A power distribution network intelligent planning method based on multi-scenario net load

Abstract

The application discloses a kind of based on multi-scene net load distribution network intelligent planning method, it is related to distribution network planning technical field, including the following steps: based on distribution network hierarchical topological relation, after the spatial coupling of the node reference power calculated, combined with multi-dimensional adjustment boundary generates multi-scene net load demand sequence;Power supply redundancy grid and load gap grid are parameterized as probability measure, and the optimal joint distribution matrix between probability measure is solved, and the optimal transmission cost is obtained;According to the capacity reduction relationship, the optimal transmission cost is converted into grid mutual aid capacity, and the substation capacity gap is obtained from the demand sequence;Composite measurement space is constructed by fusing space and electrical impedance characteristics, extract the node subset with sustained harmonic characteristics and aggregate as station address coordinates, combined with substation capacity gap to generate planning scheme.The application is used to solve the problem that capacity evaluation error and site selection deviate from real topology, leading to resource mismatch of planning scheme and insufficient network frame disturbance resistance.

Description

Technical Field

[0001] This invention relates to the field of distribution network planning technology, and more specifically, to a smart distribution network planning method based on net load in multiple scenarios. Background Technology

[0002] Existing power distribution network planning schemes mostly adopt hierarchical load forecasting or two-layer optimization configuration technology to cope with the network expansion and capacity configuration requirements under the high proportion of distributed resource access.

[0003] For example, the invention application CN121307828A discloses a hierarchical refined net load forecasting and lean distribution network planning system, which generates net load curves according to spatial levels and identifies weak links based on the power flow entropy index to generate network optimization schemes. Another example is the invention application CN114862040A, which discloses a two-layer distribution network planning method considering system flexibility requirements. This method uses forecast results to calculate the flexibility deficit at each time node and employs a particle swarm optimization algorithm to solve the two-layer model to obtain the planning scheme.

[0004] However, the aforementioned existing technologies still have the following technical problems in practical applications: 1. Load aggregation between hierarchical levels often relies on conventional numerical aggregation methods. When conducting cross-regional balance assessments, it is difficult to accurately quantify the dynamic losses and capacity reduction effects during long-distance cross-grid energy transfer, which leads to low accuracy of the calculated capacity gap in actual engineering. 2. Substation site selection and grid generation rely heavily on spatial geometric features, making it difficult to uniformly measure the actual physical power flow distribution and network spatial structure. This results in insufficient topological support capability of the planning scheme when dealing with dynamic power transfer of source and load.

[0005] To address the above problems, this invention proposes a solution. Summary of the Invention

[0006] To overcome the aforementioned deficiencies in the prior art, embodiments of the present invention provide a distribution network intelligent planning method based on multi-scenario net load. This method combines hierarchical spatial topology coupling and optimal transmission capacity replacement with topological feature clustering for site selection in composite space to solve the problems of distorted capacity demand calculation and site selection deviation from the power grid topology in the prior art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A smart distribution network planning method based on net load in multiple scenarios includes the following steps: Based on the hierarchical topology of the distribution network, the node baseline power calculated from historical operating data is spatially coupled, and then combined with the pre-constructed multi-dimensional adjustment boundary to generate a multi-scenario net load demand sequence. The power supply redundancy grid and the load gap grid are parameterized as a first probability measure and a second probability measure, respectively. The optimal joint distribution matrix between the first probability measure and the second probability measure is solved to obtain the optimal transmission cost. Based on the pre-established capacity reduction relationship, the optimal transmission cost is converted into grid mutual assistance capacity, and then the demand sequence is subjected to equivalent capacity replacement to obtain the substation capacity gap; A composite metric space integrating spatial and electrical impedance characteristics is constructed. After extracting a subset of nodes with continuous coherence characteristics and aggregating them into site coordinates, a planning scheme is generated in conjunction with the substation capacity gap.

[0008] In a preferred embodiment, the multidimensional adjustment boundary includes at least a safety adjustment boundary and an operational adjustment boundary; the safety adjustment boundary includes: fitting the equivalent thermal parameters of the temperature-controlled load based on meteorological data and historical electricity consumption data, and constructing a generalized energy function characterizing the heat exchange process; under preset thermodynamic state evolution constraints, solving the generalized energy function to obtain the steady-state response range of the adjustable power, which serves as the safety adjustment boundary.

[0009] In a preferred embodiment, the operational adjustment boundary includes: constructing power transmission constraints that take into account equipment capacity margin and topology connectivity using logical variables characterizing line open / closed states as decision variables; and performing topology reconfiguration optimization under the power transmission constraints with the goal of minimizing the highest node load, using the power flow transfer before and after topology reconfiguration as the operational adjustment boundary.

[0010] In a preferred embodiment, generating the multi-scenario net load demand sequence includes: performing multi-scale decomposition and phase alignment on the node reference power; nonlinearly coupling the aligned node reference power to the substation level based on pre-established power conservation and topology correlation constraints and distribution network hierarchical topology relationships to obtain the substation level power; and correcting the substation level power using the multi-dimensional adjustment boundary under preset differentiated scenarios to obtain the multi-scenario net load demand sequence.

[0011] In a preferred embodiment, parameterizing the power supply redundancy grid and the load gap grid into a first probability measure and a second probability measure includes: allocating the multi-scenario net load demand sequence to each power supply grid according to a preset ratio to obtain the grid net load demand base; comparing the demand base with the total power output in each power supply grid, and dividing the power supply redundancy grid and the load gap grid based on the comparison result; calculating the proportion of the excess power of each power supply redundancy grid to the total excess power of the entire network, and combining them to obtain the first probability measure; calculating the proportion of the power gap of each load gap grid to the total power gap of the entire network, and combining them to obtain the second probability measure.

[0012] In a preferred embodiment, the step of solving for the optimal joint distribution matrix between the first probability measure and the second probability measure to obtain the optimal transmission cost includes: constructing a transmission cost matrix based on the spatial distance between grids and electrical impedance characteristics; performing entropy-regularized optimal transmission calculation with the first probability measure and the second probability measure as marginal distribution constraints, combined with the cost matrix, to obtain the optimal joint distribution matrix; and calculating the global expectation of the distribution matrix and the cost matrix as the optimal transmission cost.

[0013] In a preferred embodiment, the construction of a composite metric space that integrates spatial and electrical impedance characteristics includes: calculating the electrical potential difference between nodes based on the spatial location and equivalent electrical impedance of network nodes, and identifying the power transmission direction between nodes in combination with the power flow distribution; and performing nonlinear weighting on the electrical potential difference and the power transmission direction to obtain the composite metric space.

