A power distribution network inverter voltage optimization control method, system, device and medium

Abstract

The present application relates to the technical field of distribution network voltage control, and discloses a distribution network inverter voltage optimization control method, system, device and medium, a layered Koopman observation function containing a node layer, a line layer and a global layer is constructed based on the topology structure data of the distribution network, and when a topology displacement signal is monitored, local decoupling updating of the Koopman operator matrix is realized, so as to convert the voltage control barrier function corresponding to the original nonlinear voltage constraint into a linear safety constraint in the Koopman linear characteristic space, and the projection coefficient vector is analytically solved to determine the voltage feasible region; then, a quadratic programming model is constructed and solved with the minimum sum of squares of inverter power regulation as the target, and the inverter power regulation instruction satisfying the safety constraint is obtained, without relying on an accurate physical model, realizing voltage safety control of the medium and low voltage distribution network under high proportion of distributed photovoltaic access.

Description

Technical Field

[0001] This invention relates to the field of distribution network voltage control technology, and in particular to a method, system, equipment and medium for optimizing voltage control of distribution network inverters. Background Technology

[0002] When distributed photovoltaic power sources are connected to medium and low voltage distribution networks through inverters, the randomness of their output will cause node voltage fluctuations. These voltage fluctuations not only affect the power quality on the user side, but also trigger the inverter's low voltage ride-through protection, causing large-scale grid disconnection accidents.

[0003] Currently, voltage regulation methods for distributed photovoltaic (PV) grid connection are mainly divided into three categories: Q(U) strategy based on reactive power control, Q(P) strategy based on active power correlation, and APC (Active Power Curtailment) strategy based on active power reduction. While existing methods can achieve localized voltage regulation in specific scenarios, they have significant limitations in multi-point distributed PV grid connection scenarios. For example, with Q(U) control: the voltage level difference between the beginning and end of the line leads to uneven distribution of regulation capacity. The voltage at the beginning node near the transformer is usually within limits, and the inverter does not participate in voltage regulation; however, the voltage at the end node is prone to exceeding limits and must bear the main reactive power regulation burden, resulting in resource waste of "end inverter overload and beginning capacity idle," and single-point regulation may trigger a chain reaction of voltage fluctuations. With Q(P) control, the logic is disconnected from the voltage state. When PV active power output is high but load demand is simultaneously at its peak, the inverter still absorbs reactive power according to a fixed curve, increasing reactive power flow and grid losses, and reducing system operating efficiency. The APC strategy sacrifices active power output, resulting in high curtailment rates (especially during periods of abundant sunshine), which contradicts the goal of maximizing photovoltaic power consumption. Furthermore, a single APC control cannot coordinate reactive power regulation capabilities, potentially leading to excessive curtailment in high-penetration scenarios and significant economic losses.

[0004] It is evident that existing methods are insufficient to meet the voltage control requirements of large-scale distributed photovoltaic grid connection. Summary of the Invention

[0005] This invention provides a voltage optimization control method, system, device, and medium for distribution network inverters, solving the problem that existing voltage optimization control methods for distribution network inverters cannot meet the voltage control requirements of large-scale distributed photovoltaic grid connection.

[0006] To address the aforementioned technical problems, the first aspect of this invention provides a voltage optimization control method for a distribution network inverter, comprising: Collect real-time measurement data and topology data of the power distribution network; A hierarchical Koopman observation function is constructed based on the aforementioned topology data; the hierarchical Koopman observation function includes node-level observation sub-functions, line-level observation sub-functions, and global-level observation sub-functions; The Koopman operator matrix is ​​estimated based on the real-time measurement data and the hierarchical Koopman observation function, and the Koopman operator matrix is ​​locally decoupled and updated when a topological displacement signal is detected to obtain the current Koopman operator. Based on the current Koopman operator, the pre-constructed voltage control barrier function is transformed into a linear safety constraint in the Koopman feature space; An observation matrix is ​​constructed using the real-time measurement data, and the projection coefficient vector is obtained by solving the least squares method. The projection coefficient vector is then used to transform the upper and lower voltage limits of the distribution network into the voltage feasible region in the Koopman feature space. With the goal of minimizing the sum of squares of inverter power regulation, and with the linear safety constraints and the voltage feasible region as constraints, a quadratic programming model is constructed and solved to obtain the power regulation commands for each inverter for execution.

[0007] A second aspect of the present invention provides a voltage optimization control system for a power distribution network inverter, comprising: The data acquisition module is used to collect real-time measurement data and topology data of the power distribution network; The function construction module is used to construct hierarchical Koopman observation functions based on the topology data; the hierarchical Koopman observation functions include node-level observation sub-functions, line-level observation sub-functions, and global-level observation sub-functions; The operator update module is used to estimate the Koopman operator matrix based on the real-time measurement data and the hierarchical Koopman observation function, and to perform local decoupling update of the Koopman operator matrix when a topological displacement signal is detected, so as to obtain the current Koopman operator. The constraint transformation module is used to transform the pre-built voltage control barrier function into a linear safety constraint in the Koopman feature space based on the current Koopman operator. The feasible region generation module is used to construct an observation matrix through the real-time measurement data, obtain the projection coefficient vector by the least squares method, and use the projection coefficient vector to transform the voltage upper and lower limit constraints of the distribution network into the voltage feasible region in the Koopman feature space. The voltage control module is used to construct and solve a quadratic programming model with the goal of minimizing the sum of squares of inverter power regulation, using the linear safety constraints and the voltage feasible region as constraints, to obtain power regulation commands for each inverter for execution.

[0008] A third aspect of the present invention provides an electronic device including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements the voltage optimization control method for a power distribution network inverter as described above.

[0009] A fourth aspect of the present invention provides a computer-readable storage medium comprising a stored computer program, wherein when the device containing the computer-readable storage medium executes the computer program, it implements the voltage optimization control method for power distribution network inverters as described above.

[0010] Compared with the prior art, the beneficial effects of the embodiments of the present invention are as follows: By introducing Koopman operator theory and constructing hierarchical observation functions, the nonlinear dynamic process is elevated to a linear characteristic space, transforming the complex voltage control problem into a mature linear system control problem. This approach retains the nonlinear dynamic characteristics of the original system while avoiding the sharp decline in accuracy of traditional linearization methods when far from the steady-state point. Through a local decoupling update strategy, the Koopman operator matrix is ​​updated only in local sub-regions associated with topology displacement nodes, eliminating the need to recalculate the global operator. This significantly reduces the computation time for online identification, meeting the real-time requirements of second-level or even millisecond-level control in distribution networks. The combination of the voltage control barrier function and the Koopman characteristic space transforms the originally non-convex and nonlinear voltage safety boundary into a linear inequality constraint, ensuring the uniqueness of the global optimal solution and providing a fast solution speed. It does not rely on an accurate physical model and combines topology adaptation capability, strict voltage safety guarantees, and millisecond-level real-time computation efficiency, making it suitable for voltage safety control in medium- and low-voltage distribution networks with a high proportion of distributed photovoltaic access. Attached Figure Description

