Power distribution network multistage reconstruction method and device considering main-distribution linkage under uncertain source load

Abstract

The invention belongs to the technical field of power distribution network reconstruction, and particularly relates to a power distribution network multistage reconstruction method and device considering main-distribution linkage under source load uncertainty, and the method comprises the steps: constructing a photovoltaic output model and a load model, and forming a source load uncertainty model; constructing a source load scene, and performing optical load scene generation based on Latin hypercube sampling and Cholesky decomposition through correlation modeling; performing scene reduction; dividing time periods by adopting a fuzzy C-means clustering method; constructing a reconstruction level evaluation double-layer model with the goal of minimizing light abandoning and load loss cost; converting the reconstruction level evaluation double-layer model into a single-layer model through integrated correlation modeling, and setting constraint conditions; and solving the single-layer model to obtain a multi-stage dynamic reconstruction scheme of the power distribution network. According to the method, through multi-scene modeling, time period dynamic division and multi-level collaborative optimization, efficient and stable operation of the power distribution network is realized, and the light abandoning and load loss cost is remarkably reduced.

Description

Technical Field

[0001] The present invention belongs to the technical field of distribution network reconfiguration, and particularly relates to a multi-level reconfiguration method and device for a distribution network considering main-distribution linkage under source-load uncertainty. Background Art

[0002] With the transformation of the energy structure and the wide application of green energy, especially the large-scale access of renewable energy such as photovoltaic, the operation of traditional power grids faces unprecedented challenges. In this context, the dispatching and operation of distribution networks have become more complex and dynamic, and new technologies are urgently needed to ensure the efficient and stable operation of the power grid. However, the biggest characteristic of photovoltaic is its uncertainty, that is, its output is greatly affected by natural conditions and cannot be accurately predicted. This makes the power flow (current flow) of the distribution network more difficult to control.

[0003] To solve these problems, the distribution network needs to be dynamically reconfigured. Real-time adjustment of the operation state of the power grid, automatic adjustment of the structure and working mode of the power grid to cope with different loads and power generations. Optimize the current flow, reduce energy losses, reduce the phenomenon of light abandonment, and ensure the reliability and stability of power supply.

[0004] Although dynamic reconfiguration technology has been widely studied and applied, existing distribution network reconfiguration methods usually have some problems. For example, traditional reconfiguration methods mostly rely on local decision rules, only consider the load fluctuations of local networks, and ignore the linkage effect with the main grid (such as the main substation). This results in the inability to achieve the optimal dispatching of the entire network when facing source-load uncertainty, thus affecting the overall efficiency of the distribution network, leading to low resource utilization rate of the power grid and large power losses.

[0005] Among existing related patents, CN118449146A does not consider the power interaction and topological linkage between the main power grid and the distribution network, and it is difficult to achieve the optimal power flow and cost minimization. CN116826847A does not cooperate with the photovoltaic output modeling, resulting in an increase in the light abandonment rate and low photovoltaic accommodation efficiency; moreover, the optimization goal is single, and multi-objective optimization is not carried out for the reconfiguration cost (such as switch operation cost, network loss). CN114825348A has a single reconfiguration level and does not establish a multi-level linkage model of substation-transformer-feeder. CN115642592A does not consider the randomness of photovoltaic output and load and does not establish a multi-scenario model, which may lead to insufficient real-time calculation efficiency when the source-load fluctuation exceeds the preset range. Summary of the Invention

[0006] According to the above deficiencies in the prior art, the purpose of the present invention is to provide a multi-level reconfiguration method and device for a distribution network considering main-distribution linkage under source-load uncertainty. Through multi-scenario modeling, time-period dynamic division, and multi-level collaborative optimization, the efficient and stable operation of the distribution network is achieved, and the light abandonment and load loss costs are significantly reduced.

[0007] To achieve the above objectives, the present invention provides a multi-level distribution network reconstruction method considering the main and distribution linkage under the uncertainty of power sources and loads, including the following steps: S1. Construct a photovoltaic output model and a load model to form a power source and load uncertainty model; S2. Based on the power source and load uncertainty model, construct power source and load scenarios, and through correlation modeling, generate light and load scenarios based on Latin hypercube sampling and Cholesky decomposition; S3. Use a heuristic synchronous back substitution scenario reduction method to reduce scenarios; S4. Use the fuzzy C-means clustering method to divide time periods; S5. Construct a two-layer model for evaluating the reconstruction level with the goal of minimizing the cost of abandoned light and load shedding, including an upper-level decision-making layer model and a lower-level reconstruction layer model; S6. Through integrated correlation modeling, convert the two-layer model for evaluating the reconstruction level into a single-layer model; S7. Set the constraint conditions of the single-layer model; S8. Solve the single-layer model to obtain a multi-level dynamic reconstruction plan for the distribution network.

[0008] As a preferred solution of the present invention, in the above S1, the construction process of the photovoltaic output model is as follows: S1.1. The photovoltaic output power depends on the light intensity, and the photovoltaic probability density function is expressed as: (1); In the formula, E is the current light intensity; is the maximum light intensity; represents the Gamma function; , The scale parameters of the Beta distribution are respectively expressed as: (2); In the formula, , are respectively the average value and standard deviation of the light intensity, which are obtained through historical light data; S1.2. The relationship between the photovoltaic active power output and the illuminance is approximately expressed as: (3); In the formula, is the photovoltaic output active power; is the rated active power output of the photovoltaic; The construction process of the load model is as follows: S1.3. The load prediction follows a normal distribution, and its model is constructed as follows: (4); (5); Wherein, and are the active load and reactive load model functions respectively; and are the active load and reactive load respectively; and are the expected value and standard deviation of the active load respectively; and are the expected value and standard deviation of the reactive load respectively; exp represents the exponential function with the natural constant as the base.

[0009] As a preferred solution of the present invention, in S2, the construction method of the source-load scenario is as follows: The magnitudes of the photovoltaic output and the load within a certain time interval are represented in the form of a time series, and this time series is called a scenario S within this time interval, expressed as: (6); Wherein, T is the total number of time periods included in this time interval; represents transpose; is the sample matrix of the light intensity and the load at time t, expressed as: (7); Wherein, and and are the sample values of the light intensity at time t and the active and reactive powers of the load respectively; Suppose the number of scenarios generated by sampling is N, then the scenario matrix constituted by the original scenario set containing N scenarios is as follows: (8); Wherein, is the column vector containing all the time period sample data in the scenario generated by the nth sampling; is the sample matrix at time t in the scenario generated by the nth sampling; the scenario probability value of each scenario in the original scenario set is 1 / N, and the sum of the probabilities of all scenarios is 1.

