Multi-time scale edge-end collaborative voltage treatment method based on output control of photovoltaic inverter

Abstract

A multi-time scale edge-end collaborative voltage treatment method based on output control of a photovoltaic inverter comprises the following steps: constructing a long-term scale power grid model: dynamically collecting operation data of a power distribution network, and constructing an accurate power grid model; generating a short-term-scale centralized optimization instruction: based on the accurate power grid model, taking the minimum voltage deviation of a measurement point as a target, combining a power flow constraint and a voltage safety constraint, calculating a voltage control instruction of each photovoltaic inverter through a mixed integer second-order cone optimization model, and issuing the voltage control instruction to an edge terminal; and local voltage regulation of a real-time scale: monitoring a node voltage state through an edge terminal, and executing multi-stage response: preferentially responding and executing a control instruction issued by the short-term scale; if the control instruction does not exist, reactive output of the inverter is adjusted according to a variable parameter droop control strategy based on local real-time voltage data; wherein the periods of the three scales are shortened in sequence, and an edge-end collaborative data interaction process is formed.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of power systems, and particularly relates to power distribution network operation control technology, and especially relates to a multi-time scale edge-end collaborative voltage control technology based on photovoltaic inverter output control in a high proportion distributed photovoltaic grid-connected scenario, and particularly relates to a multi-time scale edge-end collaborative voltage control method based on photovoltaic inverter output control, which is used to solve the voltage fluctuation problem caused by high proportion distributed photovoltaic access to the power distribution network, and improve the stability of power grid operation and new energy consumption capacity. BACKGROUND

[0002] With the gradual advancement of the "double carbon" goal and the rapid development of new energy generation technology, the penetration rate of distributed photovoltaic in the power distribution network is continuously increasing. However, the intermittency and volatility of distributed photovoltaic lead to frequent voltage out-of-limit of the power distribution network, seriously affecting the power quality and power supply reliability of the power distribution network. Traditional voltage regulation methods mainly rely on on-load tap-changing transformers (OLTC) and shunt capacitor bank switching in substations, but this kind of centralized regulation method has slow response speed and large information processing capacity, and is difficult to adapt to the demand for rapid change of photovoltaic output in the distribution network. In addition, with the transformation of the power distribution network from a single radial structure to a multi-source interactive active power distribution network, centralized control by the main station alone cannot meet the requirements of fine voltage regulation.

[0003] At present, domestic and foreign researches mainly adopt two ways for photovoltaic grid-connected voltage control: one is local control strategy based on inverter reactive power regulation, such as constant power factor (PF) control, constant voltage (VQ) control, etc., but this kind of method lacks global optimization and is prone to cause multi-inverter regulation conflict; the other is centralized voltage control strategy based on main station optimization, which calculates the global regulation instruction by the main station through state estimation and optimal power flow, but is easily limited by communication delay and calculation complexity, and is difficult to achieve second-level fast response. In addition, most of the existing methods only focus on real-time regulation and lack the ability to predict load and photovoltaic output, and cannot realize forward-looking voltage preventive control.

[0004] At the same time, with the steady advancement of new-type power system construction at home and abroad, the power distribution network presents the complexity characteristics of "source-grid-load-storage" multi-element interaction. On the one hand, photovoltaic inverters, energy storage systems and other distributed resources have the ability of fast regulation, providing new means for voltage control; on the other hand, the collaborative optimization of a large number of distributed devices is limited by communication bandwidth, computing resources and other factors. In summary, the traditional centralized or completely distributed control architecture is difficult to balance the demand for global optimization and fast response. Therefore, how to build a multi-time scale edge-end collaborative voltage control system and realize the whole process voltage control from hour-level prediction optimization to second-level dynamic regulation has become a key problem to be solved at present.

[0005] In domestic and foreign research, the prediction technology based on artificial intelligence and the edge computing architecture provide new ideas for the above problems. However, the existing methods still have deficiencies in long-term and short-term prediction cooperation, edge-end control right distribution, multi-objective optimization and the like. Especially in the high proportion photovoltaic access scene, the voltage out-of-limit problem is more prominent, and an intelligent voltage management method capable of fusing prediction optimization and real-time regulation, and taking into account global optimization and local rapid response is urgently needed to improve the safety of power distribution network operation and the new energy consumption capacity. SUMMARY

[0006] In view of the defects and deficiencies of the prior art, the present application provides a multi-time scale edge-end collaborative voltage management method based on photovoltaic inverter output control, which fuses dynamic topology identification, intelligent prediction optimization and edge computing technology by constructing a three-layer collaborative architecture of "long-term model dynamic updating-short-term prediction centralized optimization-real-time local rapid regulation", and realizes accurate voltage control in the high proportion photovoltaic access scene.

[0007] On the long-term scale, the present application dynamically identifies the power distribution network topology through voltage curve feature clustering analysis and graph theory algorithm, optimizes the line electrical parameters in combination with the sequence quadratic programming model, and constructs an accurate power grid model which is self-adaptively updated according to the operating state; on the short-term scale, the load data is classified and predicted by K-means clustering and long short-term memory neural network, and the photovoltaic inverter control instructions of global collaboration are generated based on the mixed integer second-order cone optimization model with the minimum voltage deviation as the target; on the real-time scale, the multi-level response mechanism is executed by the edge terminal, the instructions issued by the main station are preferentially responded, and when there is no instruction, the photovoltaic inverter reactive power output is autonomously regulated based on the variable parameter droop control strategy with preset multi-level voltage threshold, forming a collaborative closed loop of "edge-side global optimization-terminal local rapid response".

