Partition-based mesh power transmission and distribution network optimization system and method

Abstract

The invention discloses a partition-based mesh power transmission and distribution network optimization system and method. The system comprises seven units of topological structure sensing and partition division, power flow data acquisition and feature extraction and the like. The topological structure sensing and partition dividing unit is used for dynamically dividing sub-regions according to a power grid node connection relation and load characteristics; the power flow data acquisition and feature extraction unit deeply extracts data features such as voltage and current, the improved depth Q network strategy generation unit combines power grid parameter construction state and action space to generate a preliminary strategy, and the optimized attention mechanism weight distribution unit distributes weights according to node electrical distance and line transmission capacity to perform strategy fusion. The multi-region collaborative decision fusion unit synthesizes sub-region power interaction and stability indexes to perform global decision, the control instruction generation and issuing unit issues instructions in combination with equipment constraint, and the operation state feedback and updating unit realizes state real-time updating, so that the power grid operation optimization efficiency and safety are improved.

Description

Technical Field

[0001] The present invention relates to the field of meshed transmission and distribution network optimization, and in particular to a partition-based meshed transmission and distribution network optimization system and method. Background Art

[0002] With the rapid development of smart grids, meshed transmission and distribution networks, with their complex topologies and diverse operating scenarios, are placing higher demands on grid optimization and management. Traditional transmission and distribution network optimization methods are gradually exposing their limitations when faced with high penetration rates of distributed generation (DGs) and increased load fluctuations.

[0003] The first major drawback of existing technologies is their lack of adaptability to complex grid topologies. Traditional optimization systems, often based on simple network models and fixed partitioning schemes, struggle to accurately perceive the dynamically changing topology of the transmission and distribution network. Consequently, they are unable to quickly and appropriately optimize partitions in response to grid restructuring and the commissioning of new branches, causing optimization strategies to lag behind the actual grid operation.

[0004] The second shortcoming lies in the efficiency and accuracy of data processing and decision-making. Traditional methods have limited ability to extract data features when processing massive amounts of power flow data, failing to fully tap into the valuable information contained within. Furthermore, the decision-making process lacks coordinated consideration of all parts of the power grid. When formulating optimization strategies, it's impossible to effectively balance power interactions and stability across regions, making it difficult to achieve ideal optimization results and meet the requirements for safe and efficient operation of modern meshed transmission and distribution networks. Summary of the Invention

[0005] In order to overcome the shortcomings and deficiencies of the prior art, the present invention provides a partition-based meshed transmission and distribution network optimization system and method.

[0006] The technical solution adopted by the present invention is a partition-based meshed transmission and distribution network optimization system, including: a topology structure perception and partition division unit, a power flow data acquisition and feature extraction unit, an improved deep Q network strategy generation unit, an optimized attention mechanism weight allocation unit, a multi-region collaborative decision fusion unit, a control instruction generation and issuance unit, and an operation status feedback and update unit;

[0007] The topology structure perception and partitioning unit is connected to the power flow data acquisition and feature extraction unit through the network communication module, and is used to transmit the perceived meshed transmission and distribution network topology structure information to the power flow data acquisition and feature extraction unit;

[0008] The power flow data collection and feature extraction unit is connected to the improved deep Q network strategy generation unit through a data processing bus, and sends the collected and feature-extracted power flow data to the improved deep Q network strategy generation unit;

[0009] The improved deep Q network strategy generation unit is connected to the optimized attention mechanism weight allocation unit through the data interaction interface, and transmits the generated preliminary strategy information to the optimized attention mechanism weight allocation unit;

[0010] The optimized attention mechanism weight allocation unit is connected to the multi-region collaborative decision fusion unit through a decision fusion channel, and transmits the weighted strategy information to the multi-region collaborative decision fusion unit;

[0011] The multi-region collaborative decision fusion unit is connected to the control instruction generation and issuance unit via an instruction transmission line, and transmits the fused decision information to the control instruction generation and issuance unit;

[0012] The control instruction generation and issuance unit is connected to the meshed transmission and distribution network equipment through a control signal channel for issuing control instructions, and is connected to the operation status feedback and update unit through a state feedback line for feeding back the operation status of the equipment to the operation status feedback and update unit;

[0013] The operation status feedback and update unit is connected to the topology structure perception and partitioning unit, the flow data collection and feature extraction unit, the improved deep Q network strategy generation unit, the optimized attention mechanism weight allocation unit, and the multi-region collaborative decision fusion unit through the parameter update link to feedback the operation status and update the parameters of each unit;

[0014] The improved deep Q network strategy generation unit includes a state space S constructed based on the voltage amplitude and phase angle parameters of the meshed transmission and distribution network nodes, and its expression is S={(V1, θ1), (V2, θ2), ..., V n ,θ n )}, where V i represents the voltage amplitude of the i-th node, θ i represents the voltage phase angle of the ith node, n is the total number of nodes in the meshed transmission and distribution network; the action space A is constructed by the branch power regulation range, that is, A={a1,a2,…,a m}, a j represents the power regulation action for the j-th branch, m is the total number of branches; the reward function R is constructed based on the node voltage deviation and branch power loss, α and β are weight coefficients, is the rated voltage amplitude of node i, is the power loss of the j-th branch; through the improved deep Q network formula Q π (s, a; θ) = r + γmax a' Q π' (s', a'; θ') for strategy generation, where Q π(s, a; θ) is the value function of taking action a in state s, θ is the network parameter, r is the immediate reward, γ is the discount factor, s' is the next state, and θ' is the target network parameter.