[0014] In a preferred embodiment, the step of extracting a subset of nodes with continuous homology characteristics and aggregating them into site coordinates includes: performing multi-scale topological connectivity analysis on distribution network nodes within the composite metric space to identify a subset of connected nodes with continuous homology characteristics; and performing weighted aggregation on the subset of nodes based on the spatial distribution characteristics of the network nodes corresponding to the substation capacity gap to obtain the site coordinates.

[0015] In a preferred embodiment, identifying a subset of connected nodes with persistent homology features includes: constructing a nested spatial sequence reflecting the node topology clustering process within the composite metric space; identifying the evolution period of each topological connectivity component in the nested spatial sequence and screening persistent homology features that satisfy preset multi-scale stability conditions; mapping the persistent homology features to the distribution network physical topology map and combining them to obtain a subset of connected nodes.

[0016] An electronic device includes: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of any of the methods.

[0017] The technical effects and advantages of this invention's intelligent distribution network planning method based on multi-scenario net load are as follows: 1. This invention generates a demand sequence by spatially coupling the node reference power, then solves for the optimal transmission cost and converts it into grid mutual assistance capacity, and uses this to extract the substation capacity gap. This process takes into account the dynamic loss of long-distance energy transfer, reduces the calculation error introduced by traditional simple numerical aggregation between levels and direct cross-regional offsetting, improves the accuracy of substation capacity gap calculation, and enhances the rationality and physical supply guarantee capability of the distribution network planning scheme throughout its entire life cycle.

[0018] 2. This invention constructs a composite metric space that integrates spatial and electrical impedance characteristics, and extracts a subset of nodes with continuous homology characteristics to aggregate as site coordinates; it unifies the measurement of spatial geometric features and actual power flow characteristics of the power grid, effectively avoiding the limitations of traditional site selection planning that is prone to deviating from the electrical connectivity characteristics of the network, enhancing the anti-disturbance capability of the physical network structure, and improving the global topology support and security defense capability of the distribution network in response to complex source and load situation evolution. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the distribution network intelligent planning method based on multiple scenario net loads provided in an embodiment of the present invention.

[0020] Figure 2 The graph shows the convergence comparison of the large-scale grid optimal transport algorithm provided in the embodiments of the present invention.

[0021] Figure 3 This is a structural block diagram of an exemplary electronic device provided for implementing embodiments of the present disclosure. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0023] Example 1, Figure 1 This invention presents a smart distribution network planning method based on net load in multiple scenarios, comprising the following steps: S1. Based on the distribution network hierarchical topology, spatially couple the node baseline power calculated from historical operating data, and then combine it with a pre-constructed multi-dimensional adjustment boundary to generate a multi-scenario net load demand sequence, including: S101. Based on historical operational data, calculate the baseline power of the underlying physical nodes, as follows: Through the data acquisition and monitoring control system (SCADA) and advanced metering infrastructure (AMI) of the power distribution network, the active and reactive power of each node over the past year are extracted, and the corresponding meteorological environmental characteristics are obtained by calling the meteorological bureau interface to serve as historical operating data. The 3σ criterion is used to identify and remove outliers in the historical operating data sequence, and the missing values ​​are smoothed by Lagrange interpolation to complete the data cleaning. The extreme value normalization method is used (i.e., by calculating the difference between the current value of each feature and the minimum value of the feature in the historical data, and dividing by the difference between the historical maximum value and the minimum value) to uniformly map the cleaned multidimensional feature data sequence to the dimensionless interval [0,1] to obtain the normalized feature sequence. During the extreme value normalization process, the system synchronously records and saves the historical maximum and minimum values ​​corresponding to each feature as feature extreme value parameters. Subsequently, the normalized feature sequence is input into a standard Long Short-Term Memory (LSTM) network model to calculate the baseline power of nodes. The LSTM model employs a single-layer hidden layer LSTM structure with 64 hidden nodes. This model controls information flow through input, forget, and output gates to extract deep temporal features from historical data. During the LSTM model training phase, the cleaned and normalized dataset is divided into training and validation sets at a ratio of 80% to 20%. The Adam optimizer is used to update network parameters, with an initial learning rate of 0.001 and a maximum training epoch of 500 epochs. The training convergence condition is set to the mean squared error of the validation set being less than a set threshold (e.g., 1e-4) for 10 consecutive epochs. In the model inference and prediction phase, the target prediction time is... Weather forecast data and A previously preset sliding time window (such as) Hours to Historical power data within the specified time frame (within hours) is processed using the same extreme value normalization method with the aforementioned characteristic extreme value parameters, and then input into the trained LSTM model to output the target prediction time. The normalized result of the nodal reference power prediction is obtained; finally, the normalized result is denormalized by the eigenvalue extreme value parameter (that is, the normalized result is multiplied by the extreme value difference and the minimum value is added) to restore the nodal reference power prediction value with true physical dimensions. The calculation model is expressed as follows: (1) in, For underlying physical nodes At the target prediction time The node reference power prediction value; This is an inverse normalization mapping function performed using eigenvalue parameters; For the forward propagation mapping function of the LSTM network; For underlying physical nodes After normalization, including the target prediction time The meteorological characteristics and the multidimensional input feature vector of historical power characteristics within the sliding time window; This is the set of parameters, including the weight matrix and bias terms, after the LSTM network has been trained and converged.

