A Real-Time Decision-Making Method and System for Power Balance in Distribution Networks Based on Topology Sensing

Abstract

This invention discloses a real-time decision-making method and system for power balance in distribution networks based on topology awareness. The method includes: collecting distribution network topology data, real-time power flow data, and source-load prediction data, and preprocessing the collected data; performing chromosome encoding and initial population construction based on a topology awareness strategy; constructing a multi-objective function and dynamic constraint penalty weights, and constructing an individual fitness function; adopting a hierarchical tournament selection mechanism, designing a topology-preserving crossover operator and a hierarchical mutation operator, updating the population, and maintaining an external archive set; performing real-time decision optimization based on a rolling time-domain window and prediction compensation, determining whether the convergence criterion is met, and outputting the optimal decision instruction if the convergence criterion is met. This invention solves the problems of many invalid solutions, poor constraint adaptability, high topology destruction rate, and insufficient real-time performance in traditional methods, and realizes multi-objective collaborative optimization and rapid decision-making for power balance in distribution networks.

Description

Technical Field

[0001] This invention relates to the field of power distribution network operation control technology, specifically to a real-time decision-making method and system for power balance in power distribution networks based on topology-aware multi-objective evolution and dynamic constraint relaxation. Background Technology

[0002] Driven by the "dual carbon" goals, the penetration rate of distributed power sources (such as photovoltaics and wind power) in distribution networks continues to increase. This has led to bidirectional and fluctuating power flows in distribution networks, resulting in frequent problems such as main transformer back-feeding overload, excessive voltage deviation, and difficulties in integrating new energy sources, seriously threatening the safe and stable operation of distribution networks. Real-time power balance decision-making in distribution networks, as a core means of ensuring grid operation, must simultaneously consider multi-objective optimization and topology constraint satisfaction. Its decision-making accuracy and real-time performance directly determine the operational quality of the distribution network.

[0003] However, traditional power balance decision-making methods for distribution networks have significant limitations: Too many invalid solutions in population initialization: Traditional multi-objective evolutionary algorithms use random initialization strategies, failing to consider the radial topological constraints of the distribution network. This leads to a large number of initial individuals violating topological rules, resulting in a high proportion of invalid solutions, slow algorithm convergence, and difficulty meeting the time requirements of real-time decision-making. Lack of dynamic adaptability in constraint handling: Using fixed penalty weights to handle constraint violations cannot adjust constraint rigidity according to the real-time operating state of the grid (e.g., heavy load, light load). In emergency situations, constraint relaxation can easily lead to safety risks, while in light load situations, overly strict constraints can easily lead to missed optimization space. High topology disruption rate in mixed variable processing: Distribution network decision-making involves a mixed space of equipment switching (discrete variables) and power regulation (continuous variables). The genetic operations of traditional NSGA (Non-dominated Sorting Genetic Algorithms) easily disrupt the integrity of the grid topology, leading to the loss of feasible solutions during the evolution process and deteriorating optimization performance. The balance between real-time performance and convergence is difficult to achieve: Traditional algorithms have not been specifically optimized for distribution network decision-making scenarios, have slow convergence speeds, and have not adopted effective acceleration strategies, making it difficult to output optimal decision results in a short period of time and unable to meet the real-time control requirements of minute-level fluctuations in distribution networks.

[0004] Therefore, there is an urgent need for a real-time decision-making method for power balance in distribution networks that integrates topology sensing, dynamic constraint relaxation, and efficient evolutionary optimization to improve the feasibility, accuracy, and real-time nature of decision-making. Summary of the Invention

[0005] The purpose of this invention is to provide a real-time decision-making method and system for power balance in distribution networks based on topology awareness, so as to solve the problems of multiple invalid solutions, poor constraint adaptability, high topology destruction rate and insufficient real-time performance in existing methods.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solution: On one hand, this invention provides a real-time decision-making method for power balance in distribution networks based on topology awareness, comprising: collecting distribution network topology data, real-time power flow data, and source-load prediction data, and preprocessing the collected data to obtain preprocessed data; performing chromosome encoding and initial population construction on the preprocessed data based on a topology awareness strategy; constructing a multi-objective function and dynamic constraint penalty weights, and constructing an individual fitness function based on the multi-objective function and the dynamic constraint penalty weights; adopting a hierarchical tournament selection mechanism, designing a topology-preserving crossover operator and a hierarchical mutation operator, updating the population, and maintaining an external archive set; performing real-time decision optimization based on a rolling time-domain window and prediction compensation to obtain optimization results, and determining whether the optimization results meet the convergence criterion; if the convergence criterion is met, outputting the optimal decision instruction in the current non-dominated solution set.