[0010] The Spearman rank correlation coefficient can accurately describe the correlation between random variables that follow a non-normal distribution. Among the various uncertain factors generated during the operation of the distribution network, the light intensity follows a Beta distribution, which is a non-normal distribution. Therefore, the Spearman rank correlation coefficient is used to describe the correlation between random variables.

[0011] As a preferred solution of the present invention, in S2, the process of generating the optical charge scenario based on Latin hypercube sampling and Cholesky decomposition through correlation modeling is as follows: S2.1. Describe the correlation between random variables using the Spearman rank correlation coefficient, which is expressed as: (9); In the formula, is the Spearman rank correlation coefficient between random variables and ; , are the ranks corresponding to random variables and respectively; is and The covariance between; , are and Standard deviation; S2.2. Establish a Spearman rank correlation coefficient matrix to describe the correlation between photovoltaic output and load. The rank correlation coefficient matrix of M random variables is expressed as: (10); The elements in are , o = 1, 2,..., M; e = 1, 2,..., M; when o = e, ; S2.3. The combination of Latin hypercube sampling and Cholesky decomposition can generate a sample matrix of the random variables to be sampled under the given rank correlation coefficient matrix , Calculated based on historical data through formula (9), and its form is as follows: (11); In the formula, Is the rank correlation coefficient between load and light intensity; Is the rank correlation coefficient between light intensity and load, which is equal to the value of ; The process of generating the optical charge scenario consists of two parts: sampling and sorting. During the sampling process, Latin hypercube sampling is performed on each random variable to obtain a uniform sample matrix; sorting is to reorder the sample values so that the correlation between the sample values satisfies the given rank correlation coefficient matrix. Specifically: S2.4.1. Sampling, the sample size is Z, is the m-th random variable, and its cumulative probability distribution function is: (12); wherein, is 's cumulative probability distribution function, mapping to the interval [0, 1]; is the cumulative distribution value calculated by according to the value of each , and the value of is within the interval [0, 1]; is monotonically increasing in the interval [0, 1], and it is divided into Z sub-intervals (the number of scenarios N is the same as the sample size, that is, Z = N numerically). Median Latin hypercube sampling is adopted, that is, the midpoint of the sub-interval is selected, and the corresponding sampling value is calculated through the inverse function, and the calculation formula is: (13); wherein, z is the index of the sub-interval, which determines the sampling position (probability interval position); Sample the M random variables Z times to obtain the sample matrix of : (14); wherein, 's rows represent Z sampling values of a random variable, and the columns represent the sample values of the M random variables at the z-th draw; S2.4.2. Sorting to generate an -order order matrix L, each row of which is randomly arranged by integers 1, 2,..., Z, and calculate the rank correlation coefficient matrix of L according to formula (9); is a symmetric positive definite matrix. Perform Cholesky decomposition on , which is expressed as: (15); wherein, is the lower triangular matrix obtained by Cholesky decomposition ; Eliminate the correlation caused by random permutation through formula (16) to obtain matrix G: (16); wherein, the rank correlation coefficient matrix of G is the identity matrix; Perform Cholesky decomposition on , which is expressed as: (17); Wherein, W is the lower triangular matrix obtained by Cholesky decomposition; The adjusted sample matrix is made to have the rank correlation coefficient approximately equal to that of through Equation (18): Equal: (18); Change the element order of each row in the sample matrix to make it the same as the corresponding element order in ; At this time, the rank correlation coefficient matrix of is approximately equal to that of , that is, a sample matrix with a given correlation is obtained; S2.5. Perform sampling based on Latin hypercube sampling and Cholesky decomposition in the t period to obtain the sample matrix of this period, that is, one row in Equation (8) , and then through Equation (3), use the light intensity of the nth sampling in the t period to calculate the photovoltaic output of the nth sampling in the t period, and count as , expressed as: (19); Wherein, is the column vector containing the sample data of the t period in the scenario generated by the nth sampling; , respectively represent the active and reactive power sampling values of the load in the nth sampling in the t period; Sampling for T periods can obtain the sample matrix of the complete time interval.

[0012] is the row vector of the matrix in the t period in Equation (8), and collecting at each t moment can obtain .

[0013] LHS sampling needs to reach a certain sample size to approach the probability distribution of random variables and meet the accuracy requirements. However, if all scenarios are retained in the solution, it will make the solution efficiency low and even unable to be applied in practice. Therefore, scenario reduction is required to remove redundant scenarios and improve the calculation efficiency. At the same time, the finally retained scenarios are representative and the overall still meets the probability distribution. The basic idea of scenario reduction is that the probability distance between the scenario set before reduction and the finally retained scenario set is the smallest.

[0014] As a preferred solution of the present invention, in the above S3, the method for performing scenario reduction is: S3.1. Define the scenario and the scenarios have probabilities of and respectively. Then the probability distance between scenarios is given by: (20); where the subscript 2 represents the Euclidean norm; Under the condition that the number of remaining scenarios is determined, minimize Equation (21): (21); where J is the set composed of the finally pruned scenarios; S3.2. Scenario pruning is performed based on the following steps: S3.2.1. Define D as the initial scenario set, set J as an empty set, i.e., , and the iteration number u = 1; S3.2.2. Calculate the scenarios to be pruned in the u-th iteration to minimize Equation (22): (22); where , are the sets J in the u-th and (u - 1)-th iterations respectively; S3.2.3. Delete from D, add to the set J, , u = u + 1, is the union symbol; S3.2.4. Add the probability of the pruned scenario to the scenario with the closest probability distance to it; S3.2.5. Repeat steps S3.2.2 to S3.2.4 until the number of scenarios in D is equal to the number of finally retained scenarios (set the number of scenarios according to requirements).

[0015] The dynamic reconfiguration of the distribution network often needs to study dozens or hundreds of consecutive time periods. The switching states between time periods affect each other, the solution space is extremely large, the optimization technology is complex. Dividing the time periods for the complete reconfiguration time interval reduces the solving difficulty and is more in line with the actual operation of the distribution network. Based on this, the fuzzy C-means clustering method is used to divide the time periods.

[0016] The fuzzy C-means clustering method changes the inherent 0-1 attribution criterion, assigns membership degree values corresponding to multiple class centers to each sample of the given data set respectively, and its value range is within [0, 1]. It performs clustering division through a clustering loss function based on a fuzzy membership matrix, and can achieve that the difference between data in the same class is as small as possible, and the difference between data in different classes is as large as possible. The multi-level reconstruction mode during the full optimization period is unique, and it is easy to increase the computational difficulty in the time and space dimensions for large-scale systems. Therefore, considering that the spatio-temporal distribution of the system net load can represent the flexibility requirements for multi-level reconstruction, the multi-level reconstruction time period is divided according to the spatio-temporal characteristics of the net load.

