updates the matching parameters and complementarity weights, generates a grouping scheme based on the above-mentioned panoramic theoretical energy model, iteratively adjusts the grouping of the resource cluster, gradually reduces the system energy, and selects the group with lower energy as the current optimal dynamic aggregation result.

[0079] Furthermore, in step 3, the clusters are hierarchically optimized based on the aggregation results, and the objective function and constraints are constructed:

[0080] Overall, a two-level coordinated planning approach is employed, primarily divided into an upper-level investment decision module and a lower-level production and operation module. The entire planning model transforms the flexibility resource planning problem into a two-level optimization problem. The upper-level optimization objective is to minimize total investment costs, while the lower-level optimization objectives include minimizing operating costs and voltage offsets. In the lower-level production and operation model, the output of various resources is obtained, and calculations are performed based on this information to obtain relevant constraint indicators. These indicators are then fed back into the upper-level investment decision model for constraint verification.

[0081] The optimization goal of the upper layer problem is to minimize the total investment cost, which includes the investment cost of flexibility resources, as well as the operating costs, network loss costs, and wind and solar power curtailment costs at the lower layer. The objective function is:

[0082] minC=C Buy +C DG-cons +C DG-oper +C Loss +C Waste

[0083] Where C Buy is the cost of power purchase from the upper grid by the distribution network, C DG-cons 、C DG-oper Flexibility

[0084] The power supply is converted into annual construction cost and annual operation and maintenance cost, C Loss is the network loss cost, C Waste is the cost of curtailing wind and solar power. Each part is calculated as follows:

[0085] Distribution network operating costs (power purchase costs):

[0086]

[0087] Where N s Indicates the number of typical scenes; t s represents the number of time periods in a typical scenario; P sub,t,s is the power purchased by the distribution network from the upper power grid at time t in the typical scenario s; t is the electricity purchase price at time t.

[0088] Flexible power supply construction cost:

[0089]

[0090] Where r is the discount rate (fixed interest rate); n represents the total planning period; N pv and N wt , N st Represents the number of installed photovoltaic, wind power and energy storage respectively; C pv and C wt , Cst are the unit capacity investment costs of photovoltaic power, wind power and energy storage respectively; P PVi and P WTi , P STi is the installed capacity of photovoltaic and wind power installed at node i.

[0091] Flexible power supply operating costs:

[0092]

[0093] Where, P PVi,t,s 、P WTi,t,s 、P STi,t,s They represent the output of the i-th photovoltaic, wind turbine, and energy storage unit at time t under scenario s, and τ represents their respective operating cost prices.

[0094] Distribution network loss costs:

[0095]

[0096] Where C Loss is the network loss cost per unit electricity; P Loss,t,s is the total active network loss of the system at time t in the typical scenario s.

[0097] Cost of curtailing wind and solar power:

[0098]

[0099] Where C Waste P is the cost of curtailing wind and solar power per unit of electricity; Waste,t,s is the total amount of wind and solar power curtailment in the system at time t in a typical scenario s.

[0100] The constraints are the flexibility resource capacity constraint and the flexibility resource installation node constraint, and the mathematical model is:

[0101]

[0102] Where N X Indicates the node location where the flexible resource is to be installed. It must be on the installable node N. i Inside; P i,x The installable capacity must not exceed the maximum installation capacity.

[0103] The underlying problem is based on a time series simulation model, with operating costs and voltage offset as targets, including wind and solar curtailment costs, operation and maintenance costs, power purchase costs, and network loss costs. The objective function is shown below:

[0104]

[0105] Where Vi Represents the voltage value of node i, V i,n Represents the rated value of node i.

[0106] The main constraints include photovoltaic mechanism constraints, wind turbine mechanism constraints, energy storage mechanism constraints, power flow constraints, voltage constraints, and point of common coupling (PCC) power fluctuation constraints.

