Micro-grid multi-objective optimization configuration method and system based on double-layer modeling

Abstract

The invention discloses a micro-grid multi-objective optimization configuration method and system based on double-layer modeling, and relates to the technical field of micro-grid planning optimization, and the method comprises the following steps: obtaining equipment operation parameters, branch connection relations and typical daily multi-source power time sequence data of a micro-grid; constructing a first configuration space and constructing a configuration layer model according to the first configuration space; generating a physical topology based on the configuration layer model and performing time sequence simulation to obtain a power flow path reconstruction sequence; extracting loss time sequence distribution of each branch and constructing a loss migration feature set; constructing a branch loss fluctuation constraint based on the loss migration feature set, and feeding back the branch loss fluctuation constraint to the configuration layer model as a feasibility constraint to obtain a corrected configuration layer model; the optimal configuration scheme meeting the feasibility constraint is searched in the first configuration space by adopting the multi-objective optimization algorithm, so that the problem of frequent reconstruction of a power flow path caused by dynamic change of power injection of a distributed power supply and an energy storage device along with time in the operation process of the micro-grid is solved.

Description

Technical Field

[0001] This invention relates to the field of microgrid planning and optimization technology, and more specifically, to a method and system for multi-objective optimization configuration of microgrids based on two-layer modeling. Background Technology

[0002] With the rapid development of distributed energy technology, various types of distributed generation resources, including photovoltaic, wind power, and gas-fired micro-combined heat and power, are gradually being introduced into the distribution network. At the same time, flexible resources such as adjustable loads, energy storage devices, and electric vehicle charging and discharging stations are also being widely deployed on the distribution side. This allows microgrids to no longer rely on a single energy source for stable power supply, but instead achieve dynamic allocation and optimized utilization of electricity through the coordinated scheduling of multiple energy sources and flexible loads. As a result, microgrids have evolved from a traditional single energy supply structure into a coordinated energy supply system with multi-source, multi-mode, and adaptive operation capabilities. Under different operating conditions, they can flexibly respond to load fluctuations and changes in renewable energy output, improve energy utilization efficiency, and ensure power supply reliability.

[0003] However, during microgrid operation, the power injection from distributed generation and energy storage devices exhibits fluctuating characteristics over time, causing dynamic adjustments in the direction and amplitude of current flow in each branch of the network. This leads to reconfiguration of previously relatively stable power flow paths. Frequent reconfiguration of these paths results in a non-static power distribution across microgrid branches, further causing losses to migrate along the branches over time, creating a dynamic distribution pattern where some branches experience concentrated losses while others experience relatively reduced losses. In actual microgrid operation, without effective identification and constraint, this loss migration can leave some branches in a state of high loss for extended periods, leading to localized overload risks, reduced grid operating efficiency, and potentially adverse effects on energy storage dispatch, equipment lifespan, and reliability assessment, further increasing the complexity of microgrid planning and operation optimization.

[0004] To address the above problems, this invention proposes a solution. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a multi-objective optimization configuration method and system for microgrids based on two-layer modeling. By configuring vectors and power flow simulation constraints, the method addresses the problem of frequent power flow path reconfiguration caused by the dynamic changes in power injection from distributed power sources and energy storage devices over time during microgrid operation.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A multi-objective optimization configuration method for microgrids based on two-layer modeling includes the following steps: acquiring the microgrid's equipment operating parameters, branch connection relationships, and typical daily multi-source power time-series data; constructing a first configuration space and building a configuration layer model based on the first configuration space; generating a physical topology based on the configuration layer model, and injecting time-varying power disturbance sequences into the physical topology in a preset operating layer model for time-series simulation to obtain a power flow path reconstruction sequence; extracting the loss time-series distribution of each branch based on the power flow path reconstruction sequence, and identifying the loss migration trajectory during the reconstruction process based on the loss time-series distribution to obtain a loss migration feature set; constructing branch loss fluctuation constraints based on the loss migration feature set, and feeding the branch loss fluctuation constraints back to the configuration layer model as feasibility constraints to obtain a corrected configuration layer model; and using a multi-objective optimization algorithm to search for the optimal configuration scheme that satisfies the feasibility constraints within the first configuration space based on the corrected configuration layer model.

[0007] In a preferred embodiment, the construction of the first configuration space and the construction of the configuration layer model based on the first configuration space specifically involves: based on the device operating parameters and branch connection relationships, combining the distributed power types and capacity ranges that each node is allowed to configure using a Cartesian product to generate several configuration particles, and forming the first configuration space with all configuration particles; extracting the power capacity configuration value and energy storage capacity configuration value of each configuration particle from the first configuration space, and arranging them sequentially according to the node number order to obtain an initial configuration vector; and inputting the initial configuration vector into a preset configuration layer network structure to obtain the configuration layer model.

[0008] In a preferred embodiment, the step of combining the distributed power source types and capacity ranges that each node can be configured with based on the equipment operating parameters and branch connection relationships by performing a Cartesian product to generate several configuration particles and forming a first configuration space by all configuration particles specifically involves: determining the discrete value set of distributed power source capacity that each node can be configured with based on the rated capacity of the distributed power source in the equipment operating parameters, and determining the discrete value set of energy storage device capacity that each node can be configured with based on the load power demand range of each node in the branch connection relationships; pairing the elements in the discrete value set of distributed power source capacity of each node with the elements in the discrete value set of energy storage device capacity to obtain the node configuration subspace of each node; performing a Cartesian product operation on the node configuration subspaces of all nodes to obtain a configuration particle set composed of the configuration combinations of all nodes, and using the configuration particle set as the first configuration space.

[0009] In a preferred embodiment, the step of generating a physical topology based on a configuration layer model and injecting a time-varying power disturbance sequence into the physical topology in a preset operation layer model for time-series simulation to obtain a power flow path reconstruction sequence specifically involves: resolving the configuration vector output by the configuration layer model to determine the power access capacity and energy storage access capacity of each node; adjusting the injected power value of the corresponding node in the preset benchmark topology according to the power access capacity and energy storage access capacity of each node to obtain the physical topology; extracting photovoltaic power output time-series segments and wind power output time-series segments from typical daily multi-source power time-series data; and superimposing the photovoltaic power output time-series segments and wind power output time-series segments time-by-time to obtain the superimposed time-series segments. Power disturbances and power pulse disturbances are randomly inserted into the superimposed time series segments to obtain a time-varying power disturbance sequence. The physical topology is transferred to a preset operating layer model. The time-varying power disturbance sequence is injected into each node of the physical topology in the preset operating layer model segment by segment using a sliding window algorithm. After each injection, the power flow path reconstruction sequence of the physical topology is calculated using a preset power flow algorithm.

[0010] In a preferred embodiment, the step of injecting the time-varying power disturbance sequence into each node of the physical topology in the preset operating layer model segment by segment using the sliding window algorithm, and calculating the power flow path reconstruction sequence of the physical topology using a preset power flow algorithm after each injection, specifically involves: setting the window length and sliding step size of the sliding window, and dividing the time-varying power disturbance sequence into several disturbance segments according to the sliding step size, wherein the length of the disturbance segment is the window length; sequentially selecting each disturbance segment, and superimposing the disturbance value of each moment in the current disturbance segment onto the original injected power of the corresponding node in the physical topology to obtain the injected power time series of the current disturbance segment; performing power flow calculation on the injected power time series at each moment using the preset power flow algorithm to obtain the power flow direction identifier and power amplitude of each branch at each moment; and arranging the branch power flow direction identifier and power amplitude at multiple consecutive moments in chronological order to form the power flow path reconstruction sequence.