[0024] S102. Perform multi-scale decomposition and phase alignment on the node reference power. Based on the pre-established power conservation and topological correlation constraints and the distribution network hierarchical topology, nonlinearly couple the aligned node reference power to the substation level to obtain the substation level power, as follows: The variational mode decomposition algorithm is used to set the expected number of modes based on the physical periodicity of the load spectrum characteristics (such as daily periodicity, weekly periodicity, and high-frequency random disturbances). (Can be set to 5) and secondary penalty factor (Can be set to 2000), the reference power of each node is decomposed into multiple intrinsic mode functions with different center frequencies to remove high-frequency noise and extract low-frequency trends; then, a dynamic time warping algorithm is used, with the standard reference clock sequence of the substation as a reference, to solve the minimum distortion path of the cumulative distance matrix between the low-frequency trend mode function decomposed from the reference power of each node and the reference clock sequence through dynamic programming, and to perform micro-level scaling and alignment on the time axis to eliminate phase shift caused by data asynchrony; finally, the aligned mode functions are reconstructed and summed to obtain the aligned node reference power that has eliminated asynchronous errors and is in the same physical time section. This corrects the measurement clock drift while retaining the objective physical peak and natural asymmetry characteristics of each underlying physical node, ensuring that the subsequent spatial coupling satisfies the synchronization premise of Kirchhoff's laws; Next, the hierarchical topology of the distribution network is extracted from the power geographic information system (GIS), and the distribution network is abstracted into a graph model, where nodes (such as transformers and buses) are regarded as vertices and physical lines are regarded as edges, thus transforming it into an adjacency matrix representing the connection state (if the underlying physical nodes...). , If there is a physical line between them (Otherwise it is 0); based on the adjacency matrix, the power limit is established only when... Topological association constraints on the flow on the branch are established, and power conservation constraints are established based on Kirchhoff's law that the algebraic sum of the inflow and outflow power of the nodes is zero. To account for the actual line losses and electrical interactions between nodes during power transmission on physical lines, a nonlinear spatial coupling method is adopted using the standard DistFlow branch power flow equations combined with a forward-backward algorithm. Specifically, it is first assumed that the voltage amplitude of each bottom-level physical node in the entire network is its rated per-unit value, and the aligned node reference power is used as the initial load of the terminal leaf node. The power of each branch is calculated by back-pushing upstream along the topology path. Then, starting from the actual monitored voltage amplitude of the substation root node, the actual voltage amplitude of each bottom-level physical node is calculated by forward-pushing downstream along the topology path. The system continuously alternates between the back-pushing power and forward-pushing voltage steps until the maximum value of the deviation of the total network node voltage amplitude obtained from two adjacent iterations is less than the set convergence accuracy threshold (which can be preset to 10). -4 When the algorithm converges, the substation-level power considering the nonlinear losses of the entire network can be obtained; the calculation formula for the nonlinear spatial coupling is as follows: (2) (3) (4) in, and These are the underlying physical nodes obtained after processing the LSTM output. At the target prediction time The predicted active power and reactive power base values; and The target prediction time is respectively During the iterative solution process, from the upstream underlying physical nodes Flow to Node The active power and reactive power of the branch; and The target prediction time is respectively From node Active power and reactive power of branches flowing to more downstream branches; and The target prediction time is respectively The underlying physical nodes obtained after convergence of the forward and backward iterations. , The voltage amplitude is set to a per-unit value of 1.0 during the initial iteration; and These are the underlying physical nodes. , The equivalent resistance and equivalent reactance of the connecting branches are obtained by mapping from the GIS equipment parameter ledger through the adjacency matrix index.

[0025] S103. Based on the equivalent thermal parameters of the temperature-controlled load, a safe regulation boundary is constructed in the multidimensional regulation boundary, as follows: Ambient temperature and light meteorological data are obtained through the meteorological bureau's public API interface. At the same time, historical electricity consumption and room temperature response data of typical temperature-controlled loads in the distribution network (such as large cold storage facilities and data center temperature control systems on the industrial side, and central air conditioning systems in commercial or user-side buildings) are obtained through the electricity consumption information collection interface of the power marketing system. The response trends of various temperature-controlled loads under step signals are extracted, and curve fitting is performed using the least squares method to obtain the equivalent thermal parameters of specific temperature-controlled facilities (such as the insulated chambers of industrial cold storage facilities or commercial buildings), namely, equivalent thermal resistance and equivalent heat capacity. Based on the equivalent thermal parameters and according to the RC equivalent thermodynamic law, a first-order differential equation for the dynamic change of indoor temperature is established as a generalized energy function. The expression of the generalized energy function is as follows: (5) in, Equivalent heat capacity; and Each of these is a historical moment The user's actual indoor temperature (such as the temperature inside the cold storage or the room temperature in the building) and the outdoor / outdoor ambient temperature; Equivalent thermal resistance; The cooling or heating energy efficiency ratio constant is obtained from the equipment nameplate for the main temperature-controlled load (such as industrial refrigeration units or air conditioning equipment); For the main body of temperature control at a historical moment The measured electrical power; The sign switching between cooling and heating modes: When the temperature control load is working in cooling mode (such as summer air conditioning, industrial cold storage), this item takes a negative sign to indicate a temperature drop; when working in heating mode (such as winter heat pump), this item takes a positive sign to indicate a temperature rise. Thermodynamic state evolution constraints are set according to the type of temperature-controlled load. Specifically: for commercial and user-side loads, the internal temperature is constrained to be within the temperature range corresponding to the recommended index range based on the PMV (Predicted Average Votes) human thermal comfort standard (-0.5 to +0.5, such as corresponding to summer room temperature of 22℃ to 26℃); for industrial-side loads, the internal temperature is constrained to be within the temperature range defined by the upper and lower limits of the extreme safe temperature of industrial production processes. Under the constraints of the thermodynamic state evolution, the steady-state response range of the adjustable power is obtained by solving the generalized energy function, which serves as the safe adjustment boundary; specifically, the derivative term... We set the value to zero to represent thermodynamic equilibrium steady state, and then successively substitute the upper and lower limits of the temperature range corresponding to each temperature-controlled load into the generalized energy function equation. In this process, the maximum and minimum electrical power allowed for the operation of the temperature control device under the constraints of thermodynamic state evolution can be solved by algebraic rearrangement. This is the steady-state response range of the temperature control body. Finally, the response ranges of each temperature control body under the same node are summed algebraically to obtain the safety regulation boundary.

[0026] To further illustrate the quantitative manifestation of the thermodynamic state evolution constraints in practical engineering and to intuitively verify the physical reality of the fitted parameters, this embodiment takes a typical regional power distribution network as an example and extracts the equivalent thermodynamic parameters and steady-state response boundaries of three typical temperature control subjects. The data are shown in Table 1. Table 1

