A topology optimization method, system, device and medium of a flexible interconnected low-voltage distribution network under new energy access
Patent Information
- Application Number
- CN202610938301.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-06-26
- Publication Date
- 2026-10-13
Smart Images
Figure CN122908308A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of power system distribution network optimization technology, specifically to a topology optimization method, system, equipment, and medium for a flexible interconnected low-voltage distribution network with new energy access. Background Technology
[0002] With the large-scale integration of distributed photovoltaic and wind power into low-voltage distribution networks, the intermittency and volatility of renewable energy output have exacerbated voltage fluctuations at distribution network nodes and increased the risk of feeder overload. Traditional radial static topology connections are fixed and cannot proactively adjust power flow in response to real-time changes in renewable energy output. Furthermore, existing topology optimization methods fail to model the dynamic regulation characteristics of power electronic devices within the same framework as the topology structure. The computation cycle of centralized optimization models is also much longer than the fluctuation cycle of renewable energy output on a minute-scale timescale, leading to frequent topology mismatches in actual operation. Summary of the Invention
[0003] In view of the above-mentioned problems, the present invention provides a topology optimization method, system, equipment and medium for flexible interconnected low-voltage distribution networks with new energy access.
[0004] Therefore, the technical problem solved by this invention is: how to solve the problem that the existing low-voltage distribution network topology optimization method separates the network topology switch state and the power regulation of flexible interconnection devices into separate models, resulting in a mismatch between the topology decision results and the actual feasible working range of the flexible interconnection devices, and that the calculation cycle of the centralized optimization model is much longer than the minute-level fluctuation cycle of the new energy output, resulting in frequent mismatch of the topology scheme in actual operation.
[0005] To address the aforementioned technical problems, this invention provides the following technical solution: a topology optimization method for a flexible interconnected low-voltage distribution network under new energy access, comprising, Taking a low-voltage distribution network with flexible interconnection devices as the modeling object, a joint optimization model containing discrete and continuous variables is established; The joint optimization model is solved offline to obtain the optimized configuration parameters of the flexible interconnection device; Based on the optimized configuration parameters, an online reinforcement learning decision-making method is adopted to reconstruct the topology switch state of the low-voltage distribution network according to the real-time power output of new energy sources and the real-time load distribution. The reconstruction result is then converted into control commands for the flexible interconnection device to be issued and executed.
[0006] As a preferred embodiment of the topology optimization method for a flexible interconnected low-voltage distribution network under new energy access described in this invention, the offline solution of the joint optimization model includes: With minimum loss as the optimization objective, the topology connection space of the low-voltage distribution network is searched based on the minimum spanning tree algorithm to generate an initial feasible topology scheme. Starting with the initial feasible topology, the node power balance constraint and voltage safety constraint are iteratively solved based on column generation technology to obtain the optimized configuration parameters of the flexible interconnection device.
[0007] As a preferred embodiment of the topology optimization method for a flexible interconnected low-voltage distribution network under renewable energy access described in this invention, the step of reconstructing the topology switching state of the low-voltage distribution network based on real-time renewable energy output and real-time load distribution using an online reinforcement learning decision-making approach includes: The reconstruction process of the topology switch state is modeled as a Markov decision process, with the real-time renewable energy output, the real-time load distribution, and the current topology switch state of the low-voltage distribution network as the state space, and the topology switch action as the action space. The strategy and value function of the Markov decision process are expressed by a function approximator, and the Markov decision process is solved to obtain the topology reconstruction strategy.
[0008] As a preferred embodiment of the topology optimization method for a flexible interconnected low-voltage distribution network under new energy access described in this invention, the strategy and value function expressed by the function approximator for the Markov decision process include: The soft actor-critic algorithm is adopted, which expresses the policy function through the actor network and the value function through the critic network. The state-action pair transition structure generates a post-state representation, which encodes different state-action pairs that result in the same post-state using a unified post-state representation. This post-state representation is then used as the evaluation criterion for the critic network.
[0009] As a preferred embodiment of the topology optimization method for a flexible interconnected low-voltage distribution network under new energy access described in this invention, the method of expressing the strategy function through the actor network includes: Using the node feature matrix and adjacency matrix of the low-voltage distribution network as input, the node feature matrix and adjacency matrix are encoded through a shared coding layer containing linear layers and graph convolutional network modules to obtain the node embedding feature state. The node's embedded feature state is input into the attention network to calculate the node weights of the low-voltage distribution network and generate a weighted node state representation. The node embedding feature state is input into the actor network to generate topological actions; the weighted node state representation and the topological actions are input into the critic network to score the topological actions.
[0010] As a preferred embodiment of the topology optimization method for a flexible interconnected low-voltage distribution network under new energy access described in this invention, the method of reconstructing the topology switching state of the low-voltage distribution network based on real-time new energy output and real-time load distribution using an online reinforcement learning decision-making approach further includes: After the reinforcement learning training is completed, the node embedding feature weights of the low-voltage distribution network are extracted, and the power system node distribution uniformity index is calculated based on the node embedding feature weights. The topological connectivity status of the low-voltage distribution network is quantitatively evaluated based on the node distribution uniformity index.
[0011] As a preferred embodiment of the topology optimization method for a flexible interconnected low-voltage distribution network under new energy access described in this invention, the establishment of a joint optimization model containing discrete and continuous variables includes: The voltage deviation is constrained by the node voltage deviation being within a preset threshold range, the feeder capacity is constrained by the feeder transmission power being within the rated capacity range, and the new energy output fluctuation is constrained by the new energy output being within the predicted fluctuation range. The voltage deviation constraint, the feeder capacity constraint, and the new energy power output fluctuation constraint are introduced as constraints into the joint optimization model.
[0012] This invention provides a topology optimization system for a flexible interconnected low-voltage distribution network with new energy access.
[0013] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a topology optimization system for a flexible interconnected low-voltage distribution network under new energy access, comprising: The modeling module is used to establish a joint optimization model containing discrete and continuous variables, taking a low-voltage distribution network with flexible interconnection devices as the modeling object. An offline solution module is used to solve the joint optimization model offline to obtain the optimized configuration parameters of the flexible interconnection device; The online reconfiguration module is used to reconfigure the topology switch state of the low-voltage distribution network based on the optimized configuration parameters and using an online reinforcement learning decision-making method, according to the real-time power output of new energy sources and the real-time load distribution. The reconfiguration result is then converted into control commands for the flexible interconnection device to be issued and executed.
