Distributed voltage control method under variable topology of active power distribution network based on data driving

Abstract

The invention discloses a distributed voltage control method under variable topology of an active power distribution network based on data driving, and the method comprises the steps: employing a Koopman dimension raising linearization method to be combined with least square estimation based on historical operation data under different topologies of the active power distribution network, and carrying out the training, and obtaining a linearization power flow matrix corresponding to each topology; establishing a mapping relation between the filial generation matrix and the corresponding linearized power flow matrix, and generating a topological adaptive power flow element model; when the topology of the active power distribution network changes, generating a linearized power flow matrix corresponding to the new topology; solving the linearized power flow matrix of the new topology through differential operation to obtain a sensitivity matrix of the voltage to the reactive power of the distributed power supply; the voltage amplitude and the reactive power of a local node are measured by taking minimization of the reactive power regulation quantity of the distributed power supply as a target; and each node adjusts the reactive power of the distributed power supply in situ by combining the measurement information of the node and the adjacent node, and loop iteration is carried out until the voltage of each node is stabilized in an allowable range.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power system automation, and particularly relates to a distributed voltage control method for active distribution network variable topology based on data driving. BACKGROUND

[0002] With the increasing penetration of distributed generation in active distribution network, the power flow change has significantly changed the operating characteristics of the power grid. On the one hand, the distributed generation changes the original power supply structure of the power grid; on the other hand, the uncertainty of the distributed generation output increases the complexity of the power system operation. Therefore, it is necessary to effectively control the operation of the distributed generation. Some scholars use the linearized power flow model to control the optimal power flow of the distribution network. However, the accuracy of this method depends largely on the feeder parameters, and in the medium and low voltage distribution network, these parameters are difficult to accurately obtain, thereby increasing the complexity of the operation of the power grid.

[0003] The neural network voltage control method based on data driving obtains the optimal operation strategy through historical data, which provides an alternative solution to the inaccuracy of the power flow model. Some scholars use the neural network method to solve the voltage over-limit problem in the distribution network. However, the existing machine learning methods lack adaptability to changing conditions, and need to retrain the model under different topological structures, but there is a lack of operation data under the new topology, making it difficult to implement. In order to solve the topology switching problem, scholars use long short-term memory network to establish the mapping relationship between the operating state and the optimal topology reconstruction scheme of the distribution network. Some scholars also use graph convolution network to adapt to the topology change in the voltage control of the power system. However, these methods are usually "black box" models, which lack interpretability.

[0004] The iterative control method based on data driving provides a feasible solution for voltage control of active distribution network under different scenarios. Some scholars have constructed a coordinated voltage control model based on data and physical hybrid driving, which effectively alleviates voltage fluctuations at multiple time scales. Another scholar combines model linearization with data-driven methods and uses a large amount of historical data to train the sensitivity relationship between bus voltage and injected power. However, under maintenance or fault conditions, the topology of the active distribution network may undergo active or passive reconstruction, and the above methods fail due to the lack of operation data under the new topology. Therefore, it is necessary to propose a topology adaptive power flow model construction method, which can quickly generate a power flow model under a given new topology. In addition, with the explosive growth of the number of control objects in the distribution network, centralized voltage control methods have difficulties in calculation and prediction accuracy. Therefore, it is necessary to propose a distributed voltage control method to effectively distribute the heavy centralized voltage control task to each distribution substation, thereby realizing coordinated control of large-scale distributed generation. SUMMARY

[0005] In view of the defects and deficiencies of the prior art, the present application provides a kind of distributed voltage control method under the variable topology of active distribution network based on data driving, to solve the problems that traditional distribution network voltage control depends on accurate feeder parameters, topology changes need to retrain model, centralized control is difficult to adapt to large-scale distributed power.