[0027] S104. Introduce logical variables characterizing the topological connection state of the distribution network to construct the operation regulation boundary in the multidimensional regulation boundary, as follows: Boolean 0-1 logic variables are introduced to represent the physical open / closed states of distribution network tie switches and sectionalizing switches (1 for closed, 0 for open). Based on these logic variables, power transmission constraints considering equipment capacity margin and topological connectivity are established: These constraints are implemented using the spanning tree constraint equation in graph theory (i.e., the total number of constrained closed branches is always equal to the total number of nodes minus one, specifically...). In conjunction with the Laplace matrix of the distribution network graph constructed based on the aforementioned logical variables (i.e., constraining the algebraic connectivity of the Laplace matrix to be greater than zero to ensure that there is a reachable path from any node to the root node of the substation), it ensures that the network maintains a radial connected topology without islands or loops at any switching time; at the same time, it constrains the current and node voltage on all closed branches to not exceed the physical limits of the line's safe current carrying capacity and the transformer's rated capacity. Construct an objective function that aims to minimize the highest load on the node, as follows: (6) in, The highest load rate across the entire network. This is the set of all candidate branches in the distribution network; The actual current value between branches ij; To represent the open / closed state of a branch as a 0-1 logic variable, so as to ensure that the objective function is evaluated only for the valid branches in the closed state; The maximum thermally stable throttling capacity allowed for branch ij; Within the solution space satisfying the power transmission constraints, the Discrete Particle Swarm Optimization (BPSO) algorithm is used to perform topology reconstruction optimization on the objective function. The particle swarm size in the BPSO algorithm is set to 50, and both the individual learning factor and the social learning factor are preset to 2.0. The inertia weight decreases linearly from 0.9 to 0.4 with each iteration, and the 0-1 state sequence of all switches is encoded as the particle position vector. During the algorithm's iterative optimization process, considering that the distribution network needs to maintain a radial topology constraint free of circulating currents and islands, direct updates of particle positions are prone to generating illegal topological solutions. After each particle position update, a topological validity verification and repair mechanism based on the basic loop matrix in graph theory is implemented. Specifically: for the 0-1 state sequence generated by the update, if a closed loop exists in the corresponding topological graph, then any switch branch in the loop except the backbone network is disconnected to eliminate the loop network; if an isolated node exists in the corresponding topological graph, then the connecting switch associated with the isolated node is forcibly closed based on the adjacency matrix to restore network connectivity; through the verification and repair mechanism, all particles that generate out-of-bounds or illegal topologies after the update are forcibly remapped back to the valid solution space that satisfies the radial running characteristics. When the variance of the fitness of the global optimal solution after 20 consecutive iterations is less than a set minimum value (e.g., 10), -5 When the algorithm converges (on the order of magnitude of the original load) or reaches the set maximum number of iterations (e.g., 200 times), it outputs the optimal reconfiguration topology state. By calculating the algebraic difference between the original total load within the substation's power supply range before the optimal reconfiguration topology state operation and the actual total load borne by the substation after reconfiguration (when some loads are transferred to neighboring substations), the power flow transfer amount brought about by the topology reconfiguration can be obtained, which serves as the operation and adjustment boundary of the distribution network.

[0028] S105. Under the preset differentiated scenarios, the substation-level power is corrected using the multi-dimensional adjustment boundary to obtain the multi-scenario net load demand sequence, as follows: Since the actual capacity gap of substations is often exposed to extreme environmental conditions, the system, based on historical meteorological data, pre-sets multiple differentiated operating scenarios representing physical extremes, such as "summer extreme heat peak load scenario," "winter extreme cold valley scenario," and "high proportion of renewable energy extreme increase scenario." Under each specific extreme scenario, the substation-level power corresponding to the scenario calculated in S102 is used as the base, and the calculated safety regulation boundary (the temperature control capacity that can be reduced on the user side) and operation regulation boundary (the power flow load that can be transferred on the grid side) are added or subtracted. The power load remaining after multi-dimensional boundary correction, which the system itself can no longer digest, is the actual rigid demand under the scenario. By traversing multiple differentiated scenarios, multiple sets of sequence results corresponding to different extreme physical sections are finally output, which are the net load demand sequences of the multiple scenarios.

[0029] This step, by reconstructing the spatial coupling process from the bottom-level nodes to the substation and the multi-dimensional operational boundary constraints, takes into account transmission loss, thermodynamic hysteresis, and network topology flexibility. It reduces the risk of electrical condition assessment distortion caused by simple numerical superposition, improves the accuracy of demand load evolution projection, and provides a rigorous and physically realistic data foundation for subsequent capacity resource optimization and network reconfiguration at the entire network level.

[0030] S2. Parameterize the power supply redundancy grid and the load gap grid into a first probability measure and a second probability measure, respectively. Solve for the optimal joint distribution matrix between the first probability measure and the second probability measure to obtain the optimal transmission cost, including: S201. Divide the power supply redundancy grid and the load gap grid, and parameterize them into a first probability measure and a second probability measure, as follows: Based on spatial geographic coordinate boundaries and administrative divisions, the power supply range of the physical distribution network is divided into multiple independent power supply grids. For example, a 10kV distribution substation or an independent power supply area physically isolated by a tie circuit breaker is divided into a single grid unit. For each divided power supply grid, the distributed power sources (such as photovoltaic and wind power) contained within the grid are directly obtained at the target prediction time through the distribution network's new energy dispatch management system or microgrid control platform. The total output data is compared with the multi-scenario net load demand sequence generated in step S1: Specifically, for each differentiated operating scenario in the multi-scenario net load demand sequence, the substation's output at the target prediction time is independently extracted. The total net load value of the scenario is calculated, and based on the ratio of the historical highest load peak value of a single independent grid to the sum of the historical highest load peak values ​​of all grids in the entire network, the total net charge value is proportionally allocated to each independent power supply grid to obtain the target prediction time for each independent power supply grid under this operating scenario. The grid's net load demand baseline is used to determine the grid's net load demand baseline. If the total power output is greater than the grid's net load demand baseline, the grid is determined to be a power supply redundancy grid in this operating scenario. Conversely, if the total power output is less than the grid's net load demand baseline, the grid is determined to be a load deficit grid in this operating scenario. For each differentiated operating scenario, the total power output of the redundant power grids is calculated to exceed the net load demand base of the corresponding grid. The proportion of excess power in each redundant grid to the total excess power is calculated as a probability measure. Combining the probability measures corresponding to all redundant power grids yields the first probability measure. Similarly, the power gap where the total power output of each load gap grid is lower than the net load demand base of the corresponding grid is calculated. The proportion of the power gap of each load gap grid to the total power gap of the entire network is calculated as a probability measure. Combining the probability measures corresponding to all load gap grids yields the second probability measure. Since the source-load situation of each grid may reverse and evolve under different scenarios, the first and second probability measures exhibit differentiated distributions under different scenarios.

[0031] In this embodiment, solving for the optimal joint distribution matrix between the first probability measure and the second probability measure to obtain the optimal transmission cost includes: S202. Construct the transmission cost matrix based on the spatial distance and electrical impedance characteristics between grids, as follows: The latitude and longitude geographic coordinates of each distribution transformer or load node within the grid are obtained through the power geographic information system (GIS), and the rated capacity of each node is obtained from the power distribution management system as a weight. The latitude and longitude geographic coordinates of each node are weighted and averaged to obtain the equivalent geographic center point of the grid. The spatial Euclidean distance between the equivalent geographic center points of the power supply redundancy grid and the load gap grid is calculated. The algorithm calls upon the distribution network topology connectivity map and line ledger, and employs Dijkstra's shortest path search algorithm to calculate the shortest electrical path between the equivalent geographic center points of the grid. Specifically, the search algorithm uses the equivalent geographic center point of the starting grid as the source node, initializes the cumulative impedance from the source node to all other nodes to infinity, and sets the source node itself to zero. During the iterative search, each time, the node with the smallest current cumulative impedance is selected from the set of unvisited nodes as the working node, and a relaxation operation is performed: if the sum of the cumulative impedance of the working node and the impedance of the connected branch is less than the current recorded impedance of its neighboring nodes, the cumulative impedance of the neighboring nodes is updated with this smaller value. The search terminates when the equivalent geographic center point of the target grid is marked as visited. Finally, the algorithm backtracks and outputs the set of connected branches with the smallest cumulative impedance distance between the equivalent geographic center points of the grid as the shortest electrical path, and accumulates the impedance values ​​of all branches on the shortest electrical path as the equivalent impedance. The calculated spatial distance and equivalent impedance are respectively subjected to extreme value normalization processing. Each of the spatial distance and equivalent impedance constitutes a value corresponding to... A power supply redundancy grid and Each load gap grid is generated by combining two pairs of grids. The extreme value normalization process involves subtracting the minimum value from each original value in its set and then dividing by the difference between the maximum and minimum values, thereby linearly mapping all values ​​to the dimensionless interval [0, 1]. Subsequently, the normalized spatial distance and equivalent impedance are linearly weighted and fused to obtain the transmission cost matrix. The calculation formula for the transmission cost matrix is ​​as follows: (7) in, The first in the transmission cost matrix Line number Column (i.e., the first) The power supply redundancy grid and the first Matrix elements between load gap grids; For power supply redundancy grid With load gap grid Dimensionless spatial distance after normalization; For connected meshes With grid The dimensionless equivalent impedance after normalization; and The weighting coefficients, which respectively characterize the influence of spatial distance and electrical impedance on power transmission, can be preset to 0.3 and 0.7, respectively.