[0007] Preferably, it further includes: after the optimal decision instruction is issued, monitoring the response deviation of the distribution network operation status and feeding it back to the next optimization cycle.

[0008] Preferably, the topology-aware strategy for chromosome encoding and initial population construction of the preprocessed data specifically includes: Establish a two-dimensional chromosome structure containing topological and power information. : in, This is a Boolean adjacency matrix used to encode the switch states of the distribution network. An element with a value of "1" indicates a closed matrix. A value of "0" in the middle element indicates a disconnection; It is a continuous power regulation vector used to encode adjustable parameters, which include: main transformer output, tie line transmission power and distributed generation output; A set of topology trees satisfying radial constraints is generated by traversing the distribution network topology data in the preprocessed data using a depth-first search algorithm. Individuals are selected from the set of topological trees to construct an initial population, and the adjacency matrix for each individual is... All are connected acyclic graphs, and duplicate topological individuals are removed by graph isomorphism detection. Initial population size Number of distribution network nodes satisfy ,in: This represents the initial population size. This represents the number of nodes in the distribution network.

[0009] Preferably, the construction of the multi-objective function and dynamic constraint penalty weights, and the construction of the individual fitness function based on the multi-objective function and the dynamic constraint penalty weights, specifically includes: Construct a multi-objective optimization function based on "safety, economy, and efficiency" to achieve a coordinated balance among multiple objectives: The core objective is to minimize the main transformer backfeed overload rate. The formula is: in, For the first The reverse transmission power of the main transformer; For the first Rated capacity of the main transformer; To minimize distribution network losses Maximize photovoltaic absorption rate Minimize node voltage deviation As a secondary objective, a multi-objective optimization function is established; Risk sensitivity function based on real-time power flow calculation: in, The degree of impact of time-unit adjustment actions on heavy load risks. This represents the risk value of overload on the power grid. The moderating factor for the decision variable; The formula for constructing the dynamic constraint penalty weight is as follows: in for Dynamically constrain penalty weights at all times. Based on the penalty weight, , To adjust the coefficients, dynamic coupling between the constraint penalty weights and the real-time risk gradient of the power grid is achieved; The individual fitness function is constructed as follows: in, For individuals Overall adaptability; For the first The weights of each objective function; Indicates the first One objective function, ; for One constraint violation quantity; This indicates a penalty for violating a constraint, ensuring that punishment is only imposed on the individual who violates the constraint.

[0010] Preferably, the hierarchical tournament selection mechanism includes: the first layer sorting by non-dominant level, the second layer sorting by crowding distance, and the third layer introducing a topological complexity penalty term.

[0011] Preferably, the design of the topology-preserving crossover and hierarchical mutation operators, updating the population, and maintaining the external archive set specifically includes: For a mixed discrete-continuous decision space, a topology-preserving crossover operator is designed, which considers the Boolean adjacency matrix of the parent individuals. , After subgraph swapping, loops and isolated nodes are automatically detected using a depth-first search algorithm, and topological defects are repaired using the minimum edge deletion method to ensure that the adjacency matrix of the offspring still satisfies the radial constraint. The mutation operator adopts a hierarchical strategy to adjust the continuous power vector. Polynomial mutation is employed to ensure adjustment accuracy for Boolean adjacency matrices. An edge-flipping mutation based on degree distribution is employed; An external archive set is constructed to store non-dominated solutions during the evolution process. A hash table is used to record the topological feature values ​​of each non-dominated solution, and redundant solutions are removed through graph isomorphism detection.

[0012] Preferably, the real-time decision optimization based on rolling time-domain windows and prediction compensation specifically includes: Set a rolling time window, perform an optimization solution once within each rolling time window, and output only the decision instructions for the first time period; Introducing short-term source-load forecast data For the main transformer reverse transmission heavy load rate Perform feedforward correction, the correction formula is: in, The corrected core objective function value; To predict compensation weights, for Predicted source-load power at time point. for Actual value of source load power at any given time. The distance is Euclidean.