[0011] In a preferred embodiment, the step of extracting the loss time-series distribution of each branch based on the power flow path reconstruction sequence and identifying the loss migration trajectory during the reconstruction process based on the loss time-series distribution to obtain a loss migration feature set specifically involves: extracting the initial power value and the final power value of each branch at each moment from the power flow path reconstruction sequence, and taking the difference between the initial power value and the final power value as the instantaneous loss of the branch at that moment; arranging the instantaneous losses at all moments in chronological order to obtain the loss time-series distribution of each branch; traversing the loss time-series distributions of two adjacent moments, calculating the loss change of each branch between the previous moment and the next moment, and marking branches whose loss change exceeds a preset change threshold as loss migration nodes; connecting the loss migration nodes that appear sequentially in multiple consecutive moments in chronological order with time as the axis to form a loss migration trajectory, and summarizing the loss migration trajectories of all branches in different time segments to obtain a loss migration feature set.

[0012] In a preferred embodiment, the step of connecting loss migration nodes appearing sequentially in multiple consecutive time intervals according to time to form a loss migration trajectory, and summarizing the loss migration trajectories of all branches in different time intervals to obtain a loss migration feature set, specifically involves: extracting the node identifier and migration direction of loss migration nodes from the loss change between two adjacent time intervals, and dividing the loss migration nodes into positive migration nodes and negative migration nodes according to their migration direction; starting from the initial time, sequentially finding the positions where positive migration nodes appear, connecting branches with consecutive positive migration nodes in time order to form a positive migration trajectory, and connecting branches with consecutive negative migration nodes in time order to form a negative migration trajectory; aligning the positive and negative migration trajectories according to time, and extracting the branches where the positive and negative migration trajectories intersect as loss migration intersection branches; and summarizing all positive migration trajectories, negative migration trajectories, and loss migration intersection branches to obtain a loss migration feature set.

[0013] In a preferred embodiment, the step of constructing branch loss fluctuation constraints based on the loss migration feature set and feeding these constraints back to the configuration layer model as feasibility constraints to obtain a modified configuration layer model specifically involves: extracting the number of times each branch participates in the loss migration trajectory from the loss migration feature set, comparing the number of times each branch participates in the loss migration trajectory with a preset upper limit for the number of migrations, identifying the first fluctuating branch based on the comparison result, extracting the instantaneous loss of each first fluctuating branch at all times it appears in the loss migration trajectory, and calculating the difference between the maximum and minimum instantaneous loss of each first fluctuating branch to obtain the loss fluctuation amplitude of that branch; comparing the loss fluctuation amplitude with a preset fluctuation amplitude threshold, and identifying the constraint-applied branch based on the comparison result; obtaining the node numbers of the two nodes connected to the constraint-applied branch to obtain a branch constraint pair; and adding the branch constraint pair as a feasibility constraint to the output of the configuration layer model to obtain the modified configuration layer model.

[0014] In a preferred embodiment, the step of using a multi-objective optimization algorithm to search for the optimal configuration scheme that satisfies the feasibility constraints within the first configuration space based on the modified configuration layer model specifically involves: initializing the population of the multi-objective optimization algorithm, mapping each individual in the population to a configuration particle in the first configuration space, and inputting each configuration particle into the modified configuration layer model; performing feasibility constraint verification on the configuration particles based on the modified configuration layer model, marking configuration particles that pass the feasibility constraint verification as feasible particles and those that fail as infeasible particles; calculating multiple objective function values ​​for feasible particles under the multi-objective optimization algorithm, calculating penalty terms for infeasible particles and adding the penalty terms to the objective function values; performing non-dominated sorting and crowding distance calculation based on the objective function values ​​of each particle, selecting the next generation of the population according to the level of non-dominated sorting and crowding distance, iteratively executing the selection process until a preset number of iterations is reached, and extracting the particle with the highest non-dominated level from the final population as the optimal configuration scheme.

[0015] The technical effects and advantages of the microgrid multi-objective optimization configuration method and system based on two-layer modeling in this invention are as follows: This invention acquires the equipment operating parameters, branch connection relationships, and typical daily multi-source power time-series data of a microgrid to construct a first configuration space and generate a configuration layer model based on the configuration space, enabling a systematic expression of the distributed power generation and energy storage capacity configuration of each node in the microgrid. Subsequently, a physical topology is generated based on the configuration layer model, and a time-varying power disturbance sequence is injected into it for time-series simulation, thereby obtaining a power flow path reconfiguration sequence and realizing quantitative analysis of the frequent power flow path reconfiguration process caused by the time-varying power injection of distributed power generation and energy storage. Further, the loss time-series distribution of each branch is extracted, and the loss migration trajectory during the reconfiguration process is identified to form a loss migration feature set, providing quantifiable constraints for branch loss fluctuations. Based on this feature set, branch loss fluctuation constraints are constructed and fed back to the configuration layer model as feasibility constraints, resulting in a corrected configuration layer model, thereby effectively constraining potentially high-fluctuation branches during the capacity configuration stage. Finally, based on the corrected configuration layer model, a multi-objective optimization algorithm is used to search for the optimal configuration scheme in the first configuration space, realizing the coordinated optimization of microgrid capacity configuration and operation scheduling. This invention can effectively control the loss migration caused by power flow path reconfiguration, improve the overall operating efficiency of microgrids, ensure power supply reliability, optimize energy storage usage strategies, extend energy storage life, and significantly improve the economy and reliability of microgrids under complex operating conditions. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the multi-objective optimization configuration method for microgrids based on two-layer modeling according to the present invention.

[0017] Figure 2 This is a schematic diagram of the structure of the microgrid multi-objective optimization configuration system based on two-layer modeling of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0019] Example 1, Figure 1 This invention presents a multi-objective optimization configuration method for microgrids based on two-layer modeling, comprising the following steps: S1. Obtain the equipment operating parameters, branch connection relationships and typical daily multi-source power time series data of the microgrid, construct the first configuration space and construct the configuration layer model based on the first configuration space; In this embodiment, a first configuration space is constructed and a configuration layer model is constructed based on the first configuration space, specifically as follows: Based on the device operating parameters and branch connection relationships, the distributed power types and capacity ranges that each node is allowed to configure are combined by Cartesian product to generate several configuration particles, and all configuration particles constitute the first configuration space. Extract the power capacity configuration value and energy storage capacity configuration value of each configuration particle from the first configuration space, and arrange them in order according to the node number to obtain the initial configuration vector; The initial configuration vector is input into the preset configuration layer network structure to obtain the configuration layer model.

[0020] It should be noted that equipment operating parameters refer to the basic data set used to characterize the operating capabilities and constraints of various electrical devices in a microgrid. Specifically, these include, but are not limited to, the rated capacity, minimum output ratio, ramping capability, start-stop constraints, rated charging and discharging power, capacity upper and lower limits, allowable state of charge range, and power consumption range of load nodes of distributed power sources. Branch connection relationship refers to the set of data relationships used to describe the electrical connection topology between nodes in a microgrid. It is composed of node numbers and the start and end node identifiers of the corresponding branches, which are used to clarify the power transmission path and the connection method between adjacent nodes. Secondly, typical daily multi-source power time series data refers to the set of power sequences of various power sources and loads that change over time on a representative day, extracted based on historical operating data or forecast data. Specifically, it includes photovoltaic power output curves, wind power output curves, and load power curves. It is usually sampled discretely at fixed time intervals, such as forming a time point every 15 minutes or every hour, thus forming multi-dimensional time series data covering the whole day to reflect the fluctuation characteristics of different energy sources in the time dimension.