[0017] As a preferred solution of the present invention, in S4, the process of dividing the time period by using the fuzzy C-means clustering method is as follows: S4.1. For the photovoltaic output and load of node j at time t under the scenario, a net load time series data set is formed, and each sample represents the spatial distribution of the net load at a certain time period, that is, , where represents the active power of the net load of node j at time t under the scenario; S4.2. Through fuzzy C-means clustering, F is fuzzily divided into C classes, corresponding to the class center set , is the center of the c-th class; the fuzzy membership matrix of the sample for the divided classes is , and the element in represents the membership degree corresponding to the c-th class at time t; S4.3. Construct a clustering planning model based on the Euclidean distance between the sample and the center of the c-th class and the fuzzy membership matrix as follows: In the formula, Y represents the weighted sum of the clustering errors; (24); In the formula, q is the weighted exponent; is the universal quantifier, and the sample t is the time period t; S4.4. By constructing the Lagrangian function, the optimal and V that satisfy formula (23) are obtained: (25); (26); In the formula, is the Euclidean distance between the sample and the center of the m-th class; S4.5. Consider the continuity of the switch reconstruction operation, and its clustering process is as follows: S4.5.1. Initialize C, and the allowable error ; S4.5.2. Calculate according to Equation (26), which is at the v-th iteration; S4.5.3. Update according to Equation (25), which is at the v-th iteration; S4.5.4. If , the iteration ends; otherwise, go to S4.5.3 to continue the iteration until the requirement is met, which is at the (v + 1)-th iteration.

[0018] First, apply Latin hypercube sampling, Cholesky decomposition, and scenario reduction techniques to generate representative scenarios, and perform priority local reconstruction for each scenario to reduce the costs of light curtailment and load shedding and construct a two-layer model for evaluating the reconstruction level. Convert it into a single-layer model through integrated correlation modeling for solution. At the same time, based on the two-layer model for evaluating the reconstruction level, elaborate the method for identifying the multi-period multi-level reconstruction level. Guided by the elimination of light curtailment and load shedding phenomena in the distribution network, adopt the principle of giving priority to local autonomy and lagging global coordination. Minimize the costs of light curtailment and load shedding, and finally minimize the expected value of the comprehensive cost under multiple scenarios in the converted single-layer model.

[0019] As a preferred solution of the present invention, in S5, the method for constructing the upper-level decision-making layer model is that the upper-level decision-making layer model first quantifies the cost of local autonomy. The size of the reconstruction level cost in the scenario will reflect the degree of localization, which is expressed as: (27); In the formula, is the set of substation nodes; is the total number of scenarios; is the set of transformers under substation b; is the substation-level reconstruction variable of substation b in the scenario; is the transformer-level reconstruction variable under substation b in the scenario; is the feeder-level reconstruction variable of transformer f under substation b in the scenario; , , represent the costs of single feeder - level reconstruction, transformer - level reconstruction, and substation - level reconstruction respectively, and ; After dividing the time periods using the fuzzy C - means clustering method, each time period belonging to the same class center can adopt the same reconstruction mode. Then: (28); In the formula, is the substation - level reconstruction demand of substation b at time period t under the th scenario; is the transformer - level reconstruction demand of substation b at time period t under the th scenario; is the feeder - level reconstruction demand of transformer f of substation b at time period t under the th scenario; is the substation - level reconstruction demand of substation b belonging to the c - th class of time periods under the th scenario; is the transformer - level reconstruction demand of substation b belonging to the c - th class of time periods under the th scenario; is the feeder - level reconstruction demand of transformer f of substation b belonging to the c - th class of time periods under the th scenario; The upper - level decision - making layer model is expressed as: (29).

[0020] After dividing the time periods using fuzzy C - clustering, each time period belonging to the same class center can adopt the same reconstruction mode. The multi - level reconstruction under time - period division can reduce the time - calculation dimension while reducing unnecessary space - calculation costs. At the same time, a unified reconstruction mode will be adopted for time periods belonging to the same class, and the principle of taking the higher value is used, which can meet the requirements of each hierarchical level for all time periods in the same class. At the same time, a reasonable total number of time - period divisions should be determined to ensure that the time - period - based multi - level reconstruction scheme can improve the calculation efficiency without affecting the economy of distribution network reconstruction.

[0021] After time - period division, the reconstruction - level requirements of each substation and transformer in different categories of the distribution network are different. Therefore, the feeder - level reconstruction, transformer - level reconstruction, and substation - level reconstruction models that a certain class may include in the scenario are elaborated one by one. Note: The substation - level, transformer - level, and feeder - level reconstruction models corresponding to different substations and different transformers are generally the same, with the differences lying in the network topology and load scenarios.

[0022] The objective function can be decomposed into three parts: feeder - level reconstruction, transformer - level reconstruction, and substation - level reconstruction. Their basic forms are similar, but the decision variables and constraint conditions are slightly different.

[0023] As a preferred embodiment of the present invention, in S5, the lower-level reconstruction layer model includes a substation-level reconstruction model, a transformer-level reconstruction model, and a feeder-level reconstruction model; The substation-level reconstruction model ( ) takes multiple substations as the analysis object, and realizes reliable power supply for the multi-station net load through the state combination of feeder connection switches, transformer connection switches, inter-substation connection switches, and sectionalizing switches; for simplicity of expression, except outside, the clustering identifier c and the multi-substation identifiers h and l in the subscripts of the remaining symbols are omitted, and the substation-level reconstruction objective function is expressed as: (30); In the formula, represents the objective function of substation-level reconstruction of substations h and l in the cth type of time period in the scenario; , , , respectively represent the network loss cost, the curtailment cost of light, the load shedding cost, and the switch reconstruction cost; represents the set of time periods belonging to the cth type; represents the active power of load shedding at node j at time t in the scenario; represents the actual output of PV at node j at time t in the scenario; represents the square of the current of branch ij at time t in the scenario; , , , are the single operation costs of the feeder connection switch, the transformer connection switch, the inter-substation connection switch, and the sectionalizing switch respectively, and ; , represent the switch state change flag of branch ij at time t. If , then the switch of branch ij changes from the off state to the on state at time t in the scenario. If , then the switch of branch ij changes from the on state to the off state at time t in the scenario; , , , respectively represent the branch sets of the four types of connection switches, namely the feeder connection switch, the transformer connection switch, the inter-substation connection switch, and the sectionalizing switch in the reconstruction area, which can be adjusted according to different substation-level reconstructions; B, respectively represent the node set and the PV node set in the reconstruction area; Denote the set of switch branches in the optimization region; , are the load shedding and curtailment costs at time period t in the th scenario respectively; is the resistance of line ij; Transformer-level reconstruction model ( , ), taking the substation as the analysis object, realizes the autonomous operation of the substation unit through the state combination of feeder tie switches, transformer tie switches and sectionalizing switches; for simplicity of expression, except , the clustering identifier c and substation identifier b in the subscripts of the remaining symbols are omitted, and the objective function of the transformer-level reconstruction model is expressed as: (31); In the formula, represents the objective function of transformer-level reconstruction of substation b in the th scenario and the cth time period; Feeder-level reconstruction model ( , , ), taking the transformer as the analysis object, realizes reliable power supply for the feeder load in the transformer and full accommodation of PV power through the state combination of feeder tie switches and sectionalizing switches; for simplicity of expression, except , the clustering identifier c, substation identifier b and transformer identifier f in the subscripts of the remaining symbols are omitted, and the objective function of the feeder-level reconstruction model is expressed as: (32); In the formula, represents the objective function of feeder-level reconstruction of transformer f of substation b in the th scenario and the cth time period; The total number of reconstruction entities in the lower layer in the th scenario is expressed as: (33); The lower-layer level reconstruction layer model in the th scenario is expressed as: (34); In the formula, a is the entity identifier, and the entities include multi-substations, substations, transformers, and feeders; , are the cost of the lower-layer entity a and the total lower-layer cost in the th scenario respectively.