[0013] Preferably, the step of determining whether the convergence criterion is met based on the optimization result, and outputting the optimal decision instruction if the convergence criterion is met, specifically includes: Using the hypervolume index of the non-dominated solution set as the core convergence evaluation index, when the hypervolume index of the non-dominated solution set increases by less than 0.1% for three consecutive generations and the population constraint violation rate is less than 1%, the algorithm is determined to have converged, the iteration is stopped, and the optimal decision scheme in the current non-dominated solution set is output.

[0014] Preferably, the method further includes: dividing the initial population into clusters based on topological similarity, and using a distributed computing architecture to compute the fitness of individuals in each cluster in parallel based on the topological structure of individuals within the same cluster, so as to reduce redundant computation. Lightweight online deployment: Train a graph neural network agent model using offline samples, and use the trained graph neural network agent model to replace the initial population individual fitness calculation in the preset part of the online iteration process.

[0015] On the other hand, the present invention also provides a topology-aware real-time power balance decision-making system for distribution networks, which is implemented using the above-mentioned topology-aware real-time power balance decision-making method for distribution networks. The system includes: The data input layer collects distribution network topology data, real-time power flow data, and source load prediction data, and preprocesses the collected data to obtain preprocessed data. A topology-aware module, which is connected to the data input layer, performs chromosome encoding and initial population construction on the preprocessed data based on a topology-aware strategy. A multi-objective optimization module, connected to the topology sensing module, constructs a multi-objective function and dynamic constraint penalty weights, and constructs an individual fitness function based on the multi-objective function and the dynamic constraint penalty weights; The evolutionary computation module employs a hierarchical tournament selection mechanism, designs topology-preserving crossover operators and hierarchical mutation operators, updates the population, and maintains an external archive set. A real-time decision-making module, which is connected to the evolutionary computation module, is used to perform real-time decision optimization based on a rolling time-domain window and prediction compensation to obtain optimization results. The decision output layer is connected to the real-time decision module. It determines whether the convergence criterion is met based on the optimization result. If the convergence criterion is met, it outputs the optimal decision instruction.

[0016] Compared with the prior art, the present invention has the following beneficial effects: This invention provides a real-time decision-making method and system for power balance in distribution networks based on topology awareness, which improves the feasibility of initial population: the topology-aware population initialization strategy generates a radial topology tree through a depth-first search algorithm and performs graph isomorphism detection, laying a time foundation for real-time decision-making; This invention enhances the constraint adaptability: the dynamic constraint penalty mechanism realizes real-time linkage between constraint weight and power grid risk level, ensuring power grid safety in emergency situations, improving optimization space in light load situations, and significantly improving the scenario adaptability of decision-making schemes; This invention reduces the risk of topology destruction: an improved evolutionary operator for mixed variables ensures the integrity of the power grid topology during genetic operations, reducing the topology destruction rate to below 1%, avoiding the loss of feasible solutions, and improving optimization performance; This invention ensures real-time performance and accuracy: through parallel computing and lightweight design of graph neural network surrogate models, the online solution time is shortened to the minute level, meeting the needs of rolling decision-making; multi-objective optimization and predictive feedback correction ensure that the decision-making scheme takes into account safety, economy and efficiency. Attached Figure Description

[0017] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description will be briefly introduced below. Obviously, the drawings described below are one embodiment of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort: Figure 1 A flowchart of a real-time decision-making method for power balance in distribution networks based on topology-aware multi-objective evolution and dynamic constraint relaxation, provided in an embodiment of the present invention; Figure 2 This is an architecture diagram of a real-time power balance decision-making system for distribution networks based on topology-aware multi-objective evolution and dynamic constraint relaxation, provided in an embodiment of the present invention. Figure 3 A flowchart of topology-aware chromosome encoding and initial population construction provided in an embodiment of the present invention; Figure 4 This is a flowchart of an improved non-dominated sorting and genetic operation provided in an embodiment of the present invention; Figure 5 This is a flowchart of rolling time-domain real-time decision-making and deviation feedback optimization provided in an embodiment of the present invention. Detailed Implementation

[0018] The following is in conjunction with the appendix Figure 1-5 The following detailed description further illustrates the real-time power balance decision-making method and system for distribution networks based on topology sensing proposed in this invention. The advantages and features of this invention will become clearer from the following description. It should be noted that the accompanying drawings are in a very simplified form and use non-precise scales, used only to facilitate and clearly illustrate the purpose of the embodiments of this invention. Please refer to the accompanying drawings to make the objectives, features, and advantages of this invention more apparent and understandable. It should be understood that the structures, scales, sizes, etc., depicted in the accompanying drawings are only used to complement the content disclosed in the specification, for those skilled in the art to understand and read, and are not intended to limit the implementation conditions of this invention. Therefore, they have no substantial technical significance. Any modifications to the structure, changes in the proportional relationships, or adjustments to the size, without affecting the effects and objectives achieved by this invention, should still fall within the scope of the technical content disclosed in this invention.