[0014] The present invention provides a computer device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of the topology optimization method for a flexible interconnected low-voltage distribution network under new energy access.
[0015] The present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the topology optimization method for a flexible interconnected low-voltage distribution network under new energy access.
[0016] The beneficial effects of this invention are as follows: By simultaneously introducing discrete variables representing the network topology connection state and continuous variables representing the adjustment range of the flexible interconnection device into the joint optimization model, and solving both together under the same constraint framework, this invention ensures that the decision result of the topology switching state is always within the physically feasible working range of the flexible interconnection device, thus avoiding the problem of mismatch between the topology decision and the actual adjustable domain of the device caused by separate modeling.
[0017] This invention decouples offline solution and online reinforcement learning decision-making into two independent layers. The online layer directly calls the topology reconstruction strategy obtained from offline training to reconstruct the topology switch state based on real-time renewable energy output and real-time load distribution. This eliminates the need to resolve the optimization model online, enabling the adjustment of the topology switch state to be completed within minutes. This avoids the problem of frequent mismatch between the topology scheme and the actual operating state caused by the long calculation cycle of centralized optimization models. Attached Figure Description
[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0019] Figure 1 This is a flowchart illustrating the overall process of a topology optimization method for a flexible interconnected low-voltage distribution network under new energy access, as provided in one embodiment of the present invention.
[0020] Figure 2 This is a flowchart illustrating the offline solution of the joint optimization model and the online reconstruction of the topology switch state in a topology optimization method for a flexible interconnected low-voltage distribution network under new energy access, provided as an embodiment of the present invention.
[0021] Figure 3 This is a flowchart illustrating the topology reconstruction strategy generation based on the soft actor-commentator algorithm in a topology optimization method for a flexible interconnected low-voltage distribution network under new energy access, provided as an embodiment of the present invention. Detailed Implementation
[0022] To make the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.
[0023] Example 1, referring to Figures 1-3 This is one embodiment of the present invention, which provides a topology optimization method for a flexible interconnected low-voltage distribution network under new energy access, comprising: S1: Taking a low-voltage distribution network with flexible interconnection devices as the modeling object, establish a joint optimization model containing discrete and continuous variables.
[0024] In this embodiment of the application, the low-voltage distribution network includes distribution transformers, lines, loads and distributed power sources, and adopts a three-phase four-wire structure. The loads and distributed power sources are connected in a single-phase manner, and the line voltage is affected by both the line self-impedance and the line mutual impedance.
[0025] The flexible interconnection device is a power electronic conversion device deployed between the feeder nodes of the low-voltage distribution network, including intelligent soft switches and back-to-back converters, which builds a controllable power interaction channel between adjacent feeders.
[0026] The discrete variable is a binary variable that characterizes the on / off state of the network topology switches in the low-voltage distribution network. A value of 0 indicates that the corresponding switch is in the off state, and a value of 1 indicates that the corresponding switch is in the closed state. The topology connection structure of the low-voltage distribution network is determined by the combination of values of the discrete variable.
[0027] The continuous variable is a continuous variable characterizing the power regulation of the flexible interconnect device within its feasible operating range, reflecting the power regulation capability of the flexible interconnect device under different topological connection structures.
[0028] The discrete variables and the continuous variables are both constrained in the joint optimization model, so that the network topology switching state represented by the discrete variables and the power regulation represented by the continuous variables are solved under the same constraint framework.
[0029] The constraints of the joint optimization model are established based on the electrical relationships between the nodes of the low-voltage distribution network. To incorporate the line electrical constraints into the joint optimization model, the network nodes of the low-voltage distribution network are used as modeling units. Based on Kirchhoff's voltage law, the linear mapping relationship between node current vectors and node voltage vectors is used to characterize adjacent nodes. With nodes The electrical relationship between them, this linear mapping relationship is determined by the line impedance matrix. and the admittance matrix between the line and the ground Jointly determined: in, For nodes The three-phase and neutral current vectors; For nodes The three-phase and neutral current vectors; For nodes The three-phase and neutral line voltage vectors; For nodes The three-phase and neutral line voltage vectors; superscript , , Corresponding to three phases respectively, Corresponding to the neutral line, superscript Indicates transpose; This is the fourth-order impedance matrix of a three-phase circuit with a neutral line; This is the admittance matrix between the line and the ground.
[0030] Through the above-mentioned node electrical relationships, the joint optimization model can ensure that the voltage of each node and the line current of the low-voltage distribution network are within a safe range under the topological connection structure determined by the discrete variables, thereby preventing voltage overruns and line overloads in the low-voltage distribution network.
[0031] Furthermore, the line impedance matrix Including the self-impedance and phase-to-phase mutual impedance of a three-phase line, to fully characterize the electrical coupling between phases in a three-phase four-wire line, its expression is: in, , , , They are respectively Mutually, Mutually, The self-impedance of the phase and neutral line; , , , , , , , , , , , The mutual impedances between each phase and the neutral line; all impedance components are in ohms (Ω).
[0032] The aforementioned line impedance matrix provides complete electrical coupling constraints for the three-phase four-wire line in the joint optimization model, ensuring that the discrete variables can reflect the true impedance characteristics of each phase line when determining the topology connection structure.
[0033] Furthermore, the joint optimization model also needs to establish power constraint relationships at transformer nodes. The distribution transformers in the low-voltage distribution network use Dyn11 or Yyn0 connection methods, and their phase shift relationships differ from ordinary line nodes, requiring a separate electrical model for the transformer nodes.
[0034] Transformer side node With load-side nodes The electrical relationships between them are described by the nodal admittance matrix: in, , , , These are the four sub-blocks of the transformer node admittance matrix; For flow through the transformer side node The current vector; Transformer side node The line voltage vector; For flow through load-side nodes The current vector; For load-side nodes The phase voltage vector.
[0035] The aforementioned transformer node admittance matrix establishes an electrical quantity mapping relationship between the transformer side and the load side at the transformer node, enabling the joint optimization model to correctly handle the power constraints at the transformer node.