[0006] The method first trains linearized power flow matrix corresponding to each topology based on historical operation data of active distribution network under different topologies (including node active power injection, reactive power injection and voltage amplitude), with active power and reactive power injection as input, voltage amplitude as output, using Koopman dimensionality linearization method combined with least square estimation, to get rid of the dependence on distribution network model parameters;Then, the child matrix representing the upstream and downstream connectivity of the node is used as the topology feature matrix, and the mapping relationship between the child matrix and the linearized power flow matrix is established to generate a topology adaptive power flow element model, which realizes that no new topology historical data is needed when the topology changes, and the corresponding linearized power flow matrix can be quickly generated only through the child matrix of the new topology, and the sensitivity matrix of voltage to distributed power reactive power is updated through differential operation;Finally, a distributed control scheme based on subgradient iteration is constructed to minimize the deviation of distributed power reactive regulation or node voltage from the rated value, each node measures local voltage and reactive power, exchanges measurement information with adjacent nodes, adjusts distributed power reactive power on site, and iterates until the voltage is stable, while meeting the constraints of node voltage, distributed power capacity, power factor and power flow balance.

[0007] The present application does not need to retrain the model after the topology changes, saves computing resources and time, and adapts to the coordination needs of large-scale distributed power through distributed control, ensuring the stable operation of active distribution network under variable topology scenario.

[0008] The present application specifically adopts the following technical solutions:

[0009] A kind of distributed voltage control method under the variable topology of active distribution network based on data driving, comprising:

[0010] Based on the historical operation data of active distribution network under different topologies, active power injection and reactive power injection are used as input samples, and voltage amplitude is used as output sample, and linearized power flow matrix corresponding to each topology is trained using Koopman dimensionality linearization method combined with least square estimation;

[0011] The child matrix representing the upstream and downstream connectivity of the node is used as the topology feature matrix, and the mapping relationship between the child matrix and the corresponding linearized power flow matrix is established to generate a topology adaptive power flow element model;

[0012] When the active power distribution network topology changes, the corresponding child matrix of the new topology is determined according to the node connection relationship of the new topology, and is input into the topology adaptive power flow element model to generate a linearized power flow matrix corresponding to the new topology;

[0013] The linearized power flow matrix of the new topology is solved by differential operation to obtain a sensitivity matrix of voltage to distributed power reactive power;

[0014] With the minimum distributed power reactive regulation amount as the target, each node measures the local node voltage amplitude and reactive power based on the sensitivity matrix, and exchanges measurement information with adjacent upstream and downstream nodes;

[0015] The next gradient iteration method is adopted, and each node adjusts the reactive power of the distributed power on site in combination with the measurement information of the node itself and adjacent nodes, and the cycle iteration is performed until the voltage of each node is stabilized in the allowed range.

[0016] Further, the dimension increasing processing of the Koopman dimension increasing linearization method adopts a nonlinear dimension increasing function based on Euclidean distance to expand the input sample into a combination variable containing the original sample and the dimension increasing feature; and the linearized power flow matrix is solved by least square estimation based on the combination variable and the output sample.

[0017] Further, the child matrix is N*N dimension, N is the total number of nodes of the active power distribution network, and the elements thereof are uniquely determined by the upstream and downstream connection relationship of the nodes: the element value is 1 when the node j is downstream of the node i, otherwise it is 0; the construction process of the topology adaptive power flow element model includes: converting the child matrix and the corresponding linearized power flow matrix of different topologies into column vector forms respectively, constructing a training sample set, and then establishing the mapping relationship between the column vector of the child matrix and the column vector of the linearized power flow matrix by least square estimation combined with pseudo-inverse operation.

[0018] Further, the sensitivity matrix of voltage to distributed power reactive power is derived by differential operation on the linearized power flow matrix of the new topology; the differential operation is realized based on the partial derivative of the dimension increasing function to the reactive power, the partial derivative is solved based on the association relationship between the nonlinear dimension increasing function and the reactive power parameter, and finally the sensitivity matrix representing the corresponding relationship between the voltage and the reactive power regulation amount is formed.

[0019] Further, the specific implementation of the next gradient iteration method includes: each node constructs a local target function related to the voltage deviation based on the voltage and reactive power information of the node itself and adjacent nodes; the gradient of the target function is calculated through the sensitivity matrix; the node reactive power adjustment amount is updated based on the gradient result, the iteration step length adopts an adaptive strategy, and is dynamically adjusted according to the voltage deviation degree to balance the convergence speed and control accuracy; and the cycle iteration is performed until the voltage variation of each node in continuous multiple iterations is less than a preset threshold.