[0019] On the one hand, refer to Figure 1As shown, this embodiment provides a real-time power balance decision-making method for distribution networks based on topology awareness, including: Step S1: Collect distribution network topology data, real-time power flow data, and source-load prediction data, and preprocess the collected data to obtain preprocessed data. The preprocessing includes topology data cleaning and standardization.

[0020] Step S2: Reference Figure 3 As shown, based on the topology-aware strategy, chromosome encoding and initial population construction are performed on the preprocessed data, specifically including: Establish a two-dimensional chromosome structure containing topological and power information. : in, This is a Boolean adjacency matrix used to encode the switch states of the distribution network. An element with a value of "1" indicates a closed matrix. A value of "0" in the middle element indicates a disconnection; It is a continuous power regulation vector used to encode adjustable parameters, including: main transformer output, tie-line transmission power, and distributed generation output.

[0021] The distribution network topology data in the preprocessed data is traversed using the DFS (Depth First Search) algorithm to generate a set of topology trees that satisfy radial constraints. Individuals are selected from the set of topological trees to construct an initial population, and the adjacency matrix for each individual is... All are connected acyclic graphs, and duplicate topological individuals are removed by graph isomorphism detection.

[0022] Initial population size Number of distribution network nodes satisfy ,in: This represents the initial population size. This represents the number of nodes in the distribution network.

[0023] Step S3: Construct a multi-objective function and dynamic constraint penalty weights, and construct an individual fitness function based on the multi-objective function and the dynamic constraint penalty weights, specifically including: Construct a multi-objective optimization function based on "safety, economy, and efficiency" to achieve a synergistic balance among multiple objectives.

[0024] With minimizing the main transformer backfeed overload rate as the core objective (safety first), the main transformer backfeed overload rate... The formula is: in, For the first The reverse transmission power of the main transformer; For the first Rated capacity of the main transformer.

[0025] To minimize distribution network losses Maximize photovoltaic absorption rate Minimize node voltage deviation As a secondary objective, a multi-objective optimization function is established.

[0026] Breaking through the limitations of fixed weights, a risk sensitivity function based on real-time power flow calculation is proposed. in, The degree of impact of time-unit adjustment actions on heavy load risks. This represents the risk value of overload on the power grid. This is the adjustment amount for the decision variable.

[0027] The formula for constructing the dynamic constraint penalty weight is as follows: in, for Dynamically constrain penalty weights at all times. Based on the penalty weight, , To adjust the coefficients, dynamic coupling between the constraint penalty weights and the real-time risk gradient of the power grid is achieved. In emergency situations... Larger Automatically increases, enhancing constraint rigidity; under light load conditions Smaller Reduce appropriately and relax constraints appropriately.

[0028] The individual fitness function is constructed as follows: in, For individuals Overall adaptability; For the first The weights of each objective function; Indicates the first One objective function, ; for A constraint violation (such as exceeding the line transmission power limit, exceeding the voltage limit, etc.); This indicates a penalty for violating a constraint, ensuring that punishment is only imposed on the individual who violates the constraint.

[0029] Step S4: Reference Figure 4As shown, a hierarchical tournament selection mechanism is adopted, and topology-preserving crossover operators and hierarchical mutation operators are designed to update the population and maintain an external archive set; specifically including: A tiered tournament selection mechanism is adopted. The first tier is sorted by non-dominance level (the lower the level, the higher the priority). The second tier is sorted by crowding distance (the greater the distance, the better the individual diversity). The third tier introduces a topological complexity penalty (the simpler the topological structure, the smaller the penalty) to ensure that the selected individuals have both excellent target performance and topological feasibility.

[0030] For a mixed discrete-continuous decision space, a topology-preserving crossover operator is designed. Topo-Crossover Boolean adjacency matrix of parent individuals , After subgraph swapping, loops and isolated nodes are automatically detected using the DFS algorithm, and topological defects are repaired using the minimum edge deletion method to ensure that the adjacency matrix of the offspring still satisfies the radial constraint. The mutation operator adopts a hierarchical strategy to adjust the continuous power vector. Polynomial mutation is employed to ensure adjustment accuracy for Boolean adjacency matrices. We employ edge flipping mutation based on degree distribution (prioritizing the mutation of edges with moderate node degree) to reduce the risk of topology disruption.