[0024] Different entities all improve the power flow distribution by changing the working states of the tie switches and sectionalizing switches within the area, thereby reducing the amount of curtailed light and load shedding. The lower-layer model focuses on improving the problems of curtailed light and load shedding in the distribution network, thus assisting the upper-layer reconstruction level decision-making layer to quickly identify the reconstruction level requirements of each substation and transformer, and quickly providing the level identification results for system operators.

[0025] The lower layer under the

[0026] scenario is reconstructed. Different entities all improve the power flow distribution by changing the working states of the tie switches and sectionalizing switches within the area, thereby reducing the amount of curtailed light and load shedding. The lower-layer model focuses on improving the problems of curtailed light and load shedding in the distribution network. According to the output of the level decision-making layer, a multi-level reconstruction optimization scheme is formulated, and reconstruction is carried out with the goal of minimizing the cost of curtailed light and load shedding within the area and considering network losses. The feeder-level reconstruction, transformer-level reconstruction, and substation-level reconstruction respectively take the minimum amount of curtailed light and load shedding of the nodes within the transformer, within the substation, and within multiple substations as the objective functions.

[0027] As a preferred solution of the present invention, in the said S6, the process of converting the double-layer model for reconstructing level evaluation into a single-layer model is as follows: The upper and lower layer correlation constraints are as follows: (35); In the formula, , respectively represent the branch set and the total number of branches of the inter-substation tie switches of substation b; , respectively represent the branch set and the total number of branches of the transformer tie switches of transformer f under substation b; , respectively represent the branch set and the total number of branches of the feeder tie switches of transformer f under substation b; , respectively represent the branch set and the total number of branches of the sectionalizing switches of the feeder of transformer f under substation b; represents the switch state of branch ij at time t under the scenario. If , it means that the switch of branch ij is closed.

[0028] It can be seen from formula (35) that: when the scenario is , in the c-th clustering case: When 、 、 , When , , , the SS (substation interconnection switch), TS (transformer interconnection switch), FS (feeder interconnection switch), and BS (section switch) of substation b can all adjust their own on / off states to achieve substation-level reconstruction between substation b and associated substations; When 、 、 , When , , , , the TS, FS, and BS of substation b can all adjust their own on / off states to achieve transformer-level reconstruction of each transformer f under substation b; When 、 、 When , , , the FS and BS of transformer f under substation b can adjust their own on / off states to achieve feeder-level reconstruction within transformer f under substation b.

[0029] Through the correlation constraint of formula (35), the unification of the level decision layer and the reconstruction layer can be achieved, and the double-layer model for reconstructing level evaluation is converted into a single-layer model, expressed as: (36); In the formula, represents the expected value of the comprehensive cost; 、 are the upper and lower layer decision variables respectively, includes the substation-level and transformer-level reconstruction selection variables of all substations in the whole network and the feeder-level reconstruction selection variables of each transformer, includes the state variables of the "feeder-transformer-substation" different-level interconnection switches and branch section switches; is the scenario probability of the

[0030] In S7, the set constraint conditions include reconstruction level decision constraints, second-order cone power flow constraints, PV output constraints, load shedding constraints, security constraints, and network reconstruction constraints.

[0031] A multi-level reconstruction device for a distribution network considering main and distribution linkage under source-load uncertainty includes a memory, a processor, and a computer program stored on the memory and executable on the processor. The above method is implemented by the processor executing the program.

[0032] The beneficial effects of the present invention are: Significantly reduce the computational complexity: The present invention introduces a heuristic synchronous back substitution scenario reduction method to eliminate redundant scenarios and retain a representative set of scenarios. Optimizing the model solution scale reduction can meet the real-time dynamic reconfiguration requirements of the distribution network, improve the dispatching response speed, and reduce the risks of light curtailment and load loss caused by delays.

[0033] Achieve global-local collaborative optimization: The present invention proposes a multi-level linkage reconfiguration model for substations-transformers-feeders. The upper layer decides the reconfiguration level (substation / transformer / feeder), and the lower layer optimizes the specific switch operations and power flow distribution. The local optimization is consistent with the global objective (such as minimizing the light curtailment cost of the whole network), avoiding the power imbalance between the main and distribution networks caused by local optimization, and reducing the light curtailment rate.

[0034] Dynamically adapt to the real-time fluctuations of power sources and loads: The present invention divides time periods based on fuzzy C-means clustering and dynamically fuses the characteristics of photovoltaic power output and loads (such as sunny days, cloudy days, load peaks / valleys). The results of time period division are dynamically adjusted with the fluctuations of power sources and loads, and the reconfiguration strategy is adapted to different operating scenarios in real time. When the light intensity changes suddenly (such as from sunny to cloudy) or the load suddenly increases, the reconfiguration plan responds quickly, reducing the probability of load loss.