[0008] The management periods of the three time scales are shortened in turn, wherein the short-term optimization period is once per hour, the real-time regulation period is in the order of seconds to minutes, and the edge terminal uploads the monitoring data and control results at a fixed period, supporting the long-term and short-term model iterative optimization and the main station system control effect evaluation (including the inverter regulation amount compliance rate, node voltage fluctuation amplitude and the like indicators). Through the multi-time scale cooperation and edge-end architecture innovation, the present application effectively fuses the global optimization and local rapid response capability, can significantly suppress the voltage fluctuation caused by high proportion photovoltaic access, and improves the safety of power distribution network operation and the new energy consumption capacity.

[0009] The present application specifically adopts the following technical means:

[0010] A multi-time scale edge-end collaborative voltage management method based on photovoltaic inverter output control, comprising:

[0011] Long-term scale power grid model construction: dynamically collect power distribution network operation data, abstract the power distribution network as an undirected weighted graph, determine the transformer and feeder attribution based on node voltage curve feature clustering analysis, determine the line topology connection relationship using graph theory algorithm, and optimize the line impedance and reactance parameters through a sequential quadratic programming model to construct an accurate power grid model;

[0012] Short-term scale centralized optimization instruction generation: after standardizing and preprocessing the historical load data, K-means clustering classification is used to establish a long short-term memory neural network prediction model for each type of load to obtain the next period node load. Based on the accurate power grid model, the measured point voltage deviation is minimized as the target, combined with the power flow constraint and voltage safety constraint, the voltage control instruction of each photovoltaic inverter is calculated through a mixed integer second-order cone optimization model and is sent to the edge terminal;

[0013] And real-time scale local voltage regulation: the edge terminal monitors the node voltage state and performs multi-level response: preferentially responding to and executing the control instruction at the short-term scale; if there is no control instruction, based on the local real-time voltage data, the inverter reactive power output is adjusted according to the variable parameter droop control strategy, and the strategy dynamically adjusts the reactive power output through a pre-set multi-level voltage threshold;

[0014] Wherein, the periods of the three scales are shortened in turn, and the edge terminal uploads the monitoring data and control results to the edge processing unit at a fixed period, and the edge processing unit integrates the data for long-term scale model updating and short-term scale optimization calculation, forming a data interaction process of edge-end cooperation.

[0015] Further, in the long-term scale power grid model construction, the determination of the line topology connection relationship using graph theory algorithm specifically includes:

[0016] Based on the clustering results of the voltage curve features, the Kruskal algorithm is used to calculate the maximum connected tree of the power distribution network topology by calculating the joint probability distribution maximum value of the power distribution network topology; and based on the transformer and feeder attribution relationship determined by the clustering analysis, the maximum connected tree is cut to obtain the topology connection relationship between the lines;

[0017] And the voltage curve feature clustering analysis further includes: constructing an adjacency matrix based on the node voltage curve, determining the adjacency node set and out-degree of each node to construct an out-degree matrix, and then forming a Laplacian matrix; after normalizing the Laplacian matrix, its eigenvalues are calculated, the first k eigenvectors are taken to form a sample matrix, and the attribution relationship of the transformer and feeder is obtained through clustering.

[0018] Further, in the long-term scale power grid model construction, the line impedance and reactance parameters are optimized based on a one-dimensional equivalent model of the low-voltage distribution network through a sequence quadratic programming model, and the effects of line admittance and conductance on voltage are ignored, and only the line resistance and reactance are taken as optimization objects;

[0019] The optimization objective of the sequence quadratic programming model is to minimize the average value of the upstream node voltage variance in the full time scale, and the constraint condition is that the line resistance and reactance are greater than 0; before optimization, the line impedance parameter initial value is set, wherein the line resistance is the product of the unit length resistance and the line length, and the line reactance is the product of the unit length reactance and the line length.

[0020] Further, in the short-term scale centralized optimization instruction generation, the historical load data standardization preprocessing specifically includes: first, the historical load data is scaled based on the Z-score standardization method, so as to eliminate the dimension influence; then, the historical load data set is divided into a training set and a test set, so as to train and verify the generalization ability of the long short-term memory neural network prediction model;

[0021] The long short-term memory neural network prediction model takes tanh as the activation function and the mean absolute error as the loss function, adjusts the model hidden layer number, neuron number and learning rate through offline training, and stops until the prediction error of the model on the test set reaches the optimum.

[0022] Further, in the short-term scale centralized optimization instruction generation, the minimum voltage deviation at the measurement point as the target specifically refers to a double-objective optimization of minimizing the sum of the branch network loss of the distribution network and the square of the voltage deviation of each node from the rated reference voltage;

[0023] The power flow constraint specifically includes: a node power balance constraint based on node net active power = photovoltaic active power output - load active demand and node net reactive power = photovoltaic reactive power output - load reactive demand; a branch current constraint based on the square of the branch current being less than the square of the maximum allowed branch current, and the branch current being calculated based on the branch active power, the reactive power and the node voltage;

[0024] The voltage safety constraint specifically refers to the square of the node voltage being not less than the square of the minimum allowed voltage and not greater than the square of the maximum allowed voltage.

[0025] Further, in the real-time scale local voltage regulation, the variable parameter droop control strategy dynamically adjusts the reactive power output by presetting a multi-level voltage threshold value, which specifically includes:

[0026] The preset multi-level voltage threshold value is U1, U2, U3 and U4, wherein U2 and U3 are the voltage reactive power droop control starting threshold value, and U1 and U4 are the maximum reactive power output boundary;

[0027] When the node voltage is less than or equal to U1, the inverter outputs maximum reactive power; when U1 is less than the node voltage and the node voltage is less than U2, the inverter reactive output is linearly adjusted according to the product of the maximum reactive power and (voltage-U2) / (U1-U2); when U2 is less than or equal to the node voltage and the node voltage is less than or equal to U3, the inverter reactive output is 0; when U3 is less than the node voltage and the node voltage is less than U4, the inverter reactive output is linearly adjusted according to the product of the minimum reactive power and (voltage-U3) / (U4-U3); when the node voltage is greater than or equal to U4, the inverter outputs minimum reactive power.

