An automatic voltage control method and system based on data mining

Abstract

A kind of automatic voltage control method and system based on data mining, method includes the following steps: collecting the voltage data and power data of each node in power grid and processing, form voltage sample and power sample, and obtain key voltage parameter and corresponding power sample by clustering;Voltage control objective function is constructed and constraint condition is set;The voltage data of each node at current time is obtained, and the corresponding cluster is determined;Based on the key voltage parameter and power sample corresponding to cluster, the sensitivity of total active power loss to each node reactive power is calculated;The voltage data of each node at current time is substituted into voltage control objective function, the active power loss of current operating point is calculated, combined with sensitivity, voltage control objective function is linearized and solved.The present application extracts typical power grid operating mode by data mining, and realizes automatic voltage optimization control by combining reactive power sensitivity analysis, improves voltage control precision and operating stability.

Description

Technical Field

[0001] This invention belongs to the field of power system automation control technology, specifically relating to an automatic voltage control method and system based on data mining. Background Technology

[0002] In power grids and systems, Automatic Voltage Control (AVC) systems have become an indispensable and crucial component. The application of AVC systems enables stable voltage control of the power grid, improving its operational stability. However, with the continuous integration of various renewable energy loads, the power grid has expanded significantly and become increasingly complex, making it difficult to achieve ideal control performance from AVC systems and leading to voltage stability deviations. The main reason is the difficulty in fully extracting key parameters from massive amounts of voltage data, resulting in AVC system control outcomes that fail to meet the actual operational needs of the power grid.

[0003] To address this issue, existing technologies have proposed various solutions, but all have certain limitations. For example, voltage control methods based on network topology resources have a relatively complex control process, resulting in unsatisfactory control performance; voltage control methods based on model predictive control suffer from insufficient control accuracy; and voltage control methods based on reduced-order resonant controllers require improvements to the controller itself, leading to long voltage control times and failing to meet the requirements for fast control. In addition, patent application CN108376986A provides a reactive power and voltage control method and device for distribution networks. The method obtains the reactive power limit and network loss objective function of each substation in the distribution network, solves the optimal power factor value and determines the power factor control range, realizing the coordination of automatic voltage control and automatic switching of reactive power compensation equipment, and coordinating the cooperation between high-voltage and medium- and low-voltage distribution networks. Patent application CN113541184A discloses a secondary voltage control method for power grids in offshore wind power gathering areas. Based on real-time inspection of the voltage status of the high-voltage side bus of the booster station in the offshore wind power gathering area, it can adaptively introduce voltage constraints or correction targets, thereby ensuring that the offshore wind power gathering area meets the voltage operation requirements and eliminates voltage over-limit. Patent application CN118214003A provides a distributed voltage control method for distribution networks. It uses the PJ-ADMM algorithm to perform distributed decoupling and solution of the optimal power flow model of the distribution network, thereby effectively reducing the communication requirements of centralized control while meeting the information privacy protection requirements.

[0004] Therefore, existing technologies do not take into account the low voltage control accuracy caused by line losses, and it is difficult to balance control speed with line losses. Summary of the Invention

[0005] To address the technical problems in existing technologies, such as low voltage control accuracy due to neglecting line losses and the difficulty in balancing control speed, this invention provides an automatic voltage control method and system based on data mining. The method includes the following steps: collecting and processing voltage and power data from each node in the power grid to form voltage and power samples, and obtaining key voltage parameters and corresponding power samples through clustering; constructing a voltage control objective function and setting constraints; acquiring the voltage data of each node at the current moment and determining the corresponding clusters; calculating the sensitivity of total active power loss to reactive power at each node based on the key voltage parameters and power samples corresponding to the clusters; substituting the voltage data of each node at the current moment into the voltage control objective function to calculate the active power loss at the current operating point, and linearizing and solving the voltage control objective function based on the sensitivity. This invention extracts typical power grid operating modes through data mining and combines reactive power sensitivity analysis to achieve automatic voltage optimization control, improving voltage control accuracy and operational stability.

[0006] The present invention adopts the following technical solution: A first aspect of the present invention provides an automatic voltage control method based on data mining, comprising the following steps: Voltage and power data from each node in the power grid are collected and processed to form voltage and power samples, and sample matching is performed. Key voltage parameters and corresponding power samples are obtained by mining the voltage samples through clustering. Based on key voltage parameters and corresponding power samples, a voltage control objective function is constructed with the goal of minimizing the total active power loss of the power grid, and constraints are set. Obtain the voltage data of each node at the current moment, calculate the distance between the key voltage parameters of each cluster and the voltage data of each node at the current moment, and determine the cluster with the smallest distance; based on the key voltage parameters and power samples corresponding to the clusters, calculate the sensitivity of the total active power loss to the reactive power of each node; substitute the voltage data of each node at the current moment into the voltage control objective function, calculate the active power loss at the current operating point, and linearize the voltage control objective function by combining the sensitivity; under constraints, solve the linearized voltage control objective function.