[0015] Furthermore, in the optimized attention mechanism weight distribution unit, the electrical distance D between nodes is used ij and line transmission capacity C ij Construct an attention weight calculation model, attention weight ω ij The calculation formula is Among them, f(D ij , C ij )=σ(W1D ij +W2C ij +b), σ is the activation function, W1 and W2 are weight matrices, and b is the bias vector; the attention weight is applied to improve the strategy information generated by the deep Q network, through the weighted fusion formula Get the fused strategy information, S ij is the policy information associated with node i and node j.

[0016] Furthermore, the topology structure perception and partitioning unit divides the grid into multiple sub-areas through a partitioning algorithm according to the node connection relationship matrix G of the meshed transmission and distribution network and the node load characteristic parameters. The partitioning constraint condition is Among them, Ω k represents the kth sub-region, L i is the load of node i, L max is the maximum allowable load of the sub-area; and the boundary node identification formula is Determine the boundary nodes of each sub-region, E is the edge set, and v is the node.

[0017] Furthermore, the power flow data acquisition and feature extraction unit extracts features from the collected voltage, current, and power data, obtains frequency domain features through fast Fourier transform, and constructs a feature vector X=[X1, X2, ..., X p ], where X p is the pth feature component; and constructs a data sample set D = {(x1, y1), (x2, y2), ..., (x N ,y N )},x i is the input feature vector, y i is the corresponding output label, and N is the number of samples, which is used to improve the training of the deep Q network.

[0018] Furthermore, the multi-region collaborative decision fusion unit determines the decision strategy of each sub-region by the power interaction amount P between sub-regions. inter and the subregion stability index Sk Construct collaborative decision-making fusion model and the decision-making strategy after fusion Among them, D k is the decision strategy for the kth sub-region, K is the total number of sub-regions, τ is the activation function, W3, W4, W5 are weight matrices, and b' is the bias vector.

[0019] Furthermore, the control instruction generation and issuing unit constructs a control instruction generation model based on the decision information output by the multi-region collaborative decision fusion unit through the device control parameter constraints, and generates the circuit breaker control instruction to meet S on / off ∈{0, 1}, 0 means open, 1 means closed; for transformer tap adjustment instructions, satisfy T tap ∈[T min , T max ], T min and T max is the tap adjustment range; the control instructions are sorted and issued through the priority sorting algorithm, and the sorting formula is Priority i is the priority of the i-th control instruction, is the influence of the instruction on the current, is the voltage risk after the instruction is executed, and η1 and η2 are weight coefficients.

[0020] Furthermore, the operation status feedback and update unit constructs a state update model based on the equipment operation status feedback data, and updates the node voltage using the formula is the updated voltage amplitude, is the voltage amplitude before updating, is the measured voltage amplitude, δ is the update coefficient; for branch power loss update, the formula is used is the change in power loss; and the updated status information is fed back to each unit for parameter update.

[0021] Furthermore, an adaptive learning rate adjustment module is set between the improved deep Q network strategy generation unit and the optimized attention mechanism weight allocation unit to generate the error E policy and weight distribution error E weight Constructing a learning rate adjustment formula η is the adjusted learning rate, η0 is the initial learning rate, and λ is the adjustment coefficient, which is used to dynamically adjust the learning process of the improved deep Q network and optimized attention mechanism.

[0022] The partition-based meshed transmission and distribution network optimization method includes the following steps:

[0023] Step S1: using a topology sensing and partitioning unit to sense the topology information of the meshed transmission and distribution network, and performing sub-area division based on a node connection relationship matrix and load characteristic parameters, and determining the boundary nodes of each sub-area;

[0024] Step S2: The power flow data acquisition and feature extraction unit collects voltage, current, and power data of each node in the power grid, extracts frequency domain features through fast Fourier transform, and constructs a data sample set;

[0025] Step S3: The improved deep Q network strategy generation unit generates a preliminary optimization strategy using the improved deep Q network algorithm through the constructed state space, action space and reward function;

[0026] Step S4: Optimize the attention mechanism weight distribution unit to calculate the attention weight according to the electrical distance between nodes and the line transmission capacity, and perform weighted fusion on the preliminary strategy;

[0027] Step S5: The multi-region collaborative decision fusion unit performs collaborative decision fusion on the decision strategies of each sub-region based on the power interaction between sub-regions and the sub-region stability index;

[0028] Step S6: The control instruction generation and issuance unit generates control instructions based on the integrated decision information and the device control parameter constraints, and sorts the instructions using a priority sorting algorithm before issuing them to the meshed transmission and distribution network equipment;

[0029] Step S7: The operation status feedback and update unit receives the equipment operation status feedback data, updates the node voltage and branch power loss status information through the status update model, and feeds back the updated information to each unit to complete an optimization cycle.

[0030] Beneficial Effects: The present invention proposes a partition-based meshed transmission and distribution network optimization system and method. The topology perception and partitioning unit in the present invention can perceive changes in the meshed transmission and distribution network topology in real time and dynamically partition subregions based on node connectivity and load characteristics. Compared to traditional fixed partitioning methods, this method can rapidly respond to grid structure adjustments and ensure that the optimization strategy aligns with actual operating conditions. To address the shortcomings of low data processing and decision-making efficiency and accuracy, the power flow data acquisition and feature extraction unit deeply mines the frequency domain characteristics of data such as voltage and current, providing rich information for subsequent decision-making. The improved deep Q network strategy generation unit and the optimized attention mechanism weight allocation unit work together. The former constructs a state and action space generation strategy based on parameters such as node voltage and power, while the latter allocates weights based on node electrical distance and line transmission capacity, enhancing attention to key information and achieving precise strategy optimization. The multi-region collaborative decision fusion unit considers power interaction and stability between subregions to generate a global optimization strategy. The control instruction generation and issuance unit prioritizes and issues instructions based on device parameter constraints. The operating status feedback and update unit updates status information in real time, forming a closed-loop optimization loop. This significantly improves data processing efficiency and decision-making accuracy, ensuring the safe and efficient operation of the meshed transmission and distribution network. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is a diagram of the system unit composition of the present invention;