[0021] Furthermore, after constructing the first configuration space, each configuration particle corresponds to the capacity configuration result of all nodes in a certain combination state. The configuration result of each node includes two parts: the distributed power supply capacity value and the energy storage capacity value, as follows: First, any configuration particle is analyzed. The distributed power capacity and energy storage capacity values ​​corresponding to each node are read one by one according to the node number order. For example, when the node numbers are 1 to N, the power capacity and energy storage capacity values ​​of node 1, node 2, and so on are extracted sequentially until node N. Then, the above extraction results are concatenated to form a one-dimensional data sequence with a fixed dimension and sequential structure, which is the initial configuration vector. In practical applications, for example, for a microgrid with 3 nodes, if the capacity values ​​corresponding to a certain configuration particle are 50kW power and 20kWh energy storage for node 1, 30kW power and 10kWh energy storage for node 2, and 40kW power and 15kWh energy storage for node 3, an initial configuration vector of length 6 can be formed for subsequent processing.

[0022] After obtaining the initial configuration vector, it is passed as the input data carrier to the pre-built configuration layer network structure, as follows: First, the initial configuration vector is standardized to ensure its numerical range meets the input requirements of the preset network structure, for example, by scaling the capacity value to a uniform range. Then, according to the order of the input nodes in the network structure, each element in the initial configuration vector is mapped to the corresponding input node in sequence, and passed level by level through the hierarchical connection relationship within the network structure, so that the input information is feature-mapped and combined between layers. In this process, the configuration layer network structure performs response calculations on the input vector, thereby outputting the evaluation result or feature representation corresponding to the configuration particle. By repeating the above input process for each configuration particle in the first configuration space, the configuration layer model has the ability to map and constrain the input configuration vector.

[0023] It should be noted that the construction of the configuration layer model includes multiple steps such as data preparation, structure definition, mapping relationship establishment, and output definition. Specifically, firstly, a first configuration space is constructed based on device operating parameters and branch connection relationships, and all configuration particles in the first configuration space are transformed into a unified format initial configuration vector set as the input data source for the model. Secondly, the input dimension is determined according to the number of nodes and the number of configuration variables corresponding to each node, and an input node set consistent with the input dimension is constructed. At the same time, several intermediate processing layers are set according to preset rules, each layer is used to perform step-by-step feature extraction and combination expression of the input data. Subsequently, a mapping relationship is established from the input nodes to the intermediate processing layers. The connection relationship from the configuration layer to the output node ensures that each input variable can participate in the construction of the overall configuration features. After the structure definition is completed, the initial configuration vectors are input into the structure one by one, and the output results are marked or filtered according to the preset constraint rules, such as whether the output meets the capacity configuration range, node balance requirements, and subsequent constraint conditions. Finally, the mapping relationship between the input vectors and the corresponding output results is solidified into the configuration layer model, which enables it to quickly determine and output features for any input configuration particle in the subsequent optimization process. The input is the vector data formed by the combination of capacity configurations of each node, and the output is the feasibility identifier and feature description information of the corresponding configuration.

[0024] In this embodiment, based on the device operating parameters and branch connection relationships, the Cartesian product combination of the distributed power types and capacity ranges allowed to be configured for each node is performed to generate several configuration particles, and all configuration particles constitute the first configuration space, specifically: Based on the rated capacity of the distributed power source in the equipment operating parameters, determine the discrete set of distributed power source capacity values ​​that can be configured for each node, and based on the load power demand range of each node in the branch connection relationship, determine the discrete set of energy storage device capacity values ​​that can be configured for each node. Pair the elements in the discrete value set of the distributed power capacity of each node with the elements in the discrete value set of the energy storage device capacity to obtain the node configuration subspace of each node. Perform a Cartesian product operation on the node configuration subspaces of all nodes to obtain a configuration particle set composed of all node configurations, and use the configuration particle set as the first configuration space.

[0025] It should be noted that, using the rated capacity as the upper limit reference, and combining the minimum output ratio, access level, and capacity step requirements for project implementation, the continuous capacity range is divided into several discrete capacity levels with practical engineering significance, thus forming a discrete set of distributed power supply capacity values ​​for each node. For example, when the rated capacity of the photovoltaic power source allowed to be connected to a node is 100kW, and the capacity step is 20kW, discrete values ​​such as 0kW, 20kW, 40kW, 60kW, 80kW, and 100kW can be obtained. At the same time, based on the node power supply path and load distribution reflected by the branch connection relationship, and combined with the load power demand range of each node, the capacity of the energy storage device is discretized for matching, as follows: Based on the peak demand, valley demand, and load fluctuation range of the node load, a reasonable configuration range for energy storage capacity is determined, and the capacity is discretized according to a preset capacity granularity. For example, when the load demand of a node fluctuates between 30kW and 70kW, multiple discrete values ​​such as 10kWh, 20kWh, and 30kWh can be set to meet the needs of peak shaving, valley filling, and power regulation. In this way, a discrete set of distributed power supply capacity and a discrete set of energy storage capacity are formed for each node.

[0026] After obtaining the discrete value set of distributed power capacity and discrete value set of energy storage capacity for each node, the two value sets are combined and paired to construct the configuration subspace of that node. The specific construction steps are as follows: Each capacity value in the discrete set of distributed power generation capacity is sequentially combined with each capacity value in the discrete set of energy storage capacity, so that any power generation capacity value can form a one-to-one correspondence with all energy storage capacity values, thus obtaining multiple capacity combination pairs. Each capacity combination pair represents the capacity allocation scheme of the node in a specific configuration state. For example, when the discrete set of distributed power generation capacity of a node is 0kW, 50kW, and 100kW, and the discrete set of energy storage capacity is 10kWh and 20kWh, six combination states can be obtained by pairing them up, each corresponding to a different node configuration method. Summarizing all the above combination states forms the node configuration subspace of the node, which is used to describe all the possible schemes of the node at the capacity configuration level.

[0027] Furthermore, after constructing the node configuration subspaces for all nodes, the configuration subspaces of each node are combined and expanded to form a set of configuration particles covering the entire network. Specifically, according to the node numbering order, a configuration combination is selected from the configuration subspace of each node in turn, and the selected configuration combinations of each node are concatenated to form a complete network configuration scheme, which is a configuration particle. By traversing and combining the configuration subspaces of all nodes, all possible configurations of each node participate in the combination process, ultimately obtaining a configuration particle composed of all possible node configuration combinations. The set; for example, when a microgrid contains 3 nodes, and each node has 2, 3, and 2 configuration combinations respectively, a total of 12 different network configuration particles can be formed, each configuration particle corresponding to a complete capacity configuration scheme; the unified aggregation of all the above configuration particles constitutes the first configuration space; wherein, the first configuration space is used to characterize the range of all achievable capacity configuration combinations of the microgrid under given equipment parameters and network connection conditions, its essence is a discrete configuration search domain, providing a set of candidate solutions for the subsequent optimization process, and serving as the source of input data for the configuration layer model.

[0028] S2 generates a physical topology based on the configuration layer model, and injects a time-varying power disturbance sequence into the physical topology in the preset operation layer model to perform time-series simulation and obtain the power flow path reconstruction sequence. In this embodiment, a physical topology is generated based on the configuration layer model, and a time-varying power disturbance sequence is injected into the physical topology in a preset operation layer model for timing simulation to obtain a power flow path reconstruction sequence, specifically: The configuration vectors output by the configuration layer model are parsed to determine the power access capacity and energy storage access capacity of each node. Based on the power access capacity and energy storage access capacity of each node, the injected power value of the corresponding node in the preset benchmark topology is adjusted to obtain the physical topology. The photovoltaic power output time series and wind power output time series are extracted from the typical daily multi-source power time series data, and the photovoltaic power output time series and wind power output time series are superimposed on each time series to obtain the superimposed time series. Power perturbations and power pulse perturbations are randomly inserted into the superimposed time segments to obtain a time-varying power perturbation sequence; The physical topology is transferred to the preset operating layer model. The time-varying power disturbance sequence is injected into each node of the physical topology in the preset operating layer model segment by segment using the sliding window algorithm. After each injection, the power flow path reconstruction sequence of the physical topology is calculated using the preset power flow algorithm.