[0020] Further, the control target of the method is to minimize the reactive power regulation amount of the distributed power supply or to minimize the square sum of the deviation of the node voltage amplitude from the rated value;

[0021] The measurement information exchanged between each node and adjacent nodes includes the real-time voltage amplitude, real-time reactive power injection amount and current reactive power adjustment amount of the node, and the information is exchanged through the power distribution network communication system, and the exchange frequency matches the iteration frequency;

[0022] The reactive power regulation of the distributed power supply needs to meet multiple operation constraints, including that the node voltage amplitude is within the upper and lower limit ranges, the capacity of the distributed power supply does not exceed the limit, the power factor is not lower than the lower limit value, and the global balance of the active and reactive power of the power distribution network.

[0023] Further, the collection of the historical operation data needs to meet the sample representativeness requirement: covering typical operation scenarios of the power distribution network including load peak, load valley and different output levels of the distributed power supply, and the collection time length and section period need to ensure that the number of training samples is sufficient to support the reliable training of the linearized power flow matrix.

[0024] Further, the continuous multiple iterations in the voltage stability judgment standard are 2-5 consecutive iterations, and the preset threshold meets the voltage control accuracy requirement of the power system; the adaptive step length of the sub-gradient iteration is dynamically adjusted according to the mean value of the voltage deviation in the iteration process: when the voltage deviation is large, the step length is increased to speed up the convergence, and when the voltage deviation is small, the step length is reduced to improve the control accuracy.

[0025] And a data-driven distributed voltage control system for active power distribution networks with variable topology, comprising:

[0026] A data acquisition and model training module is configured to obtain historical operation data of active power injection, reactive power injection and voltage amplitude of nodes in different topologies of an active power distribution network, use the active power injection and the reactive power injection as input samples, the voltage amplitude as output samples, and train a linearized power flow matrix corresponding to each topology by using a Koopman dimensionality linearization method combined with least squares estimation; the module is also configured to construct a child matrix representing the upstream and downstream connectivity relationship of a node as a topology feature matrix, establish a mapping relationship between the child matrix and the corresponding linearized power flow matrix, and generate a topology adaptive power flow element model;

[0027] A topology adaptation and sensitivity updating module is configured to, when the topology of the power distribution network changes, determine the corresponding child matrix according to the node connectivity relationship of the new topology, input the child matrix into the topology adaptive power flow element model to generate a linearized power flow matrix of the new topology, and solve the sensitivity matrix of the voltage to the reactive power of the distributed power supply through differential operation;

[0028] The distributed control module includes multiple node control units. Each node control unit is used to measure the local voltage amplitude and reactive power based on the sensitivity matrix with the goal of minimizing the reactive power regulation of the distributed power source. It exchanges measurement information with adjacent node control units and uses a sub-gradient iterative method to adjust the reactive power of the distributed power source locally. The process is iterated until the voltage of each node stabilizes within the allowable range.

[0029] And a computer device, characterized in that it includes a processor and a memory, the memory storing a computer program, wherein when the processor executes the computer program, it implements the method described above.

[0030] A non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described above.

[0031] Compared with the prior art, the present invention and its preferred embodiments have at least the following beneficial effects:

[0032] This invention effectively eliminates the reliance on precise feeder parameters for voltage control in traditional distribution networks. By using the Koopman linearization method combined with historical operating data to train a linearized power flow matrix, a reliable voltage control model can be constructed without relying on specific model parameters of the distribution network. This adapts to scenarios where grid parameters are difficult to obtain accurately due to increased distributed generation penetration, thus improving the universality and practicality of the voltage control scheme.

[0033] This invention significantly improves the voltage control adaptation efficiency during distribution network topology changes. Addressing the issues of traditional methods requiring model retraining after topology changes and incurring data dependencies and time delays, this invention utilizes a topology-adaptive power flow meta-model. This model can quickly generate the corresponding linearized power flow matrix and update sensitivity information based solely on the node connectivity relationships (child matrix) of the new topology, eliminating the need to collect historical operating data under the new topology. This significantly saves computational resources and model adjustment time, ensuring real-time voltage control in topology change scenarios.