[0007] Preferably, the process of clustering voltage samples is as follows: Set the number of clusters and the weighting index, and randomly initialize the membership matrix and cluster centers; the membership matrix satisfies the membership constraint, that is, the sum of the membership degrees of all samples to each cluster center is always 1; Under the premise of satisfying the membership degree constraint, the dissimilarity index objective function is calculated based on the membership degree of each sample to the corresponding cluster center and the voltage sample. The cluster centers and membership degrees are iteratively updated by minimizing the dissimilarity index objective function; based on the updated cluster centers and membership degrees, the dissimilarity index objective function value at the current iteration number is calculated. The sum of the squares of the Euclidean distances between all samples in each cluster and the cluster center is calculated and accumulated as the convergence error. When the change in convergence error over adjacent iterations is less than a preset threshold, or the difference in the objective function value of the dissimilarity index is less than a predetermined difference threshold, or the maximum number of iterations is reached, the mining is terminated, and the cluster centers and membership matrix are output.

[0008] Preferably, the process of calculating the objective function of the dissimilarity index is as follows: For any cluster, the membership degree of each sample to its corresponding cluster center is used as the base and the predefined weighted index is used as the index to calculate the exponential membership degree. The square of the Euclidean distance between each sample and its corresponding cluster center is calculated, multiplied by the corresponding exponential membership degree, and then summed to obtain the dissimilarity index of each cluster center. The dissimilarity indices of all cluster centers are summed to obtain the objective function of the dissimilarity index.

[0009] Preferably, the process of updating cluster centers and membership is as follows: In each iteration, the cluster centers are updated based on the current membership matrix. For any cluster, the data vector of each sample is multiplied by the exponential membership of the corresponding sample to the corresponding cluster center and the sum is used as the numerator. The exponential memberships of all samples to the corresponding cluster centers are summed as the denominator to calculate the cluster center of the corresponding cluster at the current iteration number.

[0010] Preferably, the process of updating the membership degree is as follows: For any membership degree The Euclidean distance between the sample j corresponding to the membership degree and the updated cluster center is used as the numerator, and the Euclidean distance between the corresponding sample j and each updated cluster center is used as the denominator. The ratio of the numerator to the denominator is used as the base, and the reciprocal of the predefined weighted index minus 1 is used as the exponent to calculate the membership degree ratio index of the corresponding sample j to each cluster center. The reciprocal of the sum of all the membership degree ratio indices of the corresponding sample j is used as the updated membership degree.

[0011] Preferably, after the clustering process converges, the output cluster centers are the key voltage parameters that have been mined. If the cluster center is an arbitrary voltage sample, the corresponding matching power sample is output. If the cluster center is not an arbitrary voltage sample, the power sample data is replaced with the corresponding voltage sample data by using the cluster center update formula based on the final membership matrix to obtain the power data corresponding to each cluster center.

[0012] Preferably, in S2, for any node, the squares of the voltage amplitudes of the corresponding node and the other node in each branch are added together, and then the product of twice the voltage amplitude and the cosine of the voltage phase difference is subtracted. This product is then multiplied by the conductance of the corresponding branch to obtain the active power loss of the corresponding node and the corresponding branch. The active power losses of all nodes are accumulated and used as the voltage control objective function. The constraints of the voltage control objective function include: node power balance constraints, node voltage amplitude constraints, generator reactive power output constraints, branch current constraints, control equipment operation frequency constraints, and transformer turns ratio constraints.

[0013] Preferably, the process of calculating the sensitivity of total active power loss to reactive power at each node is as follows: Using the key voltage parameters and power samples corresponding to the cluster as initial values, power flow calculations are performed to determine the corresponding stable operating point; based on the stable operating point data and grid parameters, the Jacobian matrix is ​​calculated; and the key voltage parameters corresponding to the cluster are substituted into the voltage control objective function to determine the corresponding total active power loss expression. For any node, calculate the partial derivatives of active power loss with respect to phase angle and voltage; establish the relationship equations between the change in active power injection and the changes in voltage amplitude and phase angle, as well as the relationship equations between the change in reactive power injection and the changes in voltage amplitude and phase angle; set the change in active power injection at each node to 0, and solve for the voltage-reactive power expression and the phase angle-reactive power relationship expression; substitute the Jacobian matrix into the relationship expressions to obtain the partial derivatives of phase angle with respect to reactive power and the partial derivatives of voltage with respect to reactive power. Multiply the partial derivative of active power loss with respect to phase angle by the partial derivative of phase angle with respect to reactive power to obtain the partial derivative of phase angle term; calculate the product of the partial derivative of active power loss with respect to voltage and the partial derivative of voltage with respect to reactive power to obtain the partial derivative of voltage term; add the partial derivative of phase angle term and the partial derivative of voltage term to obtain the partial derivative of active power loss with respect to reactive power, which is the sensitivity of active power loss to reactive power at the node.