[0036] To parameterize the short-circuit impedance of each phase winding and to express the electrical relationship of each phase winding in a unified manner, a short-circuit admittance matrix is introduced. It is a diagonal matrix composed of the reciprocals of the short-circuit impedances of each phase: in, , , They are respectively Mutually, Mutually, The short-circuit impedance referred to the single-turn winding side is expressed in ohms (Ω).
[0037] The short-circuit admittance matrix The short-circuit impedance of each phase winding is expressed in a unified diagonal matrix form, providing a parameter basis for establishing the current-voltage relationship of each winding in the future.
[0038] Based on the short-circuit admittance matrix Primary current vector Secondary current vector With the primary side single-turn voltage vector Secondary side single-turn voltage vector The relationship between them is: in, This is the current vector flowing through a single-turn coil on the primary side; This is the current vector flowing through the single-turn coil on the secondary side; This represents the single-turn voltage vector of each phase coil on the primary side; This represents the single-turn voltage vector of each phase coil on the secondary side.
[0039] The above relationship establishes a current-voltage mapping at the winding level, providing a constraint basis for mapping the electrical relationships at the winding level to the voltage and current vectors at the node level.
[0040] To establish a unified transformer constraint and map the electrical relationships at the winding level to the voltage and current vectors at the node level, the current and voltage transformation relationships of an ideal transformer are further introduced: in, , This is the current transformation matrix; , This is the voltage transformation matrix; Transformer side node The line voltage vector; This is the current vector flowing through the corresponding winding; This refers to the turns ratio of the primary side of the transformer to the voltage of a single turn; This refers to the turns ratio of the transformer secondary side to the voltage of a single turn; , Depend on , constitute.
[0041] It should be noted that when the distribution transformer adopts the Dyn11 connection method, there is a phase shift relationship, and the phase correction of the primary current vector is required. With the corrected current vector The relationship is: in, This is the phase-shifting transformation matrix, used to characterize the phase correction relationship of the primary current vector after phase shifting of the transformer.
[0042] Through the above transformer electrical model, the joint optimization model can correctly establish the power constraint relationship at the transformer node under the topological connection structure determined by the discrete variables, ensuring that the solution results of the joint optimization model at the transformer node conform to the physical constraints of the transformer connection method.
[0043] In one optional implementation, the distribution transformer may be a 10kV / 0.4kV three-phase Dyn11 transformer, which is composed of three single-phase transformers. Each single-phase transformer is equivalent to two ideal transformers connected in series with the short-circuit impedance.
[0044] In another optional implementation, the distribution transformer can adopt a Yyn0 connection method, with the neutral line leading out to form a three-phase four-wire system, which is suitable for power distribution scenarios where single-phase loads and three-phase loads are connected at the same time.
[0045] Furthermore, the joint optimization model also needs to establish power balance constraints at load nodes. The loads in the low-voltage distribution network are modeled as PQ-type loads, meaning that the active and reactive power of the loads are known quantities that can be obtained through prediction or measurement, and are directly converted into node injection currents to participate in the constraint modeling of the joint optimization model.
[0046] For the nodes connected to the low-voltage distribution network For a single-phase load, the node injection current for: in, for The active power of the phase load, measured in watts (W). for The reactive power of the phase load, measured in var. for Phase node voltage amplitude, in volts (V); for Phase angle of phase node voltage, in radians (rad). Imaginary unit; superscript Indicates taking the complex conjugate; These correspond to the three phases respectively.
[0047] By injecting current into the aforementioned nodes, the joint optimization model can ensure the power supply and demand balance of each load node in the low-voltage distribution network under the topological connection structure determined by the discrete variables.
[0048] Furthermore, to determine the low-voltage distribution network topology represented by the discrete variables, it is necessary to obtain the upstream and downstream relationships of each node in the low-voltage distribution network. The topology of the low-voltage distribution network consists of four levels of nodes: transformers, branch boxes, metering boxes, and users. Each level of nodes is connected through a three-phase line, and the user node is connected to one phase in a single-phase manner.
[0049] The electrical energy measurement data of each node in the low-voltage distribution network are collected to construct an electrical energy matrix. : in, For the first The node at the th The electrical energy measurement value within each measurement cycle is expressed in kilowatt-hours (kWh). This represents the total number of nodes in the low-voltage distribution network. The total number of measurement cycles is 1. The measuring equipment collects power data once per night, and each measurement cycle corresponds to one nighttime data collection.
[0050] The power matrix By unifying the measurement data from multiple nodes and multiple periods, a data foundation is provided for the subsequent extraction of topological connection information through linear constraint relationships.
[0051] Based on Kirchhoff's current law, under steady-state conditions, the energy outflow from the upstream node is equal to the sum of the energy inflows to all directly downstream nodes. The electrical energy measurement data between all directly connected subordinate nodes satisfy the following linear constraint relationship: in, For the parent node In the Electrical energy measurement values within a measurement cycle; For the first The lower-level node in the _th ... Electrical energy measurement values within a measurement cycle; The value range covers the parent node. All directly connected Each subordinate node The number of subordinate nodes directly connected to each parent node; The range of values is to , This represents the total number of upstream nodes in the low-voltage distribution network.
[0052] By using the above linear constraint relationships, the upstream and downstream affiliation relationships of each node in the low-voltage distribution network are determined, thereby determining the topological connection structure represented by the discrete variables.
[0053] In an optional implementation, when the linear constraint relationship of the electrical energy measurement data is insufficient to uniquely determine the upstream and downstream relationship between adjacent nodes, the upstream and downstream relationship between two adjacent metering box nodes under the same phase can be determined by comparing them. and At the measurement time voltage difference Further judgment is made. The reason for using voltage difference judgment is that in radial distribution networks, due to voltage drop caused by line impedance, the voltage of upstream nodes is always higher than that of downstream nodes. Therefore, the continuous sign of the voltage difference between adjacent nodes under the same phase can reflect the power flow direction and the upstream-downstream relationship.
[0054] in, For nodes At the measurement time The voltage value, in volts (V); For nodes At the measurement time The voltage value, in volts (V); This represents the in-phase voltage difference between two nodes, expressed in volts (V).