[0034] This invention solves the "curse of dimensionality" problem commonly encountered in traditional centralized voltage control, and is adapted to the coordination needs of large-scale distributed power sources. The distributed subgradient iterative control scheme constructed in this invention allows each node to complete reactive power regulation and voltage iterative optimization locally by combining local measurement information with data exchange with adjacent nodes, without relying on global centralized calculations. This reduces data transmission pressure and control delay, and can efficiently coordinate the reactive power output of multi-node distributed power sources, ensuring voltage stability of the distribution network when large-scale distributed power sources are connected.

[0035] This invention balances the accuracy of voltage control with the safety of equipment operation. It incorporates multiple constraints during the control process, including node voltage upper and lower limits, distributed generation capacity, power factor, and power flow balance. Furthermore, it allows for flexible selection of control objectives between "minimizing reactive power regulation by distributed generation" and "minimizing node voltage deviation." This ensures that the distribution network voltage remains stable within the allowable range while preventing distributed generation from exceeding safe operating boundaries due to excessive regulation, thus achieving synergy between voltage control and equipment protection.

[0036] In summary, this invention optimizes the voltage control scheme under the changing topology of active distribution networks from multiple aspects, including modeling adaptability, topology response efficiency, large-scale coordination capability, and operational safety, providing reliable support for the stable and efficient operation of active distribution networks. Attached Figure Description

[0037] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0038] Figure 1 This is a flowchart illustrating the implementation of the method in an embodiment of the present invention. Detailed Implementation

[0039] In the following, specific embodiments of this application will be described in detail with reference to the accompanying drawings. Based on these detailed descriptions, those skilled in the art will be able to clearly understand and implement this application. Without departing from the principles of this application, features from various embodiments can be combined to obtain new implementations, or certain features from some embodiments can be substituted to obtain other preferred implementations.

[0040] To make the features and advantages of the present invention more apparent and understandable, specific embodiments are described below in conjunction with the accompanying drawings:

[0041] In view of this, this invention provides a data-driven distributed voltage control method for active distribution networks under changing topologies. First, an active distribution network voltage control model is established. The Koopman dimensionality-increasing linearization method is used to obtain a linearized power flow matrix independent of the distribution network model parameters, and global sensitivity information is updated, thereby constructing a linear optimization solution model for the distribution network voltage. Second, to address the problem of the original sensitivity information becoming invalid due to distribution network topology changes, a topology-adaptive power flow meta-model method is adopted. A meta-model is generated by establishing a mapping relationship between topology graph data (adjacency matrix) and the linearized power flow matrix. Under the new topology conditions, the corresponding linearized power flow matrix can be quickly generated using only the topology graph data and the meta-model, and online updates of the distribution network's global sensitivity can be achieved. Finally, to address the dimensionality curse problem that may be encountered in traditional centralized optimization control, this invention uses a subgradient iterative distributed voltage optimization control method. Through subgradient feedback iteration, it utilizes the information of each node to achieve local iterative control of reactive power and voltage in the active distribution network under the new topology. This method can avoid retraining the linearized power flow model after distribution network topology changes, thereby saving computational resources and training time. Simultaneously, distributed voltage control meets the needs of large-scale distributed power source coordination.

[0042] The implementation process of the present invention is further described below through an embodiment. This embodiment employs the Koopman dimensionality-upgrading linearization method to update global sensitivity information and constructs a linear optimization solution model for the distribution network voltage. Using the topology adaptive power flow meta-model method, under new topology conditions, online updates of the distribution network's global sensitivity can be achieved solely based on topology graph data and the meta-model. Finally, a distributed voltage optimization control method based on subgradient iteration is used to achieve local iterative control of reactive power voltage in the active distribution network under the new topology, utilizing information from each node. Specifically, the following steps are included:

[0043] Step 1: Voltage optimization control of the distribution network needs to ensure that the voltage at each node does not exceed the set range while maintaining voltage balance within the distribution network, and to rationally allocate reactive power between distributed generation (DG) and compensation equipment. The main objective of voltage optimization control is to minimize the reactive power regulation of distributed generation (DG), which can be expressed as:

[0044] (1)

[0045] In the formula, This represents the optimal reactive power injection vector for each DG. The 2-norm of a vector. The reactive power adjustment at node i represents the amount of reactive power adjustment compared to before adjustment, and N represents the total number of active distribution network nodes.