[0014] Preferably, the process of solving the linearized voltage control objective function is as follows: Under constraints, the reactive power change at each node is solved to generate the optimal control strategy, which is the amount of reactive power that the control equipment corresponding to the node needs to adjust.

[0015] A second aspect of the present invention provides an automatic voltage control system based on data mining, using an automatic voltage control method based on data mining, comprising: The data filtering module collects and processes voltage and power data from various nodes in the power grid to form voltage and power samples, and performs sample matching; it also mines the voltage samples through clustering to obtain key voltage parameters and corresponding power samples. The voltage control model construction module, based on key voltage parameters and corresponding power samples, constructs a voltage control objective function with the goal of minimizing the total active power loss of the power grid, and sets constraints. The voltage control model solution module acquires the voltage data of each node at the current moment, calculates the distance between the key voltage parameters of each cluster and the voltage data of each node at the current moment, and determines the cluster with the smallest distance. Based on the key voltage parameters and power samples corresponding to the clusters, it calculates the sensitivity of the total active power loss to the reactive power of each node. Substituting the voltage data of each node at the current moment into the voltage control objective function, it calculates the active power loss at the current operating point, and linearizes the voltage control objective function by combining the sensitivity. Under constraints, it solves the linearized voltage control objective function.

[0016] Compared with the prior art, the beneficial effects of the present invention include at least the following: 1. This invention utilizes data mining technology to cluster voltage samples from historical operating data, extracting representative key voltage parameters and corresponding power samples from a large number of complex power grid operating states. This enables automatic identification and classification of typical operating conditions. Compared to traditional automatic voltage control methods that rely on fixed operating scenarios for optimization, this invention can fully exploit the operating patterns contained in historical data, accurately reflect the operating characteristics of the power grid under different load levels and power flow distributions, and improve the adaptability and scenario coverage of the automatic voltage control model to actual operating conditions.

[0017] 2. This invention aims to minimize the total active power loss of the power grid. Combining key voltage parameters and power samples obtained through clustering, it establishes a quantitative relationship between active power loss and reactive power regulation through power flow calculations, Jacobian matrix analysis, and sensitivity calculations of active power loss to reactive power at nodes. The objective function is then linearized. Under the premise of ensuring that constraints such as node power balance, node voltage, generator reactive power output, branch current, and transformer turns ratio are met, the reactive power regulation of each node can be solved rapidly. This method can effectively reduce active power loss in the power grid, improve voltage quality and reactive power optimization control accuracy, while reducing the computational complexity in traditional nonlinear optimization processes and improving online control response speed and real-time optimization capabilities.

[0018] 3. This invention matches typical operating modes obtained through data mining with real-time operating states. By calculating the distance between the current node voltage and the key voltage parameters of each cluster, it quickly determines the closest operating scenario and calls upon the corresponding power samples and sensitivity information for optimized control. This eliminates the need to rebuild the model and calculate sensitivity from global operating data each time, significantly reducing the computational load for online optimization and improving the operating efficiency and stability of the automatic voltage control system. Simultaneously, this invention can dynamically generate optimal reactive power control strategies based on real-time operating states, guiding the coordinated adjustment of generators, reactive power compensation devices, and on-load tap-changing transformers to achieve precise control of the grid voltage level, enhancing the grid's adaptability and operational reliability under load fluctuations and changes in operating modes. Attached Figure Description

[0019] Figure 1 This is a flowchart of an automatic voltage control method based on data mining proposed in this invention; Figure 2 This is a comparison of line loss before and after voltage control in an embodiment of the present invention; Figure 3 This is a comparison of control time under different control methods in the embodiments of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this invention. The embodiments described in this application are merely some embodiments of this invention, and not all embodiments. Based on the spirit of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of this invention.

[0021] Example 1 Embodiment 1 of the present invention provides an automatic voltage control method based on data mining, the overall process of which is as follows: Figure 1 As shown. This method uses data mining techniques to pre-extract voltage operating characteristics, and then uses these characteristics as a basis for rapid and precise optimization control. The specific steps are as follows: S1. Collect and process voltage and power data from each node in the power grid to form voltage and power samples, and perform sample matching; mine the voltage samples through clustering to obtain key voltage parameters and corresponding power samples.