[0055] when When the value remains positive throughout the continuous measurement period, the node is determined. Located at node Upstream; when When the value remains negative throughout the continuous measurement period, the node is determined. Located at node Downstream. Based on the voltage difference, when the linear constraint relationship of the electrical energy measurement data cannot uniquely determine the topological connection relationship, the upstream and downstream node relationships of the low-voltage distribution network are completed to ensure that the topological connection structure represented by the discrete variables is completely determined.
[0056] In some embodiments, the specific implementation of establishing a joint optimization model containing discrete and continuous variables in step S1 includes step S11, wherein: S11: The voltage deviation of the node is within a preset threshold range as a voltage deviation constraint, the feeder transmission power is within the rated capacity range as a feeder capacity constraint, and the new energy output is within the predicted fluctuation range as a new energy output fluctuation constraint; the voltage deviation constraint, the feeder capacity constraint, and the new energy output fluctuation constraint are introduced as constraints into the joint optimization model.
[0057] The voltage deviation constraint limits the feasible value range of the discrete variables in the joint optimization model to a combination of topology switch states that makes the voltage deviation of each node in the low-voltage distribution network within the preset threshold range, thereby preventing node voltage exceedance caused by fluctuations in new energy output.
[0058] The feeder capacity constraint limits the feasible range of values for the continuous variable to a power adjustment amount that keeps the feeder transmission power within the rated capacity range, thereby preventing the power adjustment of the flexible interconnect device from causing feeder overload.
[0059] The new energy output fluctuation constraint incorporates the prediction deviation of new energy output into the solution range of the joint optimization model. The three types of constraints jointly limit the feasible solution range of the discrete variables and the continuous variables, ensuring that the solution result of the joint optimization model simultaneously meets the physical constraints of voltage safety, feeder capacity and new energy output fluctuation.
[0060] S2: Solve the joint optimization model offline to obtain the optimized configuration parameters of the flexible interconnection device.
[0061] In this application embodiment, the offline solution refers to pre-solving the joint optimization model before the low-voltage distribution network is put into operation, and determining the optimal configuration scheme of the flexible interconnection device in the low-voltage distribution network for the long-term investment configuration decision of the flexible interconnection device.
[0062] The reason for adopting the offline solution method is that the joint optimization model contains both discrete and continuous variables, which is a mixed integer nonlinear programming problem with high computational complexity. Offline pre-solution can decouple the computational burden from the online operation stage while ensuring optimization quality, so that the online stage can quickly execute topology reconstruction decisions based on the obtained optimization configuration parameters.
[0063] In some embodiments, the specific implementation of solving the joint optimization model offline in step S2 includes steps S21-S22, wherein: S21: With minimum loss as the optimization objective, the topology connection space of the low-voltage distribution network is searched based on the minimum spanning tree algorithm to generate an initial feasible topology scheme.
[0064] In this embodiment, the topology connection space refers to the search space consisting of all possible combinations of the discrete variables. Its size grows exponentially with the number of nodes in the low-voltage distribution network, making direct traversal of all topology combinations computationally infeasible. Therefore, an improved Prim algorithm is used to perform a compressed search of the topology connection space.
[0065] The improved Prim algorithm uses line loss as edge weight and constructs a minimum loss connected tree in the node connection graph of the low-voltage distribution network with minimum loss as the optimization objective, thereby generating the initial feasible topology scheme.
[0066] The reason for adopting the improved Prim algorithm is that the improved Prim algorithm utilizes the minimum spanning tree characteristic of the graph structure, which can compress the search space from all topological combinations to the minimum spanning tree neighborhood while ensuring the connectivity of the low-voltage distribution network, thereby reducing the amount of computation required for iterative solution in step S22.
[0067] S22: Taking the initial feasible topology as the starting point, the node power balance constraint and voltage safety constraint are iteratively solved based on column generation technology to obtain the optimized configuration parameters of the flexible interconnection device.
[0068] In the embodiments of this application, such as Figure 2 As shown, the node power balance constraint refers to the constraint condition that the difference between the injected power and the outflow power of each node in the low-voltage distribution network is within the allowable range, and the voltage safety constraint refers to the constraint condition that the voltage amplitude of each node in the low-voltage distribution network is within the safe operating range.
[0069] Starting with the initial feasible topology, a column generation technique is introduced to process the node power balance constraints and voltage safety constraints. This technique decomposes the solution of the joint optimization model into two levels: a main problem and sub-problems. The main problem solves the relaxation problem of the joint optimization model using the existing set of columns to obtain the current optimal solution. The sub-problems check whether the current optimal solution satisfies all the node power balance constraints and voltage safety constraints. If any constraints are violated, the corresponding constraint column is added to the main problem and solved again. This process is iterated until all constraints are satisfied.
[0070] It should be noted that during the iterative solution process of the column generation technique, the dual secant plane technique is combined to accelerate the convergence of the integer relaxation gap. The dual secant plane technique dynamically adds secant plane constraints during the solution process, narrowing the feasible region of the discrete variables, allowing the optimal solution of the integer relaxation problem to approach the integer optimal solution more quickly. Under the complex coupling conditions of the node power balance constraint and the voltage safety constraint, the joint optimization model is solved, yielding the optimized configuration parameters of the flexible interconnect device.
[0071] S3: Based on the optimized configuration parameters, an online reinforcement learning decision-making method is adopted to reconstruct the topology switch state of the low-voltage distribution network according to the real-time new energy output and real-time load distribution, and the reconstruction result is converted into control commands for the flexible interconnection device to be issued and executed.
[0072] In this embodiment of the application, the online reinforcement learning decision-making method refers to, during the operation of the low-voltage distribution network, based on the optimized configuration parameters obtained in step S2, sensing the changes in the real-time output of new energy sources and the real-time load distribution in real time, and making a reconfiguration decision on the topology switch state of the low-voltage distribution network.
[0073] The reason for adopting the online reinforcement learning decision-making method is that the output of new energy sources fluctuates frequently on a minute time scale, and traditional centralized optimization models are difficult to solve within this time scale. However, after the reinforcement learning agent forms the topology reconstruction strategy through offline training, it can quickly respond to changes in the real-time output of new energy sources and the real-time load distribution online, and reconstruct the switching state of the low-voltage distribution network topology in a timely manner.