[0032] Figure 2 Flowchart for the method of the present invention. DETAILED DESCRIPTION

[0033] It should be noted that, unless there is a conflict, the embodiments in this application and the features described in the embodiments can be combined with each other. The application is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0034] like Figure 1 As shown in the figure, the partition-based meshed transmission and distribution network optimization system includes: a topology structure perception and partition division unit, a flow data acquisition and feature extraction unit, an improved deep Q network strategy generation unit, an optimized attention mechanism weight allocation unit, a multi-region collaborative decision fusion unit, a control instruction generation and issuance unit, and an operation status feedback and update unit;

[0035] The topology structure perception and partitioning unit is connected to the power flow data acquisition and feature extraction unit through the network communication module, and is used to transmit the perceived meshed transmission and distribution network topology structure information to the power flow data acquisition and feature extraction unit;

[0036] Specifically, in a partition-based meshed transmission and distribution network optimization system, the topology perception and partitioning unit is the starting point of the entire optimization process. Its core role is to accurately grasp the grid structure and rationally divide the regions, providing a basic framework for subsequent optimization decisions. This unit monitors the connection relationships between nodes in real time to obtain a node connection matrix. This matrix details the line connection status between nodes and is key data for analyzing grid topology. At the same time, it combines node load characteristic parameters, including active and reactive load information, to divide the grid into sub-regions.

[0037] Reasonable zoning can divide the large and complex mesh transmission and distribution network into multiple relatively independent and easy-to-manage sub-areas, making the subsequent optimization strategy formulation more targeted and efficient. For example, when facing load fluctuations or distributed power access in different areas, each sub-area can be optimized separately to avoid the complexity and inefficiency caused by overall optimization. In terms of implementation, the unit adopts a specific partitioning algorithm to ensure that the total load in each sub-area is within a safe and controllable range based on the pre-set maximum allowable load constraints of the sub-area. At the same time, the connection nodes between the sub-areas are determined through the boundary node identification method, laying the foundation for collaborative optimization and power interaction between sub-areas.

[0038] The power flow data collection and feature extraction unit is connected to the improved deep Q network strategy generation unit through a data processing bus, and sends the collected and feature-extracted power flow data to the improved deep Q network strategy generation unit;

[0039] Specifically, the power flow data acquisition and feature extraction unit is responsible for collecting key data from the power grid's operation and performing in-depth processing to extract effective features that reflect the grid's operating status. The data collected by this unit covers voltage, current, power, and other information at each grid node. This raw data contains rich details about power grid operation but requires further processing before it can be used for optimized decision-making. During feature extraction, techniques such as fast Fourier transforms are used to convert the raw time-domain data into the frequency domain to obtain the frequency-domain features of the data.

[0040] Feature extraction can filter valuable information for grid optimization decisions from massive amounts of raw data, remove redundant data, and improve data processing efficiency and decision accuracy. For example, frequency domain features can reflect the distribution of harmonics in the grid, helping to identify power quality issues and providing a basis for optimization strategies. In terms of implementation, this unit constructs a data sample set, associates the extracted feature vectors with the corresponding output labels, and forms data samples for improving deep Q network training. By continuously accumulating and updating the data sample set, the model's adaptability and predictive capabilities for grid operating conditions are improved.

[0041] The improved deep Q network strategy generation unit is connected to the optimized attention mechanism weight allocation unit through the data interaction interface, and transmits the generated preliminary strategy information to the optimized attention mechanism weight allocation unit;

[0042] Specifically, the Improved Deep Q Network Strategy Generation Unit is one of the system's core decision-making units. It utilizes an improved Deep Q Network algorithm, combined with the actual operating parameters of the power grid, to generate an optimization strategy. This unit first constructs a state space based on the voltage amplitude and phase angle parameters of the meshed transmission and distribution network nodes. This state space comprehensively describes the voltage state at each node in the grid and serves as the basis for the algorithm's decision-making. Furthermore, an action space is constructed based on the branch power regulation range, clarifying the power regulation actions that the algorithm can take.

[0043] This unit transforms grid operation problems into algorithmically manageable decision-making problems by constructing a rational state space and action space. Leveraging the learning capabilities of deep Q networks, it searches for optimal power regulation strategies to achieve optimization goals such as reducing grid power loss and maintaining node voltage stability. During implementation, the unit sets a reward function and evaluates different strategies based on node voltage deviations and branch power losses, guiding the algorithm to learn the optimal strategy. By continuously interacting with the grid environment and iteratively updating strategies using an improved deep Q network formula, it gradually generates an optimized strategy that adapts to the dynamic changes of the grid.

[0044] The optimized attention mechanism weight allocation unit is connected to the multi-region collaborative decision fusion unit through a decision fusion channel, and transmits the weighted strategy information to the multi-region collaborative decision fusion unit;

[0045] Specifically, the optimized attention weight allocation unit further refines the initial strategy generated by the improved deep Q-network, improving its accuracy and effectiveness. This unit calculates attention weights based on parameters such as the electrical distance between nodes and the transmission capacity of the line. Electrical distance reflects the closeness of the electrical connection between nodes, while line capacity reflects the power transmission capacity of the line. These two parameters are key to determining attention weights.