[0022] By setting a fixed-duration sliding time window (e.g., 1 hour), the continuously collected node voltage and reactive power time-series data are truncated, and the time-series data of multiple nodes within each window are stitched together to form a sample. The samples cover different time periods and operating conditions. After anomaly cleaning and normalization, voltage and power samples are formed for pattern mining. Voltage and power samples under the same time and operating conditions are then matched.

[0023] In this embodiment, the data acquisition interval can be set to 15 minutes, with a total acquisition duration of 365 days to ensure the comprehensiveness and representativeness of the data. The sample division is automatically performed based on the temporal continuity of the sliding window and data integrity, without relying on manually predefined categories. Each sample represents a complete snapshot of the power grid's operation within a specific time period, ensuring that the dataset includes steady-state, transient, and various transitional operating conditions, providing comprehensive input for unsupervised clustering.

[0024] In the unsupervised clustering framework, samples are not pre-labeled. By extracting composite features such as voltage dynamic characteristics and reactive power sensitivity for each sample and then clustering them, the samples are classified into different voltage operation mode categories according to their membership degree. This mode category serves as its data-driven dynamic label, which is used for accurate matching of subsequent control strategies.

[0025] As a preferred implementation method, fuzzy C-means clustering is used to mine the dataset; the specific process is as follows: Set the number of clusters and weighting index, and randomly initialize the membership matrix and cluster centers; the membership matrix satisfies the membership constraint, that is, the sum of the membership degrees of all samples to each cluster center is always 1; the specific formula is: ; In the formula, This refers to the number of clusters, i.e., the number of predefined clusters. Indicates the first j The nth sample pair i The membership degree of each cluster center; n This represents the number of samples in the voltage dataset. Under the premise of satisfying the membership degree constraint, for any cluster, the exponential membership degree is calculated using the membership degree of each sample to its corresponding cluster center as the base and a predefined weighted index as the exponent. The square of the Euclidean distance between each sample and its corresponding cluster center is calculated, multiplied by the corresponding exponential membership degree, and then summed to obtain the dissimilarity index for each cluster center. The dissimilarity indices of all cluster centers are summed to obtain the objective function of the dissimilarity index; specifically expressed as: ; in, m The weighted index is usually set to 2; dij For the sample j With cluster center i The Euclidean distance between them; c i The cluster centers of the fuzzy sets are used to divide the voltage dataset into... c F Calculate the cluster center of each fuzzy subset; Cluster center i Dissimilarity index; The cluster centers and membership degrees are iteratively updated by minimizing the dissimilarity index objective function; the specific process is as follows: In each iteration, the cluster centers are updated based on the current membership matrix. For any cluster, the data vector of each sample is multiplied by the exponential membership degree of the corresponding sample to the corresponding cluster center, and the sum is used as the numerator. The exponential membership degrees of all samples to the corresponding cluster centers are summed as the denominator to calculate the cluster center of the corresponding cluster at the current iteration number. The specific formula is as follows: ; In the formula, X j Represents the first in the voltage dataset j The data vector of each sample; Based on the updated cluster centers, update the corresponding membership degrees; for any membership degree... The Euclidean distance between the sample j corresponding to the membership degree and the updated cluster center is used as the numerator, and the Euclidean distance between the corresponding sample j and each updated cluster center is used as the denominator. Using the ratio of the numerator to the denominator as the base, and the reciprocal of the predefined weighted index minus 1 as the exponent, the membership degree ratio of the corresponding sample j to each cluster center is calculated. The reciprocal of the sum of all membership degree ratios of the corresponding sample j is used as the updated membership degree. The specific update formula is as follows: ; In the formula, when That is, when the k-th cluster center coincides with the j-th sample, the corresponding membership degree Updated to 1; Based on the updated cluster centers and membership degrees, calculate the objective function value of the dissimilarity index at the current iteration number; The sum of the squares of the Euclidean distances between all samples in each cluster and the cluster center is calculated and then accumulated to represent the convergence error; specifically, it is expressed as: ; In the formula, This is the convergence error; Let i be the set of samples in the i-th cluster; For clustersi The first in l One sample; For this is the first i The cluster centers of each cluster; Indicates Euclidean distance; When the change in convergence error over adjacent iterations is less than a preset threshold, or the difference in the objective function value of the dissimilarity index is less than a predetermined difference threshold, or the maximum number of iterations is reached, clustering is considered converged, and the mining process is terminated. The output cluster centers and membership matrix are the key voltage parameters, providing an accurate data foundation for subsequent control.