[0029] It should be noted that, firstly, the configuration vector output by the configuration layer model is subjected to structured parsing. Specifically, according to the predetermined node numbering order, each element in the configuration vector is interpreted sequentially, and adjacent elements are respectively mapped to the power supply capacity and energy storage capacity of the same node, thereby recovering the complete capacity configuration result node by node. After parsing, the power supply capacity of each node is regarded as the power generation injection capacity of that node, and combined with the charging and discharging regulation capacity corresponding to the energy storage capacity, the preset benchmark topology is subjected to injection power correction processing. Specifically, on the basis of the original load power in the benchmark topology remaining unchanged, the power supply capacity is superimposed to the corresponding node in a forward injection manner, and the energy storage capacity is converted into equivalent injection or absorption power according to the preset operating state, thereby forming a new injection power distribution for each node. Through the above adjustments, the power injection state of each node in the benchmark topology is kept consistent with the capacity configuration corresponding to the current configuration vector, thus obtaining the physical topology for subsequent analysis. The physical topology refers to the electrical operating structure that comprehensively maps the actual power supply access, energy storage configuration, and load distribution without changing the established network connection structure. It not only includes the connection relationship between nodes and branches, but also reflects the power injection characteristics of each node in the current configuration state, and is used to reflect the actual flow of electrical energy in the network.

[0030] Secondly, photovoltaic (PV) power output data and wind power output data are extracted from typical daily multi-source power time-series data. Specifically, the original data is segmented according to a unified time scale, for example, dividing the entire day's data into several consecutive time points with a 15-minute time interval. PV power output and wind power output values ​​are then read at each time point to form corresponding PV power output time-series segments and wind power output time-series segments. After extraction, the two types of time-series segments are processed moment-by-moment according to the time alignment principle. That is, at the same time index position, the PV power output value and wind power output value are superimposed to obtain the comprehensive renewable energy power output value at that time point, thus forming a new set of superimposed time-series segments. For example, if the PV power output is 30kW and the wind power output is 20kW at a certain moment, the superimposed output at that moment is 50kW. By performing the above processing on all time points throughout the day, superimposed time-series segments reflecting the multi-source coordinated power output characteristics can be obtained for subsequent perturbation construction.

[0031] Furthermore, after obtaining the superimposed time series segment, to enhance the diversity and uncertainty of the operating scenario, power disturbances and power pulse disturbances are introduced into this time series segment to construct a time-varying power disturbance sequence. Specifically, the set of time points for disturbance insertion is first determined. These time points are extracted from the entire time series according to a preset probability or random selection rule; for example, several discrete time points are randomly selected from all time points as the disturbance occurrence times. Subsequently, at each selected time point, a disturbance value is generated according to a preset disturbance amplitude range, and this disturbance value is superimposed onto the original superposition. In terms of output, this results in continuous power disturbances. For power pulse disturbances, abrupt power changes are set at certain selected time points, that is, power increments or decrements significantly higher than the normal fluctuation range are introduced at a single time point or a small number of consecutive time points. For example, the output value is instantaneously increased or decreased by a certain percentage at a certain time point. In actual implementation, the disturbance type can be randomly determined first, and then the disturbance amplitude and duration can be randomly determined to ensure the uncertainty of the disturbance in time distribution and amplitude changes. After processing in the above way, a time-varying power disturbance sequence containing multiple random fluctuation characteristics is obtained.

[0032] Finally, after obtaining the physical topology, it is used as the basic data carrier for operational analysis and transferred to the preset operational layer model. Specifically, the injected power values, branch connection relationships, and related operational parameters of each node in the physical topology are organized according to a predetermined data format and imported into the calculation entry point of the operational layer model as input data. During the transfer process, the consistency of node numbers and branch identifiers is maintained to ensure that the topology information and power injection information can correspond one-to-one. At the same time, the time-varying power disturbance sequence is used as external driving data and correlated with the node injected power in the physical topology, so that the operational layer model can dynamically analyze the power flow based on the power changes in different time segments during subsequent calculations. Through the above data transfer method, the physical topology is fully expressed in the operational layer model, thus providing a foundation for subsequent power flow path reconstruction and loss analysis.

[0033] In this embodiment, a sliding window algorithm is used to inject the time-varying power disturbance sequence segment by segment into each node of the physical topology in the preset operating layer model, and a preset power flow algorithm is used to calculate the power flow path reconstruction sequence of the physical topology after each injection, specifically: Set the window length and sliding step size of the sliding window, and divide the time-varying power perturbation sequence into several perturbation segments according to the sliding step size. The length of the perturbation segment is the window length. Each perturbation segment is selected sequentially, and the perturbation value at each moment in the current perturbation segment is superimposed on the original injection power of the corresponding node in the physical topology to obtain the injection power time series of the current perturbation segment. A preset power flow algorithm is used to perform power flow calculation on the injected power timing at each moment, so as to obtain the power flow direction and power magnitude of each branch at each moment; The branch power flow direction identifiers and power amplitudes at multiple consecutive moments are arranged in chronological order to form a power flow path reconstruction sequence.

[0034] It should be noted that, firstly, based on the temporal resolution and analytical accuracy requirements of the time-varying power perturbation sequence, the window length and sliding step size of the sliding window are preset. The window length is used to limit the continuous time range covered by a single analysis, and the sliding step size is used to limit the time interval between the starting positions of two adjacent analyses. For example, it can be set according to the sampling interval of typical daily data. For instance, when the time series data is sampled at 15-minute intervals, the window length can be set to 4 time points to cover the 1-hour variation process, while the sliding step size can be set to 1 or 2 time points to achieve continuous or semi-overlapping sliding. After completing the parameter settings, using the sliding step size as the unit of advancement, starting from the starting position of the time-varying power perturbation sequence, a continuous subsequence of length equal to the window length is sequentially extracted as a perturbation segment. After each extraction, the starting position is moved backward by one sliding step size, and the above process is repeated until the entire time series range is covered. In this way, multiple perturbation segments with temporal overlap or interval relationships can be obtained, and each perturbation segment corresponds to the perturbation variation within a local time interval.

[0035] After obtaining each perturbation segment, each perturbation segment is selected sequentially according to time order and applied to the node injection power in the physical topology, specifically as follows: For each time point in the current perturbation segment, the corresponding perturbation value is read and superimposed onto the original injected power of the corresponding node in the physical topology. The correspondence between nodes is determined according to the preset allocation rules of the perturbation sequence. For example, the perturbation value can be proportionally allocated to multiple nodes or designated to act on a specific node. During the superposition process, the basis of the original injected power remains unchanged, and only adjustments are made to increase or decrease it, thereby forming a new node injected power distribution at that time point. Subsequently, the above process is repeated for all time points in the current perturbation segment, and the node injected power results of each time point are arranged in chronological order to obtain the injected power time sequence corresponding to the perturbation segment. For example, when a perturbation segment contains four consecutive time points, four sets of node injected power data can be formed sequentially, thus constituting a complete injected power change sequence.

[0036] Furthermore, after obtaining the injected power timing, power flow calculation is performed on the node injected power distribution at each time point. Specifically, the injected power of each node at that time point is used as the input condition, and combined with the branch connection relationship in the physical topology, a preset power flow calculation method is used to solve the power transmission situation in each branch, thereby obtaining the power transmission result of each branch at that time. In the calculation result, for each branch, the power flow direction is determined according to the power transmission direction between its starting node and the ending node. For example, when the power flows from the starting node to the ending node, it is marked as positive, and when it flows in the opposite direction, it is marked as negative. At the same time, the power value of the branch at that time is recorded as the power amplitude. By performing the above calculation process for each time point in the disturbance segment, the power flow direction and corresponding amplitude information of each branch under different time conditions can be obtained, thereby reflecting the dynamic distribution characteristics of power in the network.