[0011] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a flowchart of a voltage optimization control method for a power distribution network inverter provided in a certain embodiment of the present invention; Figure 2 This is a diagram of a 5-node radial distribution network topology provided in one embodiment of the present invention; Figure 3 This is a feature space dimension comparison diagram provided in a certain embodiment of the present invention; Figure 4This is a comparison chart of voltage prediction accuracy provided by a certain embodiment of the present invention; Figure 5 This is a comparison chart of operator training time provided in a certain embodiment of the present invention; Figure 6 This is a comparison chart of voltage recovery curves after topology disconnection provided in a certain embodiment of the present invention; Figure 7 This is a comparison chart of voltage recovery curves under extreme overvoltage scenarios provided in a certain embodiment of the present invention; Figure 8 This is a node 5 voltage diagram in a low-risk scenario provided by a certain embodiment of the present invention; Figure 9 This is a node 5 voltage diagram under a high-risk scenario provided in a certain embodiment of the present invention; Figure 10 This is a comparison chart of the computational complexity of topology decoupling update provided in a certain embodiment of the present invention; Figure 11 This is a structural diagram of a voltage optimization control system for a power distribution network inverter provided in a certain embodiment of the present invention; Figure 12 This is a structural diagram of an electronic device provided in a certain embodiment of the present invention; Figure label: Among them, 10 is the data acquisition module; 20 is the function construction module; 30 is the operator update module; 40 is the constraint transformation module; 50 is the feasible region generation module; 60 is the voltage control module; 5000 is the electronic equipment; 5001 is the processor; 5002 is the bus; 5003 is the memory; and 5004 is the transceiver. Detailed Implementation

[0013] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings and examples. 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 skilled in the art without creative effort are within the scope of protection of the present invention.

[0014] It should be understood that the step numbers used in the text are for ease of description only and are not intended to limit the order in which the steps are performed.

[0015] It should be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0016] The terms “comprising” and “including” indicate the presence of the described feature, whole, step, operation, element and / or component, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or collections thereof.

[0017] The term “and / or” refers to any combination of one or more of the associated listed items, as well as all possible combinations, and includes these combinations.

[0018] In one embodiment, such as Figure 1 As shown, the first aspect of the present invention provides a voltage optimization control method for a distribution network inverter, comprising: S1. Collect real-time measurement data and topology data of the power distribution network; wherein, the real-time measurement data includes the voltage amplitude of each node and the power data of each inverter; Specifically, this invention first defines a state-space representation of the voltage status of the distribution network, which is the set of voltage amplitudes of all nodes and power outputs of all distributed photovoltaic inverters in the distribution network, i.e.: In the formula, For system status, ; For the first The voltage amplitude at each node, j=1 , This represents the total number of nodes in the distribution network. , For the first The active and reactive power outputs of the inverter, i=1 , This represents the total number of inverters in the distribution network.

[0019] Define the control action vector as the power regulation amount of each inverter, that is: In the formula, As the regulatory vector, ; , For the first The active and reactive power adjustment parameters of the inverter; among which... For inverters The difference between the maximum power point tracking power and the set value (used for active power voltage regulation). For inverters The difference between the target reactive power and the measured reactive power.

[0020] Define the observed measurement vector as the voltage amplitude at each node, that is: In the formula, For the measurement vector, .

[0021] The nonlinear discrete-time state evolution of the distribution network is described by an unknown nonlinear vector field function: In the formula, k is the discrete sampling time, and the sampling period is preferably 1 second; G is a nonlinear vector field function. Since the power flow equation of the distribution network is inherently nonlinear, G has no explicit expression. Subsequently, the Koopman operator theory is used to elevate it to a linear characteristic space to indirectly characterize it. The system state at time k includes the voltage magnitude of all nodes and the power output of all inverters. Let be the control vector at time k. ,in, and These represent the active power regulation and reactive power regulation of the Nth photovoltaic inverter, respectively. This control vector is used to describe... The controllable photovoltaic inverter performs coordinated adjustment of active and reactive power within the same control cycle.

[0022] Subsequently, at 50-millisecond intervals, intelligent terminal units (FTUs / DTUs) installed at each node and line collect real-time voltage measurements (voltage amplitude) at each node, power output measurements (including active and reactive power data) of each inverter, and the open / closed status of each circuit breaker / disconnector within the distribution network. After bad data detection (using the 3σ criterion), filtering preprocessing, and time-aligned interpolation (using linear interpolation), the measurement data forms real-time measurement data for subsequent online updates of the Koopman operator. Simultaneously, node locations, branch numbers, and connection relationships within the distribution network are also collected as topology data to facilitate the construction of hierarchical Koopman observation functions.

[0023] S2. Construct a hierarchical Koopman observation function based on the topology data; the hierarchical Koopman observation function includes node-level observation sub-functions, line-level observation sub-functions, and global-level observation sub-functions; In one embodiment, constructing a hierarchical Koopman observation function based on the topology data includes: Based on the topology data, construct the node-branch correlation matrix and the branch impedance diagonal matrix to determine the node impedance matrix; For each node, the reactance-resistance ratio of the node is determined based on the node impedance matrix, and the node-level observation subfunction is constructed by combining the voltage amplitude of the node and the power data of the corresponding inverter to characterize the driving effect of the power injection of each inverter on the local node voltage. Based on the node impedance matrix, the impedance characteristic coefficients of each branch in the distribution network are determined, and combined with the voltage amplitude and power data of the nodes contained in each branch, a line-level observation sub-function is constructed to characterize the neighborhood coupling relationship between line power flow transmission and voltage drop. The nodes are partitioned according to the node impedance matrix to obtain several electrical partitions. The node weights are then determined. The voltage amplitudes of the nodes are weighted and averaged using the node weights to obtain the partition weighted average voltage. This is used to construct a global layer observation subfunction to characterize the degree to which the average voltage of the electrical partition deviates from the rated value. The node-level observation sub-function, the line-level observation sub-function, and the global-level observation sub-function are concatenated to form the hierarchical Koopman observation function.

[0024] Specifically, the core innovation of this invention, which differs from the conventional Koopman method's general RBF plus polynomial observation function, lies in: utilizing the physical prior of the radial topology of the distribution network, designing a three-layer structured observation function set, so that the Koopman feature space is naturally embedded into the hierarchical energy transfer structure of the distribution network, thereby improving the physical interpretability and topological robustness of operator learning.

[0025] A weighted graph is determined based on the topology data of the distribution network, which includes a set of nodes and a set of branches; a node-branch correlation matrix is ​​constructed based on the weighted graph. : , The number of branches in the distribution network is represented by the elements in this matrix. The rule is: if the branch road With nodes Starting from, If branch road With nodes As the endpoint, Otherwise Based on the actual line parameters, the branch impedance is obtained. : ( , (The resistance and reactance of branch e) are then used to form the branch impedance diagonal matrix. : ,in, branch road impedance, The nodal admittance matrix is ​​obtained from these two matrices, and then its inverse is used to obtain the nodal impedance matrix: In the formula, Here is the node impedance matrix. ; This is the Moore-Penrose pseudoinverse. In actual computation, the matrix construction process is completed offline by the edge terminal and stored.