[0032] S203. Using the first probability measure and the second probability measure as marginal distribution constraints, and combining the cost matrix, perform entropy-regularized optimal transmission calculation to obtain the optimal joint distribution matrix, as follows: To construct the optimal transmission edge distribution constraints, the row sum and column sum of the joint distribution matrix characterizing the power transfer scheme are required to be equal to the first probability measure (conservation of total power migration) and the second probability measure (conservation of total power reception), respectively. Next, to avoid the curse of dimensionality in large-scale grids and to smooth the transmission scheme using classical linear programming, a Wasserstein distance objective function including a Shannon entropy regularization term is constructed. Based on convex optimization theory, the optimal joint distribution matrix solution is equivalent to the product of a diagonal scaling vector and a Gibbs kernel matrix. The Wasserstein distance objective function and the... The formula for calculating the joint distribution matrix in the next iteration is as follows: (8) (9) in, To transmit the target function value; and These are the first probability measure vector and the second probability measure vector, respectively. For and The set of feasible solutions for the joint distribution matrix with marginal distribution constraints; The elements of the joint distribution matrix to be solved are physically characterized from the power supply redundancy grid. To load gap grid Probabilistic quality allocation weights for inter-regional power transfer; and These represent the total number of redundant power supply grids and load gap grids in the distribution network, respectively. The entropy regularization coefficient, which controls the smoothness of the transmission scheme and the convergence speed of the algorithm, can be set to 0.05. The first in the transmission cost matrix Line number Column matrix elements; Let Shannon entropy be the joint distribution matrix, and ; For the first Matrix elements generated by the Sinkhorn algorithm iteration; The first Gibbs kernel matrix pre-computed based on the transmission cost matrix. Line number Column elements, and ; To solve the objective function, the standard Sinkhorn alternating projection algorithm is introduced for iterative optimization, specifically including: in the optimization initialization phase, setting the initial values ​​of the left and right diagonal scaling vectors. and All are column vectors with all elements equal to 1; during the optimization process, a pre-computed Gibbs kernel matrix is ​​used. Alternately update the scaling vectors on the left and right sides. The specific update formula is as follows: , In this case, division is performed by dividing each element of the vector. and The first The left and right scaling vectors are updated after each iteration; after each alternating update of the scaling vectors, the updated scaling vectors are substituted into the calculation formula of the joint distribution matrix to obtain the joint distribution matrix for the current iteration. The iteration continues until the difference in the Frobenius norm between two consecutive iterations is less than a set convergence threshold (e.g., 10). -4 The optimization process terminates when the maximum number of iterations (e.g., 1000) is reached, and the optimal joint distribution matrix at convergence is obtained.

[0033] To visually verify the technical effect of introducing the Shannon entropy regularization term in this invention. Figure 2A comparison curve of algorithm convergence under a typical large-scale grid mutual assistance scenario is presented. As shown in the figure, traditional linear programming solution methods are prone to numerical oscillations due to the degeneracy of the solution space when dealing with high-dimensional constraints, making it difficult to achieve the expected convergence criterion. However, the entropy regularized Sinkhorn algorithm proposed in this invention shows a smooth exponential decreasing trend in the matrix update difference norm after about 120 iterations, and quickly falls below and stabilizes below the set convergence threshold. This indicates that the proposed scheme can effectively avoid the non-convergence problem caused by the curse of dimensionality when dealing with large-scale grid metric matrices, and has engineering effectiveness and computational robustness.

[0034] S204. By performing a Hadamard product (i.e., multiplying corresponding elements) on the optimal joint distribution matrix and the transmission cost matrix, and then calculating the global expectation, the optimal transmission cost that quantifies the global allocation difficulty can be obtained. The calculation formula is as follows: (10) in, This is the global expected value, i.e., the optimal transmission cost; These are the steady-state elements of the optimal joint distribution matrix.

[0035] This step abstracts the complex physical process of cross-regional power grid mutual assistance into an optimal transmission problem in probability space, quantifies the dynamic loss and reduction effect of long-distance cross-grid scheduling, and provides a rigorous quantitative theoretical basis for subsequent elimination of ineffective flexible mutual assistance capacity and extraction of the real rigid capacity expansion gap of substations.