[0037] Finally, after completing the identification of branch power flow direction and the acquisition of power amplitude at each time point, the calculation results of multiple consecutive time points are organized and arranged in chronological order to construct a power flow path reconstruction sequence. Specifically, taking the chronological order as the main line, the power flow direction identification and power amplitude of all branches at each time point are recorded as a whole unit, and the recording units of each time point are connected in series to form a sequence structure reflecting the time evolution process. In this sequence, the changes in branch power flow direction between different time points can intuitively reflect the reconstruction process of the power flow path. For example, if a branch is flowing in the forward direction at one time point and turns to flowing in the reverse direction at another time point, it means that the power flow direction of the branch has been adjusted within that time interval. By uniformly arranging and summarizing the time points corresponding to all disturbance segments, a complete power flow path reconstruction sequence is finally obtained to characterize the dynamic changes of the power transmission path in the network under the action of time-varying disturbances.

[0038] S3. Extract the loss time series distribution of each branch based on the power flow path reconstruction sequence, and identify the loss migration trajectory in the reconstruction process based on the loss time series distribution to obtain the loss migration feature set. In this embodiment, the loss time-series distribution of each branch is extracted based on the power flow path reconstruction sequence, and the loss migration trajectory during the reconstruction process is identified based on the loss time-series distribution to obtain a loss migration feature set, specifically: Extract the starting power value and ending power value of each branch at each time step from the power flow path reconstruction sequence, and use the difference between the starting power value and the ending power value as the instantaneous loss of the branch at that time step. Arrange the instantaneous losses at all times in chronological order to obtain the loss time distribution of each branch; Traverse the loss time sequence distribution between two adjacent time points, calculate the loss change of each branch between the previous time point and the next time point, and mark the branches whose loss change exceeds the preset change threshold as loss migration nodes. Using time as the axis, the loss migration nodes that appear sequentially in multiple consecutive time periods are connected in chronological order to form a loss migration trajectory. The loss migration trajectories of all branches in different time periods are then summarized to obtain a loss migration feature set.

[0039] It should be noted that for any branch, based on its starting and ending node identifiers in the topology, the power output value at the starting node and the power received value at the ending node are extracted from the power flow results at the corresponding time. The power value at the starting node represents the amount of electrical energy injected into the branch from that node, and the power value at the ending node represents the amount of electrical energy received at another node after transmission through the branch. After the above extraction is completed, the power values ​​at the starting and ending nodes of the branch at the same time are compared, and the difference between the two is taken as the instantaneous loss of the branch at that time, which reflects the degree of loss of electrical energy during transmission in the branch. For example, if a branch outputs 80kW at the starting node and receives 75kW at the ending node at a certain time, it can be determined that the branch has an instantaneous loss of 5kW at that time. By repeating the above process for all branches and all times, a complete set of branch instantaneous loss data can be obtained.

[0040] After obtaining the instantaneous loss of each branch at different time points, the instantaneous loss data of the same branch are organized and arranged in chronological order to form the time series distribution of loss for that branch, as follows: Using time series as an index, the instantaneous loss values ​​of the same branch at each consecutive time point are recorded sequentially, maintaining the same arrangement order as the original time series, so that the loss changes of the branch can be continuously presented in the time dimension. For example, if the instantaneous loss of a branch at four consecutive time points is 5kW, 6kW, 4kW, and 7kW, the above values ​​are arranged in chronological order to form the loss time series distribution sequence of the branch. By performing the above processing on all branches, a loss time series distribution set covering all branches of the network can be obtained, which can be used to characterize the loss change characteristics of each branch throughout the entire analysis period.

[0041] Finally, after obtaining the time-series distribution of losses for each branch, the data from two adjacent moments in the time series are traversed segment by segment to identify branches with significant loss changes. Specifically, for any branch, the instantaneous loss values ​​corresponding to the previous and next moments are selected sequentially, and the degree of change between the two is calculated. This degree of change is taken as the loss change of the branch within the time interval. Subsequently, this loss change is compared with a pre-set change threshold. When the change exceeds the threshold, it is determined that the branch has experienced significant loss fluctuations within the time interval, and the branch is marked as a loss migration node. For example, when the losses of a branch at two adjacent moments are 4kW and 7kW respectively, and the preset change threshold is 2kW, since the change amplitude reaches 3kW, the branch can be marked as a loss migration node. By repeating the above judgment process for all branches in all adjacent time intervals, a set of branches that have experienced significant loss changes in different time periods can be selected.

[0042] In this embodiment, using time as the axis, loss migration nodes appearing sequentially at multiple consecutive time points are connected in chronological order to form a loss migration trajectory. The loss migration trajectories of all branches in different time segments are then summarized to obtain a loss migration feature set, specifically: Extract the node identifier and migration direction of the loss migration node from the loss change between two adjacent time points, and divide the loss migration node into positive migration node and negative migration node according to the migration direction. Using time as the axis, starting from the initial moment, the positions where positive migration nodes appear are found sequentially. Branches with consecutive positive migration nodes are connected in chronological order to form a positive migration trajectory, and branches with consecutive negative migration nodes are connected in chronological order to form a negative migration trajectory. After aligning the positive and negative migration trajectories according to time, the branch where the positive and negative migration trajectories intersect is extracted as the loss migration intersection branch. By summing up all positive migration trajectories, negative migration trajectories, and loss migration intersection branches, a loss migration feature set is obtained.

[0043] It should be noted that, firstly, based on the connection relationship of the branch in the topology, the starting node number and ending node number corresponding to the branch are extracted, and this number information is used as the node identifier of the loss migration node. Then, the loss change trend of the branch in two adjacent time periods is determined. When the loss value in the later time period increases compared to the previous time period, the loss of the branch is determined to be increasing, and its migration direction is marked as positive migration. When the loss value in the later time period decreases compared to the previous time period, it is determined to be negative migration. Based on the above determination results, the loss migration nodes are divided into positive migration nodes and negative migration nodes. Positive migration nodes are used to represent the state of continuous increase in loss or accumulation in the branch during the time process, and negative migration nodes are used to represent the state of gradual decrease in loss or transfer from the branch to other branches. For example, if the loss of a branch is 3kW and 6kW in two consecutive time periods, it can be determined as a positive migration node, while if the corresponding loss decreases from 6kW to 2kW, it is determined as a negative migration node.

[0044] After classifying the migration nodes, the distribution of migration nodes at each time point is scanned sequentially from the initial time point, using the time series as the main thread. Specifically, at each time point, branches marked as positive migration nodes are found, and their positions in the time series are recorded. When a branch is identified as a positive migration node in multiple consecutive time points, its states at each time point are connected in chronological order to form a continuous positive migration trajectory. For negative migration nodes, the same method is used, i.e., negative migration nodes that appear continuously in multiple consecutive time points are found, and they are connected in chronological order to form a negative migration trajectory. For example, when a branch is continuously marked as a positive migration node from time point 2 to time point 4, a positive migration trajectory covering that time interval can be formed, while another branch is continuously marked as a negative migration node from time point 3 to time point 5, forming a corresponding negative migration trajectory. Through the above processing, two types of time evolution trajectories reflecting the loss enhancement process and the loss reduction process can be obtained respectively.