[0026] In distribution networks, local voltage variations at nodes are primarily determined by power injection at those nodes, and the impact of reactive power injection on voltage is directly related to the line reactance-resistance ratio. For radial feeders, the voltage-power sensitivity of the terminal nodes is approximately... , where X j Q is the cumulative reactance from node j to the root node. j Inject reactive power into node j of the inverter.

[0027] For node j, its cumulative impedance is calculated based on the node impedance matrix: In the formula, Let be the cumulative impedance of node j; From the root node to the node The only path; The impedance of branch e to which node j belongs is determined based on the node impedance matrix.

[0028] Then the reactance of node j is compared to for: In the formula, , Let be the imaginary and real parts of the cumulative impedance at node j.

[0029] Using the reactance-resistance ratio of node j, its voltage amplitude, and the power data of the corresponding inverter, a node-level observation subfunction is constructed: In the formula, For the node-level observation sub-function corresponding to node j, ; The reference voltage is preferred. pu;P j Q j Inject active and reactive power into node j of the inverter; As the baseline capacity, preferably MVA. Both terms on the right-hand side of this function are dimensionless, with the physical meanings being: the first term reflects the square of the voltage amplitude deviation from the reference, and the second term reflects the coupling strength between inverter capacity utilization and line reactance characteristics. This observation function captures the direct effect of inverter power injection on local node voltage, corresponding to the node autonomy layer of the distribution network. That is, each node corresponds to a local observation function.

[0030] In a radial distribution network, branch power flow acts as a bridge connecting the voltages of adjacent nodes. The relationship between the voltage difference between adjacent nodes and the reactive power flow of branches is determined by the following formula: In the formula, The voltage difference between adjacent nodes j and k; , Let e ​​be the resistance and reactance of the branch e to which nodes j and k belong; , Let e ​​be the active power flow and reactive power flow of the branch e to which nodes j and k belong.

[0031] Therefore, the weighted sum of the square of the voltage difference and the apparent power of the branches should be able to capture the neighborhood coupling dynamics. For any branch e=(j,k), its branch power flow vector : ,in and The power injected at the nodes is obtained recursively from the branch power flow equations.

[0032] To ensure that branch resistance and reactance participate in the construction of the observation function in a normalized form, they are normalized using a preset impedance reference value. This process can be expressed as: In the formula, Indicates the impedance reference value; and These represent the normalized branch resistance and branch reactance, respectively; Let J be the resistance of the branch to which nodes J and K belong; Let J be the reactance of the branch to which nodes J and K belong.

[0033] Based on this, the branch coupling coefficient can be expressed as: In the formula, Let be the branch coupling coefficient of branch e=(j,k), which is dimensionless; A very small positive number set to prevent the denominator from being zero.

[0034] Based on the calculated impedance characteristic coefficients and the voltage amplitude and power data of the nodes in the branch, a line-level observation sub-function is constructed: In the formula, For the line-level observation subfunction corresponding to branch e=(j,k), ; Let be the voltage magnitude at node k; The reference voltage difference is preferred. pu; This represents apparent power. The physical meaning of the above function is as follows: the first term captures the magnitude of the voltage drop, reflecting the line transmission capacity; the second term captures the coupling between branch load rate and impedance characteristics, reflecting the driving effect of power flow on voltage distribution. This layer of observation function captures the neighborhood coupling relationship between line power flow transmission and voltage drop, corresponding to the line coordination layer of the distribution network. (Radial network), that is, each branch has a differential observation function.

[0035] The global voltage distribution of a distribution network is determined by a few dominant patterns, which correspond to the directions of large eigenvalues ​​in the node impedance matrix. This invention extracts electrical centers through k-means clustering, enabling the construction of a low-dimensional representation of the entire network voltage. In one embodiment, partitioning the nodes according to the node impedance matrix to obtain several electrical partitions and determining node weights includes: Extract the real part of the impedance of each node in the node impedance matrix, and determine the electrical distance between nodes based on the real part of the impedance. The nodes are clustered using k-means clustering based on the electrical distance between them to obtain several electrical partitions. The node weight is determined based on the real part of the impedance corresponding to the geometric center node within each electrical partition.

[0036] Specifically, first extract the real part of the impedance from the nodal impedance matrix. : Its elements Represents a node With nodes The equivalent resistance distance between nodes. The electrical distance between nodes is calculated using electrical distance. ;in, For nodes With nodes The equivalent resistance distance between them; For nodes With nodes The equivalent resistance distance between them.

[0037] Then, using the electrical distance between nodes as the distance metric, k-means clustering was performed on the node set in the weighted graph, resulting in the number of clusters. ,get Electrical zones Each center Contains a set of nodes For each node set, the node weight is calculated by selecting the real part of the impedance corresponding to the geometric center node of each partition. In the formula, Let be the node weight of node j within electrical partition m; Let be the real part of the impedance of the geometric center node within electrical zone m. For clustering The geometric center node.

[0038] This invention extracts the real part (resistance component) of the node impedance matrix to calculate electrical distance. Since resistance directly affects the voltage drop caused by active and reactive power transmission, this distance metric can accurately characterize the strength of the voltage interaction between two nodes, ensuring that the partitioning results strictly match the electrical and physical characteristics of the power grid. The k-means clustering algorithm is used to automatically group nodes, without relying on the subjective experience of operators or fixed boundaries. When the network topology changes (such as adding lines or changing the status of sectional switches), the partitioning can be quickly updated simply by re-calling the clustering process. Moreover, the clustering process converges rapidly, making it suitable for online deployment in engineering projects.

[0039] By using the calculated node weights to perform a weighted average of the voltage amplitudes, the partitioned weighted average voltage can be obtained: In the formula, For electrical center The partition-weighted average voltage.

[0040] Therefore, the global layer observation subfunction is: In the formula, For global layer observation subfunctions, ; This is the reference voltage. Its physical meaning is: each observation function reflects the degree to which the average voltage of an electrical area deviates from its rated value, corresponding to a dominant pattern in the global voltage distribution of the distribution network. When multiple areas simultaneously experience voltage spikes or dips, this observation function is activated, triggering global coordinated control. L3 is set to 5-10, much smaller than M (M≥30 in typical distribution networks).

[0041] By concatenating these three observation sub-functions in sequence, we obtain the hierarchical Koopman observation function: In the formula, For hierarchical Koopman observation functions, , of which the total dimension It should be noted that the above function construction process can also be completed offline.

[0042] Comparison of the rigor and dimensionality of physical prior embedding: The observation functions of the conventional Koopman method are general mathematical basis functions (RBF and polynomials), which are independent of the physical structure of the distribution network. The three-layer observation functions of this invention directly correspond to the physical levels of the distribution network—node injection, branch power flow, and network-wide balancing—giving each dimension of the Koopman feature space a clear physical meaning. This structural prior embedding brings two technical advantages: 1. The effective dimension required for operator learning is reduced from L=5M+2N in the conventional Koopman method to L=M+M-1+L3=2M-1+L3. In a typical 33-node system (M=33, N=32), the conventional Koopman method requires L=5×33+64=485 dimensions, while this invention only requires L=2×33-1+8=73 dimensions, a reduction of approximately 85%; 2. When the topology changes, only the observation functions corresponding to the disconnected branches in the second layer are affected, while the structure of the observation functions in the remaining layers remains unchanged, and the operator update changes from global to local.