[0028] Further, in the local voltage regulation of the real-time scale, the inverter output also needs to meet the power constraint:

[0029] The inverter active output is between the minimum active output value and the maximum active output value; and the sum of the square of the real-time active output of the inverter and the square of the real-time reactive output does not exceed the square of the rated capacity of the inverter.

[0030] Further, the three scales are sequentially shortened as follows:

[0031] The grid model updating period of the long-term scale is periodically triggered, and the period is longer than the short-term scale of once per hour (for example, once per week); the centralized optimization instruction generation period of the short-term scale is once per hour, that is, the photovoltaic inverter control instruction is calculated and issued once per hour based on the load prediction result; the local voltage regulation period of the real-time scale is seconds to minutes, that is, the edge terminal monitors the node voltage state once every 1-5 minutes, and adjusts the inverter output as needed;

[0032] The edge terminal uploads the monitoring data and control results to the edge processing unit at a fixed period, and the fixed period is once every 15 minutes;

[0033] The data integrated by the edge processing unit includes node voltage monitoring values, inverter output power adjustment amounts, and control instruction execution states; in addition to being used for long-term grid model updating and short-term centralized optimization calculation, the integrated data is also submitted to the master station system for control effect evaluation, and the evaluation indexes include inverter adjustment amount compliance rate, node voltage fluctuation amplitude, and voltage out-of-limit times.

[0034] Further, in the grid model construction of the long-term scale, the step of calculating the voltage mutual information between nodes is further included:

[0035] The node discrete voltage curve is converted into a voltage distribution curve to represent the edge probability distribution of each node voltage; and based on the joint probability distribution and the edge probability distribution of the node voltage, the voltage mutual information between nodes is calculated, which is used to assist in determining the joint probability distribution of the distribution network topology and improve the recognition accuracy of the line topology connection relationship.

[0036] And a multi-time scale edge-end collaborative voltage management system based on photovoltaic inverter output control, comprising:

[0037] Edge terminal: deployed in the distribution transformer area, used for dynamically collecting power distribution network operation data and monitoring node voltage state; performing multi-level response: preferentially responding to the control instructions issued by the edge-side processing unit, and when there is no instruction, adjusting the reactive power output of the photovoltaic inverter according to the local real-time voltage data and the variable parameter droop control strategy; uploading the monitoring data and control results to the edge-side processing unit at a fixed period;

[0038] Edge-side processing unit: in communication connection with the edge terminal, used for: based on the data uploaded by the edge terminal, constructing and updating an accurate power grid model through voltage curve feature clustering, graph theory algorithm and sequential quadratic programming model; after standardizing and K-means clustering the historical load data, predicting the node load through a long short-term memory neural network, and then combining the accurate power grid model and a mixed integer second-order cone optimization model to generate voltage control instructions of the photovoltaic inverter and issue them to the edge terminal;

[0039] Master station system: in communication connection with the edge-side processing unit, used for receiving the integrated data of the edge-side processing unit, evaluating the control effect (including inverter adjustment amount, node voltage fluctuation amplitude, voltage out-of-limit times), and visually displaying the evaluation result;

[0040] Photovoltaic inverter: in communication connection with the edge terminal, used for receiving the adjustment instructions of the edge terminal, adjusting the active and reactive output power, and ensuring that the node voltage is stable within the allowed range;

[0041] The edge terminal, edge-side processing unit, master station system and photovoltaic inverter realize data interaction through an edge-end collaborative architecture, and correspond to voltage management scales that are successively shortened in long-term, short-term and real-time periods.

[0042] And a computer device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method as described above when executing the program.

[0043] A non-transitory computer readable storage medium having a computer program stored thereon, wherein the computer program is executable by a processor to implement the steps of the method as described above.

[0044] Compared with the prior art, the present application and the preferred schemes thereof at least have the following beneficial effects:

[0045] Firstly, the core defect of traditional voltage regulation mode that it is difficult to balance "global optimization" and "fast response" is effectively solved. Through the multi-time scale collaborative architecture of "long-term scale power grid model construction-short-term scale centralized optimization-real-time scale local regulation", combined with the edge-end collaborative mechanism, the global voltage deviation and network loss are optimized by the centralized calculation of the edge processing unit, avoiding the regulation conflict of multiple inverters caused by the lack of global information in pure local control. Through the local real-time regulation of the edge terminal, the limitation of traditional centralized control that cannot quickly respond to photovoltaic output fluctuation due to communication delay and computational complexity is broken, and the whole process of voltage management from forward-looking pre-control to sudden fluctuation response is realized.

[0046] Secondly, the accuracy and reliability of distribution network voltage regulation are significantly improved. In the long-term scale, the accurate power grid model constructed based on voltage curve feature clustering, graph theory algorithm and sequential quadratic programming model can dynamically adapt to the changes of distribution network topology and parameters, thereby providing accurate data basis for subsequent load prediction and voltage optimization, and avoiding the regulation deviation caused by parameter misalignment of traditional fixed model. In the short-term scale, the prediction method combining load clustering and long short-term memory neural network can obtain the trend of node load in advance, and the control instruction generated based on the mixed integer second-order cone optimization model realizes the forward-looking prevention of voltage out-of-limit. The variable parameter droop control strategy in real-time scale dynamically adjusts the reactive power output through multi-level voltage threshold, which can accurately match the regulation intensity according to the voltage state, avoiding the problem of excessive or insufficient regulation of traditional fixed parameter droop control, and further ensuring that the distribution network voltage is within the allowed stable operation range.