[0045] Furthermore, after obtaining the positive and negative migration trajectories, the two types of trajectories are aligned according to a unified time base. The specific operation is as follows: Using the same time scale as a reference, the branch states of different trajectories at each time point are arranged accordingly, so that the positive migration trajectory and the negative migration trajectory form a one-to-one correspondence in the time dimension. After completing the time alignment, a comparative analysis is performed on each time point. When a branch appears in both the positive and negative migration trajectories at the same time point, or appears in both the positive and negative migration trajectories in adjacent time points, the branch is determined to be a loss migration convergence branch, which is used to characterize the location where loss interacts or transforms between different migration trends. For example, if a branch changes from a positive migration state to a negative migration state at a certain time point, the branch can be identified as a migration convergence point. In the above way, key turning points in the loss change process can be identified.

[0046] Finally, after identifying the positive migration trajectories, negative migration trajectories, and loss migration intersection branches, the above results are uniformly summarized to form a loss migration feature set. Specifically, all positive migration trajectories are used as feature subsets representing loss accumulation paths, all negative migration trajectories are used as feature subsets representing loss release paths, and the identified loss migration intersection branches are included as key interaction features. During the summarization process, corresponding time interval information and node identification information are added to various trajectories and branches to ensure that the feature data has complete temporal and topological attributes. Through the above integration process, the final loss migration feature set can comprehensively reflect the migration patterns and interaction relationships of loss in each branch in time and space under time-varying disturbances.

[0047] S4. Based on the loss migration feature set, branch loss fluctuation constraints are constructed, and the branch loss fluctuation constraints are fed back to the configuration layer model as feasibility constraints to obtain the corrected configuration layer model. In this embodiment, branch loss fluctuation constraints are constructed based on the loss migration feature set, and these constraints are fed back to the configuration layer model as feasibility constraints, resulting in a modified configuration layer model, specifically: The number of times each branch participates in the loss migration trajectory is extracted from the loss migration feature set, and the number of times each branch participates in the loss migration trajectory is compared with a preset upper limit for the number of migrations. The first fluctuation branch is identified based on the comparison results, and the instantaneous loss of each first fluctuation branch at all times in the loss migration trajectory is extracted. The difference between the maximum and minimum instantaneous loss of each first fluctuation branch is calculated to obtain the loss fluctuation amplitude of the branch. The loss fluctuation amplitude is compared with the preset fluctuation amplitude threshold, and the constraint application branch is identified based on the comparison result; Obtain the node numbers of the two nodes connected by the branch to which the constraint is applied, and thus obtain the branch constraint pair; By adding branch constraint pairs as feasibility constraints to the output of the configuration layer model, the modified configuration layer model is obtained.

[0048] It should be noted that the process involves traversing all positive and negative migration trajectories in the loss migration feature set, identifying the branch identifiers contained in each trajectory, and accumulating the number of times the same branch appears in different trajectories to obtain the total number of times each branch participates in the loss migration process. After the statistics are completed, the participation count of each branch is compared with the preset upper limit of migration count. When the participation count of a branch is close to or exceeds the upper limit, it indicates that the branch frequently participates in the loss migration process in multiple time periods and has high volatility. For example, if a branch appears a total of 5 times in all migration trajectories, while the preset upper limit of migration count is 3 times, it can be determined that the branch exceeds the upper limit.

[0049] After comparing the number of participations with the maximum number of migrations, the branches are classified and filtered based on the comparison results. The specific filtering steps are as follows: Branches that participate more than or reach a preset upper limit for migration times are marked as first fluctuating branches, which are used to characterize the set of branches that exhibit frequent participation characteristics during loss migration. Branches that participate less than the upper limit are not included in this set. For example, in a network containing multiple branches, if the participation times of branch A and branch B are 5 and 4 times respectively, both exceeding the upper limit of 3 times, then branch A and branch B can be identified as first fluctuating branches. Through the above screening process, the focus can be concentrated on branches that have a more significant impact on the overall loss migration.

[0050] Furthermore, after identifying the first fluctuation branch, its loss changes during the loss migration process are further analyzed. The specific analysis steps are as follows: For each first fluctuation branch, all migration trajectories it participates in are traversed, and the instantaneous loss value of the branch at the corresponding time point of each trajectory is extracted. The instantaneous loss values ​​are then summarized in chronological order. After summarization, the maximum and minimum values ​​are identified from all instantaneous loss values ​​of the branch, and the difference between the two is taken as the loss fluctuation amplitude of the branch, which is used to characterize the range of loss change of the branch throughout the migration process. For example, if the instantaneous loss of a branch at different time points is 3kW, 7kW, 5kW and 9kW respectively, then its maximum value is determined to be 9kW and its minimum value to be 3kW, thus obtaining the loss fluctuation amplitude of the branch as 6kW. Through the above processing, the fluctuation intensity characteristics of each first fluctuation branch can be obtained.

[0051] After obtaining the loss fluctuation amplitude of each first fluctuation branch, the fluctuation amplitude is compared with a preset fluctuation amplitude threshold to further screen the key branches that need to be constrained. Specifically, when the loss fluctuation amplitude of a branch exceeds the threshold, it indicates that the branch has a large loss fluctuation during operation, which may have an adverse effect on the overall operational stability. Therefore, the branch is identified as a constraint-applied branch. If the threshold is not exceeded, no constraint processing is performed. For example, when the fluctuation amplitude threshold is set to 4kW and the fluctuation amplitude of a branch is 6kW, the branch can be included in the constraint-applied branch set. Among them, the constraint-applied branch refers to the branch whose capacity configuration or power allocation range needs to be restricted in the subsequent configuration optimization process. The purpose is to suppress the drastic fluctuation of losses on the branch, thereby improving the overall operational stability.

[0052] After determining the constraint-applying branches, a node-level mapping process is performed based on the branch connection relationship. Specifically, for each constraint-applying branch, its starting node number and ending node number in the topology are read, and these two node numbers are recorded as a pair of nodes to form the corresponding branch constraint pair. For example, when a constraint-applying branch connects node 2 and node 5, a node pair can be formed as a constraint expression unit. By repeating the above process for all constraint-applying branches, multiple branch constraint pairs can be obtained to characterize the node connection relationship that needs to be restricted.

[0053] Finally, after obtaining the branch constraint pairs, they are introduced as feasibility constraints into the output judgment stage of the configuration layer model to correct the original configuration layer model. In specific implementation, a constraint verification step is first added to the output of the configuration layer model. The capacity configuration result corresponding to each input configuration vector is checked item by item to determine whether it involves the node combination corresponding to the constraint-applied branch. When the power capacity or energy storage capacity allocated on the two nodes corresponding to the constraint-applied branch by a certain configuration result exceeds the preset limit range, or causes the branch to have excessive power transmission, the configuration result is determined to not meet the feasibility requirements and is marked or removed. Configuration results that meet the constraint conditions are retained. After completing the above constraint embedding, the updated input-output mapping relationship is solidified to obtain the corrected configuration layer model. Through this correction process, the configuration layer model can actively avoid unreasonable configurations corresponding to high-loss fluctuation branches when generating or screening configuration schemes, thereby improving the effectiveness of the final optimization result.

[0054] S5, based on the modified configuration layer model, uses a multi-objective optimization algorithm to search for the optimal configuration scheme that satisfies the feasibility constraints in the first configuration space.

[0055] In this embodiment, based on the modified configuration layer model, a multi-objective optimization algorithm is used to search for the optimal configuration scheme that satisfies the feasibility constraints within the first configuration space, specifically as follows: Initialize the population for the multi-objective optimization algorithm, map each individual in the population to a configuration particle in the first configuration space, and input each configuration particle into the corrected configuration layer model; Based on the revised configuration layer model, the configuration particles are checked for feasibility constraints. Configuration particles that pass the feasibility constraint check are marked as feasible particles, and configuration particles that fail the feasibility constraint check are marked as infeasible particles. For feasible particles, calculate their multiple objective function values ​​under the multi-objective optimization algorithm; for infeasible particles, calculate the penalty term and add the penalty term to the objective function value. Based on the objective function value of each particle, non-dominated sorting and crowding distance calculation are performed. The next generation of the population is selected according to the non-dominated sorting level and crowding distance. The selection process is iterated until the preset number of iterations is reached, and the particle with the highest non-dominated level is extracted from the final population as the optimal configuration scheme.