[0036] S3. Based on the pre-established capacity reduction relationship, after converting the optimal transmission cost into grid mutual assistance capacity, the demand sequence is subjected to equivalent capacity replacement to obtain the substation capacity gap, including: In actual operation of distribution networks, long-distance power transmission and power transmission across complex, heavily loaded topologies are often accompanied by the accumulation of active power losses due to resistance heating along the lines, and a decrease in the voltage support capacity at the end of the lines due to uneven reactive power distribution. Therefore, the transmission cost and the actual resilient capacity exhibit typical nonlinear attenuation characteristics. Based on this characteristic, a capacity reduction function based on a negative exponential attenuation law is constructed as the capacity reduction relationship. Using this capacity reduction function, the optimal transmission cost can be directly converted into the effective grid resilient capacity that the gap grid can actually receive. The functional formula for the capacity reduction relationship is as follows: (11) (12) in, For the scene Below, load gap grid After deducting transmission losses, the total grid mutual assistance capacity that can actually be extracted from the entire network of redundant grids; In the scene The sum of excess power in all redundant power grids across the entire network; The physical meaning of the product is that, under the optimal transmission strategy, from the power supply redundancy grid... Actual directional allocation to the load gap grid The physical power mutual aid reference value; The steady-state elements of the optimal joint distribution matrix are used here as topology allocation coefficients to orient redundant power to each gap grid. This is the capacity reduction function that is constructed; The spatial physical attenuation damping constant is obtained by extracting the active power loss records of cross-regional power flow transmission from the historical SCADA system of the distribution network, and using nonlinear least squares method to curve fit the nonlinear mapping relationship between transmission distance, equivalent impedance and actual network loss rate. In the scene Underloaded gap grid The scene capacity gap; In the scene Underloaded gap grid The initial power gap after deducting the output of local distributed power sources is the difference between the grid net load demand base and the total power output in step S201. The function is used to ensure that the capacity gap does not produce a negative value that has no physical meaning; Subsequently, after calculating the scenario capacity gap sequence for each load gap grid under all operating scenarios, the maximum capacity gap value corresponding to the gap grid is extracted by traversing along the scenario axis, which is taken as the final substation capacity gap of the grid. Taking the derivation process of substation capacity gap in a typical distribution network area as an example: the initial power gap of a certain load gap grid in the "summer extreme heat peak" scenario is 15 MW, at which time the photovoltaic excess power of the remote power supply redundant grid is 10 MW. Based on the capacity reduction function, due to the high electrical impedance leading to the increase in transmission cost, the actual power gap of the 10 MW after negative exponential decay is... The effective mutual assistance capacity drops to 6 MW, and the calculated capacity gap for this grid in the summer scenario is 9 MW. However, in the "winter extreme cold and low valley" scenario, the initial power gap of this load gap grid reaches 18 MW due to the increase in heating demand. At this time, the excess power of the remote power supply redundant grid drops to 8 MW due to the influence of weather. After the same attenuation reduction, the effective mutual assistance capacity is only 4 MW, and the calculated capacity gap for this gap grid in the winter scenario is 14 MW. The system traverses the scenario capacity gap sequence along the scenario axis and extracts the maximum value of 14 MW, which is taken as the substation capacity gap that needs to be planned and expanded for this gap grid in the end.

[0037] This step quantifies the available capacity loss during cross-regional power support by constructing a nonlinear capacity reduction mechanism that takes into account electrical and spatial obstacles. At the same time, it effectively removes flexible demands that can be mitigated through grid mutual assistance by using equivalent substitution in multiple scenarios, providing rigorous data support for subsequently identifying the real rigid capacity expansion gap of substations.

[0038] S4. Construct a composite metric space that integrates spatial and electrical impedance characteristics, extract a subset of nodes with continuous coherence characteristics and aggregate them into site coordinates, then combine this with the substation capacity gap to generate a planning scheme, including: S401. Based on the spatial location and equivalent electrical impedance of each node within the load gap grid, calculate the electrical potential difference between nodes and identify the power transmission direction between nodes by combining the power flow distribution, as detailed below: For each load gap grid with a substation capacity shortage, the latitude and longitude coordinates of each underlying physical node (distribution transformer, branch bus, etc.) within the load gap grid are extracted. The equivalent resistance and equivalent reactance between connected branches are obtained by mapping using GIS equipment parameter ledgers. Next, the operating scenario corresponding to the substation capacity shortage of the load gap grid is extracted, and the specific time corresponding to the load peak in the operating scenario is used. Extract the DistFlow branch power flow equations from step S102 at this specific moment. The network state (i.e., branch power and node voltage) is obtained by solving the following: Finally, combining the network state, equivalent resistance, and equivalent reactance, the electrical potential difference between the underlying physical nodes is calculated using the following formula: (13) in, For the bottom-level physical nodes within this load gap grid With nodes The electrical potential difference between them; and The values ​​calculated in step S102 at time [time] are respectively Below, from the underlying physical nodes Flow to Node The active power and reactive power of the branch; and These are the underlying physical nodes. and The equivalent resistance and equivalent reactance of the connecting branches; The underlying physical nodes obtained in step S102 At any moment The voltage amplitude; In power flow calculations for distribution networks, the algebraic sign of the active power in a branch naturally represents the actual flow direction of power relative to a given branch reference direction; based on this, the sign of the active power is directly extracted as the criterion for judgment: if Then the power transmission direction is determined to be from the underlying physical node. Flow to Node ;like Then the power transmission direction is determined to be from the node. Flow to Node After the determination is completed, the positive or negative sign feature is used as input and mapped to the sign function in the subsequent step S402. In this context, it is used to punish reverse current paths.

[0039] S402. The electrical potential difference and the power transmission direction are nonlinearly weighted to obtain a composite metric space, as follows: Based on the latitude and longitude coordinates of the underlying physical nodes, the spatial geographic Euclidean distance between each pair of underlying physical nodes is calculated. Subsequently, using an extreme value normalization method, the absolute values ​​of the spatial geographic Euclidean distance and the electrical potential difference between the nodes are linearly mapped to the [0,1] interval, respectively. Then, a nonlinear weighted fusion considering the power transmission direction penalty is performed to obtain a composite metric distance characterizing the comprehensive correlation strength between the underlying physical nodes. The formula for the nonlinear weighted fusion is as follows: (14) in, The distance is the composite metric after fusion; For underlying physical nodes , The dimensionless spatial Euclidean distance between them after extreme value normalization; and To adjust the weighting coefficients for spatial and electrical characteristics, values ​​of 0.4 and 0.6 can be used respectively. The dimensionless electrical potential difference between nodes after extracting the absolute value and normalizing the extreme values; To characterize the nonlinear penalty function in the power transmission direction, For the sign function, if the power flow is... Flow direction (Right now , ), then the exponent term The calculated result is less than 1, thus reducing the equivalent distance of forward power supply in the metric space; conversely, if the power flow is reversed (i.e., ... , The exponent term will be greater than 1, thus multiplying the distance measurement. This is the positive decay penalty parameter, which can be 2.0. For the current load gap mesh, traverse all the underlying physical nodes within it and calculate the load gap between any two nodes. and The composite distance between them is used as the calculated value as the matrix. Line number The elements of the column, thus constructing a scale of The distance matrix (where (This refers to the total number of bottom-level physical nodes within the load gap grid). Due to the difference in the exponential penalty coefficients corresponding to forward and reverse power flows, the number of bottom-level physical nodes... arrive The distance is not equal to arrive distance (i.e.) This results in the distance matrix exhibiting typical asymmetric characteristics, which preliminarily characterizes the composite metric space corresponding to the current load gap grid.