[0046] In active distribution network voltage control, operating constraints include node voltage constraints (2)-(3), DG capacity constraints (4)-(5), DG power factor constraints (6), and distribution network power flow constraints (7).

[0047] (2)

[0048] (3)

[0049] (4)

[0050] (5)

[0051] (6)

[0052] (7)

[0053] In the formula, (2) is the linearized Distflow branch power flow equation of the distribution network, and the matrix is... and Then they represent A diagonal matrix of order 1. It is represented as the sensitivity matrix of voltage to reactive power of DG. This indicates the result after removing the first row corresponding to the reference node. Node-branch incidence matrix, vector This indicates the voltage distribution of the distribution network before voltage optimization control is implemented. This represents the column vector of reactive power injection at each node before voltage optimization control. This represents the column vector formed by the optimized reactive power regulation of each node. and These represent the active power of the nodes, respectively. and reactive power The column vector formed by the injection. and These are column vectors formed by the lower and upper limits of the node voltage amplitude, respectively. This represents the column vector consisting of the upper limit of reactive power of the nodes. This represents the set of nodes in an active distribution network. and They are nodes The active and reactive power injected at the point, The power factor angle of node i is represented. This is the corresponding lower limit value. Indicates the node before voltage optimization control reactive power, Represents a node reactive power regulation, Represents a node Capacity.

[0054] Step 2: The distribution network control center collects historical operating data from each node of the distribution network. The active power injection and reactive power injection at each node are used as input samples, i.e.: The node voltage amplitude is used as the output sample, i.e.: The upgraded linear power flow model was trained offline using the Koopman method. The specific steps are as follows:

[0055] Step 2-1: Collect data samples from S historical cross-sections to construct the input sample set. and output sample set ;

[0056] The input sample is:

[0057] (8)

[0058] The output sample is:

[0059] (9)

[0060] in, Representing the The input data for each training sample, and Representing the Output data for each training sample.

[0061] Step 2-2: Upscaling the input samples yields a sample set of input variables with increased dimensionality. , can be represented as:

[0062] (10)

[0063] In the formula, The input variable for the i-th sample in higher dimensions can be represented as:

[0064] (11)

[0065] In the formula, Let represent the dimension-upgrading function of the i-th input sample in the distribution network. Assuming the dimension of the dimension-upgrading function is K, then... It can be represented in vector form as follows:

[0066] (12)

[0067] Since state-space dimensionality increase transformations require stable dimensionality increase effects, this embodiment employs a 'polyharmonic' type dimensionality increase function. The expression is:

[0068] (13)

[0069] in, To and Basis vectors of the same dimension represent The k-th dimension raising function, It represents the k-th Euclidean distance of the i-th data sample.

[0070] Steps 2-3: Based on the above training samples, and using least squares estimation, the mapping relationship between the output variables and the upgraded input variables can be obtained, i.e., the linearized power flow matrix. The result can be calculated using the following formula:

[0071] (14)

[0072] in, represent matrix transpose, represent The pseudo-inverse matrix.

[0073] Steps 2-4: Linearized power flow matrix based on equivalent values Through differential operations, the data-driven sensitivity coefficient matrix in (2) can be derived. .make Representative sensitivity coefficient matrix The Middle Line 1 The formula for calculating the elements of a column can be expressed as follows:

[0074] (15)

[0075] in, Represents a linear power flow matrix The middle corresponds to the voltage With DG reactive power adjustment elements, Representing the linearized power flow matrix The middle corresponds to the voltage With the Dimensional Upgrading Operation Function The partial derivatives on the right side of equation (15) can be calculated using the following formula:

[0076] (16)

[0077] The sensitivity coefficient matrix results derived from equations (15)-(16) A linear voltage model can then be constructed (2).

[0078] Step 3: Based on the topology adaptive power flow meta-model method, derive the topology data representing the active distribution network and the linearized power flow matrix. The mapping relationship between them. Using this method, the linearized power flow matrix under the new topology can be calculated without relying on historical operating data under the new topology.

[0079] Step 3-1: The topology data of the distribution network can be simplified into a child matrix. It can be represented as:

[0080] (17)

[0081] in, This indicates the connectivity between nodes i and j. Its value is 1 when node j is downstream of node i, and 0 otherwise. This represents the set of downstream nodes of node i.