[0046] By allocating attention weights, the system can pay more attention to nodes and lines that have a greater impact on the operation of the power grid, avoiding inefficient decision-making and waste of resources caused by averaging all information. For example, higher attention weights are given to lines with shorter electrical distances and limited transmission capacity, and they are optimized first, which helps improve the overall stability and transmission efficiency of the power grid. In terms of implementation, the unit uses a specific calculation model, takes the electrical distance between nodes and the transmission capacity of the line as input, and obtains the attention weights through operations such as activation functions. Then, using the weighted fusion method, the attention weights are applied to the preliminary strategy, which is recombined and optimized to generate strategy information that better meets the actual needs of the power grid.

[0047] The multi-region collaborative decision fusion unit is connected to the control instruction generation and issuance unit through an instruction transmission line, and transmits the fused decision information to the control instruction generation and issuance unit.

[0048] Specifically, the multi-region collaborative decision-making fusion unit is responsible for integrating the decision-making strategies of each sub-region to achieve global optimization. This unit comprehensively considers factors such as inter-region power interaction and sub-region stability indicators for each sub-region's decision-making strategy. Inter-region power interaction reflects the power flow between different sub-regions, while sub-region stability indicators are used to assess the operational stability of each sub-region. These factors are important foundations for collaborative decision-making.

[0049] This unit solves the problem of lack of coordination in regional decision-making in traditional methods. By comprehensively considering the actual conditions and mutual relationships of each sub-region, the generated global optimization strategy can better balance the power distribution between regions and improve the overall operational stability and economy of the power grid. For example, when there is excess power in some sub-regions and insufficient power in other sub-regions, the unit can coordinate the decisions of each region to achieve reasonable allocation of power. In terms of implementation, the unit adopts a collaborative decision fusion model, calculates the fusion weight based on the power interaction amount and stability index between sub-regions, and performs weighted fusion on the decision-making strategies of each sub-region to obtain the final global optimization strategy.

[0050] The control instruction generation and issuance unit is connected to the meshed transmission and distribution network equipment through a control signal channel for issuing control instructions, and is connected to the operation status feedback and update unit through a state feedback line for feeding back the operation status of the equipment to the operation status feedback and update unit;

[0051] Specifically, the control instruction generation and issuance unit's primary function is to convert optimized decisions into executable control instructions and issue them to power grid equipment. This unit generates instructions based on the decision information output by the multi-region collaborative decision fusion unit and the control parameter constraints of the power grid equipment. For switchgear such as circuit breakers, control instructions specify their open or closed state; for regulating equipment such as transformer tap changers, instructions define their adjustment range and specific gear position.

[0052] Ensure that issued control instructions meet the actual operating capabilities and safety requirements of the equipment, avoiding equipment damage or grid failures caused by unreasonable instructions. For example, when adjusting transformer taps, the adjustment range is strictly controlled within the equipment's permitted range to ensure safe operation. In terms of implementation, this unit uses a priority sorting algorithm to sort the generated control instructions. The execution priority of the instructions is determined based on factors such as the degree of impact of the instructions on current and the potential voltage risks after execution. This ensures that critical instructions are executed first, improving the real-time and effectiveness of grid optimization.

[0053] The operation status feedback and update unit is connected to the topology structure perception and partitioning unit, the flow data collection and feature extraction unit, the improved deep Q network strategy generation unit, the optimized attention mechanism weight allocation unit, and the multi-region collaborative decision fusion unit through the parameter update link to feedback the operation status and update the parameters of each unit;

[0054] Specifically, the operating status feedback and update unit is a key component in achieving closed-loop optimization of the system. Its role is to collect real-time operational status feedback data from grid equipment and update relevant system parameters. This unit receives operational data such as voltage, current, and power from grid equipment and compares it with previous operational status.

[0055] From a practical point of view, through real-time feedback and updates, the system can promptly understand the changes in the operating status of the power grid, discover the effects and existing problems after the implementation of the optimization strategy, and provide accurate data support for the next round of optimization decisions. For example, when an abnormal fluctuation in the voltage of a node is found, the unit will promptly feed back this information to other related units in order to adjust the optimization strategy. In terms of implementation, the unit adopts a state update model to update key parameters such as node voltage and branch power loss. Based on the measurement data and the preset update rules, the updated parameter values ​​are calculated, and the updated information is fed back to other units such as the topology perception and partitioning unit, the flow data acquisition and feature extraction unit, etc., to achieve dynamic optimization of the entire system.

[0056] The adaptive learning rate adjustment module, located between the improved deep Q-network policy generation unit and the optimized attention mechanism weight allocation unit, aims to dynamically optimize the learning process of both units. This module adjusts the learning rate in real time based on the policy generation error and weight allocation error. The policy generation error reflects the deviation between the policy generated by the improved deep Q-network and the actual optimal policy, while the weight allocation error reflects the accuracy of the weighted fusion of the policies by the optimized attention mechanism. This module avoids the problems associated with a fixed learning rate. If the learning rate is too high, the algorithm may not converge to the optimal solution; if the learning rate is too low, the learning process will be too slow, affecting the system's optimization efficiency. By dynamically adjusting the learning rate, the improved deep Q-network can learn the optimal policy more quickly, while the optimized attention mechanism can more accurately allocate weights, improving the learning and decision-making efficiency of the entire system. In terms of implementation, this module uses a specific learning rate adjustment formula. It takes the policy generation error and weight allocation error as inputs and calculates the adjusted learning rate through a function operation. This adjusted learning rate is then applied to the training of the improved deep Q-network and the optimized attention mechanism, achieving adaptive learning rate adjustment.