[0031] An external archive set is constructed to store non-dominated solutions during the evolution process. A hash table is used to record the topological feature values ​​of each non-dominated solution. Redundant solutions are eliminated through graph isomorphism detection to ensure the effectiveness and conciseness of the archive set, while accelerating the convergence speed of subsequent iterations.

[0032] Step S5: Reference Figure 5 As shown, real-time decision optimization based on a rolling time-domain window and prediction compensation yields the following optimization results: A rolling time-domain window is set up, and the optimization solution is performed once within each rolling time-domain window. Only the decision instructions for the first time period are output, which takes into account both the foresight and real-time nature of the decision.

[0033] Introducing short-term source-load forecast data It is used to predict the source load power for one future rolling window and to affect the main transformer backfeed overload rate. Perform feedforward correction, the correction formula is: in, The corrected core objective function value; To predict compensation weights, for Predicted source-load power at time point. for Actual value of source load power at any given time. Using Euclidean distance improves the adaptability of decision-making to source load fluctuations.

[0034] After the optimal decision command is issued, the response deviation of the distribution network operation status is monitored and fed back to the next optimization cycle. Specifically, after the decision command is issued, the response deviation between the actual operating parameters of the power grid and the decision target is monitored in real time. In the next optimization cycle, a deviation compensation term is introduced to pass the deviation value back to the objective function, thereby realizing the closed-loop optimization of "decision-execution-feedback-correction".

[0035] Step S6: Determine whether the convergence criterion is met based on the optimization results. If the convergence criterion is met, output the optimal decision instruction. Specifically, the hypervolume index of the non-dominated solution set is used as the core convergence evaluation index. When the hypervolume index of the non-dominated solution set increases by less than 0.1% for three consecutive generations and the population constraint violation rate is less than 1%, the algorithm is determined to be converged, the iteration is stopped, and the optimal decision scheme in the current non-dominated solution set is output.

[0036] The initial population is clustered according to topological similarity. Based on the topological structure of individuals within the same cluster, a distributed computing architecture is used to compute the fitness of individuals in each cluster in parallel, thereby reducing redundant computations, reducing communication overhead by more than 60%, and significantly improving computational efficiency.

[0037] This embodiment also allows for lightweight online deployment: A GNN (Graph Neural Network) surrogate model is trained using a large number of offline samples. This surrogate model can quickly approximate the individual fitness calculation, replacing the initial population individual fitness calculation in the preset part of the online iteration process. The preset part is a more complex fitness calculation selected from the existing fitness calculations. This further shortens the online solution time and meets the needs of real-time decision-making and deployment.

[0038] On the other hand, reference Figure 2As shown, this embodiment also provides a real-time power balance decision-making system for distribution networks based on topology awareness. It is implemented using the aforementioned real-time power balance decision-making method for distribution networks based on topology awareness. The system includes: a data input layer, which collects distribution network topology data, real-time power flow data, and source-load prediction data, and preprocesses the collected data to obtain preprocessed data; a topology awareness module, connected to the data input layer, which performs chromosome encoding and initial population construction on the preprocessed data based on a topology awareness strategy; a multi-objective optimization module, connected to the topology awareness module, which constructs a multi-objective function and dynamic constraint penalty weights, and constructs an individual fitness function based on the multi-objective function and the dynamic constraint penalty weights; an evolutionary calculation module, which adopts a hierarchical tournament selection mechanism, designs a topology-preserving crossover operator and a hierarchical mutation operator, updates the population, and maintains an external archive set; a real-time decision-making module, connected to the evolutionary calculation module, which performs real-time decision optimization based on a rolling time-domain window and prediction compensation to obtain optimization results; and a decision output layer, connected to the real-time decision-making module, which determines whether the convergence criterion is met based on the optimization results, and outputs the optimal decision instruction if the convergence criterion is met.

[0039] Specifically, the topology-aware distribution network power balance real-time decision-making system described in this embodiment also includes a feedback optimization layer, which is used to monitor the response deviation between the actual operating parameters of the power grid and the decision target in real time after the decision command is issued. In the next optimization cycle, a deviation compensation term is introduced to pass the deviation value back to the objective function, thereby realizing the closed-loop optimization of "decision-execution-feedback-correction".