[0082] Step 3-2: Sub-matrix and linearized power flow matrix The mapping relationship between them can be represented as:

[0083] (18)

[0084] in, Representing the meta-model, and These are the column vectors of the offspring matrix and the linearized power flow matrix, respectively:

[0085] (19)

[0086] in, Indicates will Convert to dimension Column vectors.

[0087] Step 3-3: Collect historical operating data of active distribution networks under different topologies, and obtain the linearized power flow matrix for each topology based on Step 2. Meta-model It can be obtained by retraining based on the adjacency matrix samples and the corresponding linearized power flow matrix samples, and its sample set is defined as:

[0088] (20)

[0089] in, and Let E represent the sample sets of the input variable (subsidiary matrix) and the output variable (linearized power flow matrix), respectively, and E is the number of training samples.

[0090] Steps 3-4: Meta-model based on least squares method It can be determined by the following formula:

[0091] (twenty one)

[0092] By using the Moore-Penrose pseudo-inverse, It can be represented as:

[0093] (twenty two)

[0094] Steps 3-5: Based on the meta-model The column vectors of the offspring matrix under a given new topology The linearized power flow matrix for this topology can be calculated without requiring historical operational data for that topology. :

[0095] (twenty three)

[0096] In summary, based on the power flow meta-model It can generate linear power flow matrices under new topologies. Thus, the sensitivity coefficient matrix is ​​updated online according to steps 2–4.

[0097] Step 4: Construct a data-driven distributed voltage optimization control scheme based on subgradient iteration. Through subgradient feedback iteration, local iterative control of reactive power and voltage in the active distribution network is achieved based on the information of each node.

[0098] Step 4-1: Construct a distributed iterative solution model for the active distribution network voltage optimization control.

[0099] After DG reactive power adjustment, the sum of squares of the voltages between the node voltage amplitude and the rated value should be minimized. Therefore, the objective function in (1) can be rewritten as:

[0100] (twenty four)

[0101] In the formula, This represents a column vector composed of the rated voltages of each node in an active distribution network. In this invention, the rated voltage of each node is set to 1.

[0102] The constraints to be considered for distributed voltage control are given by equations (2)-(7). In summary, the objective function (24) and equations (2)-(7) constitute a complete distributed voltage control optimization model.

[0103] Step 4-2: The main steps of distributed voltage iterative control are as follows:

[0104] (1) Measure the voltage amplitude of the local node and update the local injected reactive power;

[0105] (2) Each node exchanges voltage amplitude and reactive power injection measurements with adjacent upstream and downstream nodes;

[0106] (3) Each node adjusts the reactive power of the distributed power source based on the voltage and reactive power measured at its own node, combined with the voltage and reactive power of the adjacent nodes, to achieve one-time iterative adjustment of the voltage.

[0107] The distributed subgradient method is used to implement the iterative adjustment of each node in each step. The reactive power adjustment of node i can be expressed as:

[0108] (25)

[0109] in, This represents the reactive power adjustment amount of node i in the k-th iteration. The iteration step size represents the distributed subgradient method. Let represent the gradient of the objective function of node i with respect to the optimization variables at the k-th iteration. The objective function of node i at the k-th iteration is defined as follows:

[0110] (26)

[0111] in, Let i represent the set of node i itself and its upstream and downstream adjacent nodes. Let represent the voltage amplitude measurement of node j, the neighboring node of node i, at the k-th iteration. Based on the above formula, we can calculate... ,Right now:

[0112] (27)

[0113] (4) After the reactive power adjustment is completed, return to step (1) and repeat the process until the voltage of each node is stable.

[0114] Based on the complete design provided in this embodiment, the implementation process of distributed voltage control in a data-driven active distribution network with varying topologies is as follows: Figure 1 As shown, it is divided into two main stages: offline training and online distributed voltage control, as detailed below:

[0115] 1. Offline training phase

[0116] First, historical data samples of active / reactive power injection and voltage amplitude at distribution network nodes are collected; then, basis vectors are constructed to increase the dimensionality of the input samples; next, sub-matrix samples of the existing topology are collected, and a linearized power flow matrix is ​​obtained by training based on the least squares method; finally, the mapping relationship between the topology and the power flow matrix is ​​established through least squares regression to complete the training of the power flow meta-model.