[0047] Thirdly, the burden of distribution network operation and regulation is reliably reduced to adapt to the actual needs of high proportion of photovoltaic access. The edge-end collaborative architecture sinks part of the computing tasks to the edge terminal and edge processing unit, reducing the data interaction amount and computing pressure of the main station, and avoiding the high dependence of traditional centralized control on communication bandwidth and main station computing power. At the same time, the closed-loop feedback mechanism can timely find the regulation deficiency and optimize the strategy through the evaluation of the control effect and data visualization by the main station, and continuously improve the governance effect. Finally, this method can effectively suppress the voltage fluctuation caused by high proportion of photovoltaic access, improve the stability of power grid operation, and provide technical support for new energy consumption, which meets the development needs of "source-grid-load-storage" multi-element interaction in the construction of new power system. BRIEF DESCRIPTION OF DRAWINGS

[0048] The application will be further described in detail below in combination with the drawings and specific embodiments:

[0049] Figure 1 is the implementation flowchart of the multi-time scale edge-end collaborative voltage management method of the embodiments of the application;

[0050] Figure 2 is a flow diagram of power distribution network topology identification updating of an embodiment of the application.

[0051] Figure 3 is a variable parameter droop control curve diagram of a photovoltaic inverter of an embodiment of the application. DETAILED DESCRIPTION

[0052] In the following, specific embodiments of the present application will be described in detail with reference to the accompanying drawings, and according to these detailed descriptions, those skilled in the art can clearly understand the present application and can implement the present application. The features in each different embodiment can be combined to obtain new implementation modes, or replace some features in some embodiments to obtain other preferred implementation modes, without departing from the principles of the present application.

[0053] The present application is to solve the problems of slow response speed, serious regulation conflict, insufficient new energy consumption capacity and the like of the existing voltage regulation technology, and through a collaborative mechanism combining long-term and short-term prediction optimization and real-time dynamic adjustment, fine regulation and control of the power distribution network voltage is realized, the response speed and global optimization effect of voltage treatment are significantly improved, and communication dependence and calculation burden are reduced.

[0054] The multi-time scale edge-end collaborative voltage treatment method based on photovoltaic inverter output control provided mainly includes the following steps:

[0055] Step one: power grid topology and parameter dynamic identification updating, the power distribution network is abstracted as an undirected weighted graph, through voltage curve feature data clustering analysis, it is decomposed into mutually non-overlapping subgraphs, and the attribution relationship of transformers and feeders is determined; the Kruskal algorithm is used to calculate the maximum value of the joint probability distribution of the topology of the power distribution network to determine the topology structure; a sequential quadratic programming model is constructed, and the line impedance and reactance parameters are optimized through iterative calculation.

[0056] Step two: short-term load prediction and centralized optimization, the historical load data is standardized pretreated, the K-means clustering algorithm is used to classify the load data, and the corresponding LSTM load prediction model is established; taking the minimum voltage deviation as the objective function, combining the power flow constraint and the voltage safety constraint, the mixed integer second-order cone optimization model is solved centrally, and the result is issued to the sensing terminal, realizing the hour-level forward-looking voltage regulation and control.

[0057] Step three: real-time local optimization of photovoltaic inverters, through the deployment of the sensing and calculation terminal in the distribution transformer area, a multi-level response mechanism is set: when the master station issues control instructions, the centralized control is executed first; without master station instructions, the local optimization threshold is used to adjust the output power of the inverter. The adjustment strategy adopts a variable parameter droop control strategy to ensure that the voltage is quickly and stably within the allowed range. At the same time, the module uploads the monitoring data and control results to the concentrator, which is integrated by the concentrator and submitted to the master station for visual display, forming a closed-loop control and decision support.

[0058] This scheme realizes rapid and accurate regulation of voltage through multi-time scale cooperative control and edge-end cooperative architecture, significantly improves the stability of the power grid under high proportion of photovoltaic access, and greatly reduces the computing burden and communication demand of the master station, providing an efficient and reliable solution for new energy consumption.

[0059] The present application realizes fine regulation and control of distributed photovoltaic grid-connected voltage by combining long-term and short-term prediction optimization with real-time dynamic adjustment, improves the stability of power grid operation and new energy consumption capacity, and improves the response speed and global optimization effect of voltage control. The implementation of the scheme includes the use of a power grid topology and parameter identification update module, a short-term load prediction and centralized optimization module, and a real-time local optimization module of photovoltaic inverters.

[0060] Firstly, the power grid topology and parameter identification update module collects power distribution network operation data, updates the network topology and electrical parameters in a long-term scale, and then builds an accurate power distribution network model to provide a data basis for subsequent voltage control. The module reconstructs the topology structure by using voltage curve clustering combined with graph theory method, and calculates the parameters based on sequential quadratic programming algorithm, to ensure the accuracy of the topology model and electrical parameters.

[0061] Secondly, the short-term load prediction and centralized optimization module predicts the load of each node in the next hour based on historical data, and optimizes the local voltage control value of each photovoltaic inverter with the minimum voltage deviation of the measurement point as the target. The optimization result is issued through the sensing and calculation terminal, realizing hour-level forward-looking voltage regulation and providing pre-control support for stable operation of the power grid.

[0062] Further, the real-time local optimization module of photovoltaic inverters, through the local real-time monitoring of the voltage state by the sensing and calculation terminal, gives priority to responding to the instructions when the master station issues control instructions; otherwise, the output of the inverter is dynamically adjusted according to the droop control curve to ensure that the voltage is quickly and stably within the allowed range. At the same time, the module uploads the monitoring data and control results to the concentrator, which is integrated by the concentrator and submitted to the master station for visual display, forming a closed-loop control and decision support.

[0063] The multi-time scale edge-end collaborative voltage management method based on photovoltaic inverter output control of the application solves the voltage fluctuation problem caused by high proportion of photovoltaic access through the collaborative mechanism of "long short-term prediction optimization + real-time dynamic adjustment", and significantly improves the flexibility and reliability of the distribution network. It provides an efficient technical solution for new energy consumption and voltage management.