[0040] S403. Within the composite metric space, perform multi-scale topological connectivity analysis on the distribution network nodes to identify a subset of connected nodes with continuous homology characteristics, as follows: To accommodate the symmetry requirements of the Vietoris-Rips complex algorithm in terms of topological axioms, the system first performs a symmetric reconstruction of the asymmetric distance matrix. Specifically, it traverses all pairs of underlying physical nodes within the asymmetric distance matrix and extracts the nodes respectively. arrive Composite metric distance With nodes arrive Composite metric distance The maximum value of the two composite metric distances is taken as the symmetricized undirected equivalent distance. Then, all the undirected equivalent distances calculated through traversal are filled into the matrix. Line number The positions of the columns are used to construct a symmetric distance matrix that corresponds one-to-one with the composite metric space; Using the symmetric distance matrix as input, the Vietoris-Rips complex algorithm is used to construct a nested spatial sequence reflecting the evolution of node clustering. Specifically, a sequence is introduced that continuously increases from 0 to the maximum connected diameter. (In this embodiment) The scale parameter (preset to be the maximum element value in the distance matrix) During the incremental evolution of the scale parameter, when the undirected equivalent distance between any two bottom-level physical nodes... When a topological edge is constructed between two nodes, a topological edge is constructed between them. When multiple nodes are fully interconnected, a corresponding high-dimensional simplex (such as a triangular face or a tetrahedron) is generated by closing the loop. This geometric structure, which is continuously generated and merged from low-dimensional topological edges to high-dimensional simplexes, constructs a series of spatial geometric sets with inclusion relationships, namely the nested spatial sequence. During the evolution of scale parameters, the standard continuous homology matrix reduction algorithm in the field of topological data analysis is used to perform algebraic tracing of topological features: First, all simplexes in the nested space sequence are extracted and arranged into a simplex sequence according to their corresponding scale parameters from smallest to largest. Based on this, an initial boundary matrix is ​​constructed. The row and column indices of the initial boundary matrix uniquely correspond to each simplex in the simplex sequence, and the element values ​​in the boundary matrix are determined by the boundary inclusion relationship between simplexes (if the row simplex is the codimensional surface of the column high-dimensional simplex, i.e., the boundary constitutes the surface, then the corresponding matrix element is assigned 1; otherwise, it is assigned 0). The initial boundary matrix is ​​used as the input to the matrix reduction algorithm. The matrix reduction algorithm is based on the modulo-2 algebra operation criterion of the binary field. It performs column elimination operation on the initial boundary matrix from left to right. By adding the columns modulo-2, it iteratively eliminates the non-zero elements at the bottom of the current column until the rows containing the bottom non-zero elements (i.e., pivot elements) of each column in the simplified boundary matrix are all different. The algorithm finally outputs the set of row and column index pairs containing these pivot elements, i.e., pivot pairings. The row index of the pivot pairings corresponds to the simplex that triggers the generation of the new topological feature, and the column index corresponds to the simplex that closes and terminates the extinction of the topological feature. The system extracts the pairing results representing independent topologically connected components (i.e., 0-dimensional topological homology groups) from the principal component pairing relationships, and records the scale parameter corresponding to when the simplex (mainly 0-dimensional vertices) generated in the pairing results is incorporated into the spatial sequence as the birth time of the component, and records the scale parameter when the extinction simplex (mainly 1-dimensional topological edges) terminating the connected component is incorporated as the death time of the component; the complete duration of the topologically connected component from birth to death is recorded as the evolutionary period of the component, and the scale parameter difference corresponding to the evolutionary period (i.e., death time minus birth time) is defined as the life cycle of the topologically connected component; Subsequently, from all evolved topologically connected components, independent topologically connected components whose lifetimes satisfy a preset multi-scale stability condition are selected as highly stable persistent homology features. Specifically, the stability condition is that the lifetime is greater than a preset stability threshold (e.g., a preset value of 100%). In the theory of persistent cohomology, each highly stable persistent cohomological feature uniquely corresponds to a topologically stable generator, which is mathematically an algebraic set composed of multiple interconnected vertices. The system extracts all vertices contained in each highly stable persistent cohomological feature and performs a search and matching in the power distribution network GIS ledger database based on the device information ID of the vertex to obtain the real underlying physical nodes corresponding to these vertices. Based on the algebraic attribution relationship between vertices and topologically stable generators at the mathematical space level, the system divides the real underlying physical nodes into the same cluster in the physical space, that is, forms a subset of connected nodes after removing isolated noise points.

[0041] S404. Based on the spatial distribution characteristics of network nodes corresponding to the substation capacity gap, the node subset is weighted and aggregated to obtain the site coordinates, and a planning scheme is generated in combination with the substation capacity gap, as follows: Based on the typical power supply capacity of substations in actual engineering projects, a preset spatial power supply radius (e.g., 3 to 5 kilometers) is introduced to divide the planning subspace. Specifically, firstly, based on the latitude and longitude coordinates of each underlying physical node, for any two different connected node subsets, the actual geographic spatial distance between each node in one subset and each node in the other subset is calculated, and the minimum value among all cross-subset distance calculation results is extracted as the shortest spatial distance to measure the distance between two connected node subsets. Next, it is determined whether the shortest spatial distance is less than the spatial power supply radius. If it is less, a set union operation is performed on these closely spaced connected node subsets at the mathematical level to achieve spatial boundary fusion, thereby dividing the entire distribution network into several independent planning subspaces, and each planning subspace can contain one or more node subsets originally belonging to different grids. In the actual computer hardware and software implementation process, the above geographic distance calculation and spatial boundary fusion operation can be performed through a spatial database (such as PostGIS) deployed on a server or P The system uses ython spatial analysis components (such as the GeoPandas library) to automate the process. Specifically, the system first instantiates the latitude and longitude coordinates of the underlying physical nodes contained in each connected subset into standard multipoint geometric objects (such as the MultiPoint spatial data type) in the spatial database. In calculating the shortest spatial distance, the system calls a standard spatial distance calculation interface (such as the ST_Distance function in PostGIS), using two different multipoint geometric objects as input parameters. The underlying algorithm automatically performs cross-traversal distance measurement between nodes within the two point sets and nests a minimum value function (such as the MIN aggregation query statement) to directly output the minimum extreme value in the distance between node pairs across subsets. When the minimum extreme value is determined to be less than the spatial power supply radius, the system immediately calls a spatial merging interface (such as the ST_Union function) or an array append operation to merge the originally independent multipoint geometric objects into a unified new multipoint geometric object at the data structure level, thereby achieving the fusion of algebraic and physical spatial boundaries. Subsequently, based on the ratio of the historical load peak value of each bottom-level physical node within the grid to the sum of the historical load peak values ​​of all nodes within the grid, the substation capacity gaps of each load gap grid extracted in step S3 are proportionally refined and allocated to specific bottom-level physical nodes. This characterizes the spatial distribution characteristics of the substation capacity gaps on micro-network nodes (i.e., bottom-level physical nodes). Within each divided planning subspace, using the node-level substation capacity gaps allocated to the bottom-level physical nodes as weights, a joint weighted aggregation is performed on all node subsets contained in the subspace to obtain unique station location latitude and longitude coordinates. The weighted calculation formula for the station location coordinates is as follows: (15) (16) in, and The first The final latitude and longitude coordinates of the substation site generated within each planning subspace; For the first The set of all connected subsets of nodes within a planning subspace; To assign to the underlying physical nodes The node-level capacity gap; and These are the underlying physical nodes. The actual physical latitude and longitude coordinates; After obtaining the latitude and longitude coordinates of the substation sites in each planning subspace, the system summarizes the total substation capacity gap of all underlying physical nodes in the corresponding planning subspace, multiplies it by a preset safety margin (ranging from 1.2 to 1.5), and rounds it up to match the standard specification series of State Grid equipment (such as 31.5MVA or 50MVA) to obtain the main transformer configuration capacity of the newly built substations in the planning subspace. Based on the main transformer configuration capacity and the latitude and longitude coordinates of the substation sites determined in each planning subspace, a distribution network planning scheme is generated. The planning scheme includes at least: the latitude and longitude coordinates of each newly built substation, the corresponding main transformer configuration capacity, and the logical power supply topology affiliation relationship established between the newly built substation and the subset of connected nodes it covers.