[0057] Preferably, the improved deep Q network strategy generation unit includes a state space S constructed based on the voltage amplitude and phase angle parameters of the meshed transmission and distribution network nodes, and its expression is S={(V1, θ1), (V2, θ2), ..., (V n ,θ n )}, where V i represents the voltage amplitude of the i-th node, θ i represents the voltage phase angle of the ith node, n is the total number of nodes in the meshed transmission and distribution network; the action space A is constructed by the branch power regulation range, that is, A={a1,a2,…,a m}, a j represents the power regulation action for the j-th branch, m is the total number of branches; the reward function R is constructed based on the node voltage deviation and branch power loss, α and β are weight coefficients, is the rated voltage amplitude of node i, is the power loss of the j-th branch; through the improved deep Q network formula Q π (s, a; θ) = r + γmax a' Q π' (s', a'; θ') for strategy generation, where Q π (s, a; θ) is the value function of taking action a in state s, θ is the network parameter, r is the immediate reward, γ is the discount factor, s' is the next state, and θ' is the target network parameter.

[0058] Specifically, the improved deep Q network strategy generation unit constructs a state space and action space based on technical parameters such as the voltage amplitude and phase angle of the meshed transmission and distribution network nodes and the branch power adjustment range, and converts the grid operation status and executable operations into a form that can be processed by the algorithm. By constructing a reward function based on the node voltage deviation and branch power loss, it provides optimization direction guidance for the algorithm. The complex power grid optimization problem is converted into a decision-making process that can be learned by the deep Q network, so that the algorithm can find the optimal strategy to reduce power loss and stabilize voltage through continuous trial and learning. In terms of implementation method, the unit continuously interacts with the power grid environment and uses the improved deep Q network mechanism to iteratively update the strategy to adapt to the real-time changing operation status of the power grid.

[0059] Preferably, in the optimized attention mechanism weight distribution unit, the electrical distance D between nodes is used ij and line transmission capacity C ij Construct an attention weight calculation model, attention weight ω ij The calculation formula is Among them, f(D ij , C ij )=σ(W1D ij +W2C ij +b), σ is the activation function, W1 and W2 are weight matrices, and b is the bias vector; the attention weight is applied to improve the strategy information generated by the deep Q network, through the weighted fusion formula Get the fused strategy information, S ij is the policy information associated with node i and node j.

[0060] Specifically, the optimized attention mechanism weight allocation unit calculates attention weights based on parameters such as the electrical distance between nodes and the transmission capacity of the lines, which reflect the importance of the nodes and lines in the power grid. Nodes and lines with short electrical distances and limited transmission capacity will receive higher weights and thus be given special attention in strategy optimization. The significance of this mechanism is to avoid the average processing of all information in the power grid, improve decision-making efficiency and targeted resource utilization, and give priority to solving the optimization problems of key parts that have a large impact on the operation of the power grid. During implementation, through a specific calculation model, the relevant parameters are used as input to calculate the weights, and then the weighted fusion method is used to optimize and reorganize the preliminary strategy generated by the improved deep Q network, so that the final strategy is more in line with the actual operation needs of the power grid.

[0061] Preferably, the topology structure perception and partitioning unit divides the grid into multiple sub-areas through a partitioning algorithm according to the node connection relationship matrix G of the meshed transmission and distribution network and the node load characteristic parameters. The partitioning constraint condition is Among them, Ω k represents the kth sub-region, L iis the load of node i, L max is the maximum allowable load of the sub-area; and the boundary node identification formula is Determine the boundary nodes of each sub-region, E is the edge set, and v is the node.

[0062] Specifically, the topology perception and partitioning unit uses the node connection relationship matrix to present the grid topology. Combined with the node load characteristic parameters, the meshed transmission and distribution network is divided into sub-regions based on the maximum allowable load constraints of the sub-regions and the boundary node identification rules. The complex and large power grid is divided into multiple sub-regions that are easy to manage and optimize. According to the load fluctuations and distributed power access in different regions, more targeted optimization strategies are formulated to improve the optimization efficiency and effect. During the implementation process, the partitioning algorithm is used to determine the scope of each sub-region and identify the boundary nodes while ensuring the safety and controllability of the total load in each sub-region, creating conditions for subsequent collaborative optimization and power interaction between sub-regions.

[0063] Preferably, the power flow data acquisition and feature extraction unit performs feature extraction on the collected voltage, current and power data, obtains frequency domain features through fast Fourier transform, and constructs a feature vector X=[X1, X2, ..., X p ], where X p is the pth feature component; and constructs a data sample set D = {(x1, y1), (x2, y2), ..., (x N ,y N )},x i is the input feature vector, y i is the corresponding output label, and N is the number of samples, which is used to improve the training of the deep Q network.

[0064] Specifically, the power flow data acquisition and feature extraction unit collects operational data such as voltage, current, and power at each grid node. It then uses techniques such as fast Fourier transforms to convert the raw time-domain data into frequency-domain features, mining the data for valuable information for grid optimization. A data sample set is constructed, and the extracted feature vectors are associated with output labels to provide data support for improved deep Q network training. This process filters valid information from massive amounts of raw data, removing redundancy and improving data processing efficiency and decision-making accuracy. For example, frequency-domain features can be used to identify grid power quality issues. During implementation, data is continuously collected and updated to enrich the data sample set, enhancing the model's adaptability and predictive capabilities for grid operating conditions.

[0065] Preferably, the multi-region collaborative decision fusion unit determines the decision strategy of each sub-region through the power interaction amount P between sub-regions. inter and the subregion stability index S k Construct collaborative decision-making fusion model and the decision-making strategy after fusion Among them, D k is the decision strategy for the kth sub-region, K is the total number of sub-regions, τ is the activation function, W3, W4, W5 are weight matrices, and b' is the bias vector.