[0064] In the scheme, the power grid topology and parameter identification updating module is used to collect real-time distribution network operation data, dynamically update the network topology structure and electrical parameters, and build an accurate power grid model; the short-term load forecasting and centralized optimization module is used to predict the load demand of each node in the next period based on historical load data, and optimize the local voltage control value of each photovoltaic inverter with the minimum voltage deviation of the measurement point as the target; the photovoltaic inverter real-time local optimization module is used to adjust the inverter output locally and in real time according to the droop control curve, and ensure that the voltage is quickly and stably within the allowed range.

[0065] The power grid topology and parameter identification updating module specifically includes: abstracting the distribution network as an undirected weighted graph, clustering and analyzing the voltage curve feature data to decompose into non-overlapping subgraphs, and clearly defining the attribution relationship of transformers and feeders; the Kruskal algorithm is used to calculate the maximum value of the joint probability distribution of the topology of the distribution network to determine the topology structure; a sequential quadratic programming model is constructed, and the line impedance and reactance parameters are optimized through iterative calculation.

[0066] The short-term load forecasting and centralized optimization module specifically includes: standardizing the historical load data, classifying the load data using the K-means clustering algorithm, and establishing the corresponding LSTM load forecasting model; taking the minimum voltage deviation as the objective function, combining the power flow constraint and the voltage safety constraint, and solving the mixed integer second-order cone optimization model, and then sending the results to the sensic terminal to realize the hour-level forward-looking voltage regulation.

[0067] The photovoltaic inverter real-time local optimization module specifically includes: setting a multi-level response mechanism: preferentially executing centralized control when the master station issues control instructions; optimizing the photovoltaic inverter output power in real time based on the droop control strategy when there is no master station instruction; uploading monitoring data and control results to the concentrator in real time to form a closed-loop feedback.

[0068] The above modules realize minute-level to hour-level multi-time scale voltage control through the collaborative mechanism of "long short-term prediction optimization + real-time dynamic adjustment"; the method adopts an edge-end collaborative architecture, which not only takes advantage of the global optimization of centralized control, but also retains the fast response characteristics of local control; the method is particularly suitable for high proportion of photovoltaic access scenarios, and can effectively suppress voltage fluctuations and improve new energy consumption capacity.

[0069] The following is a more specific implementation method to further demonstrate and describe the specific implementation process of the present invention:

[0070] like Figure 1 As shown, the main steps include the following:

[0071] Step 1: Dynamic identification and updating of power grid topology and parameters.

[0072] In this invention, the dynamic identification and updating of the power grid topology and parameters is a fundamental step in the entire multi-time-scale voltage management method, and its accuracy directly affects the precision and real-time performance of subsequent predictive analysis and control decisions. The dynamic identification steps of the power grid topology are as follows:

[0073] To resolve the adjacency matrix A(i,j) corresponding to the distribution network topology, we first need to abstract the distribution network as an undirected weighted graph G(V,E,A). In graph G, V represents the nodes of the distribution transformers; E represents the connecting edges between nodes; a ij The weight of an edge represents the similarity of data between two nodes. Since transformers on the same power line exhibit similar voltage change trends, and transformers with close electrical distances have similar voltages, voltage curves are used as feature data for classification analysis. This decomposes the equivalent distribution network graph G(V,E,A) into k non-overlapping subgraphs, thus clarifying the attribution relationship between transformers and 10kV feeders, allowing for the analysis of the correct power grid topology. At this point, the problem of identifying the relationship between distribution network lines and transformers is transformed into the problem of dividing graph G into k subgraphs. Typically, the value of k is related to the number of feeders; therefore, transformers are matched with k feeders. The specific topology identification method is as follows:

[0074] First, the substation and its power supply range are determined, and a reference transformer, i.e., a reference node, is set. U, I, P, and other data for each node are obtained using measuring devices in the distribution network. For a distribution network with n nodes, the voltage values ​​on the low-voltage side of each transformer are selected to form a feature vector set. The sampling period is t. Further, an adjacency matrix A is constructed, with elements a in the matrix... ij The definition is as follows:

[0075]

[0076] Next, cluster analysis is performed on the characteristic data of the distribution transformers using graph theory. Based on the known categories of reference transformers, the matching relationship between each transformer and its corresponding 10kV feeder is inferred. Using the adjacency matrix A, we will... i = {j : a ij The set of neighboring nodes of node i is defined as d, where d is the number of neighboring nodes of node i. iiThe out-degree of all nodes in the network can form an n-dimensional diagonal matrix D, which is called the out-degree matrix D of the communication network. Thus, the Laplacian matrix L = D-W is obtained, which is normalized to calculate its eigenvalues. The eigenvalues are arranged in ascending order and the first k eigenvectors are calculated to form a sample matrix, and finally clustering is performed to obtain the clustering result, thereby directly determining the correspondence between the transformer and the feeder.

[0077] Further, the mutual information between the characteristic data of each transformer is calculated: the discrete voltage curve of node i is converted into a voltage distribution curve, thereby representing the voltage edge probability distribution function p(u i ) of node i, and the calculation formula is as follows:

[0078]

[0079] Therefore, the topology of the distribution network can be represented by the joint probability distribution p(u), and the topology of the distribution network can be calculated by solving the maximum value of p(u). Kruskal algorithm is used to reconstruct the topology of the distribution network. Based on the graph G, the total weight of the node set (maximum connected tree) connected and not forming a loop in the graph G is calculated by iteration, until the total weight is calculated to be the maximum value, the topology of the distribution network is constructed; then, based on the previous clustering result, the maximum connected tree obtained is cut to obtain the topological connection relationship between the lines, and the flow is as shown in Figure 2 .