[0035] Enhance uncertainty analysis and economy: The present invention generates multiple scenarios through Latin hypercube sampling + Cholesky sampling, covering extreme fluctuations of photovoltaic power output and loads (such as heavy rain weather, sudden load increase). The integrated modeling of the main and distribution networks optimizes the light curtailment and load loss costs based on uncertain power sources and loads, significantly improving the stability of power grid operation. Description of the Drawings

[0036] Figure 1 is the process schematic diagram of the present invention; Figure 2 is the multi-level dynamic reconfiguration flow chart of the distribution network of the present invention. Detailed Embodiments

[0037] The following further describes the embodiments of the present invention with reference to the drawings: Embodiment 1: As Figure 1 shown, the multi-level reconfiguration method for a distribution network considering main-distribution linkage under uncertain power sources and loads includes the following steps: S1. Construct a photovoltaic power output model and a load model to form a power source-load uncertainty model; S2. Based on the power source-load uncertainty model, construct power source-load scenarios, and through correlation modeling, generate light-load scenarios based on Latin hypercube sampling and Cholesky decomposition; S3. Use a heuristic-based synchronous back substitution scenario reduction method to perform scenario reduction; S4. Adopt the fuzzy C-means clustering method to divide time periods; S5. Build a two - layer model for evaluating the reconstruction level with the goal of minimizing the costs of curtailed light and load shedding, including the upper - level decision - making layer model and the lower - level reconstruction layer model; S6. Through integrated correlation modeling, convert the two - layer model for evaluating the reconstruction level into a single - layer model; S7. Set the constraint conditions of the single - layer model; S8. Solve the single - layer model to obtain the multi - level dynamic reconstruction scheme of the distribution network.

[0038] For the specific implementation methods of S1 - S6, refer to the "Summary of the Invention" section. In S7, the set constraint conditions specifically include: S7.1. Reconstruction - level decision constraints. There are coupling constraints among the selection of the three - level reconstruction modes of feeder - level, transformer - level, and substation - level, which are expressed as: (37); (38); (39); In the formula, represents the total number of transformers included in substation b; , are respectively the substation - level reconstruction requirements of substations h and l of the c - th category during the scenario; Formula (37) means that substation b cannot participate in substation - level reconstruction and transformer - level reconstruction simultaneously during the c - th category of the scenario; formula (38) means that substations h and l associated with the c - th category during the scenario either participate in or do not participate in substation - level reconstruction simultaneously; When, formula (39) means that for any transformer under substation b during the c - th category of the scenario can only be 0, that is, when the upper - level substation performs transformer - level / substation - level reconstruction, the internal transformers cannot perform feeder - level reconstruction repeatedly.

[0039] The substation reconstruction constraint conditions, transformer reconstruction constraint conditions, and feeder reconstruction constraint conditions are basically similar, except that some network reconstruction constraints and power supply constraints are different. The following similar constraints will not be repeated, and only the different parts will be elaborated.

[0040] S7.2. Second - order cone power flow constraints. The construction process is as follows: In the power flow analysis of the distribution network, the relationship between power transmission and current and impedance is usually considered. In traditional power flow calculations, there is a non - linear relationship between power and current, and the power flow model is expressed as: (40); wherein, and are the active and reactive powers of node i at time t, respectively; and are the voltages of nodes i and j at time t, respectively; and are the conductance and susceptance of branch ij, respectively; is the phase angle of branch ij at time t; represents the set of all nodes; indicates that node j is an adjacent node of node i; Introduce and demonstrate the equivalent transformation. First, apply the following formula: (41); wherein, and are the real and imaginary parts of the voltage difference between the two nodes of branch ij at time t, respectively; represents the set of all branches; is the square of the voltage of node i at time t; In an actual scenario, the phase difference between adjacent nodes is relatively small, such that usually holds; Since , then: (42); Introducing equation (41), equation (40) becomes: (43); wherein, and are the active and reactive powers of branch ij at time t; The power flow model also needs to satisfy: (44); wherein, is the square of the voltage of node j at time t; Due to the quadratic equality constraint, equation (44) is still non-linear. Relax the equality constraint to an inequality constraint and describe it with the following second-order cone function: (45); After the original non-linear power flow constraint composed of the above formula is transformed by second-order cone relaxation and combined with the scenario and node power balance constraint, a second-order cone power flow constraint is formed, expressed as: (46); wherein, is the set of the terminal nodes of the branches with node j as the initial node; is the set of the initial nodes of the branches with node j as the terminal node; represents the reactance of the branch; and are respectively the squared voltages of node i and node j at time t in the th scenario; and respectively represent the active and reactive injection powers of the substation at node j at time t in the th scenario; and respectively represent the active powers transmitted by branches ij and jk at time t in the th scenario; and respectively represent the reactive powers transmitted by branches ij and jk at time t in the th scenario; and respectively represent the active injection power and active load of node j at time t in the th scenario; and respectively represent the reactive injection power and reactive load of node j at time t in the th scenario; is the penalty coefficient, which is a very large constant; represents the reactive power output of the load curtailment at node j at time t in the th scenario; is the voltage of node j at time t in the th scenario; is the current of branch ij at time t in the th scenario; S7.3. PV output constraint, expressed as: (47); S7.4. Load loss constraint, expressed as: (48); In the formula, represents the load curtailment status flag of node j in the th scenario, = 1 indicates that load loss is allowed; S7.5. Security constraint, expressed as: (50); In the formula, and are respectively the upper and lower limits of the current of branch ij; and are respectively the upper and lower limits of the voltage of node j; S7.6. Network reconfiguration constraint, including: Substation reconfiguration constraint, expressed as: (51); (52); Wherein, represents the total number of non - adjustable branches that are always in a closed state in the grid; represents the total number of substations; represents the switching state of branch ij at time t - 1 in the Equation (51) represents a node system with substations, and the number of network closed branches is the number of nodes minus the number of substations.

[0041] Transformer and feeder reconfiguration constraints are expressed as: (53); A small node injection power is introduced at non - substation nodes, and the network connectivity is ensured by adding auxiliary power flow constraints.

[0042] (54); Wherein, represents the auxiliary power flow active power injected at node j at time t in the scenario; , respectively represent the auxiliary power flow active powers on branch jk and branch ij; represents the auxiliary injection power; represents the maximum value of the active power transmitted by branch ij; Power supply constraints are expressed as: (55); (56); Wherein, , respectively represent the active and reactive powers transmitted by transformer f under substation b at time t in the scenario; , respectively represent the minimum and maximum values of the active power transmitted by transformer f under substation b; , respectively represent the minimum and maximum values of the reactive power transmitted by transformer f under substation b; represents the node set of transformer f under substation b.

[0043] The constraint conditions of Equation (36) include Equation (37), Equation (38), Equation (39) and Equations (46) - (56).