[0066] Specifically, the multi-region collaborative decision-making fusion unit considers factors such as inter-regional power interaction and sub-regional stability indicators for each sub-region's decision-making strategy, performs a weighted fusion of the sub-regional strategies, and generates a global optimization strategy. These factors reflect the interrelationships between sub-regions and their own operating status. Through collaborative decision-making, it is possible to effectively balance power distribution across regions, address the lack of coordination in regional decision-making in traditional methods, and improve the overall operational stability and economic efficiency of the power grid. During implementation, the collaborative decision-making fusion model is used to calculate fusion weights based on relevant factors, and the sub-regional strategies are weighted and integrated. This allows the final global strategy to take into account the needs of each region and achieve overall optimization of the power grid.

[0067] Preferably, the control instruction generation and issuing unit, based on the decision information output by the multi-region collaborative decision fusion unit, constructs a control instruction generation model through the device control parameter constraints, and generates a control instruction for the circuit breaker to meet S on / off ∈{0, 1}, 0 means open, 1 means closed; for transformer tap adjustment instructions, satisfy T tap ∈[T min , T max ], T min and T max is the tap adjustment range; the control instructions are sorted and issued through the priority sorting algorithm, and the sorting formula is Priority i is the priority of the i-th control instruction, is the influence of the instruction on the current, is the voltage risk after the instruction is executed, and η1 and η2 are weight coefficients.

[0068] Specifically, the control instruction generation and issuance unit generates executable control instructions based on the decision information output by the multi-region collaborative decision-making fusion unit and the control parameter constraints of power grid equipment such as circuit breakers and transformer tap changers. Simultaneously, a priority sorting algorithm is used to sort and issue instructions based on factors such as the impact of the instructions on current and the potential voltage risks associated with their execution. This ensures that control instructions meet the operational capabilities and safety requirements of the equipment, preventing equipment damage and power grid failures. Prioritizing prioritizes critical instructions, improving the real-time and effectiveness of power grid optimization. During implementation, instructions are generated strictly in accordance with equipment parameter constraints, and a sorting algorithm is used to scientifically arrange the order of instruction execution to ensure smooth power grid optimization operations.

[0069] Preferably, the operation status feedback and update unit constructs a state update model based on the equipment operation status feedback data, and updates the node voltage using the formula is the updated voltage amplitude, is the voltage amplitude before updating, is the measured voltage amplitude, δ is the update coefficient; for branch power loss update, the formula is used is the change in power loss; and the updated status information is fed back to each unit for parameter update.

[0070] Specifically, the operating status feedback and update unit collects operational status feedback data such as voltage, current, and power of power grid equipment in real time. After comparing and analyzing the data with the previous status, it uses the state update model to update key parameters such as node voltage and branch power loss, and feeds back the updated information to other units in the system. This unit realizes the closed-loop optimization of the system, which is significant in that it enables the system to timely grasp the changes in the operating status of the power grid, understand the implementation effect and existing problems of the optimization strategy, and provide accurate data support for the next round of optimization decisions. During the implementation process, the equipment operation data is continuously monitored, the updated parameter values ​​are calculated according to the preset update rules, and the information is promptly fed back to relevant units such as the topology structure perception and partitioning unit to promote continuous optimization of the system.

[0071] Preferably, an adaptive learning rate adjustment module is set between the improved deep Q network strategy generation unit and the optimized attention mechanism weight allocation unit to generate the error E through the strategy. policy and weight distribution error E weight Constructing a learning rate adjustment formula η is the adjusted learning rate, η0 is the initial learning rate, and λ is the adjustment coefficient, which is used to dynamically adjust the learning process of the improved deep Q network and optimized attention mechanism.

[0072] Specifically, the adaptive learning rate adjustment module dynamically adjusts the learning rates of the improved deep Q network policy generation unit and the optimized attention mechanism weight allocation unit based on the policy generation error and weight allocation error. The policy generation error reflects the deviation between the policy and the optimal solution, and the weight allocation error reflects the accuracy of policy fusion. By adjusting the learning rate, the problem of algorithm non-convergence or slow learning caused by a fixed learning rate is avoided. Its significance lies in improving the speed at which the improved deep Q network learns the optimal policy and the accuracy of the weight allocation of the optimized attention mechanism, thereby improving the learning and decision-making efficiency of the entire system. During implementation, a specific learning rate adjustment method is used, and the two errors are used as input for calculation to obtain the adjusted learning rate and apply it to the training process of the relevant units to achieve adaptive adjustment of the learning rate.

[0073] like Figure 2 As shown, the partition-based meshed transmission and distribution network optimization method includes the following steps:

[0074] Step S1: using a topology sensing and partitioning unit to sense the topology information of the meshed transmission and distribution network, and performing sub-area division based on a node connection relationship matrix and load characteristic parameters, and determining the boundary nodes of each sub-area;

[0075] Step S2: The power flow data acquisition and feature extraction unit collects voltage, current, and power data of each node in the power grid, extracts frequency domain features through fast Fourier transform, and constructs a data sample set;

[0076] Step S3: The improved deep Q network strategy generation unit generates a preliminary optimization strategy using the improved deep Q network algorithm through the constructed state space, action space and reward function;

[0077] Step S4: Optimize the attention mechanism weight distribution unit to calculate the attention weight according to the electrical distance between nodes and the line transmission capacity, and perform weighted fusion on the preliminary strategy;

[0078] Step S5: The multi-region collaborative decision fusion unit performs collaborative decision fusion on the decision strategies of each sub-region based on the power interaction between sub-regions and the sub-region stability index;

[0079] Step S6: The control instruction generation and issuance unit generates control instructions based on the integrated decision information and the device control parameter constraints, and sorts the instructions using a priority sorting algorithm before issuing them to the meshed transmission and distribution network equipment;

[0080] Step S7: The operation status feedback and update unit receives the equipment operation status feedback data, updates the node voltage and branch power loss status information through the status update model, and feeds back the updated information to each unit to complete an optimization cycle.