[0080] The main parameters of the distribution line include resistance, reactance, conductance and susceptance, and a π-type equivalent model is used for analysis and calculation. Because the length of the low-voltage line is short and the voltage level is low, the voltage loss caused by the susceptance and the conductance is small, and their influence can be ignored, so the model is simplified to a one-letter equivalent model. Therefore, in the present application, the parameter identification step of the distribution network topology is as follows:

[0081] (1) Set the impedance parameter initial value of each power line in the distribution network , wherein , .

[0082] (2) Substitute the resistance value and the reactance value into the following formula, and calculate the voltage value U U of the uppermost node of the distribution network using each distribution transformer node as the starting node;

[0083]

[0084] wherein . U1, P1 and Q1 are measurement data.

[0085] (3) Calculate the voltage average value U of the upstream node according to the following formula and the variance :

[0086]

[0087]

[0088] (4) Calculate the average value of variance on the entire time scale data set;

[0089] (5) Change the set impedance parameter value, and iterate the following formula as the objective function to calculate the line resistance value and reactance value in the power distribution network . .

[0090]

[0091] The above problem is a nonlinear function of variables, and the highest order of the decision variables resistance and reactance is quadratic, and it is continuous and differentiable, so solving the inequality constrained optimization problem is a nonlinear programming problem. Sequential quadratic programming algorithm can be used to convert it into a quadratic programming subproblem for solving.

[0092] Through this step, the system constructs a unified mathematical model for control and prediction calculation, fully reflects the structural characteristics and operating state of the power distribution network, and provides solid data support for short-term load forecasting, voltage regulation optimization and collaborative control.

[0093] Step two: short-term load forecasting and centralized optimization.

[0094] In order to carry out short-term load forecasting, first of all, the historical load data and related voltage, power and other characteristic parameters need to be standardized and preprocessed, and the normalization method is used to scale each input feature to the [0, 1] interval. The normalization process of is as follows:

[0095]

[0096] Before training the model, the normalized historical data set is divided into training set and test set according to the ratio of 7:3, to ensure that the model has sufficient sample coverage in the training stage, and at the same time, the generalization ability and prediction performance of the model can be fully verified in the test stage.

[0097] Then, aiming at the problem of obvious difference in user load characteristics in distribution network, K-means clustering algorithm is used for clustering analysis of load data, and sample data is divided into K categories with similar load fluctuation rules. Based on the clustering results, a corresponding long short term memory (LSTM) load prediction model is established for each category to realize accurate modeling and prediction of different types of load curve characteristics. The model parameters include the number of hidden layers, the number of neurons, and the learning rate.

[0098] During the model training process, modeling and parameter optimization are carried out based on the TensorFlow deep learning framework. The LSTM unit uses tanh as the activation function to enhance the model's expression ability for the nonlinear characteristics of the load time series. The loss function selects the mean absolute deviation (MAE), that is:

[0099]

[0100] In the formula, is the true value, is the predicted value.

[0101] Through offline training, the load samples of each category are input into the corresponding LSTM model for sufficient training. The validation set is used to repeatedly adjust the hyperparameter configuration until the prediction error of the model on the test set reaches the optimal value. Finally, a high-precision prediction model suitable for the load characteristics of each category is obtained, providing reliable load prediction results for subsequent voltage control and reactive power optimization.

[0102] To solve the problem of grid-connected point voltage out-of-limit caused by the access of photovoltaic system and improve the stability of power grid voltage, the grid-connected photovoltaic system should send a certain amount of reactive power to provide reactive voltage support to the power grid. Therefore, after the prediction is completed, the voltage control of the photovoltaic inverter needs to be carried out through the concentrator.

[0103] Based on the distribution network topology and parameter dynamic identification results in step one and the short-term load prediction results in step two, the concentrator takes the minimum voltage deviation of the measurement point as the optimization objective function:

[0104]

[0105] Where N is the number of nodes in the distribution network; N i is the set of adjacent nodes of node i obtained from the network topology graph in step one; r ij is the resistance value of branch ij; I ij is the current value between node i and node j; U i is the voltage value of node i; U0 is the rated reference voltage value of the node.

[0106] The constraint is the branch power flow constraint, as shown below:

[0107]

[0108]

[0109]

[0110]

[0111] wherein, is the actual voltage of node i; is the active power and reactive power flowing from node i to node j at time t, respectively; is the active power and reactive power output of the distributed photovoltaic of node i, respectively; is the active load and reactive load prediction value of node i, respectively; is the upper limit of the branch ij current. In addition, the safe operation constraint of voltage also needs to be considered:

[0112]

[0113]

[0114] According to the established mathematical model of the power distribution network system, it can be converted into a mixed integer second-order cone optimization model and directly solved by CPLEX for centralized solution, and finally the solution result is put down to the corresponding feeling terminal.

[0115] Step three: real-time local optimization of photovoltaic inverter.

[0116] In the present application, a multi-level response mechanism is set up: when the master station issues a control instruction, the centralized control is preferentially executed; when there is no master station instruction, the output power of the inverter is adjusted according to the variable parameter droop control strategy. For the local optimization method, the feeling terminal deployed in the distribution transformer area is used to monitor the state quantities such as voltage amplitude and phase of each node in real time, and the photovoltaic inverter is locally and real-timely optimized based on the droop control strategy. The local optimization model is as follows:

[0117]

[0118] The constraints are as follows:

[0119]

[0120] In the formula, U2 and U3 are the starting control threshold values of voltage reactive droop control, and U1 and U4 are the maximum reactive output boundaries. ​​​Upper limit of reactive power output of photovoltaic inverter i. and Upper and lower limits of active power output of photovoltaic inverter i, respectively, Capacity of photovoltaic inverter i, the obtained control curve is as shown in Figure 3

[0121] Based on the above established local optimization model of photovoltaic inverter, the output of photovoltaic inverter without master station instruction is locally optimized. Moreover, the edge node integrates the regulation results of each inverter, the sampling voltage and the control state; the summary data is uploaded once every 15 minutes; the master station system evaluates the control effect, including the adjustment amount, the node voltage stability and other indexes. In this way, the problems of large amount of calculation data and communication delay faced by the traditional centralized control in processing frequent fluctuations of photovoltaic output can be solved, the real-time regulation is improved, and the shortage of centralized control is made up.