[0043] This invention obtains the node impedance matrix by explicitly constructing the node-branch correlation matrix and the branch impedance diagonal matrix, and calculates the reactance-resistivity ratio for each node accordingly. This directly reflects the sensitivity of reactive power changes to local voltage amplitude. Based on this, the node-level observation sub-function can accurately characterize the driving effect of inverter power injection (especially reactive power) on node voltage, avoiding the blindness of "black box" observation in pure data-driven methods. The line-level observation sub-function utilizes the branch impedance characteristic coefficient, combined with the voltage and power data of the nodes at both ends of the branch, to accurately quantify the voltage drop and phase shift generated during the line power flow transmission process. This allows the Koopman operator to more faithfully reproduce the nonlinear coupling dynamics between adjacent nodes in high-dimensional space, thereby improving the accuracy of voltage prediction and control. The physical meanings of the node layer, line layer, and global layer are clear. When a topology change occurs, the subset of observation functions that need to be updated can be quickly located based on the branches or nodes involved in the change. Combined with a local decoupling update strategy, the online recalculation time is greatly reduced, meeting the millisecond-level real-time control requirements of the distribution network.

[0044] S3. Estimate the Koopman operator matrix based on the real-time measurement data and the hierarchical Koopman observation function, and perform local decoupling update on the Koopman operator matrix when a topological displacement signal is detected to obtain the current Koopman operator; In one embodiment, estimating the Koopman operator matrix based on the real-time measurement data and the hierarchical Koopman observation function includes: Based on the real-time measurement data and the hierarchical Koopman observation function, an observation matrix is ​​constructed; Based on the observation matrix, the initial Koopman operator is estimated using the extended dynamic mode decomposition method and the least squares method. Based on the hierarchical structure of the hierarchical Koopman observation function, the initial Koopman operator is divided into blocks to form the Koopman operator matrix.

[0045] Specifically, this invention utilizes the structural characteristics of hierarchical observation functions to achieve local decoupling updates of operators during topological changes, replacing the global batch retraining of the conventional Koopman method.

[0046] For real-time arrival Each sample (i.e., the result obtained by sliding window processing of real-time measurement data) is combined with the hierarchical Koopman observation function to calculate the three-layer observation function values, forming the observation matrix at the current time. and the observation matrix at the next time step Both of these observation matrices belong to .

[0047] Based on the constructed observation matrix, the initial Koopman operator is estimated from W samples using batch EDMD (Extended Dynamic Mode Decomposition) and least squares, as shown in the following equation: In the formula, K is the initial Koopman operator. .

[0048] Based on the hierarchical structure of the hierarchical Koopman observation function, the initial Koopman operator is divided into three blocks according to the three-layer structure, resulting in the Koopman operator matrix: In the formula, For the internal evolution of the local autonomous layer; The impact of neighborhood coordination on local autonomy; The impact of global equilibrium on local autonomy; The impact of local autonomy on neighborhood coordination; For the internal evolution of the neighborhood coordination layer; The impact of global equilibrium on neighborhood coordination; The impact of local autonomy on global equilibrium; The impact of neighborhood coordination on global equilibrium; This is the internal evolution of the global equilibrium layer.

[0049] This invention utilizes Extended Dynamic Mode Decomposition (EDMD) combined with least squares estimation to directly extract the optimal linear evolution relationship in high-dimensional space from measurement snapshots without requiring explicit differential equations of the system. The resulting operator is further decomposed into sub-matrix structures adapted to the node layer, line layer, and global layer, allowing independent tracing of the identification error at each layer's dynamics. The overall estimation accuracy is superior to a single black-box global fitting. After cutting the complete Koopman operator into several sub-matrixes according to the hierarchical observation function, when topological changes or local parameter changes occur in the distribution network, only a few affected sub-matrix blocks (such as the cross sub-blocks related to the line layer) need to be recalculated, without recalculating the entire high-dimensional matrix.

[0050] In one embodiment, the step of locally decoupling and updating the Koopman operator matrix to obtain the current Koopman operator includes: The target level is determined based on the topological displacement signal, and a Gram rearrangement matrix is ​​constructed based on the target level and the observation matrix. The Schur complement matrix and Schur complement correction are calculated using the Gram rearrangement matrix, and the inverse matrix is ​​determined by the Woodbury matrix identity. The target level of the Koopman operator matrix is ​​updated based on the inverse matrix to obtain the current Koopman operator.

[0051] Specifically, when a circuit breaker sends a topology change signal (such as a line being disconnected due to maintenance), the full-dimensional matrix is ​​not recalculated; that is, when the branch... When disconnected, the first layer Unaffected because of node injected power With voltage The definition does not depend on specific branch connections; Layer 2 Only affected branch roads Corresponding observation function Failure (the branch no longer exists, power flow) (Undefined), the branch adjacent to this branch (shared node) or The observation function of the third layer may change due to power flow redistribution, but the function form remains unchanged; If the cluster center does not involve the end nodes that break off branches, it is unaffected; if it does, the weighting coefficient of the cluster will be adjusted. Recalculation is required, but the functional form remains unchanged. In other words, topology changes only directly affect one or a few observation functions in the second layer, while the first and third layers retain their structural integrity. This property is unique to radial topologies—any branch breakage only affects the local neighborhood and does not disrupt the definition of network node injection and global patterns.

[0052] In summary, the local decoupling update during topology changes is as follows: when a branch is disconnected, only the observation function corresponding to the disconnected branch in the second-layer observation function becomes invalid. The observation functions of the branches adjacent to this branch remain unchanged in form but are recalculated numerically. The first and third layers maintain their structural integrity. Using Schur complement decomposition, only the Schur complement and its inverse matrix of the affected second-layer submatrix are recalculated. The inverse matrix of the outer layer Gram matrix does not need to be recalculated.

[0053] Define time Observation function value matrix and the next time step matrix Define the Gram matrix. And divide it into three layers: In the formula, p and q can both be 1, 2, or 3. When the branch... When the connection is broken, only the second layer is affected; therefore, the second layer is the target layer. The submatrix of the affected second layer is defined as follows: ,in A low-rank correction matrix (rank is ) (i.e., the number of nodes in the affected neighborhood), the rest of the submatrix remains unchanged: (Only involving) Adjacent nodes) In order to seek The matrix is ​​rearranged into blocks centered on the second layer, which means constructing a Gram rearranged matrix based on the target layer and the observation matrix: In the formula, This is a Gram rearrangement matrix; The outermost Gram matrix is ​​unaffected; This is the coupling matrix.