[0107] Photovoltaic mechanism constraints:

[0108]

[0109] Where, P PV (t) is the output power of the photovoltaic power generation system at time t, k T is the temperature coefficient, P SET is the rated output power of the photovoltaic power generation system, S SET is the standard light intensity of the photovoltaic power generation system, S(t) is the light intensity of the photovoltaic power generation system at time t, T(t) is the ambient temperature at time t, T SET is the standard ambient temperature.

[0110] Mechanical constraints of wind turbines:

[0111] The key factor affecting the power of a wind turbine is the real-time wind speed. To ensure the safe operation of the wind turbine, it is necessary to set an upper limit on the wind speed that the blades can withstand. The functional relationship between the output power of a wind turbine and the wind speed is:

[0112]

[0113] Where, P rated is the rated power of the fan, v is the real-time wind speed, v in is the cut-in wind speed, v rated is the rated wind speed, v out is the cut-out wind speed. in When the wind speed is low, the fan has no power output. in ≤ν≤ν rated When ν rated ≤ν≤ν out The wind turbine still maintains rated power output. If the wind speed continues to increase and exceeds the cut-out wind speed, the wind turbine will exit operation and be disconnected from the grid to avoid accidents.

[0114] Energy storage mechanism constraints:

[0115] Taking batteries as the research object of energy storage systems, the energy storage system (ESS) is charged and discharged according to the system's scheduling instructions. Its charging and discharging power is related to the system's remaining power and charging and discharging efficiency. In addition, the power of the ESS involves temporal coupling. The power of the energy storage device at adjacent moments satisfies the following relationship (ignoring the self-discharge coefficient):

[0116]

[0117] Among them, P ch,i,t , P dch,i,t , are the charging and discharging powers of the energy storage device on node i at time t, respectively, and should meet the charging and discharging power constraints per unit time, namely:

[0118]

[0119] Where, E i,t 、E i,t+1 are the remaining power of the energy storage device on node i at time t and time t+1 respectively; η ch ,η dch are the charging and discharging efficiencies of the energy storage device respectively; Δt is the scheduling time interval; B ESS is the set of nodes equipped with energy storage devices; α ch,i,t , α dch,i,t are 0-1 variables, representing the charging and discharging states of the energy storage device respectively.

[0120] To ensure that the energy storage device cannot be charged and discharged at the same time, add the following constraints:

[0121]

[0122] To ensure that the energy storage system operates normally and prevent overcharging or discharging, a constraint on the remaining energy storage capacity should be added:

[0123]

[0124] Where, E max is the maximum capacity of the energy storage device.

[0125] In addition, to ensure the energy conservation of the energy storage device within the cycle, a periodic energy conservation constraint needs to be added:

[0126]

[0127] Where, E i,begin is the initial capacity of the energy storage device at node i; E i,end is the remaining power of the energy storage device at node i after one cycle of operation.

[0128] Power flow constraints:

[0129]

[0130] Where, P G,sub,i and Q G,sub,i P is the active power and reactive power transmitted by the substation during time t in scenario s; Li,t,s and Q Li,t,s G is the active load and reactive load power of node i in time period t at scene s; ij and B ij The equivalent conductance and susceptance of the line between nodes i and j.

[0131] Voltage Constraints:

[0132] V i min ≤V i,t,s ≤V i max i∈{1,2,...,N b}

[0133] Where V i max and V i min The upper and lower limits of the voltage at node i.

[0134] PCC power fluctuation constraint:

[0135]

[0136] Where, is the maximum value of PCC power fluctuation.

[0137] Furthermore, in step 3, the particle swarm algorithm is used to solve the hierarchical optimization model. The specific steps are as follows:

[0138] Step 1: Initialize the particle position and velocity, i.e., the position and capacity of the flexibility resources of the decision layer, as input to the operation layer; set parameters such as particle population size, maximum number of iterations, inertia weight, and learning factor;

[0139] Step 2: Using the optimization objectives and constraints of the operation layer as input, optimize the output of flexible resources in each period and provide feedback to the decision-making layer;

[0140] Step 3: The decision layer receives feedback from the operation layer, calculates the decision layer constraints and objective function, and optimizes and updates the location and capacity of the flexible resources;

[0141] Step 4: When the maximum number of iterations is reached or the rate of change of the objective function value is less than the threshold, the optimal resource allocation solution is output. Otherwise, the particle population is updated and the above steps are repeated.