[0040] The following specific application examples further illustrate the above-mentioned topology-aware real-time power balance decision-making method and system for distribution networks: The effectiveness of this method is verified using a 110kV distribution network area as an application scenario. The specific parameters and implementation steps are as follows: Scenario parameter settings: Number of distribution network nodes It includes three main transformers (each with a rated capacity of 50MVA), five interconnection lines, ten photovoltaic distributed power sources (total installed capacity of 20MW), and a controllable load (air conditioning, electric vehicles) capacity of 8MW; rolling time domain window Target Net Present Value 10,000 yuan, target internal rate of return The comparison method is the traditional NSGA-Ⅲ power balance decision method.

[0041] Basic data acquisition: One-year distribution network topology data, real-time power flow data (line power, node voltage), photovoltaic output data, and load data were collected, and outliers were removed through filtering and interpolation. An LSTM (Long Short-Term Memory) model was used for short-term source-load forecasting to obtain source-load forecast data for the next 15 minutes. .

[0042] Initial population construction and fitness calculation: Based on the DFS algorithm, the original topology of the distribution network is traversed to generate a set of topology trees that satisfy the radial constraints. , ( By removing 12 duplicate topological individuals through graph isomorphism detection, 48 valid initial individuals were obtained; a basic penalty weight was set. adjustment coefficient , Objective function weights Calculate the fitness of individuals in the initial population.

[0043] Evolutionary Iteration and Real-Time Decision Making: Executing Improved Non-Dominated Sorts and Genetic Operations: Selecting 24 Superior Individuals through a Hierarchical Tournament Topo-Crossover Crossover generates 24 offspring individuals, which, after topological repair, have no loops or isolated nodes; for continuous variables... P Polynomial mutation (mutation probability 0.1) is used for discrete variables. G We employ edge-flipping mutation based on degree distribution (mutation probability 0.05); update the population and maintain an external archive set, with the size of the archive set controlled to within 20 individuals.

[0044] Rolling optimization and bias correction: An iteration is performed every 15 minutes, incorporating source load prediction data to correct the core objective function. When iterating to the 7th generation, the hypervolume index HV of the non-dominated solution set has increased by 0.08% for three consecutive generations, the constraint violation rate is 0.8%, the convergence criterion is met, the iteration stops and the optimal decision scheme is output: the main transformer backfeed heavy load rate is controlled within 5%, the photovoltaic absorption rate is 92%, the grid loss rate is 3.1%, and the node voltage deviation is ≤±3%.

[0045] Comparative Experiments and Result Analysis: The method of this embodiment is compared with the traditional method. The experiment covers three typical operating scenarios. The evaluation indicators include the proportion of initial invalid solutions, the number of convergence iterations, the decision-making time, the main transformer back-feeding overload rate, and the photovoltaic absorption rate. The comparison results between the method of this embodiment and the traditional method are shown in Table 1 below: Table 1 Therefore, this embodiment significantly reduces the proportion of initial invalid solutions through a topology-aware initialization strategy. Combined with parallelization and surrogate model acceleration, the decision-making time is shortened to less than 8 minutes, meeting the requirement of 15-minute rolling real-time decision-making, and significantly improving real-time performance compared to traditional methods. The dynamic constraint penalty mechanism and improved evolutionary operator ensure that the decision-making scheme can balance safety and optimization objectives under different operating scenarios. The main transformer backfeeding overload rate is significantly reduced, and the photovoltaic absorption rate and network loss optimization effects are superior to traditional methods, improving the safety and economy of distribution network operation. Comparative experiments verify the stability and effectiveness of this method under different load and photovoltaic output scenarios, demonstrating its wide applicability and strong engineering application value.

[0046] In summary, this embodiment proposes a real-time power balance decision-making method and system for distribution networks based on topology awareness, improving the feasibility of the initial population: the topology-aware population initialization strategy generates a radial topology tree through the DFS algorithm and performs graph isomorphism detection, reducing the proportion of initial invalid solutions by more than 90% and improving the algorithm's convergence speed by more than 50%, laying a time foundation for real-time decision-making. It enhances constraint adaptability: the dynamic constraint penalty mechanism achieves real-time linkage between constraint weights and grid risk levels, ensuring grid safety in emergency situations and improving the optimization space under light load conditions, significantly improving the scenario adaptability of the decision scheme. It reduces the risk of topology destruction: the improved evolutionary operator for mixed variables ensures the integrity of the grid topology during genetic operations, reducing the topology destruction rate to below 1%, avoiding the loss of feasible solutions, and improving optimization performance. Ensuring real-time performance and accuracy: Through parallel computing and lightweight design of GNN proxy model, online solution time is shortened to minutes, meeting the 15-minute rolling decision-making requirements; multi-objective optimization and predictive feedback correction ensure that the decision scheme takes into account safety, economy and efficiency, reducing the main transformer back-feeding heavy load rate by more than 30%, increasing the photovoltaic absorption rate by more than 15%, and reducing grid loss by more than 10%.