[0122] Based on the same inventive concept, the application further provides a computer device, which comprises one or more processors and a memory for storing one or more computer programs; the program comprises program instructions, and the processor is configured to execute the program instructions stored in the memory. The processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc., which are the computing core and control core of the terminal, and are configured to implement one or more instructions, and are specifically configured to load and execute one or more instructions in the computer storage medium to implement the above method.

[0123] ​It should be further explained that based on the same inventive concept, the present application also provides a computer storage medium, which stores a computer program, and the computer program is executed by a processor to perform the above method. The storage medium can adopt any combination of one or more computer readable media. The computer readable medium can be a computer readable signal medium or a computer readable storage medium. The computer readable storage medium may, for example, but is not limited to, an electrical, magnetic, optical, infrared, or semiconductor system, device or apparatus, or any combination of the above. More specific examples (non-exhaustive list) of the computer readable storage medium include: electrical connections having one or more wires, portable computer disks, hard drives, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above. In the present application, the 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, device or apparatus.

[0124] In the description of the present application, the description of the terms "one embodiment", "an example", "a specific example" and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present disclosure. In the present specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0125] The basic principles, main features and advantages of the present disclosure are shown and described above. It should be understood by those skilled in the art that the present disclosure is not limited by the above embodiments, and the above embodiments and descriptions in the specification are only to illustrate the principles of the present disclosure. Without departing from the spirit and scope of the present disclosure, various changes and improvements can be made to the present disclosure, and these changes and improvements all fall within the scope of the claimed present disclosure.

[0126] The present application is not limited to the above best mode, and anyone can derive other various forms of a multi-time scale edge-end collaborative voltage management method based on photovoltaic inverter output control under the inspiration of the present application. Any equivalent changes and modifications made within the scope of the present application application patent range shall fall within the scope of the present application.

Claims

1. A multi-time scale edge-end collaborative voltage management method based on photovoltaic inverter output control, characterized in that, Comprise: Long-term scale power grid model construction: dynamically collect power distribution network operation data, abstract the power distribution network as an undirected weighted graph, determine the transformer and feeder attribution based on node voltage curve feature clustering analysis, determine the line topology connection relationship using graph theory algorithm, optimize the line impedance and reactance parameters through a sequential quadratic programming model, and construct an accurate power grid model; Short-term scale centralized optimization instruction generation: after standardizing and preprocessing historical load data, use K-means clustering classification to establish a long short-term memory neural network prediction model for each type of load to obtain the next period node load, based on the accurate power grid model, minimize the measurement point voltage deviation as the target, combine the power flow constraint and voltage safety constraint, calculate the voltage control instruction of each photovoltaic inverter through a mixed integer second-order cone optimization model and issue it to the edge terminal; And, real-time scale local voltage regulation: monitor the node voltage state through the edge terminal, and perform multi-level response: preferentially respond to and execute the control instruction issued in the short-term scale; If there is no control instruction, adjust the inverter reactive power output according to the variable parameter droop control strategy based on the local real-time voltage data, and the strategy dynamically adjusts the reactive power output through a pre-set multi-level voltage threshold; Wherein, the periods of the three scales are shortened in turn, and the edge terminal uploads the monitoring data and control results to the edge processing unit at a fixed period, and the edge processing unit integrates the data for long-term scale model updating and short-term scale optimization calculation, forming an edge-end collaborative data interaction process.

2. The multi-time scale edge-end collaborative voltage management method based on photovoltaic inverter output control according to claim 1, characterized in that: In the long-term scale power grid model construction, the determination of the line topology connection relationship using graph theory algorithm specifically comprises: Based on the clustering results of voltage curve features, the Kruskal algorithm is used to calculate the maximum value of the joint probability distribution of the power distribution network topology to obtain the maximum connected tree of the power distribution network; then, based on the transformer and feeder attribution relationship determined by the clustering analysis, the maximum connected tree is cut to obtain the topology connection relationship between the lines; And the voltage curve feature clustering analysis further comprises: constructing an adjacency matrix based on the node voltage curve, determining the adjacency node set and out-degree of each node to construct an out-degree matrix, and then forming a Laplacian matrix; after normalizing the Laplacian matrix, calculate its eigenvalues, take the first k eigenvectors to form a sample matrix, and obtain the attribution relationship of the transformer and feeder through clustering.

3. The multi-time scale edge-end collaborative voltage management method based on photovoltaic inverter output control according to claim 1, characterized in that: In the long-term scale power grid model construction, the optimization of line impedance and reactance parameters through a sequential quadratic programming model is specifically based on a one-dimensional equivalent model of a low-voltage power distribution network, and the effects of line susceptance and conductance on voltage are ignored, only taking line resistance and reactance as optimization objects; The optimization target of the sequence quadratic programming model is to minimize the average value of the upstream node voltage variance in the full time scale, and the constraint condition is that the line resistance and reactance are greater than 0; before optimization, the line impedance parameter initial value is set, wherein the line resistance is the product of the unit length resistance and the line length, and the line reactance is the product of the unit length reactance and the line length.

4. The multi-time scale edge-end collaborative voltage management method based on photovoltaic inverter output control according to claim 1, characterized in that: In the centralized optimization instruction generation of the short-term scale, the historical load data standardization preprocessing specifically includes: first, the historical load data is scaled based on the Z-score standardization method, so as to eliminate the dimension influence; then, the historical load data set is divided into a training set and a test set, so as to train and verify the generalization ability of the long short-term memory neural network prediction model; The long short-term memory neural network prediction model takes tanh as the activation function and the mean absolute error as the loss function, and adjusts the model hidden layer number, neuron number and learning rate through offline training until the prediction error of the model on the test set reaches the optimum.