[0081] The partition-based mesh transmission and distribution network optimization system and method has achieved a key breakthrough through an innovative combination of architecture and algorithms. To address the problem that traditional methods are difficult to adapt to dynamic changes in grid topology, the system uses topology structure perception and partitioning units to monitor the connection relationships and load characteristics of grid nodes in real time, dynamically divide sub-areas based on actual operating conditions, and accurately identify boundary nodes. Compared with fixed partitioning modes, this dynamic adjustment mechanism can quickly respond to changes in grid structure. Whether it is the commissioning of new branches or the maintenance of existing lines, the system can quickly optimize the partitioning strategy to ensure that subsequent decisions always match the actual operating status of the grid.

[0082] To address the inefficiencies in data processing and decision-making efficiency and accuracy, the optimization system has built a complete data-driven decision-making chain. The flow data acquisition and feature extraction unit deeply mines real-time data such as voltage and current, extracting frequency domain features with the help of techniques such as fast Fourier transform, providing a rich and valuable information basis for subsequent decision-making. The improved deep Q network strategy generation unit constructs state and action spaces based on parameters such as node voltage amplitude and phase angle, sets a reward mechanism based on node voltage deviation and branch power loss, and generates a preliminary optimization strategy. The optimized attention mechanism weight allocation unit performs weighted fusion of preliminary strategies based on the electrical distance between nodes and line transmission capacity, focusing on the optimization needs of key nodes and lines to improve strategy accuracy.

[0083] The multi-region collaborative decision-making fusion unit further coordinates decisions across sub-regions, comprehensively considering inter-sub-region power interactions and stability indicators to form a global optimization strategy. The control instruction generation and issuance unit prioritizes instructions based on device control parameter constraints and accurately issues them. The operation status feedback and update unit collects device operation data in real time, updates grid operation status information, and drives the entire system into the next round of optimization. This model of multi-unit collaboration, deep data mining, and intelligent decision-making significantly improves data processing efficiency and decision-making accuracy, ensuring the safe and efficient operation of the meshed transmission and distribution network and effectively addressing the shortcomings of traditional technologies.

[0084] In the description of the present invention, it should be noted that, unless otherwise expressly specified or limited, the terms "disposed," "installed," "connected," "connected," and "fixed" should be understood in a broad sense. For example, they may refer to fixed connections, detachable connections, or integral connections; they may refer to mechanical connections or electrical connections; they may refer to direct connections or indirect connections through an intermediate medium; and they may refer to internal connections between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0085] While embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various equivalent changes, modifications, substitutions and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A partition-based meshed transmission and distribution network optimization system, characterized in that: include: Topology perception and partitioning unit, flow data collection and feature extraction unit, improved deep Q network strategy generation unit, optimized attention mechanism weight allocation unit, multi-region collaborative decision fusion unit, control instruction generation and issuance unit, operation status feedback and update unit; The topology structure perception and partitioning unit is connected to the power flow data acquisition and feature extraction unit through the network communication module, and is used to transmit the perceived meshed transmission and distribution network topology structure information to the power flow data acquisition and feature extraction unit; The power flow data collection and feature extraction unit is connected to the improved deep Q network strategy generation unit through a data processing bus, and sends the collected and feature-extracted power flow data to the improved deep Q network strategy generation unit; The improved deep Q network strategy generation unit is connected to the optimized attention mechanism weight allocation unit through the data interaction interface, and transmits the generated preliminary strategy information to the optimized attention mechanism weight allocation unit; The optimized attention mechanism weight allocation unit is connected to the multi-region collaborative decision fusion unit through a decision fusion channel, and transmits the weighted strategy information to the multi-region collaborative decision fusion unit; The multi-region collaborative decision fusion unit is connected to the control instruction generation and issuance unit through an instruction transmission line, and transmits the fused decision information to the control instruction generation and issuance unit.

2. The partition-based meshed transmission and distribution network optimization system according to claim 1, characterized in that: The control instruction generation and issuance unit is connected to the meshed transmission and distribution network equipment through a control signal channel for issuing control instructions, and is connected to the operation status feedback and update unit through a state feedback line for feeding back the operation status of the equipment to the operation status feedback and update unit; The operation status feedback and update unit is connected to the topology structure perception and partitioning unit, the flow data collection and feature extraction unit, the improved deep Q network strategy generation unit, the optimized attention mechanism weight allocation unit, and the multi-region collaborative decision fusion unit through the parameter update link to feedback the operation status and update the parameters of each unit; The improved deep Q network strategy generation unit includes a state space S constructed based on the voltage amplitude and phase angle parameters of the meshed transmission and distribution network nodes, and its expression is S={(V1, θ1), (V2, θ2), ..., (V n ,θ n )}, where V i represents the voltage amplitude of the i-th node, θ i represents the voltage phase angle of the i-th node, n is the total number of nodes in the meshed transmission and distribution network, i = 1, 2, ..., n; the action space A is constructed by the branch power regulation range, that is, A = {a1, a2, ..., a m }, a j represents the power regulation action for the j-th branch, m is the total number of branches, j = 1, 2, ..., m,; the reward function R is constructed based on the node voltage deviation and branch power loss, α and β are weight coefficients, is the rated voltage amplitude of node i, is the power loss of the j-th branch; through the improved deep Q network formula Q π (s, a; θ) = r + γmax a' Q π' (s', a'; θ') for strategy generation, where Q π (s, a; θ) is the value function of taking action a in state s, θ is the network parameter, r is the immediate reward, γ is the discount factor, s ' is the next state, θ ' are the target network parameters.

