Power grid day-ahead regulation and control method and device based on distributed photovoltaic virtual cluster dynamic division

Abstract

The invention provides a power grid day-ahead regulation and control method and device based on distributed photovoltaic virtual cluster dynamic division, and the method comprises the steps: obtaining a distributed photovoltaic virtual cluster through the dynamic division of an improved community theory of weight clustering based on distributed photovoltaic characteristic data and a regulation and control target; based on a pre-trained distributed photovoltaic prediction model, predicting future output data of each distributed photovoltaic according to historical output data, historical meteorological data and meteorological prediction data of the distributed photovoltaic in the distributed photovoltaic virtual cluster; and regulating and controlling the distributed photovoltaic virtual cluster according to the regulation and control target and the corresponding regulation and control target function based on the future output data of each distributed photovoltaic in the distributed photovoltaic virtual cluster. According to the method, distributed photovoltaic can be effectively classified, so that the regulation and control capability and the stable operation capability of a low-voltage power grid are improved; meanwhile, the economical efficiency, the safety and the reliability of low-voltage power grid operation are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid regulation and control, and particularly to a day-ahead regulation and control method and device for a power grid based on dynamic division of distributed photovoltaic virtual clusters. Background Art

[0002] In recent years, China has actively promoted the realization of the goals of carbon peak and carbon neutrality. Distributed photovoltaic has gradually become an important power source, with a relatively high proportion in the regional power grid and an increasingly significant impact on the operation of the power grid.

[0003] However, the output of distributed photovoltaic is greatly affected by external factors such as weather changes, with intermittency, volatility and uncertainty, and is geographically dispersed with a wide distribution space range. These characteristics will cause problems such as voltage fluctuations and voltage over-limit when large-scale distributed photovoltaic is connected to the regional power grid, seriously affecting the economy and stability of the power grid.

[0004] Existing research mainly focuses on the time scale and distribution space range of distributed photovoltaic output, with less research on the spatio-temporal differences of distributed photovoltaic output, and it is difficult to take into account the real-time nature of power grid regulation and control and the complex network structure.

[0005] Therefore, there is an urgent need for more efficient, accurate and intelligent methods to deal with the impact of distributed photovoltaic on the operation of the power grid and improve the operation economy and stability of the regional power grid. Summary of the Invention

[0006] The present invention provides a day-ahead regulation and control method and device for a power grid based on dynamic division of distributed photovoltaic virtual clusters, aiming to overcome the defects that the prior art does not consider the spatio-temporal differences of distributed photovoltaic and cannot take into account the real-time nature of power grid regulation and control and the network complexity, so as to effectively improve the economy, safety and reliability of low-voltage power grid operation.

[0007] On the one hand, the present invention provides a day-ahead regulation and control method for a power grid based on dynamic division of distributed photovoltaic virtual clusters, including: obtaining characteristic data of distributed photovoltaic, where the characteristic data includes active voltage sensitivity, reactive voltage sensitivity, active regulation capacity, reactive regulation capacity and regulation cost; dynamically dividing distributed photovoltaic virtual clusters through an improved community theory of weight clustering based on the characteristic data of distributed photovoltaic and regulation and control objectives; predicting the future output data of each distributed photovoltaic according to the historical output data, historical meteorological data and meteorological prediction data of distributed photovoltaic within the distributed photovoltaic virtual cluster based on a pre-trained distributed photovoltaic prediction model; wherein, the distributed photovoltaic prediction model is constructed based on an attention mechanism and a long short-term memory network; regulating the distributed photovoltaic virtual cluster according to the future output data of each distributed photovoltaic within the distributed photovoltaic virtual cluster, the regulation and control objective and the corresponding regulation and control objective function.

[0008] Furthermore, based on the characteristic data and regulation objectives of distributed photovoltaic, a dynamically partitioned distributed photovoltaic virtual cluster is obtained through an improved community theory of weight clustering, including: determining cluster partition indicators from the characteristic data of distributed photovoltaic according to the regulation objectives; calculating a similarity matrix according to the cluster partition indicators; and using the modularity index to correct the weights of the cluster partition indicators, optimizing the similarity matrix, so as to obtain the distributed photovoltaic virtual cluster.