[0047] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0048] It should be noted that the apparatus and methods disclosed in the embodiments herein can also be implemented in other ways. The apparatus embodiments described above are merely illustrative; for example, the block diagrams in the accompanying drawings show apparatuses and methods according to various embodiments herein. It should be noted that combinations of blocks in the block diagrams can be used to perform specific functions or actions.

[0049] Although the present invention has been described in detail through the preferred embodiments above, it should be understood that the above description should not be considered as a limitation of the present invention. Various modifications and substitutions to the present invention will be apparent to those skilled in the art after reading the above description. Therefore, the scope of protection of the present invention should be defined by the appended claims.

Claims

1. A real-time power balance decision-making method for distribution networks based on topology sensing, characterized in that, include: Collect distribution network topology data, real-time power flow data, and source load prediction data, and preprocess the collected data to obtain preprocessed data; Based on a topology-aware strategy, chromosome encoding and initial population construction are performed on the preprocessed data; a multi-objective function and dynamic constraint penalty weights are constructed, and an individual fitness function is constructed based on the multi-objective function and the dynamic constraint penalty weights; a hierarchical tournament selection mechanism is adopted, and topology-preserving crossover operators and hierarchical mutation operators are designed to update the population and maintain an external archive set; real-time decision optimization is performed based on a rolling temporal window and prediction compensation to obtain optimization results; the optimization results are used to determine whether the convergence criterion is met, and if the convergence criterion is met, the optimal decision instruction in the current non-dominated solution set is output.

2. The real-time power balance decision-making method for distribution networks based on topology sensing as described in claim 1, characterized in that, Also includes: After the optimal decision command is issued, the response deviation of the distribution network operation status is monitored and fed back to the next optimization cycle.

3. The real-time power balance decision-making method for distribution networks based on topology sensing as described in claim 1, characterized in that, The aforementioned topology-aware strategy for chromosome encoding and initial population construction of the preprocessed data specifically includes: Establish a two-dimensional chromosome structure containing topological and power information. : in, This is a Boolean adjacency matrix used to encode the switch states of the distribution network. In this context, a value of "1" indicates a closed loop, and "0" indicates an open loop. It is a continuous power regulation vector used to encode adjustable parameters, which include: main transformer output, tie line transmission power and distributed generation output; A set of topology trees satisfying radial constraints is generated by traversing the distribution network topology data in the preprocessed data using a depth-first search algorithm. Individuals are selected from the set of topological trees to construct an initial population, and the adjacency matrix for each individual is... All are connected acyclic graphs, and duplicate topological individuals are removed by graph isomorphism detection. Initial population size Number of distribution network nodes satisfy ,in: This represents the initial population size. This represents the number of nodes in the distribution network.

4. The real-time power balance decision-making method for distribution networks based on topology sensing as described in claim 3, characterized in that, The construction of the multi-objective function and dynamic constraint penalty weights, and the construction of the individual fitness function based on the multi-objective function and the dynamic constraint penalty weights, specifically includes: Construct a multi-objective optimization function based on "safety, economy, and efficiency" to achieve a coordinated balance among multiple objectives: The core objective is to minimize the main transformer backfeed overload rate. The formula is: in, For the first The reverse transmission power of the main transformer; For the first Rated capacity of the main transformer; To minimize distribution network losses Maximize photovoltaic absorption rate Minimize node voltage deviation As a secondary objective, a multi-objective optimization function is established; Risk sensitivity function based on real-time power flow calculation: in, The degree of impact of time-unit adjustment actions on heavy load risks. This represents the risk value of overload on the power grid. The moderating factor for the decision variable; The formula for constructing the dynamic constraint penalty weight is as follows: in for Dynamically constrain penalty weights at all times. Based on the penalty weight, , To adjust the coefficients, dynamic coupling between the constraint penalty weights and the real-time risk gradient of the power grid is achieved; The individual fitness function is constructed as follows: in, For individuals Overall adaptability; For the first The weights of each objective function; Indicates the first One objective function, ; for One constraint violation quantity; This indicates a penalty for violating a constraint, ensuring that punishment is only imposed on the individual who violates the constraint.