[0142] Furthermore, in step 4, a finite time preset performance algorithm is used to constrain the system state error, further shorten the decision time, and improve the control performance:

[0143] Design an asymmetric adaptive finite-time performance function ρ(t) to constrain the system state error and ensure that it converges to the allowable range within time T, specifically:

[0144]

[0145] Where ρ0 is the upper limit of the initial error, which is determined based on the historical data of each cluster; ∞ is the final allowable steady-state error, which is set according to the grid demand; t∈[0,T] is a preset finite time window; n,m are positive real numbers that control the asymmetry of the convergence speed (usually n>m to accelerate the initial convergence).

[0146] In the specific implementation process, the controller input dead zone and actuator failure will significantly affect the system performance and the selection of the finite time performance function, so a flexible performance function is introduced. To balance the relationship between the system's input restrictions and output constraints:

[0147]

[0148] Where λ(t) is a continuous smooth function, satisfying 0≤λ(t)<ρ0+ρ ∞ -ρ(t), and λ(0)=0. Dynamically expand or shrink the error boundary to flexibly deal with uncertainties such as input dead zone and actuator failure, thereby improving system robustness. is the regulation function generated by the auxiliary system:

[0149]

[0150] Where, k1, k2 are set constants, k1>0 and k2>0; u i,a is the control input, The input threshold is set.

[0151] The system state x(t) (such as voltage deviation, power fluctuation) must satisfy

[0152]

[0153] That is, the error trajectory is constrained to the performance function In the envelope of desired is the expected system state.

[0154] A time penalty term is introduced into the objective function of the hierarchical optimization model to accelerate convergence. Specifically,

[0155]

[0156] Where, J new is the new objective function, J is the original objective function (such as total investment cost, operating cost), and γ is the penalty coefficient, which is dynamically adjusted to balance the convergence speed and optimization accuracy.

[0157] In the aforementioned particle swarm algorithm, the inertia weight factor and learning factor adjustment are introduced to improve it, and the following time-varying inertia weight and learning factor are designed:

[0158] ω(t)=ω initial ·e -εt

[0159]

[0160] Where ω(t) is the inertia weight, which decreases over time to accelerate convergence, ω initial is the initial value of the inertia weight, ε is the convergence coefficient; c1(t), c2(t) are learning factors that adjust the weights of individual and group experience.

[0161] Reference Figure 2 , the specific implementation steps of the algorithm are:

[0162] Step 1: Set the performance function parameters ρ0, ρ ∞ , n, m and optimization time T;

[0163] Step 2: Update the inertia weight ω(t) and learning factors c1(t), c2(t) in each iteration;

[0164] Step 3: Real-time calculation of system state error |x(t)-x desired |, if the performance function is exceeded Then increase the penalty item weight γ;

[0165] Step 4: When t≥T or state error |x(t)-x desired |≤ρ ∞ Output the optimization results.

[0166] The specific implementation effect of the algorithm is referenced Figure 3 , e1, e2, and e3 are the system state errors in three different scenarios respectively. are the upper and lower limits of the performance function. Figure 3 As shown in Figure 3, the error trajectory is constrained within the preset envelope and converges to a stable state within the preset time (5 minutes).

[0167] Through the above design, the optimization process is strictly limited to the preset time T, ensuring that the system achieves the required performance within the limited time T. Through the coordinated adjustment of parameters ω(t), c1(t), and c2(t), the algorithm widely explores the solution space in the early stage and converges in a refined manner in the later stage, avoiding premature convergence and reducing redundant calculations. The performance function ρ(t) forces the system to converge to the steady-state error ρ within the preset time T. ∞ , combined with dynamic parameter adjustment, it ensures that the optimization results meet both timeliness and reliability, significantly improving the dynamic performance of cluster aggregation operation, thereby shortening decision-making time and improving control performance.