[0042] This step, based on a composite metric space and a continuous coherence algorithm, achieves a unified metric for spatial geometry and physical power flow constraints. Furthermore, by combining power supply radius constraints and a gap-weighted aggregation mechanism, it outputs site locations and capacities, effectively reducing the electrical isolation risks caused by pure geometric site selection, ensuring the engineering feasibility of the planning scheme and the overall grid support capability, and improving the power supply resilience of the distribution network.

[0043] See Figure 3 An electronic device includes: a memory for storing a computer program; and a processor for implementing the steps of the method when executing the computer program.

[0044] The above formulas are all dimensionless calculations. The formulas are derived from software simulations using a large amount of collected data, and are the closest to the real situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0045] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0046] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0047] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0048] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0049] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A distribution network intelligent planning method based on multi-scenario net load, characterized in that, Includes the following steps: Based on the hierarchical topology of the distribution network, the node baseline power calculated from historical operating data is spatially coupled, and then combined with the pre-constructed multi-dimensional adjustment boundary to generate a multi-scenario net load demand sequence. The power supply redundancy grid and the load gap grid are parameterized as a first probability measure and a second probability measure, respectively. The optimal joint distribution matrix between the first probability measure and the second probability measure is solved to obtain the optimal transmission cost. Based on the pre-established capacity reduction relationship, the optimal transmission cost is converted into grid mutual assistance capacity, and then the demand sequence is subjected to equivalent capacity replacement to obtain the substation capacity gap; A composite metric space integrating spatial and electrical impedance characteristics is constructed. After extracting a subset of nodes with continuous coherence characteristics and aggregating them into site coordinates, a planning scheme is generated in conjunction with the substation capacity gap.

2. The method according to claim 1, characterized in that, The multidimensional adjustment boundary includes at least a safety adjustment boundary and an operational adjustment boundary; the safety adjustment boundary includes: The equivalent thermal parameters of the temperature-controlled load are fitted based on meteorological data and historical electricity consumption data, and a generalized energy function characterizing the heat exchange process is constructed. Under the preset thermodynamic state evolution constraints, the generalized energy function is solved to obtain the steady-state response range of the adjustable power, which serves as the safe adjustment boundary.

3. The method according to claim 2, characterized in that, The operational adjustment boundary includes: Using logical variables representing the line opening and closing state as decision variables, power transmission constraints that take into account equipment capacity margin and topological connectivity are constructed. Under the power transmission constraint, topology reconfiguration optimization is performed with the goal of minimizing the highest node load, and the power flow transfer before and after topology reconfiguration is used as the operation adjustment boundary.

4. The method according to claim 1, characterized in that, The generation of the multi-scenario net load demand sequence includes: The node reference power is decomposed and phase aligned at multiple scales. Based on the pre-established power conservation and topological correlation constraints and the distribution network hierarchical topology, the aligned node reference power is nonlinearly coupled to the substation level to obtain the substation level power. Under the preset differentiated scenarios, the substation hierarchical power is corrected by the multi-dimensional adjustment boundary to obtain the multi-scenario net load demand sequence.

5. The method according to claim 1, characterized in that, The parameterization of the power supply redundancy grid and the load gap grid into a first probability measure and a second probability measure includes: Based on a preset ratio, the multi-scenario net load demand sequence is allocated to each power supply grid to obtain the grid net load demand base. The demand baseline is compared with the total power output in each power grid, and the power supply redundancy grid and load gap grid are divided based on the comparison results; Calculate the proportion of excess power in each power supply redundancy grid to the total excess power of the entire network, and combine them to obtain the first probability measure; calculate the proportion of power deficit in each load deficit grid to the total power deficit of the entire network, and combine them to obtain the second probability measure.

6. The method according to claim 1, characterized in that, The process of solving for the optimal joint distribution matrix between the first probability measure and the second probability measure to obtain the optimal transmission cost includes: Based on the spatial distance and electrical impedance characteristics between grids, a transmission cost matrix is ​​constructed; Using the first probability measure and the second probability measure as marginal distribution constraints, and combining the cost matrix, entropy regularization optimal transmission calculation is performed to obtain the optimal joint distribution matrix; Calculate the global expectation of the distribution matrix and the cost matrix, which is taken as the optimal transmission cost.

7. The method according to claim 1, characterized in that, The construction of the composite metric space integrating spatial and electrical impedance characteristics includes: Based on the spatial location and equivalent electrical impedance of network nodes, the electrical potential difference between nodes is calculated and the power transmission direction between nodes is identified by combining the power flow distribution. By applying nonlinear weighting to the electrical potential difference and the power transmission direction, a composite metric space is obtained.

8. The method according to claim 1, characterized in that, The extraction and aggregation of a subset of nodes with continuous coherence characteristics into station coordinates includes: Within the composite metric space, multi-scale topological connectivity analysis is performed on distribution network nodes to identify a subset of connected nodes with continuous homology characteristics. Based on the spatial distribution characteristics of the network nodes corresponding to the substation capacity gap, the node subset is weighted and aggregated to obtain the station location coordinates.

9. The method according to claim 8, characterized in that, The identification of a subset of connected nodes possessing continuous homology features includes: Within the composite metric space, a nested spatial sequence reflecting the node topological clustering process is constructed; Identify the evolution period of each topologically connected component in the nested spatial sequence, and screen out continuous homology features that satisfy preset multi-scale stability conditions; The continuous homology features are mapped to the physical topology of the distribution network and combined to obtain a subset of connected nodes.

10. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the steps of the method as claimed in any one of claims 1 to 9.

Images