[0026] Once the clustering process converges, the output cluster centers are the key voltage parameters mined from massive historical data, and the corresponding power samples are also output. Each cluster center... c i It is d The dimensional vector represents a typical, global steady-state voltage distribution pattern in a power system; where d is the feature dimension of the sample, for example, if data is collected every 15 minutes throughout the day, then d is 96; each cluster center contains voltage data from all nodes in the distribution network; if the cluster center is any voltage sample, the corresponding matching power sample is output; if the cluster center is not any voltage sample, based on the final membership matrix, the cluster center update formula is used to replace the corresponding voltage sample data with the power sample data to obtain the power data corresponding to each cluster center.

[0027] In this embodiment, the clustering process only clusters voltage samples, but each voltage sample has a corresponding power sample for a given time. Optionally, after clustering is completed, the power sample corresponding to each voltage cluster center can be retrieved; for example, for each voltage mode, the average power vector is calculated using the power data of all samples within the cluster (or the power data of the nearest sample to the cluster center is matched within the cluster), thereby obtaining the power data corresponding to that voltage mode.

[0028] The active and reactive power data used in the objective function are derived from these matched power samples.

[0029] S2. Based on key voltage parameters and corresponding power samples, construct a voltage control objective function with the goal of minimizing the total active power loss of the power grid; set constraints on the voltage control objective function, including: node power balance constraints, node voltage amplitude constraints, generator reactive power output constraints, branch current constraints, control equipment operation frequency constraints, and transformer turns ratio constraints.

[0030] In a preferred implementation, for any node, the active power loss of the corresponding node and the other node in each branch is obtained by adding the squares of their voltage amplitudes, subtracting twice the product of the voltage amplitude and the cosine of the voltage phase difference, and multiplying by the conductance of the corresponding branch. The active power losses of all nodes are then summed to obtain the voltage control objective function. The voltage control objective function is: ; in, Total active power loss; branch road UV electrical conductivity, branch road UV The resistance, branch road UV The reactance; The number of nodes in the power grid; v For nodes u Another node on the branch road, For nodes u The set of nodes directly connected by branches; , They are nodes u and v The voltage amplitude; , They are nodes u and v The voltage phase angle.

[0031] As a preferred implementation, the node power balance constraint is: ; In the formula, For injection nodes u The active power; For injection nodes u reactive power; Indicates at the compute node u When calculating the power flow equations, summation is performed on all nodes directly connected to it; Let be the conductance of the corresponding node in the nodal admittance matrix; Let be the susceptance of the corresponding node in the node admittance matrix; To represent a node u With nodes v The offset of the voltage waveform on the time axis, i.e., the node u With nodes v The voltage phase angle difference between them; The node voltage amplitude constraint is: ;in, , These are the minimum and maximum values ​​of the node voltage amplitude, respectively; in this embodiment, the range is set to [0.95, 1.05]. The reactive power output constraint of the generator is: ;in, For reactive power output from the generator; , These are the minimum and maximum values ​​of the generator output, respectively. Transformer turns ratio constraint: ;in, The transformer turns ratio of the transformer on branch line uv; , These are the minimum and maximum values ​​of the transformer turns ratio, respectively. Branch current constraints: ;in, Let be the branch current of branch uv; , These are the minimum and maximum values ​​of the branch current, respectively; Constraints on the number of control device actions: ;in, The number of times the control equipment on the branch UV is activated; This represents the maximum number of times the device can perform actions.

[0032] S3. Obtain the voltage data of each node at the current moment, calculate the distance between the key voltage parameters of each cluster in S1 and the voltage data of each node at the current moment, and determine the cluster with the smallest distance; based on the key voltage parameters and power samples corresponding to the clusters, calculate the sensitivity of the total active power loss to the reactive power of each node; substitute the voltage data of each node at the current moment into the voltage control objective function, calculate the active power loss at the current operating point, and linearize the voltage control objective function by combining the sensitivity; under the constraints, solve the linearized voltage control objective function.

[0033] Directly solving the nonlinear voltage control objective function is complex and time-consuming. Therefore, this embodiment linearizes it.