[0054] Therefore, the Schur complement matrix can be calculated using the Schur complement formula: In the formula, To complete the Schur matrix, the dimension is... .because If the topology remains unchanged before and after the change, then No need to recalculate, only the Schur complement needs to be recalculated. and its reverse Therefore, the Schur correction is: In the formula, Add a correction amount to Schur; A low-rank matrix (rank) ).

[0055] Using the Woodbury matrix identity: In the formula, It is a low-rank decomposition; It is obtained by performing truncated singular value decomposition on ΔS and absorbing the square root of the singular value: before extracting ΔS The singular values ​​and their corresponding left and right singular vectors are obtained by multiplying the square roots of the singular values ​​into the left and right singular vector matrices, respectively.

[0056] The inverse of Schur's complement is calculated using the above formula. Substituting this inverse matrix into the block matrix inversion formula yields the updated inverse. This updates the Koopman operator K (by recalculating only the operator blocks directly corresponding to the affected input and output dimensions) to obtain the current Koopman operator.

[0057] The conventional Koopman method requires calculating the Gram matrix for global retraining. (complexity) )and (complexity) The total complexity is .exist It takes about time For sub-floating-point operations, the local decoupling update in this invention only requires calculating the Schur correction. (complexity) ) and low-rank correction (complexity) The total complexity is .exist It takes about time This involves a floating-point operation. The complexity reduction is approximately [percentage missing]. That is, a reduction of about four orders of magnitude.

[0058] Adaptive layer activation: During normal operation, only the observation functions of the first and third layers are used ( (Dimension) performs operator estimation and control calculations, corresponding to a local autonomous plus global equilibrium mode; when a node voltage is detected to be close to the over-limit threshold (e.g. When a node exceeds its limit, the second-layer neighborhood observation function associated with that node is automatically activated, entering neighborhood coordination mode; when multiple nodes exceed their limits simultaneously, the second-layer observation function for the entire network is activated. This adaptive layer activation allows the operator dimension to dynamically change with the risk level, minimizing computational cost in low-risk situations and maximizing accuracy in high-risk situations.

[0059] This invention locates the affected level by rearranging the Gram matrix and transforms the global matrix inversion into an inversion of only the low-dimensional weighted variable subspace using Schur complement and Woodbury identity. The computation time is compressed from hundreds of milliseconds to within a few milliseconds, fully meeting the needs of rapid control in distribution networks. Based on the hierarchical observation structure, Gram rearrangement can accurately focus the observation index corresponding to the displacement branch / node on a specific sub-block. Schur complement calculation effectively eliminates the coupling interference of unchanged subsystems on the update quantity, so that the updated operator only changes the value on the row / column block of the target level, while the rest retains the prior identification accuracy. This prevents the cumulative error introduced by global recalculation and avoids unnecessary numerical jitter. The Woodbury matrix identity combined with the Schur complement form utilizes the known information of the original matrix inversion, avoiding direct inversion of potentially ill-conditioned Gram matrices. At the same time, the calculation of the correction quantity only involves small-scale matrix operations, and the condition number is controllable. Even in scenarios with measurement noise or data redundancy, the solution accuracy of the inverse matrix can still be guaranteed, which is significantly better than direct global recalculation.

[0060] S4. Based on the current Koopman operator, the pre-constructed voltage control barrier function is transformed into a linear safety constraint in the Koopman feature space; This invention constructs a voltage control barrier function in the Koopman feature space, transforming voltage over-limit constraints in the original space into linear forward invariant set constraints in the feature space. First, the voltage control barrier function is defined, which includes nodes... Voltage over-limit barrier function and lower limit barrier function : In the formula, For nodes The projection coefficient vector of voltage in the Koopman feature space is determined by the constraint embedding in step S5.

[0061] In one embodiment, the step of transforming the voltage control barrier function into a linear security constraint in the Koopman feature space based on the current Koopman operator includes: Based on the current Koopman operator, the voltage control barrier function is subjected to differential operation in the Koopman feature space, and a voltage safety set is constructed based on the differential operation result; The voltage safety set is processed by introducing a barrier function attenuation coefficient to form the linear safety constraint.

[0062] In the Koopman feature space, the distribution network dynamics are strictly linear: ,in To regulate the coupling block, based on the current Koopman operator, a difference operation is performed on the voltage control barrier function in the Koopman feature space to obtain the time derivative (difference form) of the barrier function: In the formula, It is an L-dimensional identity matrix.

[0063] Based on the above differential operation results, a voltage safety set is constructed. Based on the control barrier function theory, a barrier decay rate is introduced to process the voltage safety set, requiring the barrier function to satisfy the exponential decay condition, thus forming a control variable. Linear safety constraints: In the formula, Barrier function decay rate; safety set It is forward invariant, meaning that if the initial state is within the safe set, then the state at any subsequent time step will not exceed the safe set. This inequality is a linear constraint in the Koopman feature space because... and Given a matrix, Given a vector, the unknowns are only... This is fundamentally different from nonlinear CBF design in the original space—the original space requires calculating the nonlinear Lie derivative, while Koopman linearization degenerates the Lie derivative into matrix multiplication. Parameters This is the projection coefficient vector, with the dimension of voltage (V); Let be the barrier function decay rate, with dimensions . Typical value .

[0064] This invention utilizes the linear evolution characteristics of the Koopman feature space to expand the barrier function into a strict linear inequality with respect to the control quantity after difference operations. This ensures the convexity of the subsequent quadratic programming problem, guaranteeing the uniqueness of the solution and the efficient convergence of the solver. Based on the accurate linear prediction of the future observation by the current Koopman operator, the difference operation of the voltage control barrier function (CBF) actually measures the voltage safety trend at the next moment. Combined with the attenuation coefficient to process the safety set, the controller not only passively responds to the current over-limit but also actively prevents voltage instability at future moments, significantly improving the robustness of the system in dynamic processes. The introduction of the attenuation coefficient to tighten the safety set allows for a flexible balance between the aggressiveness of the control and the conservatism of safety.

[0065] S5. Construct an observation matrix using the real-time measurement data, obtain the projection coefficient vector by solving the least squares method, and use the projection coefficient vector to transform the voltage upper and lower limit constraints of the distribution network into the voltage feasible region in the Koopman feature space. This invention rigorously embeds the voltage compliance constraints in the original state space into the Koopman feature space, and analytically solves for the projection coefficients based on the observation matrix constructed from real-time measurement data. Node voltage It can be represented as a linear combination of the Koopman characteristic functions. Coefficient vector The following is obtained by least-squares projection of the characteristic function onto the voltage measurement data: In the formula, This is a voltage measurement sequence. Unlike the conventional Koopman method, this step involves solving... Subsequently, instead of using it only for power flow linearization, it is used to construct the barrier function in step S4.

[0066] In one embodiment, the step of transforming the voltage upper and lower limit constraints of the distribution network into the voltage feasible region in the Koopman feature space using the projection coefficient vector includes: Using the projection coefficient vector, the voltage upper and lower limit constraints of the distribution network are transformed into a set of linear inequalities in the Koopman feature space; Stack the projection coefficient vectors row by row to obtain the constraint matrix; Based on the constraint matrix, the system of linear inequalities is transformed into a linear voltage feasible region in the Koopman feature space.