[0009] Furthermore, determining cluster partition indicators from the characteristic data of distributed photovoltaic according to the regulation objectives includes: when the regulation objective is economic optimization, determining the reactive voltage sensitivity, reactive power regulation capacity, and regulation cost in the characteristic data as the cluster partition indicators; when the regulation objective is the safe and stable operation of the power grid, determining the active voltage sensitivity, reactive voltage sensitivity, active regulation capacity, and reactive power regulation capacity in the characteristic data as the cluster partition indicators.

[0010] Furthermore, based on a pre-trained distributed photovoltaic prediction model, the future output data of each distributed photovoltaic is predicted according to the historical output data, historical meteorological data, and meteorological prediction data of the distributed photovoltaic in the distributed photovoltaic virtual cluster, including: inputting the historical output data, historical meteorological data, and meteorological prediction data into the pre-trained distributed photovoltaic prediction model to obtain hidden state vectors of a set time series; extracting a time pattern matrix from the hidden state vectors, and defining an evaluation function according to the time pattern matrix, weight matrix, and hidden state vectors; calculating parameter weights according to the evaluation function; summing the parameter weights and the weights of the time pattern matrix to obtain attention; and adding the attention and the hidden state vectors after linear mapping to obtain the future output data of each distributed photovoltaic in the output distributed photovoltaic virtual cluster.

[0011] Further, regulating the distributed PV virtual cluster based on the future output data of each distributed PV in the distributed PV virtual cluster according to the regulation target and the corresponding regulation target function includes: determining active power commands and reactive power commands according to the future output data of each distributed PV in the distributed PV virtual cluster; when the regulation target is economic optimization, regulating the first distributed PV virtual cluster according to the active power commands and reactive power commands through the first regulation target function. When the regulation target is the secure and stable operation of the power grid, regulating the second distributed PV virtual cluster according to the active power commands and reactive power commands through the second regulation target function; wherein, the first distributed PV virtual cluster is an economic regulation cluster, the second distributed PV virtual cluster is an emergency regulation cluster, both the first distributed PV virtual cluster and the second distributed PV virtual cluster are included in the distributed PV virtual cluster, and both the first distributed PV virtual cluster and the second distributed PV virtual cluster satisfy active power constraints and reactive power constraints.

[0012] Further, the first regulation target function is defined as follows: ; The second regulation target function is defined as follows: ; The active power constraints and reactive power constraints are defined as follows: ; Wherein, represents the first regulation target function, represents the second regulation target function, represents the regulation weight of the node voltage change, represents the distributed PV virtual cluster number, represents the number of economic regulation clusters, represents the number of emergency regulation clusters, represents the main node voltage change caused by the output change of the distributed PV virtual cluster , represents the reference value of the main node voltage change obtained after optimization, represents the regulation weight of the reactive power, represents the reactive power command, represents the actual reactive power output value of the distributed PV virtual cluster , represents the active power command, represents the actual active power output value of the distributed PV virtual cluster , represents the allowable maximum active power deviation, represents the allowable maximum reactive power deviation.

[0013] In a second aspect, the present invention further provides a grid day-ahead regulation device based on dynamic partitioning of distributed photovoltaic virtual clusters, including: a distributed photovoltaic characteristic data acquisition module for acquiring the characteristic data of distributed photovoltaics, where the characteristic data includes active voltage sensitivity, reactive voltage sensitivity, active regulation capacity, reactive regulation capacity, and regulation cost; a distributed photovoltaic virtual cluster partitioning module for dynamically partitioning distributed photovoltaic virtual clusters through an improved community theory of weight clustering based on the characteristic data of distributed photovoltaics and regulation objectives; a distributed photovoltaic output data prediction module for predicting the future output data of each distributed photovoltaic based on a pre-trained distributed photovoltaic prediction model according to the historical output data, historical meteorological data, and meteorological prediction data of the distributed photovoltaics within the distributed photovoltaic virtual cluster; wherein, the distributed photovoltaic prediction model is constructed based on an attention mechanism and a long short-term memory network; a distributed photovoltaic virtual cluster regulation module for regulating the distributed photovoltaic virtual cluster based on the future output data of each distributed photovoltaic within the distributed photovoltaic virtual cluster according to the regulation objective and the corresponding regulation objective function.