[0056] It should be noted that, firstly, based on all the configuration combinations already constructed in the first configuration space, the population of the multi-objective optimization algorithm is initialized by random selection or hierarchical extraction, so that each individual in the population corresponds to a specific capacity configuration scheme. Among them, the configuration particle refers to a complete configuration unit composed of the power supply access capacity and energy storage access capacity of each node, which can uniquely represent the capacity distribution of the microgrid under a certain configuration state. During initialization, each individual in the population is mapped to a configuration particle in the first configuration space according to a predetermined encoding rule. For example, the individual encoding is directly mapped to a capacity combination arranged in sequence, such as the power capacity of node 1, the energy storage capacity of node 1, the power capacity of node 2, and the energy storage capacity of node 2. After the mapping is completed, each configuration particle is converted into a corresponding configuration vector form and input into the modified configuration layer model one by one. This enables the model to uniformly process and constrain the configuration schemes represented by different individuals, thereby establishing a one-to-one correspondence between population individuals and configuration particles.

[0057] After inputting the configuration particles into the corrected configuration layer model, each configuration particle is checked item by item using the embedded branch constraint pairs and capacity limit rules. Specifically, the capacity configuration of each node in the configuration particle is analyzed, and it is checked whether it meets the node combination restrictions corresponding to the constraint applied branch. For example, is there a situation where the nodes at both ends of the constraint branch are configured with excessively high capacity, which may lead to potential power concentration problems? At the same time, the basic constraints such as equipment capacity range and node supply and demand balance of the configuration particles are also verified. When a configuration particle meets the preset requirements in all the above verification items, it is marked as a feasible particle. Otherwise, when it violates any constraint condition, it is marked as an infeasible particle. Through this classification process, different individuals in the population are effectively screened, providing a basis for subsequent optimization calculations.

[0058] Furthermore, after classifying particles for feasibility, feasible and infeasible particles are treated differently. For feasible particles, multi-dimensional performance evaluation is performed based on preset multi-objective optimization requirements. The evaluation indicators may include total operating cost, network loss level, and power supply reliability indicators. By analyzing the corresponding operating state of the configured particles, the specific values ​​of each indicator are obtained and used as the objective function value of the particle. For infeasible particles, to avoid them being incorrectly retained during the optimization process, a penalty mechanism is introduced. That is, the penalty intensity is determined according to the degree of constraint violation, and the penalty value is added to its original objective evaluation result, making its overall evaluation result significantly worse than that of feasible particles. For example, when an infeasible particle has obvious capacity overruns on the constraint branch, a large penalty can be added to it, so that it will be preferentially eliminated in the subsequent screening process.

[0059] Finally, after obtaining the target evaluation results of all particles, the population is subjected to non-dominated sorting. Specifically, based on the superiority-inferiority relationship of each particle across multiple target dimensions, particles that are not comprehensively superior to other particles are classified into the first level, and the remaining particles are classified into lower levels accordingly. Within the same level, to maintain the diversity of solutions, the crowding degree of the particle distribution is calculated, prioritizing the retention of particles in sparse regions. Subsequently, according to the principle of non-dominated level priority and crowding degree assistance, the current population is screened, and the best-performing particles are selected to form the next generation population. New candidate particles are generated through crossover, mutation, and other methods. The above sorting and screening process is repeated until the preset number of iterations is reached. In the final population, a representative configuration scheme is selected from the set of particles with the highest non-dominated level as the optimal configuration scheme.

[0060] For example, in a microgrid with three nodes, the optimal configuration scheme obtained by the final selection can be represented as follows: Node 1 is configured with a power supply of 60kW and energy storage of 20kWh, Node 2 is configured with a power supply of 40kW and energy storage of 15kWh, and Node 3 is configured with a power supply of 50kW and energy storage of 25kWh. This scheme achieves a good balance among multiple indicators such as operating cost, loss level and power supply stability, and meets all constraints, thus serving as the final output result.

[0061] Example 2, Figure 2 The present invention provides a multi-objective optimization configuration system for microgrids based on two-layer modeling, including a model building module, a time-series simulation module, a loss identification module, a model correction module, and an optimization configuration module. The model building module is used to acquire the equipment operating parameters, branch connection relationships and typical daily multi-source power time series data of the microgrid, build the first configuration space and build the configuration layer model based on the first configuration space; The timing simulation module is used to generate physical topology based on the configuration layer model, and inject time-varying power disturbance sequence into the physical topology in the preset operation layer model to perform timing simulation and obtain power flow path reconstruction sequence; The loss identification module is used to extract the loss time series distribution of each branch according to the power flow path reconstruction sequence, and to identify the loss migration trajectory in the reconstruction process according to the loss time series distribution to obtain the loss migration feature set. The model correction module is used to construct branch loss fluctuation constraints based on the loss migration feature set, and feed the branch loss fluctuation constraints back to the configuration layer model as feasibility constraints to obtain the corrected configuration layer model. The optimization configuration module is used to search for the optimal configuration scheme that satisfies the feasibility constraints in the first configuration space based on the modified configuration layer model and a multi-objective optimization algorithm.

[0062] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0063] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0064] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0065] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0066] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-objective optimization configuration method for microgrids based on two-layer modeling, characterized in that, Includes the following steps: The system acquires the equipment operating parameters, branch connection relationships, and typical daily multi-source power time-series data of the microgrid, constructs a first configuration space, and builds a configuration layer model based on the first configuration space. It then generates a physical topology based on the configuration layer model and injects a time-varying power disturbance sequence into the physical topology in a preset operation layer model for time-series simulation, obtaining a power flow path reconstruction sequence. Based on the power flow path reconstruction sequence, it extracts the time-series loss distribution of each branch and identifies the loss migration trajectory during the reconstruction process based on the time-series loss distribution, obtaining a loss migration feature set. Branch loss fluctuation constraints are constructed based on the loss migration feature set, and these constraints are fed back to the configuration layer model as feasibility constraints to obtain the corrected configuration layer model. Based on the corrected configuration layer model, a multi-objective optimization algorithm is used to search for the optimal configuration scheme that satisfies the feasibility constraints in the first configuration space.

2. The microgrid multi-objective optimization configuration method based on two-layer modeling according to claim 1, characterized in that, The construction of the first configuration space and the construction of the configuration layer model based on the first configuration space are specifically as follows: Based on the device operating parameters and branch connection relationships, the distributed power types and capacity ranges that each node is allowed to configure are combined by Cartesian product to generate several configuration particles, and all configuration particles constitute the first configuration space. Extract the power capacity configuration value and energy storage capacity configuration value of each configuration particle from the first configuration space, and arrange them in order according to the node number to obtain the initial configuration vector; The initial configuration vector is input into the preset configuration layer network structure to obtain the configuration layer model.

3. The microgrid multi-objective optimization configuration method based on two-layer modeling according to claim 2, characterized in that, Based on device operating parameters and branch connection relationships, the Cartesian product combination of the distributed power types and capacity ranges allowed to be configured for each node is performed to generate several configuration particles, and all configuration particles constitute the first configuration space, specifically: Based on the rated capacity of the distributed power source in the equipment operating parameters, determine the discrete set of distributed power source capacity values ​​that can be configured for each node, and based on the load power demand range of each node in the branch connection relationship, determine the discrete set of energy storage device capacity values ​​that can be configured for each node. Pair the elements in the discrete value set of the distributed power capacity of each node with the elements in the discrete value set of the energy storage device capacity to obtain the node configuration subspace of each node. Perform a Cartesian product operation on the node configuration subspaces of all nodes to obtain a configuration particle set composed of all node configurations, and use the configuration particle set as the first configuration space.