[0067] Specifically, the calculated projection coefficient vector is used to constrain the upper and lower voltage limits of the distribution network. ( , The minimum and maximum values ​​of the node voltages are transformed into a system of linear inequalities in the Koopman characteristic space: Stack the projection coefficient vectors of all nodes row-wise to obtain the constraint matrix. Then the system of linear inequalities can be transformed into a linear voltage feasible region in the Koopman feature space, which is represented as a convex polyhedron: In the formula, This is the voltage feasible region; It is an M-dimensional column vector of all 1s. The strictness of this linear embedding is guaranteed by the completeness of the Koopman feature function as the intrinsic mode of the system.

[0068] This invention transforms the entire voltage feasible region into a set of closed linear inequalities in the Koopman feature space by projecting coefficient vectors, ultimately forming a convex polyhedral feasible region. This ensures that all subsequent quadratic programming (QP) problems are constrained linearly, guaranteeing the rapid and unique convergence of the global optimal solution. The construction of the constraint matrix is ​​only a matter of vector stacking and simple algebraic combination, without the need for iterative solution. When the projected coefficient vectors are recursively updated, the constraint matrix and feasible region can be updated synchronously and quickly. The computation time is usually in the microsecond range, which is perfectly compatible with the second-level control cycle.

[0069] S6. With the goal of minimizing the sum of squares of inverter power regulation, and with the linear safety constraints and the voltage feasible region as constraints, construct a quadratic programming model and solve it to obtain the power regulation commands of each inverter for execution. This invention utilizes the linear CBF constraint in step S4 to transform the control input solution into a quadratic programming problem with linear constraints, achieving strict preservation of the voltage safety set. The control objective is defined as minimizing the sum of squares of the inverter power regulation (economic objective): In the formula, It is a diagonal weight matrix, with dimensions being the reciprocals of the power dimension; typically, it is the identity matrix. If reactive power regulation is prioritized, the weight of active power regulation can be increased.

[0070] The constraint condition is the CBF safety constraint of step S4 (for all). The voltage feasible region and inverter capacity circle constraints (active and reactive power coupling constraints, N quadratic cone constraints to ensure that the adjusted power does not exceed the inverter's rated capacity): In the formula, , The measured active and reactive power outputs of the i-th inverter are the real-time measured values ​​of Pᵢ and Qᵢ in section S1. Let be the rated apparent capacity of the i-th inverter.

[0071] Since the dynamics in the Koopman feature space are linear, and the CBF constraints are linear inequality constraints, the distributed subgradient method of the conventional Koopman method cannot directly handle such coupled constraints. The aforementioned quadratic programming problem can be solved centrally (for small-scale systems) or quickly using the primal-dual interior-point method (for large-scale systems), with a solution complexity of O(n log n). This can be completed in milliseconds, solving for and outputting the optimal control vector, which is the active / reactive power adjustment command for each inverter to control the corresponding inverter to execute, thereby achieving optimized voltage control of the distribution network.

[0072] The method described in this invention is a data-driven + physical prior fusion distribution network voltage closed-loop control method. It achieves strict linearization by dynamically upscaling the nonlinear voltage-power dynamics to a feature space using a hierarchical Koopman observation function. A linear control barrier function (CBF) is then constructed in the feature space to ensure voltage safety. Finally, control commands are quickly solved using quadratic programming, while also supporting local low-rank updates after topology changes. The method is divided into two main stages: an offline design stage and an online operation stage. In the offline design stage, the hierarchical observation function structure, Koopman operator block matrix, and barrier function projection coefficients are all designed offline based on the distribution network's rated topology, line parameters, and inverter capacity. In the online operation stage, the central controller executes the following in each sampling cycle: acquiring real-time measurements → calculating the current layer's observation function → solving the quadratic programming model → issuing control commands. When the topology changes, only the local updates of the affected layers are activated, while the offline design parameters of the remaining layers remain unchanged.

[0073] This invention introduces the control barrier function into the Koopman feature space and utilizes Koopman linearization to transform the voltage safety set design in the original nonlinear space into a linear forward invariant set constraint in the feature space. This analytically and rigorously guarantees that the voltage does not exceed the limit, solving the fundamental defect of existing Koopman methods that can only approximate convergence and cannot strictly guarantee constraint satisfaction. A three-layer structured Koopman observation function based on the physical prior of the radial topology of the distribution network is designed. The general mathematical basis function is replaced with a physically meaningful node autonomy layer, neighborhood coordination layer, and global equilibrium layer, which reduces the feature space dimension by about 40%. Moreover, when the topology changes, only the affected layer needs to be locally updated, improving the computational efficiency by more than two orders of magnitude compared to global retraining. Utilizing the linear dynamic properties of the Koopman feature space, the calculation of the Lie derivative of the control barrier function is degenerated from a nonlinear operation to matrix multiplication, making the CBF constraint a linear inequality. This transforms the safety set preservation control law into a quadratic programming problem that can be solved quickly, combining strict safety and real-time performance.

[0074] In one embodiment, a simplified 5-node radial cable distribution network model is constructed based on the IEEE 33-node standard test system, and its topology is as follows: Figure 2 As shown, node 0 is the upstream grid access point, and nodes 1-5 are distributed photovoltaic access nodes. The lines are laid by cable, with a rated voltage of 0.4kV and a short-circuit capacity of 20MVA. The line parameters are selected according to GB / T 14049-2018 "Low-voltage power distribution systems" standard, as shown in the table below: Table 1 Parameter Table Model assumptions: ① The load remains constant during the simulation period; ② The photovoltaic inverter operates in MPPT mode, and its maximum output is limited by the light intensity; ③ Line-to-ground capacitance and three-phase imbalance factors are ignored; ④ The sampling period of the central controller is 50ms, and the delay in issuing control commands is ≤100ms.

[0075] Under these conditions, a comparison of the feature space dimensions of the conventional Koopman method and the present invention is shown in the figure below. Figure 3 As shown in the figure, the voltage prediction accuracy comparison chart is as follows: Figure 4 As shown in the figure, the comparison of operator training time is as follows: Figure 5 As shown in the figure, the voltage recovery curves after the topology is disconnected are compared as follows: Figure 6 As shown in the figure, the voltage recovery curves under extreme overvoltage scenarios are compared as follows: Figure 7 As shown (where the illumination is 1000Wm) 2 (PV output 2.25MW), Node 5 voltage diagram under low-risk scenarios is as follows: Figure 8 As shown (photovoltaic penetration rate 30%), the node 5 voltage diagram under high-risk scenarios is as follows. Figure 9As shown in the figure (photovoltaic penetration rate 60%), the computational complexity comparison of topology decoupling update is as follows: Figure 10 As shown (logarithmic coordinates).