[0168] The above description is only a preferred embodiment of the present invention and is not intended to further limit the present invention. All equivalent changes made using the contents of the present invention description and drawings are within the scope of protection of the present invention.

Claims

1. A method for dynamic aggregation and finite-time hierarchical optimization control of distribution network flexibility resources, characterized in that: The following steps are involved: Step 1: Build a multi-cluster model for a distributed resource distribution network that includes photovoltaic power plants, wind farms, electric vehicles, and energy storage. Cluster the clusters based on different clustering indicators, analyze the adjustable potential of each cluster, and calculate and quantify the cluster's scale parameters. Step 2: Based on the calculation and quantification results, apply panopticon theory to dynamically aggregate different resource clusters according to their complementarity; Step 3: Perform hierarchical optimization on the clusters based on the aggregation results, construct the objective function and constraints, and use the particle swarm algorithm to solve the hierarchical optimization model; Step 4: Based on the solution of step 3, a finite time preset performance algorithm is used and dynamic parameters are introduced into the particle swarm algorithm to dynamically constrain the system state error, further shorten the decision time and improve the control performance.

2. A distribution network flexibility resource dynamic aggregation and finite time hierarchical optimization control method according to claim 1, characterized in that: In step 1, the K-means++ clustering algorithm is used to cluster photovoltaic power stations, wind farms, energy storage systems, and electric vehicles based on the relevant data of photovoltaic, wind power, energy storage, and electric vehicles, respectively. The clusters are distributed photovoltaic power clusters, distributed wind power clusters, distributed electric energy storage systems, and electric vehicle clusters, including: Extract key features for different resources and construct a standardized feature matrix; Standardize features of different dimensions; The K-means++ algorithm is used to cluster the standardized features. The objective function is to minimize the sum of squared Euclidean distances from samples within a cluster to the centroid, so that resources with similar operating characteristics are clustered to the cluster corresponding to the nearest centroid. The degree of separation between the sample and its own cluster and other clusters is measured according to the silhouette coefficient, and the optimal number of clusters and clustering results are screened out.

3. A distribution network flexibility resource dynamic aggregation and finite time hierarchical optimization control method according to claim 2, characterized in that: Extract key features for different resources and build a standardized feature matrix, including: For wind farms and photovoltaic power stations, the mean, standard deviation, and peak-valley period ratio of the daily power curve are selected as the criteria for cluster division; For energy storage, rated capacity, maximum charge and discharge power, and charge and discharge power are selected as the criteria for cluster division; For electric vehicles, the average daily charging times, single charging time, and charging capacity are selected as the criteria for cluster division.

4. A distribution network flexibility resource dynamic aggregation and finite time hierarchical optimization control method according to claim 2, characterized in that: In step 1, the adjustable potential of each cluster is analyzed, and the cluster scale parameters are calculated and quantified, including: Based on the operating mechanisms and characteristics of distributed power clusters, distributed electric energy storage systems, and electric vehicle clusters, we construct the adjustable power calculation model for distributed power clusters, the adjustable power calculation model for distributed electric energy storage systems, and the adjustable power calculation model for electric vehicle clusters. The distributed power clusters include distributed photovoltaic power clusters and distributed wind power clusters. The calculation model of the adjustable power of distributed photovoltaic power cluster is: Where, ΔP PV Indicates the adjustable power of the photovoltaic cluster, n PV represents the number of photovoltaic units, Pp,i,max(t+1) represents the predicted maximum power of photovoltaic unit i at the next moment, P p,i (t) represents the output power of photovoltaic unit i under the current conditions, P N,p,i Indicates the rated power of photovoltaic unit i; The calculation model of the adjustable power of distributed wind power cluster is: Where, ΔP WT Indicates the adjustable power of the photovoltaic cluster, n WT represents the number of wind turbines, Pp,i,max(t+1) represents the predicted maximum power of photovoltaic unit i at the next moment, and P p,i (t) represents the output power of photovoltaic unit i under the current conditions, P N,p,i Indicates the rated power of photovoltaic unit i; The calculation model of the adjustable power of the distributed electric energy storage system is: Where, Indicates the power of the energy storage load group before adjustment. Indicates the power of the energy storage load group after regulation; the power of the N energy storage charging and discharging group is: Where k j (t) represents the operating status of energy storage system j, represents the charging and discharging rated power of energy storage system j; The calculation model of the adjustable power of electric vehicle cluster is: Where n EV represents the number of electric vehicles, P EV,c,i (t) represents the charging response potential of user i at time t, P EV,d,i (t) represents the discharge response potential of user i at time t.