[0074] The reconstruction result refers to the topological action output by the reinforcement learning agent in the current state, i.e., the update scheme for the values of the discrete variables. The control command refers to the specific execution instructions generated based on the reconstruction result and directed to the flexible interconnection device.
[0075] In some embodiments, the specific implementation of reconstructing the topology switch state of the low-voltage distribution network using an online reinforcement learning decision-making method in step S3 includes steps S31-S33, wherein: S31: The reconstruction process of the topology switch state is modeled as a Markov decision process, with the real-time renewable energy output, the real-time load distribution and the current topology switch state of the low-voltage distribution network as the state space, and the topology switch action as the action space.
[0076] In this embodiment, the reconstruction process is modeled as a Markov decision process because: the topology switch state of the low-voltage distribution network transitions after each reconstruction action, and the current state is completely determined by the previous state and the current action, which conforms to the Markov property. Therefore, the reconstruction process can be described as a Markov decision process including a state space, action space, state transition probability, immediate reward, and discount factor, wherein the strategy... This represents the probability distribution of actions based on the current state.
[0077] The discount factor is used to characterize the degree of decay of future rewards relative to current rewards, and its value is between 0 and 1. The closer the value is to 1, the higher the importance that the reinforcement learning agent attaches to long-term cumulative rewards.
[0078] The state space is composed of the real-time renewable energy output, the real-time load distribution, and the current topology switching state of the low-voltage distribution network, providing input information for the decision-making of the reinforcement learning agent.
[0079] The action space consists of all possible topology switch actions, and each topology switch action corresponds to an adjustment scheme for the on / off state of one or more switches in the low-voltage distribution network.
[0080] It should be noted that each substation in the low-voltage distribution network is equipped with two busbars, forming a double busbar structure. The load, generators and power lines of the substation can be allocated to either busbar. Current flows only between the equipment on the same busbar. Therefore, each substation can be equivalent to two independent nodes. The scale of the operating space increases with the combination of the number of substations and the busbar allocation method.
[0081] The immediate reward consists of a positive incentive component and a negative penalty component. The positive incentive component includes the increase in the amount of renewable energy consumption and the improvement in the load balance of the feeder; the negative penalty component includes the voltage exceeding the limit at the low-voltage distribution network node, automatic disconnection caused by line overload, and increased active power loss.
[0082] S32: Express the strategy and value function of the Markov decision process through a function approximator, solve the Markov decision process, and obtain the topology reconstruction strategy.
[0083] In this embodiment, the reason for using the function approximator to express the policy and value functions is that the action space is so large that it is impossible to store the value estimates of all state-action pairs in a tabular manner. The function approximator is introduced to parameterize the policy function and value function, so that the reinforcement learning agent can solve the Markov decision process in a large-scale action space and obtain the topology reconstruction policy.
[0084] It should be noted that the online reinforcement learning decision-making method adopts a hybrid strategy combining offline training and online application: in the offline training stage, the reinforcement learning agent performs policy training based on the simulation environment of the low-voltage distribution network to form a complete topology reconfiguration policy; in the online application stage, the reinforcement learning agent directly calls the topology reconfiguration policy obtained from offline training, and outputs the topology action decision based on the real-time perceived output of new energy sources and the real-time load distribution, without the need for online retraining.
[0085] The topology reconstruction strategy refers to the probability distribution of the topology action output by the reinforcement learning agent in any state.
[0086] In some embodiments, the specific implementation of expressing the strategy and value function of the Markov decision process through a function approximator in step S32 includes steps S321-S322, wherein: S321: Employs a soft actor-critic algorithm, expressing the policy function through the actor network and the value function through the critic network.
[0087] In this embodiment, the soft actor-critic algorithm is adopted because: through the collaborative training of the actor network and the critic network, the soft actor-critic algorithm introduces policy entropy regularization while maximizing the cumulative reward, so that the reinforcement learning agent can maintain the ability to explore multiple topology schemes in the low-voltage distribution network topology reconfiguration scenario, and avoids the policy from converging to a local optimum too early.
[0088] like Figure 3 As shown, the actor network is responsible for outputting the probability distribution of the topological actions in the current state, i.e., the policy function.
[0089] The critic network is responsible for evaluating the long-term cumulative value of the current state-action pair, i.e., the value function, and the critic network's rating guides the actor network's strategy update.
[0090] S322: Generate post-state representations based on the transition structure of state-action pairs, and encode different state-action pairs that result in the same post-state with a unified post-state representation, using the post-state representation as the evaluation basis of the critic network.
[0091] In this embodiment of the application, the reason for introducing the post-state representation is that different state-action pairs may lead to the same post-state. If the value of each state-action pair is stored independently, it will result in a large amount of redundant calculation.
[0092] The post-state representation is a specific implementation of approximating the value function through approximate dynamic programming within the Markov decision process framework. By mapping different state-action pairs that lead to the same post-state to the same post-state representation, the critic network can use the more concise post-state representation as the evaluation basis, reducing redundant computation.
[0093] In some embodiments, the specific implementation of expressing the strategy function through the actor network in step S321 includes steps S3211-S3213, wherein: S3211: Using the node feature matrix and adjacency matrix of the low-voltage distribution network as input, the node feature matrix and adjacency matrix are encoded through a shared coding layer containing linear layers and graph convolutional network modules to obtain the node embedding feature state.
[0094] In this embodiment of the application, the reason for using the node feature matrix and the adjacency matrix as inputs is that the low-voltage distribution network has graph structure characteristics. The connection relationship between nodes is represented by the adjacency matrix, and the operating state of each node is represented by the node feature matrix. Together, they constitute a complete description of the current state of the low-voltage distribution network.
[0095] The node feature matrix includes the topology, generator power, load demand, and line transmission power of each node in the low-voltage distribution network; the adjacency matrix describes the connection relationship between each node in the low-voltage distribution network.
[0096] The shared coding layer consists of a linear layer and a graph convolutional network module connected in series. The linear layer uses the ReLU activation function to transform the dimension of the node feature matrix, mapping the original features to a feature space suitable for graph convolution processing. The graph convolutional network module performs graph convolution operations on the transformed node features based on the adjacency matrix, aggregating the feature information of adjacent nodes and outputting the node embedded feature state.