4. The microgrid multi-objective optimization configuration method based on two-layer modeling according to claim 3, characterized in that, The process involves generating a physical topology based on a configuration layer model, injecting a time-varying power disturbance sequence into the physical topology within a preset runtime layer model for timing simulation, and obtaining a power flow path reconstruction sequence. Specifically: The configuration vectors output by the configuration layer model are parsed to determine the power access capacity and energy storage access capacity of each node. Based on the power access capacity and energy storage access capacity of each node, the injected power value of the corresponding node in the preset benchmark topology is adjusted to obtain the physical topology. The photovoltaic power output time series and wind power output time series are extracted from the typical daily multi-source power time series data, and the photovoltaic power output time series and wind power output time series are superimposed on each time series to obtain the superimposed time series. Power perturbations and power pulse perturbations are randomly inserted into the superimposed time segments to obtain a time-varying power perturbation sequence; The physical topology is transferred to the preset operating layer model. The time-varying power disturbance sequence is injected into each node of the physical topology in the preset operating layer model segment by segment using the sliding window algorithm. After each injection, the power flow path reconstruction sequence of the physical topology is calculated using the preset power flow algorithm.

5. The microgrid multi-objective optimization configuration method based on two-layer modeling according to claim 4, characterized in that, The process involves using a sliding window algorithm to inject the time-varying power disturbance sequence segment by segment into each node of the physical topology in the preset operating layer model, and then using a preset power flow algorithm to calculate the power flow path reconstruction sequence of the physical topology after each injection. Specifically: Set the window length and sliding step size of the sliding window, and divide the time-varying power perturbation sequence into several perturbation segments according to the sliding step size. The length of the perturbation segment is the window length. Each perturbation segment is selected sequentially, and the perturbation value at each moment in the current perturbation segment is superimposed on the original injection power of the corresponding node in the physical topology to obtain the injection power time series of the current perturbation segment. A preset power flow algorithm is used to perform power flow calculation on the injected power timing at each moment, so as to obtain the power flow direction and power magnitude of each branch at each moment; The branch power flow direction identifiers and power amplitudes at multiple consecutive moments are arranged in chronological order to form a power flow path reconstruction sequence.

6. The microgrid multi-objective optimization configuration method based on two-layer modeling according to claim 5, characterized in that, The loss time-series distribution of each branch is extracted based on the power flow path reconstruction sequence, and the loss migration trajectory during the reconstruction process is identified based on the loss time-series distribution to obtain a loss migration feature set, specifically as follows: Extract the starting power value and ending power value of each branch at each time step from the power flow path reconstruction sequence, and use the difference between the starting power value and the ending power value as the instantaneous loss of the branch at that time step. Arrange the instantaneous losses at all times in chronological order to obtain the loss time distribution of each branch; Traverse the loss time sequence distribution between two adjacent time points, calculate the loss change of each branch between the previous time point and the next time point, and mark the branches whose loss change exceeds the preset change threshold as loss migration nodes. Using time as the axis, the loss migration nodes that appear sequentially in multiple consecutive time periods are connected in chronological order to form a loss migration trajectory. The loss migration trajectories of all branches in different time periods are then summarized to obtain a loss migration feature set.

7. The microgrid multi-objective optimization configuration method based on two-layer modeling according to claim 6, characterized in that, The loss migration feature set is obtained by connecting loss migration nodes that appear sequentially in multiple consecutive time periods, using time as the axis, in chronological order. This is done by summarizing the loss migration trajectories of all branches in different time intervals. Extract the node identifier and migration direction of the loss migration node from the loss change between two adjacent time points, and divide the loss migration node into positive migration node and negative migration node according to the migration direction. Using time as the axis, starting from the initial moment, the positions where positive migration nodes appear are found sequentially. Branches with consecutive positive migration nodes are connected in chronological order to form a positive migration trajectory, and branches with consecutive negative migration nodes are connected in chronological order to form a negative migration trajectory. After aligning the positive and negative migration trajectories according to time, the branch where the positive and negative migration trajectories intersect is extracted as the loss migration intersection branch. By summing up all positive migration trajectories, negative migration trajectories, and loss migration intersection branches, a loss migration feature set is obtained.

8. The microgrid multi-objective optimization configuration method based on two-layer modeling according to claim 7, characterized in that, The branch loss fluctuation constraint is constructed based on the loss migration feature set, and the branch loss fluctuation constraint is fed back to the configuration layer model as a feasibility constraint to obtain the corrected configuration layer model, specifically: The number of times each branch participates in the loss migration trajectory is extracted from the loss migration feature set, and the number of times each branch participates in the loss migration trajectory is compared with a preset upper limit for the number of migrations. The first fluctuation branch is identified based on the comparison results, and the instantaneous loss of each first fluctuation branch at all times in the loss migration trajectory is extracted. The difference between the maximum and minimum instantaneous loss of each first fluctuation branch is calculated to obtain the loss fluctuation amplitude of the branch. The loss fluctuation amplitude is compared with the preset fluctuation amplitude threshold, and the constraint application branch is identified based on the comparison result; Obtain the node numbers of the two nodes connected by the branch to which the constraint is applied, and thus obtain the branch constraint pair; By adding branch constraint pairs as feasibility constraints to the output of the configuration layer model, the modified configuration layer model is obtained.

9. The microgrid multi-objective optimization configuration method based on two-layer modeling according to claim 8, characterized in that, The modified configuration layer model employs a multi-objective optimization algorithm to search for the optimal configuration scheme that satisfies the feasibility constraints within the first configuration space, specifically as follows: Initialize the population for the multi-objective optimization algorithm, map each individual in the population to a configuration particle in the first configuration space, and input each configuration particle into the corrected configuration layer model; Based on the revised configuration layer model, the configuration particles are checked for feasibility constraints. Configuration particles that pass the feasibility constraint check are marked as feasible particles, and configuration particles that fail the feasibility constraint check are marked as infeasible particles. For feasible particles, calculate their multiple objective function values ​​under the multi-objective optimization algorithm; for infeasible particles, calculate the penalty term and add the penalty term to the objective function value. Based on the objective function value of each particle, non-dominated sorting and crowding distance calculation are performed. The next generation of the population is selected according to the non-dominated sorting level and crowding distance. The selection process is iterated until the preset number of iterations is reached, and the particle with the highest non-dominated level is extracted from the final population as the optimal configuration scheme.

10. A microgrid multi-objective optimization configuration system based on two-layer modeling, applied to the microgrid multi-objective optimization configuration method based on two-layer modeling as described in any one of claims 1-9, characterized in that, It includes a model building module, a time series simulation module, a loss identification module, a model correction module, and an optimization configuration module: The model building module is used to acquire the equipment operating parameters, branch connection relationships and typical daily multi-source power time series data of the microgrid, build the first configuration space and build the configuration layer model based on the first configuration space; The timing simulation module is used to generate physical topology based on the configuration layer model, and inject time-varying power disturbance sequence into the physical topology in the preset operation layer model to perform timing simulation and obtain power flow path reconstruction sequence; The loss identification module is used to extract the loss time series distribution of each branch according to the power flow path reconstruction sequence, and to identify the loss migration trajectory in the reconstruction process according to the loss time series distribution to obtain the loss migration feature set. The model correction module is used to construct branch loss fluctuation constraints based on the loss migration feature set, and feed the branch loss fluctuation constraints back to the configuration layer model as feasibility constraints to obtain the corrected configuration layer model. The optimization configuration module is used to search for the optimal configuration scheme that satisfies the feasibility constraints in the first configuration space based on the modified configuration layer model and a multi-objective optimization algorithm.

Images