3. The partition-based meshed transmission and distribution network optimization system according to claim 1, characterized in that: In the optimized attention mechanism weight distribution unit, the electrical distance D between nodes is used to calculate the weight distribution of the node. ij and line transmission capacity C ij Construct an attention weight calculation model, attention weight ω ij The calculation formula is Among them, f(D ij , C ij )=σ(W1D ij +W2C ij +b), σ is the activation function, W1 and W2 are weight matrices, and b is the bias vector; the attention weight is applied to improve the strategy information generated by the deep Q network, and the weighted fusion formula is used. Get the fused strategy information, S ij is the policy information associated with node i and node j.

4. The partition-based meshed transmission and distribution network optimization system according to claim 1, characterized in that: The topology structure perception and partitioning unit divides the grid into multiple sub-areas through a partitioning algorithm according to the node connection relationship matrix G of the meshed transmission and distribution network and the node load characteristic parameters. The partitioning constraint conditions are: Among them, Ω k represents the kth sub-region, L i is the load of node i, L max is the maximum allowable load of the sub-area; and the boundary node identification formula is Determine the boundary nodes of each sub-region, E is the edge set, and v is the node.

5. The partition-based meshed transmission and distribution network optimization system according to claim 1, characterized in that: The power flow data acquisition and feature extraction unit extracts features from the collected voltage, current, and power data, obtains frequency domain features through fast Fourier transform, and constructs a feature vector X=[X1, X2, ..., X p ], where X p is the pth feature component; and constructs a data sample set D = {(x1, y1), (x2, y2), ..., (x N ,y N )},x i is the input feature vector, y i is the corresponding output label, N is the number of samples, i = 1, 2, ..., N, j = 1, 2, ..., N, used to improve the training of deep Q network.

6. The partition-based meshed transmission and distribution network optimization system according to claim 1, characterized in that: The multi-region collaborative decision fusion unit determines the decision strategy of each sub-region through the power interaction amount P between sub-regions. inter and the subregion stability index S k Construct collaborative decision-making fusion model and the decision-making strategy after fusion Among them, D k is the decision strategy for the kth sub-region, K is the total number of sub-regions, τ is the activation function, W3, W4, W5 are weight matrices, b ' is the bias vector.

7. The partition-based meshed transmission and distribution network optimization system according to claim 1, characterized in that: The control instruction generation and issuing unit constructs a control instruction generation model based on the decision information output by the multi-region collaborative decision fusion unit and the equipment control parameter constraints, and generates the circuit breaker control instruction to meet S on / off ∈{0, 1}, 0 means open, 1 means closed; Adjust the transformer tap to meet the T tap ∈[T min , T max ], T min and T max is the tap adjustment range; the control instructions are sorted and issued through the priority sorting algorithm, and the sorting formula is Priority i is the priority of the i-th control instruction, is the influence of the instruction on the current, is the voltage risk after the instruction is executed, and η1 and η2 are weight coefficients.

8. The partition-based meshed transmission and distribution network optimization system according to claim 1, characterized in that: The operation status feedback and update unit constructs a state update model based on the equipment operation status feedback data, and updates the node voltage using the formula is the updated voltage amplitude, is the voltage amplitude before updating, is the measured voltage amplitude, δ is the update coefficient; for branch power loss update, the formula is used is the change in power loss; and the updated status information is fed back to each unit for parameter update.

9. The partition-based meshed transmission and distribution network optimization system according to claim 1, characterized in that: An adaptive learning rate adjustment module is set between the improved deep Q network strategy generation unit and the optimized attention mechanism weight allocation unit to generate the error E policy and weight distribution error E weight Constructing a learning rate adjustment formula η is the adjusted learning rate, η0 is the initial learning rate, and λ is the adjustment coefficient, which is used to dynamically adjust the learning process of the improved deep Q network and optimized attention mechanism.

10. A partition-based meshed transmission and distribution network optimization method, characterized in that: The following steps are involved: Step S1: using a topology sensing and partitioning unit to sense the topology information of the meshed transmission and distribution network, and performing sub-area division based on a node connection relationship matrix and load characteristic parameters, and determining the boundary nodes of each sub-area; Step S2: The power flow data acquisition and feature extraction unit collects voltage, current, and power data of each node in the power grid, extracts frequency domain features through fast Fourier transform, and constructs a data sample set; Step S3: The improved deep Q network strategy generation unit generates a preliminary optimization strategy using the improved deep Q network algorithm through the constructed state space, action space and reward function; Step S4: Optimize the attention mechanism weight distribution unit to calculate the attention weight according to the electrical distance between nodes and the line transmission capacity, and perform weighted fusion on the preliminary strategy; Step S5: The multi-region collaborative decision fusion unit performs collaborative decision fusion on the decision strategies of each sub-region based on the power interaction between sub-regions and the sub-region stability index; Step S6: The control instruction generation and issuance unit generates control instructions based on the integrated decision information and the device control parameter constraints, and sorts the instructions using a priority sorting algorithm before issuing them to the meshed transmission and distribution network equipment; Step S7: The operation status feedback and update unit receives the equipment operation status feedback data, updates the node voltage and branch power loss status information through the status update model, and feeds back the updated information to each unit to complete an optimization cycle.

Images