[0034] Obtain the current voltage amplitude and phase angle of each node at the current moment, and substitute them into the voltage control objective function to calculate the active power loss at the current moment, which is taken as the active power loss at the current operating point; perform a first-order Taylor expansion of the voltage control objective function near the current operating point, ignoring higher-order terms, and convert it into a function based on the change in reactive power at the nodes, Δ. Q A linear function of variables; specifically expressed as: ; in, This represents the active power loss at the current operating point. Δ represents the sensitivity of active power loss to reactive power at node n, i.e., the impact of changes in reactive power at node n on active power loss; Q n This represents the change in reactive power at the node before and after voltage control. Under constraints, the reactive power change at each node is solved to generate the optimal control strategy; that is, the reactive power (in var) that the control equipment corresponding to the node needs to adjust. For continuous equipment such as generators and SVG, their reactive power setpoints can be directly modified. For discrete equipment such as capacitors / reactors, the number of switching groups is obtained by rounding down the capacity of a single group. For transformer tap changes, their tap changes can be used as independent decision variables or back-calculated through equivalent reactive power sensitivity, thereby ensuring that the optimization results can be accurately mapped to the actual control commands of each device.

[0035] As a preferred implementation, the Euclidean distance between the key voltage parameters of each cluster in S1 and the voltage data of each node at the current moment is calculated, and the cluster with the smallest distance is determined as the cluster to which the voltage data of each node at the current moment belongs; based on the key voltage parameters and power samples corresponding to the cluster, the sensitivity of active power loss to reactive power of each node is calculated; the specific process is as follows: Using the key voltage parameters and power samples corresponding to the respective cluster as initial values, power flow calculations are performed to determine the corresponding stable operating point; based on the stable operating point data and grid parameters (i.e., the node admittance matrix), the Jacobian matrix is ​​calculated; substituting the key voltage parameters corresponding to the respective cluster into the voltage control objective function, the corresponding total active power loss expression is determined; for any node... u Calculate the partial derivative of active power loss with respect to the phase angle; specifically expressed as:

[0036] In the formula, For nodes u Self-conductivity; For any node u Calculate the partial derivative of active power loss with respect to voltage; specifically expressed as: ; In the formula, branch road UV The electrical conductivity; Establish the relationship between the changes in active and reactive power injection at nodes and the changes in voltage amplitude and phase angle, that is, linearize the power flow equations as follows:

[0037] in, , These are the matrices representing the changes in active and reactive power injection, respectively. , These are matrices representing the changes in voltage magnitude and phase angle, respectively. This represents the partial derivative of active power injection with respect to the voltage phase angle; This represents the partial derivative of active power injection with respect to voltage amplitude; This represents the partial derivative of reactive power injection with respect to the voltage phase angle; This represents the partial derivative of reactive power injection with respect to voltage amplitude; the Jacobian matrix J is defined as: ; In the formula, This indicates the sensitivity of active power to the voltage phase angle, i.e. ; This indicates the sensitivity of active power to voltage amplitude, i.e. ; This indicates the sensitivity of reactive power to the voltage phase angle, i.e. ; This indicates the sensitivity of reactive power to voltage amplitude, i.e. ; It is typically assumed that active power is injected into each node. P i If the above relationship remains unchanged, i.e., ΔP=0; transforming the above relationship into equations relating the change in active power injection to the changes in voltage amplitude and phase angle, and the change in reactive power injection to the changes in voltage amplitude and phase angle, we can obtain:

[0038] Setting the change in active power injection at each node to zero, the phase angle-voltage relationship between Δθ and ΔV can be solved from the first relational equation; specifically, it is expressed as: ; Based on the phase angle-voltage relationship expression, the phase angle in the second relationship equation is converted into voltage, and the reactive power-voltage sensitivity matrix is ​​defined. This yields the voltage-reactive power expression; specifically, it is expressed as:

[0039]

[0040] Substituting the phase angle-voltage relationship expression into the voltage-reactive power expression, we obtain the phase angle-reactive power relationship expression: ; Substituting the Jacobian matrix into the phase angle-reactive power relationship and the voltage-reactive power expression, we obtain the partial derivatives of the phase angle with respect to reactive power and the partial derivatives of the voltage with respect to reactive power; specifically expressed as: ; In the formula, This is the partial derivative of the phase angle with respect to reactive power; This is the partial derivative of voltage with respect to reactive power; Multiply the partial derivative of active power loss with respect to phase angle by the partial derivative of phase angle with respect to reactive power to obtain the partial derivative of phase angle; calculate the product of the partial derivative of active power loss with respect to voltage and the partial derivative of voltage with respect to reactive power to obtain the partial derivative of voltage; add the partial derivatives of phase angle and voltage to obtain the partial derivative of active power loss with respect to reactive power, i.e., the sensitivity of active power loss to reactive power at the node; specifically expressed as: ; In the formula, This is the sensitivity matrix of active power loss to reactive power at nodes; the final result is a 1×N row vector, where the nth element is the sensitivity of node n. .