[0044] First, establish the probability models of photovoltaic and load. Apply the sampling method based on Latin hypercube and Cholesky decomposition to generate large-scale scenarios. Obtain representative scenarios and corresponding scenario probabilities through the heuristic synchronous back substitution scenario reduction method. Secondly, use fuzzy C-clustering to determine the time period division results. Subsequently, establish a two-layer model for evaluating the reconstruction level with the main goal of minimizing the costs of abandoned light and load shedding, and convert it into a single-layer model through integrated correlation modeling for solution. At the same time, based on the two-layer model for evaluating the reconstruction level, elaborate the multi-level distribution network reconstruction model at each level of the substation, transformer, and feeder. Finally, through constraints, calculate the expected value of the comprehensive cost mainly composed of abandoned light and load shedding, and solve it using the CPLEX solver. Combining the entire modeling process, the solution process in S8 is as follows: S8.1. Input the actual light data and load curve, and calculate the scale parameters of the Beta distribution for each time period within a day 、 , as well as the expected value and standard deviation of the load, and calculate ; S8.2. Use the sampling method based on Latin hypercube and Cholesky decomposition to generate the original scenario set, and then reduce it to representative scenarios through the heuristic synchronous back substitution scenario reduction method, and obtain the corresponding scenario probabilities , let ; S8.3. Read the scenario and for each corresponding node after reduction and , to form the net load time series data set F; S8.4. Initialize C, and the allowable error ; S8.5. Divide F into c classes through fuzzy C-means clustering, corresponding to the class centers ; S8.6. Calculate according to Equation (26), which is the at the v-th iteration; S8.7. Update according to Equation (25), which is the at the v-th iteration; S8.8. If , the iteration ends, otherwise go to S8.7 to continue the iteration until the requirements are met, which is the at the (v + 1)-th iteration; S8.9. Determine the class labels for each time period and the time period division plan for a day; S8.10. Initialize the network topology, including , , , , , , , and the decision variables in the scenario , , , the lower-level decision variables , , , , ; Define and initialize the objective functions at each level , , ; Define , , , , , , ; S8.11. Set the total number of clusters C, initialize c = 1, , , b = 1; S8.12. Read in the and at time period t; S8.13. Set the upper-level objective function based on the constraints in Equation (28), set the decision-level reconstruction constraints including Equations (37), (38), and (39), and the upper-level decision variables become , , ; S8.14. Determine whether is equal to 1. If so, the substation interconnection switches, transformer interconnection switches, feeder interconnection switches, and sectionalizing switches of substation b can adjust their own on / off states to achieve substation-level reconstruction between substation b and associated substations. Set the constraint conditions including Equations (46), (47), (48), (49), (50), (51), (52), (54), (55), and (56), and solve for , , , , , , and , calculate , and jump to step S8.19; if not, execute step S8.15; S8.15. Determine whether Is it equal to 1? If so, the transformer tie switch, feeder tie switch, and sectionalizing switch of substation b can adjust their own on / off states to achieve transformer-level reconstruction of each transformer f belonging to substation b. Set the constraint conditions, including equations (46), (47), (48), (50), (53), (54), (55), (56), and solve through CPLEX. and and and and and and and , calculate and jump to step S8.19; if not, execute step S8.16. S8.16: Set the number of transformers f = 1 and execute step S8.17. S8.17: Judge whether is equal to 1. If so, the feeder tie switch and sectionalizing switch of transformer f under substation b can adjust their own on / off states to achieve feeder-level reconstruction within transformer f belonging to substation b. Set the constraint conditions, including equations (46), (47), (48), (50), (53), (54), (55), (56), and solve through CPLEX. and and and and and and and , calculate , and jump to step S8.18; if not, directly execute step S8.18. S8.18: f = f + 1, and judge whether f belongs to . If so, return to step S8.17; if not, execute step S8.19. S8.19: , b = b + 1, and judge whether b belongs to . If so, return to step S8.14; if not, c = c + 1 and execute step S8.20. S8.20: Judge whether c is less than or equal to C. If so, return to step S8.12; if not, ; S8.21: Set: ; ; S8.22. Integrate and correlate the modeling by setting the constraint formula (35), and convert the two-layer model for evaluating the reconstruction level into a single-layer model; S8.23. Determine Is less than or equal to , if yes, return to step S8.3; if no, execute step S8.24; S8.24. Calculate the total cost , output the results of decision variables at each level, the optimal values of objective functions at each level, switch states, load shedding, and curtailment costs, and end the solution.

[0045] In summary, the flowchart of the multi-level dynamic reconfiguration of the distribution network with coordinated main and distribution under uncertain source and load is as shown in Figure 2 Figure [].

[0046] Embodiment 2: A multi-level reconfiguration device for a distribution network considering coordinated main and distribution under uncertain source and load, including a memory, a processor, and a computer program stored on the memory and executable on the processor. The method in Embodiment 1 is implemented by the processor executing the program.

Claims

1. A multi-level distribution network reconstruction method considering the linkage between the main and distribution networks under source and load uncertainty, characterized by The following steps are involved: S1. Construct photovoltaic output model and load model to form source-load uncertainty model; S2, based on the source-load uncertainty model, construct the source-load scenario, and generate the light-load scenario based on Latin hypercube sampling and Cholesky decomposition through correlation modeling; S3, using a heuristic-based synchronous back-generation scene reduction method to perform scene reduction; S4, using fuzzy C-means clustering method to divide the time periods; S5. Construct a two-layer reconstruction level evaluation model with the goal of minimizing the cost of abandoned light and lost load, including an upper-level decision-making layer model and a lower-level reconstruction layer model; S6. Convert the reconstruction level assessment two-layer model into a single-layer model through integrated association modeling; S7, setting constraints for the single-layer model; S8. Solve the single-layer model to obtain a multi-level dynamic reconstruction solution for the distribution network.

2. The method for multi-level reconstruction of distribution network considering main-distribution linkage under source-load uncertainty according to claim 1 is characterized in that: In S1, the construction process of the photovoltaic output model is as follows: S1.

1. Photovoltaic output power depends on light intensity, photovoltaic probability density function It is expressed as: （1）； Where E is the current light intensity; is the maximum light intensity; represents the Gamma function; , is the scale parameter of Beta distribution, respectively expressed as: （2）； In the formula, , are the mean and standard deviation of light intensity, respectively, obtained through historical light data; S1.

2. The relationship between photovoltaic active output and illumination can be expressed approximately as follows: （3）； In the formula, Output active power for photovoltaic; is the rated active power of photovoltaic output; The process of building the load model is as follows: S1.3, load forecasting follows normal distribution, and its model is constructed as follows: （4）； （5）； In the formula, , They are active load and reactive load model functions respectively; , They are active load and reactive load respectively; , are the expected value and standard deviation of active load respectively; , are the expected value and standard deviation of reactive load respectively; exp represents an exponential function with a natural constant as the base.