5. The multi-time scale edge-end collaborative voltage management method based on photovoltaic inverter output control according to claim 1, characterized in that: In the centralized optimization instruction generation of the short-term scale, the minimum measurement point voltage deviation is specifically a double-objective optimization of minimizing the distribution network branch loss and the square sum of the voltage deviation of each node from the rated reference voltage; The power flow constraint specifically includes: the node power balance constraint based on the node net active power = photovoltaic active output - load active demand and the node net reactive power = photovoltaic reactive output - load reactive demand; the branch current constraint based on the square of the branch current not exceeding the square of the maximum allowed branch current, and the branch current being calculated based on the branch active power, reactive power and node voltage; The voltage safety constraint is specifically that the square of the node voltage is not less than the square of the minimum allowed voltage and not greater than the square of the maximum allowed voltage.

6. The multi-time scale edge-end collaborative voltage management method based on photovoltaic inverter output control according to claim 1, characterized in that: In the local voltage regulation of the real-time scale, the variable parameter droop control strategy specifically includes dynamically adjusting the reactive output by pre-setting multiple voltage threshold values, when the variable parameter droop control strategy is used for the local voltage regulation of the real-time scale. The pre-set multiple voltage threshold values are U1, U2, U3 and U4, wherein U2 and U3 are the voltage reactive droop control starting threshold values, and U1 and U4 are the maximum reactive output boundaries; When the node voltage ≤ U1, the inverter outputs the maximum reactive power; when U1 < node voltage < U2, the inverter reactive output is linearly adjusted according to the product of the maximum reactive power and (voltage-U2) / (U1-U2); when U2 ≤ node voltage ≤ U3, the inverter reactive output is 0; when U3 < node voltage < U4, the inverter reactive output is linearly adjusted according to the product of the minimum reactive power and (voltage-U3) / (U4-U3); when the node voltage ≥ U4, the inverter outputs the minimum reactive power.

7. The multi-time-scale edge-end collaborative voltage management method based on photovoltaic inverter output control according to claim 1, characterized in that: in the real-time scale local voltage regulation, the inverter output also needs to meet the power constraint: the inverter active output is between the minimum active output value and the maximum active output value; and the sum of the square of the real-time active output of the inverter and the square of the real-time reactive output does not exceed the square of the rated capacity of the inverter.

8. The multi-time-scale edge-end collaborative voltage management method based on photovoltaic inverter output control according to claim 1, characterized in that: the three scales are sequentially shortened, and specifically: the long-term scale power grid model update period is triggered periodically, and the period is longer than the short-term scale every hour; the short-term scale centralized optimization instruction generation period is every hour, that is, photovoltaic inverter control instructions are calculated and issued based on load prediction results every hour; the real-time scale local voltage regulation period is seconds to minutes, that is, the edge terminal monitors the node voltage state every 1-5 minutes, and adjusts the inverter output as needed; the edge terminal uploads the monitoring data and control results to the edge side processing unit at a fixed period, and the fixed period is every 15 minutes; the data integrated by the edge side processing unit includes node voltage monitoring values, inverter output power regulation amounts, and control instruction execution states; in addition to being used for long-term scale power grid model updating and short-term scale centralized optimization calculation, the integrated data is also submitted to the master station system for control effect evaluation, and the evaluation indexes include inverter regulation amount compliance rate, node voltage fluctuation amplitude, and voltage out-of-limit times.

9. The multi-time-scale edge-end collaborative voltage management method based on photovoltaic inverter output control according to claim 2, characterized in that: in the long-term scale power grid model construction, a step of calculating the voltage mutual information between nodes is further included: the node discrete voltage curve is converted into a voltage distribution curve to represent the edge probability distribution of the voltage of each node; and based on the joint probability distribution and the edge probability distribution of the node voltage, the voltage mutual information between nodes is calculated, which is used to assist in determining the joint probability distribution of the distribution network topology and improve the identification accuracy of the line topology connection relationship.

10. A multi-time scale edge-end collaborative voltage management system based on photovoltaic inverter output control, characterized in that, including: edge terminal: deployed in the distribution transformer area, used for dynamically collecting distribution network operation data and monitoring node voltage state; performing multi-level response: preferentially responding to the control instructions issued by the edge side processing unit, and when there is no instruction, adjusting the reactive output of the photovoltaic inverter based on the local real-time voltage data according to the variable parameter droop control strategy; uploading the monitoring data and control results to the edge side processing unit at a fixed period; edge side processing unit: in communication connection with the edge terminal, used for: based on the data uploaded by the edge terminal, constructing and updating an accurate power grid model through voltage curve feature clustering, graph theory algorithm and sequential quadratic programming model; after standardizing and K-means clustering the historical load data, predicting the node load through a long short-term memory neural network, and combining the accurate power grid model and a mixed integer second-order cone optimization model, generating voltage control instructions for the photovoltaic inverter and issuing them to the edge terminal; The main station system is in communication connection with the edge-side processing unit, is used for receiving the integrated data of the edge-side processing unit, evaluating the control effect (including the inverter adjustment amount, the node voltage fluctuation amplitude, and the voltage overrun times), and performing visual display; The photovoltaic inverter is in communication connection with the edge terminal, is used for receiving the adjustment instruction of the edge terminal, adjusting the active and reactive output power, and ensuring that the node voltage is stable within the allowed range; The edge terminal, the edge-side processing unit, the main station system, and the photovoltaic inverter realize data interaction through the edge-terminal collaborative architecture, and correspond to voltage management scales that are sequentially shortened in long-term, short-term, and real-time periods.

Images