5. The real-time power balance decision-making method for distribution networks based on topology sensing as described in claim 1, characterized in that, The hierarchical tournament selection mechanism includes: the first layer sorting by non-dominance level, the second layer sorting by crowding distance, and the third layer introducing a topological complexity penalty term.

6. The real-time power balance decision-making method for distribution networks based on topology sensing as described in claim 4, characterized in that, The design topology maintains the crossover and hierarchical mutation operators, updates the population, and maintains the external archive set, specifically including: For a mixed discrete-continuous decision space, a topology-preserving crossover operator is designed, which considers the Boolean adjacency matrix of the parent individuals. , After subgraph swapping, loops and isolated nodes are automatically detected using a depth-first search algorithm, and topological defects are repaired using the minimum edge deletion method to ensure that the adjacency matrix of the offspring still satisfies the radial constraint. The mutation operator adopts a hierarchical strategy to adjust the continuous power vector. Polynomial mutation is employed to ensure adjustment accuracy for Boolean adjacency matrices. An edge-flipping mutation based on degree distribution is employed; An external archive set is constructed to store non-dominated solutions during the evolution process. A hash table is used to record the topological feature values ​​of each non-dominated solution, and redundant solutions are removed through graph isomorphism detection.

7. The real-time power balance decision-making method for distribution networks based on topology sensing as described in claim 6, characterized in that, The aforementioned real-time decision optimization based on rolling time-domain windows and prediction compensation specifically includes: Set a rolling time window, perform an optimization solution once within each rolling time window, and output only the decision instructions for the first time period; Introducing short-term source-load forecast data For the main transformer reverse transmission heavy load rate Perform feedforward correction, the correction formula is: in, The corrected core objective function value; To predict compensation weights, for Predicted source-load power at time point. for Actual value of source load power at any given time. The distance is Euclidean.

8. The real-time power balance decision-making method for distribution networks based on topology sensing as described in claim 1, characterized in that, The process of determining whether the convergence criterion is met based on the optimization result, and outputting the optimal decision instruction if the convergence criterion is met, specifically includes: Using the hypervolume index of the non-dominated solution set as the core convergence evaluation index, when the hypervolume index of the non-dominated solution set increases by less than 0.1% for three consecutive generations and the population constraint violation rate is less than 1%, the algorithm is determined to have converged, the iteration is stopped, and the optimal decision scheme in the current non-dominated solution set is output.

9. The real-time power balance decision-making method for distribution networks based on topology sensing as described in claim 8, characterized in that, Also includes: The initial population is clustered according to topological similarity. Based on the topological structure of individuals within the same cluster, a distributed computing architecture is used to compute the fitness of individuals in each cluster in parallel to reduce redundant computation. Lightweight online deployment: Train a graph neural network agent model using offline samples, and use the trained graph neural network agent model to replace the initial population individual fitness calculation in the preset part of the online iteration process.

10. A real-time power balance decision-making system for distribution networks based on topology sensing, characterized in that, It is implemented using the real-time power balance decision-making method for distribution networks based on topology awareness as described in any one of claims 1-9, and the system includes: The data input layer collects distribution network topology data, real-time power flow data, and source load prediction data, and preprocesses the collected data to obtain preprocessed data. A topology-aware module, which is connected to the data input layer, performs chromosome encoding and initial population construction on the preprocessed data based on a topology-aware strategy. A multi-objective optimization module, connected to the topology sensing module, constructs a multi-objective function and dynamic constraint penalty weights, and constructs an individual fitness function based on the multi-objective function and the dynamic constraint penalty weights; The evolutionary computation module employs a hierarchical tournament selection mechanism, designs topology-preserving crossover operators and hierarchical mutation operators, updates the population, and maintains an external archive set. A real-time decision-making module, which is connected to the evolutionary computation module, is used to perform real-time decision optimization based on a rolling time-domain window and prediction compensation to obtain optimization results. The decision output layer is connected to the real-time decision module. It determines whether the convergence criterion is met based on the optimization result. If the convergence criterion is met, it outputs the optimal decision instruction.

Images