[0041] Figure 2 The comparison of line losses before and after voltage control is further demonstrated. It can be seen that before control, the line loss fluctuates greatly, reaching a maximum of 280kW; after control, the line loss is stably controlled at a low level, with the maximum value reduced to 75kW, proving the excellent loss reduction effect of the present invention. Figure 3 The control time of this invention was compared with that of existing methods. The results show that the maximum control time of this invention does not exceed 5 seconds, while the maximum control time of methods based on network topology resources and model predictive control both exceed 43 seconds. This fully demonstrates the speed advantage brought by this invention through data mining and model linearization.

[0042] Example 2 Embodiment 2 of the present invention provides an automatic voltage control system based on data mining, which executes an automatic voltage control method based on data mining as described in Embodiment 1, including: The data filtering module collects and processes voltage and power data from various nodes in the power grid to form voltage and power samples, and performs sample matching; it also mines the voltage samples through clustering to obtain key voltage parameters and corresponding power samples. The voltage control model construction module, based on key voltage parameters and corresponding power samples, constructs a voltage control objective function with the goal of minimizing the total active power loss of the power grid, and sets constraints. The voltage control model solution module acquires the voltage data of each node at the current moment, calculates the distance between the key voltage parameters of each cluster and the voltage data of each node at the current moment, and determines the cluster with the smallest distance. Based on the key voltage parameters and power samples corresponding to the clusters, it calculates the sensitivity of the total active power loss to the reactive power of each node. Substituting the voltage data of each node at the current moment into the voltage control objective function, it calculates the active power loss at the current operating point, and linearizes the voltage control objective function by combining the sensitivity. Under constraints, it solves the linearized voltage control objective function.

[0043] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.

[0044] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination of the foregoing. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0045] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0046] Computer program instructions used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing the status information of the computer-readable program instructions to implement various aspects of this disclosure.

[0047] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the claims of the present invention.

Claims

1. An automatic voltage control method based on data mining, characterized in that, Includes the following steps: Voltage and power data from each node in the power grid are collected and processed to form voltage and power samples, and sample matching is performed. Key voltage parameters and corresponding power samples are obtained by mining the voltage samples through clustering. Based on key voltage parameters and corresponding power samples, a voltage control objective function is constructed with the goal of minimizing the total active power loss of the power grid, and constraints are set. Obtain the voltage data of each node at the current moment, calculate the distance between the key voltage parameters of each cluster and the voltage data of each node at the current moment, and determine the cluster with the smallest distance; based on the key voltage parameters and power samples corresponding to the clusters, calculate the sensitivity of the total active power loss to the reactive power of each node; substitute the voltage data of each node at the current moment into the voltage control objective function, calculate the active power loss at the current operating point, and linearize the voltage control objective function by combining the sensitivity. Under constraints, the linearized voltage control objective function is solved.

2. The automatic voltage control method based on data mining according to claim 1, characterized in that: The process of clustering voltage samples is as follows: Set the number of clusters and the weighting index, and randomly initialize the membership matrix and cluster centers; the membership matrix satisfies the membership constraint, that is, the sum of the membership degrees of all samples to each cluster center is always 1; Under the premise of satisfying the membership degree constraint, the dissimilarity index objective function is calculated based on the membership degree of each sample to the corresponding cluster center and the voltage sample. The cluster centers and membership degrees are iteratively updated by minimizing the dissimilarity index objective function; based on the updated cluster centers and membership degrees, the dissimilarity index objective function value at the current iteration number is calculated. The sum of the squares of the Euclidean distances between all samples in each cluster and the cluster center is calculated and then accumulated to form the convergence error. When the change in convergence error over adjacent iterations is less than a preset threshold, or the difference in the objective function value of the dissimilarity index is less than a predetermined difference threshold, or the maximum number of iterations is reached, the mining process terminates, and the cluster centers and membership matrix are output.

3. The automatic voltage control method based on data mining according to claim 2, characterized in that: The process of calculating the objective function of the dissimilarity index is as follows: For any cluster, the membership degree of each sample to its corresponding cluster center is used as the base and the predefined weighted index is used as the index to calculate the exponential membership degree. The square of the Euclidean distance between each sample and its corresponding cluster center is calculated, multiplied by the corresponding exponential membership degree, and then summed to obtain the dissimilarity index of each cluster center. The dissimilarity indices of all cluster centers are summed to obtain the objective function of the dissimilarity index.

4. The automatic voltage control method based on data mining according to claim 2, characterized in that: The process of updating cluster centers and membership is as follows: In each iteration, the cluster centers are updated based on the current membership matrix. For any cluster, the data vector of each sample is multiplied by the exponential membership of the corresponding sample to the corresponding cluster center and the sum is used as the numerator. The exponential memberships of all samples to the corresponding cluster centers are summed as the denominator to calculate the cluster center of the corresponding cluster at the current iteration number.