[0076] Please see Figures 3-10 The conventional Koopman method requires a feature space of L=75 dimensions, while the hierarchical observation function of this invention only requires L=5+4+5=14 dimensions. With the same sample size, the root mean square error of voltage prediction estimated by the operator of this invention is 0.003 pu, while that of the conventional Koopman method is 0.008 pu, representing a 62% improvement in accuracy. The operator training time is reduced from 12.5 seconds for the conventional Koopman method to 0.8 seconds. Simulating the disconnection of the line between nodes 3 and 4, this invention only needs to update one observation function and the correlation submatrix in the second layer, with a local update time of 0.12 seconds; the global retraining time of the conventional Koopman method is 15.2 seconds. After the update, the voltage control performance of this invention is equivalent to that of the retraining, with the voltage fluctuation of node 5 not exceeding 0.012 pu, while the voltage peak of the conventional Koopman method reaches 1.128 pu within a 15-second update window. An extreme overvoltage scenario with a light intensity of 1000 W / m² and a total photovoltaic output of 2.25 MW is set, with the initial voltage of node 5 at 1.125 pu. Traditional Q(U) control cannot reduce the voltage to the acceptable range; the conventional Koopman method reduces the voltage to 1.095 pu within 1.2 seconds, but briefly exceeds the limit to 1.118 pu during this period; the Koopman-CBF method of this invention pulls the voltage back to 1.075 pu within 0.3 seconds and maintains it strictly throughout the process. No limit violations occurred. In a medium-risk scenario with a photovoltaic penetration rate of 30%, the system only activated the first and third layer observation functions (9 dimensions), with a control cycle of 2 seconds, reducing the inverter operation frequency by 40%. When the simulated photovoltaic penetration rate suddenly increased to 60%, causing the node voltage to approach 1.075 pu, the system automatically activated the second layer observation function (expanded to 14 dimensions) within 0.5 seconds, shortening the control cycle to 0.5 seconds, and strictly controlling the voltage below 1.08 pu.

[0077] This invention achieves strict forward invariance of the voltage safety set within a data-driven framework by embedding a control barrier function in the Koopman feature space. This ensures the system will not experience voltage exceedances under any operating condition, thus resolving the safety risks caused by the approximate linearization of existing Koopman methods. Through the design of a topology-prior hierarchical observation function, Koopman operator learning fully utilizes the physical structure information of the distribution network, significantly reducing the feature space dimension. Operator updates are transformed from global retraining to local decoupling correction, greatly enhancing topology adaptability. By linearizing the CBF constraint, the control law solution degenerates from nonlinear optimization to quadratic programming, making computational complexity controllable. Strictly safe control commands can be obtained within milliseconds, a significant improvement over the second-level convergence of the distributed subgradient method.

[0078] It should be noted that although the steps in the flowchart above are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order requirement for the execution of these steps, and they can be executed in other orders.

[0079] In another embodiment, such as Figure 11 As shown, a second aspect of the present invention provides a voltage optimization control system for a power distribution network inverter, comprising: The data acquisition module 10 is used to acquire real-time measurement data and topology data of the power distribution network; Function construction module 20 is used to construct hierarchical Koopman observation functions based on the topology data; the hierarchical Koopman observation functions include node-level observation sub-functions, line-level observation sub-functions, and global-level observation sub-functions; The operator update module 30 is used to estimate the Koopman operator matrix based on the real-time measurement data and the hierarchical Koopman observation function, and to perform local decoupling update on the Koopman operator matrix when a topological displacement signal is detected, so as to obtain the current Koopman operator. The constraint transformation module 40 is used to transform the pre-built voltage control barrier function into a linear safety constraint in the Koopman feature space based on the current Koopman operator. The feasible region generation module 50 is used to construct an observation matrix through the real-time measurement data, obtain the projection coefficient vector by the least squares method, and use the projection coefficient vector to transform the voltage upper and lower limit constraints of the distribution network into the voltage feasible region in the Koopman feature space. The voltage control module 60 is used to construct and solve a quadratic programming model with the goal of minimizing the sum of squares of inverter power regulation, and with the linear safety constraints and the voltage feasible region as constraints, to obtain power regulation commands for each inverter for execution.

[0080] It should be noted that each module in the aforementioned distribution network inverter voltage optimization control system can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module. For specific limitations regarding the distribution network inverter voltage optimization control system, please refer to the limitations of the distribution network inverter voltage optimization control method described above; both have the same function and role, and will not be repeated here.

[0081] A third aspect of the present invention provides an electronic device comprising: Processor, memory, and bus; The bus is used to connect the processor and the memory; The memory is used to store operation instructions; The processor is configured to execute instructions by calling the operation instructions, causing the processor to perform operations corresponding to the voltage optimization control method for a power distribution network inverter as shown in the first aspect of this application.

[0082] In one alternative embodiment, an electronic device is provided, such as Figure 12 As shown, Figure 12 The illustrated electronic device 5000 includes a processor 5001 and a memory 5003. The processor 5001 and the memory 5003 are connected, for example, via a bus 5002. Optionally, the electronic device 5000 may also include a transceiver 5004. It should be noted that in practical applications, the transceiver 5004 is not limited to one type, and the structure of this electronic device 5000 does not constitute a limitation on the embodiments of this application.

[0083] Processor 5001 may be a CPU, a general-purpose processor, a DSP, an ASIC, an FPGA, or other programmable logic device, transistor logic device, hardware component, or any combination thereof. It may implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 5001 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0084] Bus 5002 may include a path for transmitting information between the aforementioned components. Bus 5002 may be a PCI bus or an EISA bus, etc. Bus 5002 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 12 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0085] The memory 5003 may be a ROM or other type of static storage device capable of storing static information and instructions, RAM or other type of dynamic storage device capable of storing information and instructions, or it may be an EEPROM, CD-ROM or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but is not limited thereto.

[0086] The memory 5003 is used to store application code that executes the scheme of this application, and its execution is controlled by the processor 5001. The processor 5001 is used to execute the application code stored in the memory 5003 to implement the content shown in any of the foregoing method embodiments.

[0087] Among them, electronic devices include, but are not limited to: mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), and in-vehicle terminals (such as in-vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers.

[0088] The fourth aspect of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements a voltage optimization control method for a power distribution network inverter as shown in the first aspect of this application.

[0089] Another embodiment of this application provides a computer-readable storage medium storing a computer program that, when run on a computer, enables the computer to execute the corresponding content in the aforementioned method embodiments.

[0090] Furthermore, embodiments of the present invention also provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0091] In summary, this invention relates to the field of distribution network voltage control technology, and discloses a method, system, device, and medium for optimizing voltage control of distribution network inverters. Based on the topology data of the distribution network, a hierarchical Koopman observation function including node, line, and global layers is constructed. When a topology change signal is detected, the Koopman operator matrix is ​​locally decoupled and updated to transform the voltage control barrier function corresponding to the original nonlinear voltage constraint into a linear safety constraint in the Koopman linear characteristic space. The projection coefficient vector is analytically solved to determine the voltage feasible region. Furthermore, a quadratic programming model is constructed and solved with the objective of minimizing the sum of squares of inverter power regulation, resulting in an inverter power regulation command that satisfies the safety constraints. This achieves safe voltage control of medium- and low-voltage distribution networks under high-proportion distributed photovoltaic access without relying on an accurate physical model.