[0117] 2. Online Distributed Voltage Control Stage

[0118] First, input the real-time topology data of the distribution network. If it is a new topology, update the linearized power flow matrix and voltage-reactive power sensitivity matrix through the meta-model. Then, each node measures its local operating status, constructs a distributed voltage optimization model based on the given constraints, calculates the gradient of the local objective function of each node, and adjusts the reactive power of the distributed power source locally through the sub-gradient iterative algorithm until the voltage of all nodes in the network is stable.

[0119] Based on the same inventive concept, this invention also provides a computer device, comprising: one or more processors, and a memory for storing one or more computer programs; the programs include program instructions, and the processor executes the program instructions stored in the memory. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, used to implement one or more instructions, specifically for loading and executing one or more instructions stored in a computer storage medium to implement the above-described method.

[0120] It should be further explained that, based on the same inventive concept, the present invention also provides a computer storage medium storing a computer program, which, when executed by a processor, performs the above-described method. This storage medium can be any combination of one or more computer-readable media. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In the present invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0121] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this disclosure. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0122] The foregoing has shown and described the basic principles, main features, and advantages of this disclosure. Those skilled in the art should understand that this disclosure is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of this disclosure. Various changes and modifications can be made to this disclosure without departing from its spirit and scope, and all such changes and modifications fall within the scope of this disclosure as claimed.

[0123] This invention is not limited to the preferred embodiment described above. Anyone inspired by this invention can derive various other forms of distributed voltage control method based on data-driven active distribution network topology changes. All equivalent variations and modifications made within the scope of the claims of this invention should be included within the scope of this invention.

Claims

1. A data-driven distributed voltage control method for active distribution networks with varying topologies, characterized in that, include: Based on historical operating data of different topologies of active distribution networks, with active power injection and reactive power injection as input samples and voltage amplitude as output samples, the linear power flow matrix corresponding to each topology is trained by using the Koopman dimension-up linearization method combined with least squares estimation. Using the child matrix of the distribution network that represents the upstream and downstream connectivity of nodes as the topology feature matrix, a mapping relationship between the child matrix and the corresponding linearized power flow matrix is ​​established to generate a topology adaptive power flow meta-model. When the topology of the active distribution network changes, the corresponding child matrix is ​​determined according to the node connectivity of the new topology, and input into the topology adaptive power flow element model to generate the linearized power flow matrix corresponding to the new topology. By solving the linearized power flow matrix of the new topology through differential operations, the sensitivity matrix of voltage to the reactive power of distributed generation is obtained. With the goal of minimizing the reactive power regulation of distributed power sources, each node measures the voltage amplitude and reactive power of its local node based on the sensitivity matrix, and exchanges measurement information with adjacent upstream and downstream nodes. The subgradient iterative method is adopted, in which each node combines its own and the measurement information of its neighboring nodes to adjust the reactive power of the distributed power source on the spot, and iterates in a loop until the voltage of each node stabilizes within the allowable range.

2. The distributed voltage control method for active distribution networks with varying topologies based on data-driven principles as described in claim 1, characterized in that: The Koopman dimensionality-upgrading linearization method employs a nonlinear dimensionality-upgrading function based on Euclidean distance to expand the input sample into a combined variable containing the original sample and the dimensionality-upgrading features. The linearized power flow matrix is ​​obtained by least squares estimation based on this combined variable and the output sample.

3. The distributed voltage control method for active distribution networks with varying topologies based on data-driven principles as described in claim 1, characterized in that: The child matrix has N×N dimensions, where N is the total number of nodes in the active distribution network. Its elements are uniquely determined by the upstream and downstream connectivity of the nodes: when node j is downstream of node i, the element has a value of 1, otherwise it has a value of 0. The construction process of the topology adaptive power flow meta-model includes: converting the child matrix of different topologies and the corresponding linearized power flow matrix into column vector form respectively, constructing a training sample set, and then establishing the mapping relationship between the column vector of the child matrix and the column vector of the linearized power flow matrix through least squares estimation combined with pseudo-inverse operation.