5. The automatic voltage control method based on data mining according to claim 2, characterized in that: The process of updating membership is as follows: For any membership degree The Euclidean distance between the sample j corresponding to the membership degree and the updated cluster center is used as the numerator, and the Euclidean distance between the corresponding sample j and each updated cluster center is used as the denominator. The ratio of the numerator to the denominator is used as the base, and the reciprocal of the predefined weighted index minus 1 is used as the exponent to calculate the membership degree ratio index of the corresponding sample j to each cluster center. The reciprocal of the sum of all the membership degree ratio indices of the corresponding sample j is used as the updated membership degree.

6. The automatic voltage control method based on data mining according to claim 4, characterized in that: Once the clustering process converges, the output cluster centers are the key voltage parameters that have been mined. If the cluster center is any voltage sample, the corresponding matching power sample is output. If the cluster center is not any voltage sample, the power sample data is replaced with the corresponding voltage sample data by using the cluster center update formula based on the final membership matrix to obtain the power data corresponding to each cluster center.

7. The automatic voltage control method based on data mining according to claim 1, characterized in that: In S2, for any node, the active power loss of the corresponding node and the other node of each branch is obtained by adding the squares of the voltage amplitude of the corresponding node and subtracting the product of twice the voltage amplitude and the cosine of the voltage phase difference, and multiplying by the conductance of the corresponding branch. The active power losses of all nodes are summed up and used as the target function for voltage control. The constraints of the voltage control objective function include: node power balance constraints, node voltage amplitude constraints, generator reactive power output constraints, branch current constraints, control equipment operation frequency constraints, and transformer turns ratio constraints.

8. The automatic voltage control method based on data mining according to claim 1, characterized in that: The process of calculating the sensitivity of total active power loss to reactive power at each node is as follows: Using the key voltage parameters and power samples corresponding to the cluster as initial values, power flow calculations are performed to determine the corresponding stable operating point; Based on stable operating point data and grid parameters, the Jacobian matrix is ​​calculated; the key voltage parameters corresponding to the cluster are substituted into the voltage control objective function to determine the corresponding total active power loss expression. For any node, calculate the partial derivatives of active power loss with respect to phase angle and voltage. Establish the relationship equations between the change in active power injection and the changes in voltage amplitude and phase angle, as well as the relationship equations between the change in reactive power injection and the changes in voltage amplitude and phase angle; Set the change in active power injection at each node to 0, and solve for the voltage-reactive power expression and the phase angle-reactive power relationship expression; substitute the Jacobian matrix into the relationship expression to obtain the partial derivatives of the phase angle with respect to reactive power and the partial derivatives of the voltage with respect to reactive power. Multiply the partial derivative of active power loss with respect to phase angle by the partial derivative of phase angle with respect to reactive power to obtain the partial derivative of phase angle term; calculate the product of the partial derivative of active power loss with respect to voltage and the partial derivative of voltage with respect to reactive power to obtain the partial derivative of voltage term; add the partial derivative of phase angle term and the partial derivative of voltage term to obtain the partial derivative of active power loss with respect to reactive power, which is the sensitivity of active power loss to reactive power at the node.

9. The automatic voltage control method based on data mining according to claim 1, characterized in that: The process of solving the linearized voltage control objective function is as follows: Under constraints, the reactive power change at each node is solved to generate the optimal control strategy, which is the amount of reactive power that the control equipment corresponding to the node needs to adjust.

10. An automatic voltage control system based on data mining, using the method described in any one of claims 1-9, characterized in that, include: The data filtering module collects and processes voltage and power data from various nodes in the power grid to form voltage and power samples, and performs sample matching; it also mines the voltage samples through clustering to obtain key voltage parameters and corresponding power samples. The voltage control model construction module, based on key voltage parameters and corresponding power samples, constructs a voltage control objective function with the goal of minimizing the total active power loss of the power grid, and sets constraints. The voltage control model solving module acquires the voltage data of each node at the current moment, calculates the distance between the key voltage parameters of each cluster and the voltage data of each node at the current moment, and determines the cluster with the smallest distance; based on the key voltage parameters and power samples corresponding to the clusters, it calculates the sensitivity of the total active power loss to the reactive power of each node; substitutes the voltage data of each node at the current moment into the voltage control objective function, calculates the active power loss at the current operating point, and linearizes the voltage control objective function by combining the sensitivity. Under constraints, the linearized voltage control objective function is solved.

Images