5. A method for dynamic aggregation and finite-time hierarchical optimization control of distribution network flexibility resources according to claim 1, characterized in that: Step 2 is as follows: Based on the operating status of the distributed resource system and the optimal aggregation mode of scheduling resources to participate in demand response, the following distributed resource cluster optimization scheduling strategy based on panoptic theory is proposed: Establish the loss calculation formula: In the formula, X represents the grouping situation; s j represents the scale parameter of the jth individual; p ij represents the matching degree between individuals i and j; d ij represents the distance between individuals i and j; Based on the loss degree, a system energy model including matching degree is defined: Combining the above two equations, we can get the system energy in the case of X groups: When two individuals with high matching degrees are in the same group and two individuals with low matching degrees are not in the same group, the energy of the system is lower. When the system energy reaches the minimum value, it is the best grouping situation.

6. A distribution network flexibility resource dynamic aggregation and finite time hierarchical optimization control method according to claim 1, characterized in that: In step 3, the clusters are optimized hierarchically based on the aggregation results, and the objective function and constraints are constructed, including: Adopt a two-layer coordinated planning approach, divided into an upper-layer investment decision-making module and a lower-layer production and operation module; The optimization goal of the upper-level problem is to minimize the total investment cost, which includes the investment cost of flexibility resources, as well as the operating costs, network loss costs, and wind and solar curtailment costs of the lower-level problem. The constraints are the flexibility resource capacity constraint and the flexibility resource installation node constraint. The optimization goal of the lower-level problem is to minimize operating costs and voltage deviations, which include wind and solar power curtailment costs, operation and maintenance costs, electricity purchase costs, and network loss costs; the constraints include photovoltaic mechanism constraints, wind turbine mechanism constraints, energy storage mechanism constraints, power flow constraints, voltage constraints, and power fluctuation constraints at the common connection point.

7. A method for dynamic aggregation and finite-time hierarchical optimization control of distribution network flexibility resources according to claim 6, characterized in that: In step 3, the particle swarm algorithm is used to solve the hierarchical optimization model, including: Initialize the particle position and velocity, i.e. the position and capacity of the flexibility resources in the decision layer, as input to the operation layer; Taking the optimization goals and constraints of the operation layer as input, the flexibility resource output in each period is optimized and fed back to the decision-making layer; The decision layer receives feedback from the operation layer, calculates the decision layer constraints and objective functions, and optimizes and updates the location and capacity of flexible resources; When the maximum number of iterations is reached or the rate of change of the objective function value is less than the threshold, the optimal resource allocation plan is output.

8. A distribution network flexibility resource dynamic aggregation and finite time hierarchical optimization control method according to claim 7, characterized in that: In step 4, based on the solution of step 3, a finite time preset performance algorithm is used and dynamic parameters are introduced into the particle swarm algorithm to dynamically constrain the system state error, including: An asymmetric adaptive finite-time performance function ρ(t) is designed to constrain the system state error and ensure that it converges to the allowable range within time T. Introducing a time penalty term into the objective function of the hierarchical optimization model to accelerate convergence; In the particle swarm algorithm, the inertia weight factor and learning factor adjustment are introduced to improve it, and the time-varying inertia weight and learning factor are designed.

Images