[0097] The reason for using the graph convolutional network module is that the graph convolutional network can effectively capture the adjacency relationship between nodes in the graph topology of the low-voltage distribution network, so that the embedded feature state of the node simultaneously includes the operating state information of each node in the low-voltage distribution network and the structural information of its neighboring nodes.
[0098] S3212: The node's embedded feature state is input into the attention network to calculate the node weights of the low-voltage distribution network and generate a weighted node state representation.
[0099] In this embodiment, the attention network is introduced because the nodes in the low-voltage distribution network have varying degrees of importance in topology reconfiguration decisions, and the state changes of some nodes have a more significant impact on the overall topology performance. The attention network takes the node embedding feature states as input, calculates the node weights of each node in the low-voltage distribution network, applies these weights to the state representations of each node, and generates the weighted node state representations. This enables the critic network to distinguish the differences in importance among the nodes in the low-voltage distribution network during value function evaluation.
[0100] S3213: Input the node embedding feature state into the actor network to generate a topological action; input the weighted node state representation and the topological action into the critic network to score the topological action.
[0101] In this embodiment of the application, the actor network takes the node embedding feature state as input and outputs the topology action in the current state, that is, the probability distribution of the low-voltage distribution network topology switch state adjustment scheme.
[0102] The critic network uses the weighted node state representation and the topological action output by the actor network as joint inputs to score the topological action and evaluate the cumulative value of the current topological action in long-term operation.
[0103] The rating results of the critic network are fed back to the actor network. The training environment is adjusted according to the rating results, and the new node state is obtained after the adjustment. The node attributes of the low-voltage distribution network are updated based on the new node state. The reinforcement learning agent is continuously trained and optimized, so that the reinforcement learning agent gradually forms an autonomous topology management capability in the low-voltage distribution network operation scenario.
[0104] It should be noted that each line in the low-voltage distribution network has a transmission capacity limit. When the transmission power of a line exceeds the transmission capacity limit, the line is automatically disconnected. When the critic network scores the topology actions, it includes the automatic disconnection caused by line overload in the negative penalty component, guiding the actor network to avoid the risk of line overload when generating the topology actions.
[0105] S33: After the reinforcement learning training is completed, extract the node embedding feature weights of the low-voltage distribution network, calculate the power system node distribution uniformity index based on the node embedding feature weights, and quantitatively evaluate the topology connection status of the low-voltage distribution network based on the node distribution uniformity index.
[0106] In this embodiment, after the reinforcement learning agent is trained, the node embedding feature weights corresponding to each node of the low-voltage distribution network are extracted from the attention network. The node embedding feature weights reflect the relative importance of each node in the topology reconfiguration decision, and nodes with higher node embedding feature weights indicate that their state changes have a more significant impact on the overall topology performance.
[0107] The node distribution uniformity index of the power system is calculated based on the node embedding feature weights. There is a negative correlation between the node distribution uniformity and topological vulnerability of the low-voltage distribution network. The higher the node distribution uniformity index, the stronger the resistance of the low-voltage distribution network to disturbances caused by fluctuations in new energy output and sudden load changes.
[0108] The topology connectivity of the low-voltage distribution network is quantitatively evaluated based on the node distribution uniformity index, and weak areas with low node distribution uniformity index are identified in the low-voltage distribution network, providing structural reference for the configuration optimization of the flexible interconnection device and subsequent topology reconfiguration decisions.
[0109] Through step S3, based on the optimized configuration parameters obtained in step S2, the topology switch state of the low-voltage distribution network is dynamically reconstructed using an online reinforcement learning decision-making method. This overcomes the limitation of traditional centralized optimization models that cannot respond to high-frequency fluctuations in renewable energy output on a minute-scale basis. Under the condition of ±40% renewable energy output fluctuation, the voltage qualification rate of the low-voltage distribution network is maintained at 99.3%.
[0110] Example 2 is an embodiment of the present invention, which provides a topology optimization method for a flexible interconnected low-voltage distribution network under new energy access. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through experiments.
[0112] The simulation scenario of the low-voltage distribution network is established based on the IEEE 33-node system. The initial penetration rate of distributed photovoltaic is set to 35%. Node 0 is the benchmark constant voltage node, and other nodes are connected to distributed photovoltaic according to a set ratio. The distributed photovoltaic only generates active power, and the power factor is 1.
[0113] The typical daily load characteristics of the low-voltage distribution network are as follows: within a complete day-night cycle, the load power exhibits a daily variation pattern of morning peak, noon trough, and evening peak. The difference in the real-time load distribution at different times is the basis for setting the feeder capacity constraint and the new energy output fluctuation constraint in the joint optimization model.
[0114] To verify the impact of the distributed photovoltaic (PV) grid connection ratio on the operation of the low-voltage distribution network, the PV grid connection ratio was set to 40%, 60%, 80%, and 100%, respectively. The maximum node voltage, maximum node voltage deviation, and branch active power loss were statistically analyzed for each ratio in the four seasons. The results are shown in Table 1.
[0115] Table 1. Comparison of low-voltage distribution network node data under different proportions of distributed photovoltaic access.
[0116] When the distributed photovoltaic access ratio is 0%, the maximum node voltage of the low-voltage distribution network in spring, summer, autumn and winter is 0.40273kV, 0.40261kV, 0.40270kV and 0.40275kV respectively, and the active power loss of the branch is 14.73043kW, 18.95548kW, 16.34132kW and 17.83052kW respectively. The maximum node voltage deviation is between 0.03951 and 0.03986, all within the preset threshold range of the voltage deviation constraint.
[0117] When the proportion of distributed photovoltaic (PV) grid connection increases to 40%, the active power loss of the branch circuits in the low-voltage distribution network decreases to 10.26844kW, 11.89922kW, 11.02008kW and 12.86128kW in the four seasons, respectively, which is lower than when the grid connection rate is 0%. Appropriate distributed PV grid connection reduces the line transmission power of the low-voltage distribution network and reduces the active power loss of the branch circuits by providing local power supply.
[0118] At the same time, when the distributed photovoltaic access ratio is 40%, the voltage of each node rises by more than 0.4kV during the peak photovoltaic power generation period, and the maximum node voltage deviation is 5.7%, which is close to the allowable upper limit of +7% specified by the voltage deviation constraint. The low-voltage distribution network is in a critical state of voltage over-limit.