[0092] The various embodiments in this specification are described in a progressive manner. For directly identical or similar parts of the embodiments, refer to each other. Each embodiment focuses on its differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. It should be noted that the technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification.

[0093] The embodiments described above are merely preferred embodiments of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various improvements and substitutions without departing from the technical principles of this invention, and these improvements and substitutions should also be considered within the scope of protection of this application. Therefore, the scope of protection of this patent application should be determined by the scope of the claims.

Claims

1. A voltage optimization control method for a power distribution network inverter, characterized in that, include: Collect real-time measurement data and topology data of the power distribution network; A hierarchical Koopman observation function is constructed based on the aforementioned topology data; the hierarchical Koopman observation function includes node-level observation sub-functions, line-level observation sub-functions, and global-level observation sub-functions; The Koopman operator matrix is ​​estimated based on the real-time measurement data and the hierarchical Koopman observation function, and the Koopman operator matrix is ​​locally decoupled and updated when a topological displacement signal is detected to obtain the current Koopman operator. Based on the current Koopman operator, the pre-constructed voltage control barrier function is transformed into a linear safety constraint in the Koopman feature space; An observation matrix is ​​constructed using the real-time measurement data, and the projection coefficient vector is obtained by solving the least squares method. The projection coefficient vector is then used to transform the upper and lower voltage limits of the distribution network into the voltage feasible region in the Koopman feature space. With the goal of minimizing the sum of squares of inverter power regulation, and with the linear safety constraints and the voltage feasible region as constraints, a quadratic programming model is constructed and solved to obtain the power regulation commands for each inverter for execution.

2. The voltage optimization control method for a distribution network inverter according to claim 1, characterized in that, The real-time measurement data includes the voltage amplitude of each node and the power data of each inverter; wherein, The construction of the hierarchical Koopman observation function based on the topology data includes: Based on the topology data, construct the node-branch correlation matrix and the branch impedance diagonal matrix to determine the node impedance matrix; For each node, the reactance-resistance ratio of the node is determined based on the node impedance matrix, and the node-level observation subfunction is constructed by combining the voltage amplitude of the node and the power data of the corresponding inverter to characterize the driving effect of the power injection of each inverter on the local node voltage. Based on the node impedance matrix, the impedance characteristic coefficients of each branch in the distribution network are determined, and combined with the voltage amplitude and power data of the nodes contained in each branch, a line-level observation sub-function is constructed to characterize the neighborhood coupling relationship between line power flow transmission and voltage drop. The nodes are partitioned according to the node impedance matrix to obtain several electrical partitions. The node weights are then determined. The voltage amplitudes of the nodes are weighted and averaged using the node weights to obtain the partition weighted average voltage. This is used to construct a global layer observation subfunction to characterize the degree to which the average voltage of the electrical partition deviates from the rated value. The node-level observation sub-function, the line-level observation sub-function, and the global-level observation sub-function are concatenated to form the hierarchical Koopman observation function.

3. The voltage optimization control method for a distribution network inverter according to claim 2, characterized in that, The step of partitioning each node according to the node impedance matrix to obtain several electrical partitions and determining node weights includes: Extract the real part of the impedance of each node in the node impedance matrix, and determine the electrical distance between nodes based on the real part of the impedance. The nodes are clustered using k-means clustering based on the electrical distance between them to obtain several electrical partitions. The node weight is determined based on the real part of the impedance corresponding to the geometric center node within each electrical partition.

4. The voltage optimization control method for a distribution network inverter according to claim 1, characterized in that, The step of estimating the Koopman operator matrix based on the real-time measurement data and the hierarchical Koopman observation function includes: Based on the real-time measurement data and the hierarchical Koopman observation function, an observation matrix is ​​constructed; Based on the observation matrix, the initial Koopman operator is estimated using the extended dynamic mode decomposition method and the least squares method. Based on the hierarchical structure of the hierarchical Koopman observation function, the initial Koopman operator is divided into blocks to form the Koopman operator matrix.

5. The voltage optimization control method for a distribution network inverter according to claim 4, characterized in that, The step of locally decoupling and updating the Koopman operator matrix to obtain the current Koopman operator includes: The target level is determined based on the topological displacement signal, and a Gram rearrangement matrix is ​​constructed based on the target level and the observation matrix. The Schur complement matrix and Schur complement correction are calculated using the Gram rearrangement matrix, and the inverse matrix is ​​determined by the Woodbury matrix identity. The target level of the Koopman operator matrix is ​​updated based on the inverse matrix to obtain the current Koopman operator.

6. The voltage optimization control method for a distribution network inverter according to claim 1, characterized in that, The step of transforming the voltage control barrier function into a linear safety constraint in the Koopman feature space based on the current Koopman operator includes: Based on the current Koopman operator, the voltage control barrier function is subjected to differential operation in the Koopman feature space, and a voltage safety set is constructed based on the differential operation result; The voltage safety set is processed by introducing a barrier function attenuation coefficient to form the linear safety constraint.

7. The voltage optimization control method for a distribution network inverter according to claim 1, characterized in that, The step of transforming the voltage upper and lower limit constraints of the distribution network into the voltage feasible region in the Koopman feature space using the projection coefficient vector includes: Using the projection coefficient vector, the voltage upper and lower limit constraints of the distribution network are transformed into a set of linear inequalities in the Koopman feature space; Stack the projection coefficient vectors row by row to obtain the constraint matrix; Based on the constraint matrix, the system of linear inequalities is transformed into a linear voltage feasible region in the Koopman feature space.

8. A voltage optimization control system for a power distribution network inverter, characterized in that, include: The data acquisition module is used to collect real-time measurement data and topology data of the power distribution network; The function construction module is used to construct hierarchical Koopman observation functions based on the topology data; the hierarchical Koopman observation functions include node-level observation sub-functions, line-level observation sub-functions, and global-level observation sub-functions; The operator update module is used to estimate the Koopman operator matrix based on the real-time measurement data and the hierarchical Koopman observation function, and to perform local decoupling update of the Koopman operator matrix when a topological displacement signal is detected, so as to obtain the current Koopman operator. The constraint transformation module is used to transform the pre-built voltage control barrier function into a linear safety constraint in the Koopman feature space based on the current Koopman operator. The feasible region generation module is used to construct an observation matrix through the real-time measurement data, obtain the projection coefficient vector by the least squares method, and use the projection coefficient vector to transform the voltage upper and lower limit constraints of the distribution network into the voltage feasible region in the Koopman feature space. The voltage control module is used to construct and solve a quadratic programming model with the goal of minimizing the sum of squares of inverter power regulation, using the linear safety constraints and the voltage feasible region as constraints, to obtain power regulation commands for each inverter for execution.

9. An electronic device, characterized in that, The method includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements the distribution network inverter voltage optimization control method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein when the device containing the computer-readable storage medium executes the computer program, it implements the voltage optimization control method for a distribution network inverter as described in any one of claims 1 to 7.

Images