3. The method for multi-level reconstruction of distribution network considering main-distribution linkage under source-load uncertainty according to claim 2 is characterized in that: In S2, the source-load scenario is constructed as follows: The size of photovoltaic output and load in a certain time interval is expressed in the form of a time series. This time series is called a scene S in the time interval and is expressed as: （6）； In the formula, T is the total number of time periods included in the time interval; represents transpose; is the sample matrix of light intensity and load in period t, expressed as: （7）； In the formula, , , are the sample values ​​of the light intensity in period t and the active and reactive power of the load respectively; Assume that the number of scenes generated by sampling is N, then the scene matrix consisting of the original scene set containing N scenes is as follows: （8）； In the formula, is a column vector containing sample data of all time periods in the scene generated by the nth sampling; The sample matrix of time period t in the scene generated by the nth sampling; The scene probability value of each scene in the original scene set is 1 / N, and the sum of the probabilities of all scenes is 1.

4. The method for multi-level reconstruction of distribution network considering main-distribution linkage under source-load uncertainty according to claim 3 is characterized in that: In S2, the process of generating the light charge scene based on Latin hypercube sampling and Cholesky decomposition through correlation modeling is as follows: S2.1, the Spearman rank correlation coefficient is used to describe the correlation between random variables, expressed as: （9）； In the formula, is a random variable and The Spearman rank correlation coefficient between , are random variables and The corresponding rank; for and The covariance between , They are and The standard deviation of S2.

2. Establish the Spearman rank correlation coefficient matrix to describe the correlation between photovoltaic output and load. The rank correlation coefficient matrix of M random variables is It is expressed as: （10）； The elements in are , o=1,2,…,M; e=1, 2, …, M; When o=e, ; S2.3, Latin hypercube sampling combined with Cholesky decomposition can produce a given rank correlation coefficient matrix The sample matrix of the random variable to be sampled, It is calculated based on historical data through formula (9), and its form is as follows: （11）； In the formula, is the rank correlation coefficient between load and light intensity; is the rank correlation coefficient between light intensity and load, and The values ​​are equal; S2.4, the process of generating light charge scenes consists of two parts: sampling and sorting. During the sampling process, Latin hypercube sampling is performed on each random variable to obtain a uniform sample matrix; sorting is to reorder the sample values ​​so that the correlation between the sample values ​​satisfies the given rank correlation coefficient matrix, specifically: S2.4.1, sampling, sample size is Z, is the mth random variable, and its cumulative probability distribution function is: （12）； In the formula, yes The cumulative probability distribution function of Map to the interval [0, 1]; According to each Value, through To calculate the cumulative distribution value, and The value of lies in the interval [0, 1]; It is monotonically increasing in the interval [0, 1], which is divided into Z subintervals. Median Latin hypercube sampling is used, that is, the midpoint of the subinterval is selected and the corresponding sampling value is calculated by the inverse function. , the calculation formula is: （13）； Where z is the index of the subinterval; Sampling M random variables Z times, we get The sample matrix : （14）； In the formula, The rows represent the Z sample values ​​of a random variable, and the columns represent the sample values ​​of the zth extraction of M random variables; S2.4.2, sorting, generating a The rank correlation coefficient matrix L of the order matrix L is calculated according to formula (9): ; is a symmetric positive definite matrix, Perform Cholesky decomposition, expressed as: （15）； In the formula, Cholesky decomposition The resulting lower triangular matrix; By using formula (16) to eliminate the correlation caused by random arrangement, we get the matrix G: （16）； In the formula, the rank correlation coefficient matrix of G is the unit matrix; right Perform Cholesky decomposition, expressed as: （17）； Where W is the Cholesky decomposition The obtained lower triangular matrix; through formula (18), the adjusted sample matrix The rank correlation coefficient is approximately the same as equal: （18）； Changing the sample matrix The order of elements in each row of The corresponding elements in are in the same order; in this case, The rank correlation coefficient matrix of Approximately equal, that is, a sample matrix with a given correlation is obtained; S2.

5. Sampling based on Latin hypercube sampling and Cholesky decomposition is performed in time period t. The sample matrix at this time period is Then, through formula (3), the light intensity sampled at the nth time in period t is used Calculate the photovoltaic output of the nth sampling in period t ,Will Calculated as , expressed as: （19）； In the formula, is a column vector containing sample data for period t in the scene generated by the nth sampling; , They represent the active and reactive power sampling values ​​of the load in the nth sampling in period t respectively; Sampling T time periods can obtain the sample matrix of the complete time interval .

5. The method for multi-level reconstruction of distribution network considering main-distribution linkage under source-load uncertainty according to claim 4 is characterized in that: In the above-mentioned S3, the method for performing scene reduction is as follows: S3.

1. Define the scenario and scenes The probabilities are and , then the probability distance between scenes for: （20）； In the formula, the subscript 2 represents the Euclidean norm; Under the condition that the number of retained scenes is determined, equation (21) is minimized: （21）； Where J is the set of scenes that are finally removed; S3.2, based on the following steps, perform scene reduction: S3.2.

1. Define D as the initial scene set and set J as the empty set, that is, , iteration number u=1; S3.2.

2. Calculate the scene that needs to be cut at the uth iteration So that formula (22) takes the minimum value: （22）； In the formula, , are the sets J at the u-th and u-1-th iterations respectively; S3.2.

3. Delete from D ,Will Add to set J, , u=u+1, is the union symbol; S3.2.4, Scenes to be cut The probability of is added to the scenario closest to its probability; S3.2.

5. Repeat steps S3.2.2 to S3.2.4 until the number of scenes in D is equal to the number of scenes finally retained.

6. The method for multi-level reconstruction of distribution network considering main-distribution linkage under source-load uncertainty according to claim 5 is characterized in that: In S4, the process of dividing the time periods using the fuzzy C-means clustering method is as follows: S4.

1. The photovoltaic output of node j in period t under the scenario and load , forming a net load time series data set , each sample represents the spatial distribution of net load in a certain period of time, that is, ,in Representative The net load active power of node j in period t under the scenario; S4.

2. Fuzzy C-means clustering is used to fuzzily divide F into C classes, corresponding to the class center set , is the center of the cth class; the fuzzy membership matrix of the sample to the partition class is , Elements in represents the membership degree corresponding to the cth category in time period t; S4.

3. Construct the Euclidean distance between the sample and the center of the cth class and the fuzzy membership matrix The cluster planning model is as follows: （23）； Where Y represents the weighted sum of clustering errors; （24）； Where q is the weighted index; is a universal quantifier, and sample t is the time period t; S4.