[0119] When the proportion of distributed photovoltaic grid connection increases to 80%, the maximum node voltage deviation in autumn reaches 0.07911; when the grid connection reaches 100%, the maximum node voltage deviation in spring and autumn reaches 0.08301 and 0.08304 respectively, exceeding the preset threshold range of the voltage deviation constraint.
[0120] As the proportion of distributed photovoltaic (PV) grid connection increases from 40% to 100%, the active power loss of branch lines generally shows an upward trend. At a 100% grid connection rate, the active power loss of branch lines in winter reaches 18.33368 kW, exceeding the maximum loss value at a 0% grid connection rate. When the proportion of distributed PV grid connection exceeds the rated capacity range specified by the feeder capacity constraint, the new energy output fluctuation constraint in step S11 is indispensable for ensuring the safe operation of the low-voltage distribution network.
[0121] Before and after implementing the methods described in steps S1 to S3, the active power loss of the low-voltage distribution network branches is compared: before implementation, the active power loss of the branches under maximum load conditions is 16.2202kW, and the active power loss of the branches under minimum load conditions is 14.1896kW; after implementation, it is reduced to 6.7923kW under maximum load conditions and to 6.0169kW under minimum load conditions.
[0122] Before and after implementing the methods described in steps S1 to S3, the maximum node voltage deviation of the low-voltage distribution network was compared, and the results are shown in Table 2.
[0123] Table 2 Comparison of maximum node voltage deviation before and after optimization under maximum and minimum load conditions.
[0124] As shown in Table 2, under maximum load conditions, the maximum node voltage deviation decreased from 8.6% before implementation to 5.7%; under minimum load conditions, the maximum node voltage deviation decreased from 10.2% before implementation to 5.4%, both within the preset threshold range of the voltage deviation constraint.
[0125] The changes in active power loss of branches before and after implementing the methods described in steps S1 to S3 under different distributed photovoltaic (PV) access ratios were compared: When the PV access ratio was 40%, the voltage of the low-voltage distribution network node was close to the upper limit of the voltage deviation constraint, and the online reinforcement learning decision-making method described in step S3 caused a slight increase in active power loss of branches while performing voltage control; when the PV access ratio continued to increase, the control effect of the method described in step S3 on active power loss of branches was significantly enhanced, with a maximum reduction of 62.37%.
[0126] To verify the coordinated operation effect between the energy storage system and the low-voltage distribution network, the exchange power of the energy storage system and the over-limit power of the low-voltage distribution network during the peak photovoltaic power generation period (10:00 to 15:00) were compared. The results are shown in Table 3. The exchange power of the energy storage system refers to the power injected or absorbed by the energy storage system into the low-voltage distribution network at each time period; the over-limit power of the low-voltage distribution network refers to the power value exceeding the feeder capacity constraint when the control measures described in step S3 are not implemented.
[0127] Table 3 Comparison of Energy Storage System Exchange Power and Over-Limit Power of Low-Voltage Distribution Network
[0128] As shown in Table 3, during the peak photovoltaic power generation period from 10:00 to 15:00, the total exchange power of the energy storage system in the four typical scenarios is 270.22kW, 258.84kW, 251.19kW and 281.64kW, respectively, and the total over-limit power of the low-voltage distribution network is 477.05kW, 358.13kW, 436.93kW and 464.86kW, respectively. At all times and in all scenarios, the exchange power of the energy storage system is less than the over-limit power of the low-voltage distribution network.
[0129] The daily variation of the state of charge (SOC) of energy storage batteries under different seasons was analyzed: From 0:00 to 7:00, the energy storage batteries are in a static state, and the SOC remains stable. The voltage of each node in the low-voltage distribution network is within the range specified by the voltage deviation constraint. From 8:00 to 9:00, the SOC of the energy storage batteries begins to decline, and the voltage of some nodes touches the lower limit of the voltage safety constraint, triggering the discharge of the energy storage batteries.
[0130] The minimum state of charge (SOC) limit for the energy storage battery is set to 5%; the initial SOC limit is set to 10% to prevent the battery from being in a discharge-triggered state at the start of daily operation; the battery can withstand deep discharge up to the minimum SOC limit. These SOC management parameters ensure that the battery's SOC remains above the minimum SOC limit throughout the entire daily operation.
[0131] To verify the impact of the number of input features of the node feature matrix on decision performance in the online reinforcement learning decision-making method described in step S3, comparative tests were conducted in the IEEE-5 power grid system with the number of input features of the node feature matrix set to six and nine, respectively.
[0132] When the number of input features in the node feature matrix is set to nine, the average scaling score of the reinforcement learning agent is 96.82, the average management steps are 864, and the reinforcement learning agent can perform autonomous topology management of the low-voltage distribution network for an average of about 3 days.
[0133] When the number of input features for the node feature matrix is set to six, the reinforcement learning agent achieves better average scaling scores and average management steps in optimizing the active power loss of the low-voltage distribution network branches compared to when the number of input features is nine. The reason why six input features are superior to nine is that too many input features introduce redundant information, increasing the computational burden on the graph convolutional network module in the shared coding layer of step S3211, and interfering with the accurate calculation of the weights of each node in the low-voltage distribution network by the attention network in step S3212. After simplifying to six input features, the node feature matrix can more effectively describe the key operating states of the low-voltage distribution network, helping the reinforcement learning agent to form a more efficient topology reconstruction strategy.
[0134] Compared with the traditional topology, the method described in steps S1 to S3 reduces the voltage fluctuation of the low-voltage distribution network nodes by 52.8%, and the highest node voltage deviation is reduced from 10.2% to 4.8%. Through the online reinforcement learning decision-making method described in step S3, the dynamic power allocation of the flexible interconnection device is reduced by 41.3%, eliminating the overload risk of key sections. The upper limit of renewable energy consumption of the low-voltage distribution network is increased from the initial penetration rate of 35% to 49.6%, an increase of about 14.6 percentage points, and the curtailment rate is reduced to 2.1%. The optimized configuration parameters obtained by offline solution in step S2 reduce traditional capacity expansion investment by 18.7% and equipment operating losses by 23.4%. Under the test condition of ±40% renewable energy output fluctuation, the voltage qualification rate of the low-voltage distribution network is 99.3%.