4. The distributed voltage control method for active distribution networks with varying topologies based on data-driven principles as described in claim 2, characterized in that: The voltage sensitivity matrix to the reactive power of the distributed power source is derived by performing differential operations on the linearized power flow matrix of the new topology. The differential operation is based on the partial derivative of the upgraded function with respect to reactive power. The partial derivative is solved based on the correlation between the nonlinear upgraded function and the reactive power parameter, and finally forms a sensitivity matrix characterizing the correspondence between voltage and reactive power regulation.

5. The distributed voltage control method for active distribution network with variable topology based on data-driven approach according to claim 1, characterized in that: The specific implementation of the sub-gradient iteration method includes: each node constructing a local objective function related to voltage deviation based on its own and neighboring nodes' voltage and reactive power information; calculating the gradient of the objective function through the sensitivity matrix; updating the node's reactive power adjustment based on the gradient result; adopting an adaptive strategy for the iteration step size, dynamically adjusting it according to the degree of voltage deviation to balance convergence speed and control accuracy; and iterating repeatedly until the voltage change of each node in multiple consecutive iterations is less than a preset threshold.

6. The distributed voltage control method for active distribution network with variable topology based on data-driven approach according to claim 1, characterized in that: The control objective of the method is to minimize the reactive power regulation of the distributed power source or to minimize the sum of squares of the deviations between the node voltage amplitude and the rated value. The measurement information exchanged between each node and its neighboring nodes includes the node's real-time voltage amplitude, real-time reactive power injection, and current reactive power adjustment. The information is exchanged through the distribution network communication system, and the exchange frequency is matched with the iteration frequency. The reactive power regulation of distributed generation needs to meet several operational constraints, including the node voltage amplitude being within the upper and lower limits, the distributed generation capacity not exceeding the limit, the power factor not being lower than the lower limit, and the global balance of active and reactive power in the distribution network.

7. The distributed voltage control method for active distribution network with variable topology based on data-driven approach according to claim 1, characterized in that: The collection of historical operating data must meet the sample representativeness requirement: it must cover typical operating scenarios of distribution networks including peak load, trough load, and different output levels of distributed power sources. The collection duration and cross-sectional period must ensure that the number of training samples is sufficient to support the reliable training of the linearized power flow matrix.

8. The distributed voltage control method for active distribution network with variable topology based on data-driven approach according to claim 5, characterized in that: The voltage stability judgment criterion includes 2-5 consecutive iterations, and the preset threshold meets the voltage control accuracy requirements of the power system. The adaptive step size of the sub-gradient iteration is dynamically adjusted according to the average voltage deviation during the iteration process: when the voltage deviation is large, the step size is increased to accelerate convergence, and when the voltage deviation is small, the step size is decreased to improve control accuracy.

9. A data-driven distributed voltage control system for active distribution networks with varying topologies, characterized in that, include: The data acquisition and model training module is used to acquire historical operating data of active power injection, reactive power injection, and voltage amplitude of nodes under different topologies of the active distribution network. Using the active power injection and reactive power injection as input samples and voltage amplitude as output samples, the module trains the linearized power flow matrix corresponding to each topology using the Koopman dimension-upgrading linearization method combined with least squares estimation. The module is also used to construct a child matrix representing the upstream and downstream connectivity of nodes as a topology feature matrix, establish a mapping relationship between the child matrix and the corresponding linearized power flow matrix, and generate a topology adaptive power flow meta-model. The topology adaptation and sensitivity update module is used to determine the corresponding child matrix based on the node connectivity of the new topology when the distribution network topology changes, input it into the topology adaptive power flow element model to generate the linearized power flow matrix of the new topology, and solve for the voltage sensitivity matrix to the reactive power of distributed generation through differential operations. The distributed control module includes multiple node control units. Each node control unit is used to measure the local voltage amplitude and reactive power based on the sensitivity matrix with the goal of minimizing the reactive power regulation of the distributed power source. It exchanges measurement information with adjacent node control units and uses a sub-gradient iterative method to adjust the reactive power of the distributed power source locally. The process is iterated until the voltage of each node stabilizes within the allowable range.

10. A computer device, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements the method of any one of claims 1-8.

Images