4. By constructing the Lagrangian function, we obtain the optimal and V: （25）； （26）； In the formula, is the Euclidean distance between the sample and the center of the mth class; S4.

5. Considering the continuity of switch reconstruction action, the clustering process is as follows: S4.5.

1. Initialize C. And the allowable error ; S4.5.

2. Calculate according to formula (26) , which is the vth iteration ; S4.5.

3. Update according to formula (25) , which is the vth iteration ; S4.5.4 If , the iteration ends, otherwise go to S4.5.3 and continue iterating until the requirements are met. That is, the v+1th iteration .

7. The method for multi-level reconstruction of distribution network considering main-distribution linkage under source-load uncertainty according to claim 6 is characterized in that: In the above S5, the method for constructing the upper level decision-making model is that the upper level decision-making model first quantifies the cost of local autonomy, Reconstruction level cost in the scenario The size of will reflect the degree of localization, It is expressed as: （27）； In the formula, is a set of substation nodes; is the total number of scenes; is the transformer set under substation b; For the Substation-level reconstruction variables of substation b in scenario; For the Transformer-level reconstruction variables under substation b in scenario; For the The feeder-level reconstruction variables of transformer f under substation b in the scenario; , , denote the cost of performing a single feeder-level reconstruction, transformer-level reconstruction, and substation-level reconstruction, respectively, and ; After the time periods are divided by the fuzzy C-means clustering method, each time period belonging to the same type of center can adopt the same reconstruction mode, then: （28）； In the formula, For the Substation-level reconstruction requirements for substation b during period t in the scenario; For the The transformer-level reconfiguration requirement at substation b during period t in the scenario; For the The feeder-level reconstruction requirements of transformer f under substation b during period t in the scenario; For the The substation-level reconstruction requirements of substation b in the time period of category c in the scenario; For the The transformer-level reconstruction requirements of substation b in the time period of category c in the scenario; For the The feeder-level reconstruction requirements of transformer f under substation b in the time period of category c in the scenario; The upper level decision layer model is expressed as: （29）。 8. The method for multi-level reconstruction of distribution network considering main-distribution linkage under source-load uncertainty according to claim 7 is characterized in that: In said S5, the lower level reconstruction layer model includes a substation level reconstruction model, a transformer level reconstruction model and a feeder level reconstruction model; The substation-level reconstruction model takes multiple substations as the analysis objects and realizes the reliable power supply of multiple station net loads through the state combination of feeder tie switches, transformer tie switches, substation inter-station tie switches and section switches. In addition, the cluster identifier c and multi-substation identifiers h and l in the subscripts of other symbols are omitted, and the substation-level reconstruction objective function is expressed as: （30）； In the formula, Indicates The objective function of substation-level reconstruction of substations h and l in the c-th time period under the scenario; , , , They represent network loss cost, abandoned light cost, load reduction cost, and switch reconfiguration cost respectively; represents the set of time periods belonging to the cth category; Indicates The load reduction active output of node j in period t in scenario; Indicates The actual PV output of node j in period t under the scenario; Indicates The square of the current in branch ij during period t in the scenario; , , , are the single operation costs of feeder tie switches, transformer tie switches, substation tie switches, and section switches, respectively, and ; , Indicates the switch state change flag of branch ij during period t. , then the branch ij is in the first In the scenario, the switch changes from the open state to the closed state during the t period. , then the branch ij is in the first In the scenario, the switch changes from the closed state to the open state during period t; , , , Respectively represent the branch sets of four types of tie switches: feeder tie switches, transformer tie switches, substation tie switches, and section switches in the reconstruction area; B, They represent the node set and PV node set of the reconstruction area respectively; Represents the set of switch branches in the optimization area; , Respectively The cost of load loss and abandoned light in period t under the scenario; is the resistance of line ij; The transformer-level reconstruction model takes the substation as the analysis object and realizes the autonomous operation of the substation unit through the state combination of feeder tie switches, transformer tie switches and section switches. In addition, the cluster identifier c and substation identifier b in the subscripts of other symbols are omitted, and the objective function of the transformer-level reconstruction model is expressed as: （31）； In the formula, Indicates The objective function of transformer-level reconstruction of substation b in the c-th time period under the scenario; The feeder-level reconstruction model takes the transformer as the analysis object and realizes the reliable power supply of the feeder load in the transformer and the full absorption of photovoltaic power through the state combination of the feeder tie switch and the section switch. In addition, the cluster identifier c, substation identifier b, and transformer identifier f in the subscripts of other symbols are omitted, and the objective function of the feeder-level reconstruction model is expressed as: （32）； In the formula, Indicates The objective function of feeder-level reconstruction of transformer f at substation b in the cth time period under the scenario; No. The total number of entities that underwent reconstruction in the lower layer of the scene It is expressed as: （33）； No. The lower level reconstruction layer model in the scenario is expressed as: （34）； In the formula, a is the subject identifier, and the subject includes multiple substations, transformers, transformers, and feeders; , Respectively The cost of the lower-level subject a and the total cost of the lower level in the scenario.

9. The method for multi-level reconstruction of distribution network considering main-distribution linkage under source-load uncertainty according to claim 8 is characterized in that: In S6, the process of converting the reconstruction level evaluation double-layer model into a single-layer model is as follows: The upper and lower layer association constraints are as follows: （35）； In the formula, , They represent the branch set and the total number of branches of the inter-substation tie switch of substation b respectively; , They represent the branch set and the total number of branches of the transformer tie switch of transformer f under substation b respectively; , They represent the branch set and the total number of branches of the feeder tie switch of transformer f under substation b respectively; , They represent the branch set and the total number of branches of the section switches of the feeder of transformer f under substation b respectively; Indicates The switch state of branch ij in period t in the scenario, if , it means that the switch of branch ij is closed; Through the association constraint of formula (35), the level decision layer and the reconstruction layer can be unified, and the reconstruction level evaluation two-layer model can be converted into a single-layer model, which can be expressed as: （36）； In the formula, It represents the expected value of comprehensive cost; , are the upper and lower decision variables, respectively. Including substation-level and transformer-level reconstruction selection variables of all substations in the whole network and feeder-level reconstruction selection variables of each transformer, Including the state variables of the tie switches and branch section switches at different levels of "feeder-transformer-substation"; For the scenario probability of the scenario; In S7, the set constraints include reconstruction level decision constraints, second-order cone power flow constraints, PV output constraints, load loss constraints, safety constraints, and network reconstruction constraints.

10. A multi-level distribution network reconstruction device considering the linkage between the main and the distribution network under the uncertainty of the source and load, characterized by: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the method according to any one of claims 1 to 9 is implemented by executing the program on the processor.

Images