[0135] Example 3 is an embodiment of the present invention, which provides a topology optimization system for a flexible interconnected low-voltage distribution network under new energy access, comprising: The modeling module is used to establish a joint optimization model containing discrete and continuous variables, taking a low-voltage distribution network with flexible interconnection devices as the modeling object. An offline solution module is used to solve the joint optimization model offline to obtain the optimized configuration parameters of the flexible interconnection device; The online reconfiguration module is used to reconfigure the topology switch state of the low-voltage distribution network based on the optimized configuration parameters and using an online reinforcement learning decision-making method, according to the real-time power output of new energy sources and the real-time load distribution. The reconfiguration result is then converted into control commands for the flexible interconnection device to be issued and executed.
[0136] This embodiment also provides an electronic device applicable to a topology optimization method for a flexible interconnected low-voltage distribution network under new energy access, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the topology optimization method for a flexible interconnected low-voltage distribution network under new energy access as proposed in the above embodiment.
[0137] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements a topology optimization method for a flexible interconnected low-voltage distribution network under new energy access as proposed in the above embodiments.
[0138] The storage medium proposed in this embodiment and the topology optimization method for a flexible interconnected low-voltage distribution network under new energy access proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.
[0139] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.
[0140] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.
Claims
1. A topology optimization method for a flexible interconnected low-voltage distribution network with new energy access, characterized in that, include: Taking a low-voltage distribution network with flexible interconnection devices as the modeling object, a joint optimization model containing discrete and continuous variables is established; The joint optimization model is solved offline to obtain the optimized configuration parameters of the flexible interconnection device; Based on the optimized configuration parameters, an online reinforcement learning decision-making method is adopted to reconstruct the topology switch state of the low-voltage distribution network according to the real-time power output of new energy sources and the real-time load distribution. The reconstruction result is then converted into control commands for the flexible interconnection device to be issued and executed.
2. The topology optimization method for a flexible interconnected low-voltage distribution network under new energy access as described in claim 1, characterized in that, The offline solution of the joint optimization model includes: With minimum loss as the optimization objective, the topology connection space of the low-voltage distribution network is searched based on the minimum spanning tree algorithm to generate an initial feasible topology scheme. Starting with the initial feasible topology, the node power balance constraint and voltage safety constraint are iteratively solved based on column generation technology to obtain the optimized configuration parameters of the flexible interconnection device.
3. The topology optimization method for a flexible interconnected low-voltage distribution network under new energy access as described in claim 2, characterized in that, The method of reconstructing the topology switching state of the low-voltage distribution network using online reinforcement learning decision-making based on real-time renewable energy output and real-time load distribution includes: The reconstruction process of the topology switch state is modeled as a Markov decision process, with the real-time renewable energy output, the real-time load distribution, and the current topology switch state of the low-voltage distribution network as the state space, and the topology switch action as the action space. The strategy and value function of the Markov decision process are expressed by a function approximator, and the Markov decision process is solved to obtain the topology reconstruction strategy.
4. The topology optimization method for a flexible interconnected low-voltage distribution network under new energy access as described in claim 3, characterized in that, The strategy and value functions expressed by the function approximator for the Markov decision process include: The soft actor-critic algorithm is adopted, which expresses the policy function through the actor network and the value function through the critic network. The state-action pair transition structure generates a post-state representation, which encodes different state-action pairs that result in the same post-state using a unified post-state representation. This post-state representation is then used as the evaluation criterion for the critic network.
5. The topology optimization method for a flexible interconnected low-voltage distribution network under new energy access as described in claim 4, characterized in that, The strategy function expressed through the actor network includes: Using the node feature matrix and adjacency matrix of the low-voltage distribution network as input, the node feature matrix and adjacency matrix are encoded through a shared coding layer containing linear layers and graph convolutional network modules to obtain the node embedding feature state. The node's embedded feature state is input into the attention network to calculate the node weights of the low-voltage distribution network and generate a weighted node state representation. The node embedding feature state is input into the actor network to generate topological actions; the weighted node state representation and the topological actions are input into the critic network to score the topological actions.
6. The topology optimization method for a flexible interconnected low-voltage distribution network under new energy access as described in claim 5, characterized in that, The method of reconstructing the topology switch state of the low-voltage distribution network based on real-time renewable energy output and real-time load distribution using online reinforcement learning also includes: After the reinforcement learning training is completed, the node embedding feature weights of the low-voltage distribution network are extracted, and the power system node distribution uniformity index is calculated based on the node embedding feature weights. The topological connectivity status of the low-voltage distribution network is quantitatively evaluated based on the node distribution uniformity index.
7. The topology optimization method for a flexible interconnected low-voltage distribution network under new energy access as described in claim 6, characterized in that, The establishment of a joint optimization model containing discrete and continuous variables includes: The voltage deviation is constrained by the node voltage deviation being within a preset threshold range, the feeder capacity is constrained by the feeder transmission power being within the rated capacity range, and the new energy output fluctuation is constrained by the new energy output being within the predicted fluctuation range. The voltage deviation constraint, the feeder capacity constraint, and the new energy power output fluctuation constraint are introduced as constraints into the joint optimization model.
8. A topology optimization system for a flexible interconnected low-voltage distribution network with new energy access, employing the topology optimization method for a flexible interconnected low-voltage distribution network with new energy access as described in any one of claims 1 to 7, characterized in that, include: The modeling module is used to establish a joint optimization model containing discrete and continuous variables, taking a low-voltage distribution network with flexible interconnection devices as the modeling object. An offline solution module is used to solve the joint optimization model offline to obtain the optimized configuration parameters of the flexible interconnection device; The online reconfiguration module is used to reconfigure the topology switch state of the low-voltage distribution network based on the optimized configuration parameters and using an online reinforcement learning decision-making method, according to the real-time power output of new energy sources and the real-time load distribution. The reconfiguration result is then converted into control commands for the flexible interconnection device to be issued and executed.
9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the topology optimization method for a flexible interconnected low-voltage distribution network under new energy access as described in any one of claims 1 to 7.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the topology optimization method for a flexible interconnected low-voltage distribution network under new energy access as described in any one of claims 1 to 7.