[0014] In a third aspect, the present invention further provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, where when the processor executes the computer program, it implements the grid day-ahead regulation method based on dynamic partitioning of distributed photovoltaic virtual clusters as described in any one of the above.

[0015] In a fourth aspect, the present invention further provides a non-transitory computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the grid day-ahead regulation method based on dynamic partitioning of distributed photovoltaic virtual clusters as described in any one of the above.

[0016] In a fifth aspect, the present invention further provides a computer program product, including a computer program, and when the computer program is executed by a processor, it implements the grid day-ahead regulation method based on dynamic partitioning of distributed photovoltaic virtual clusters as described in any one of the above.

[0017] The grid day-ahead regulation method based on the dynamic partitioning of distributed photovoltaic virtual clusters provided by the present invention obtains characteristic data of distributed photovoltaics, where the characteristic data includes active voltage sensitivity, reactive voltage sensitivity, active regulation capacity, reactive regulation capacity, and regulation cost; based on the characteristic data of distributed photovoltaics and the regulation target, the distributed photovoltaic virtual clusters are dynamically partitioned through an improved community theory of weight clustering; based on a pre-trained distributed photovoltaic prediction model, according to the historical output data, historical meteorological data, and meteorological prediction data of the distributed photovoltaics within the distributed photovoltaic virtual clusters, the future output data of each distributed photovoltaic is predicted; among them, the distributed photovoltaic prediction model is constructed based on the attention mechanism and the long short-term memory network; based on the future output data of each distributed photovoltaic within the distributed photovoltaic virtual clusters, according to the regulation target and the corresponding regulation target function, the distributed photovoltaic virtual clusters are regulated. This method can effectively classify distributed photovoltaics by dynamically partitioning distributed photovoltaic virtual clusters through an improved community theory of weight clustering, thereby improving the regulation ability and stable operation ability of low-voltage power grids; at the same time, by regulating distributed photovoltaic virtual clusters according to the predicted future output data and regulation targets of distributed photovoltaics, the economy, safety, and reliability of low-voltage power grid operation are improved. Description of the Drawings

[0018] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0019] Figure 1 It is a schematic flowchart of the grid day-ahead regulation method based on the dynamic partitioning of distributed photovoltaic virtual clusters provided by the embodiments of the present invention.

[0020] Figure 2 It is a schematic structural diagram of the grid day-ahead regulation device based on the dynamic partitioning of distributed photovoltaic virtual clusters provided by the embodiments of the present invention.

[0021] Figure 3 It is a schematic physical structure diagram of the electronic device provided by the embodiments of the present invention. Detailed Embodiments

[0022] To make the objectives, technical solutions, and advantages of the present invention clearer, the following will clearly and completely describe the technical solutions in the present invention in conjunction with the drawings in the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

[0023] It should be noted that, under the strategic background of carbon peak and carbon neutrality in China, the proportion of distributed photovoltaic in the power grid is increasing, and its impact on the power grid is also growing. However, distributed photovoltaic is greatly affected by external factors, and its output has intermittency, volatility and uncertainty. Moreover, distributed photovoltaic has dispersion in geographical space distribution and high penetration rate in regional power grids. Therefore, the large-scale access of distributed photovoltaic in regional power grids will cause serious problems such as voltage fluctuation and voltage over-limit in low-voltage power grids, seriously affecting the economic, safe and stable operation of the power grid.

[0024] Local control of distributed photovoltaic and global optimization based on comprehensive measurement play an important role in improving the operation economy and stability of regional power grids. Existing research mainly focuses on the output time scale of distributed photovoltaic and the distribution space range of distributed photovoltaic, and there is less research on the spatio-temporal difference of distributed photovoltaic, which cannot take into account the real-time nature of power grid regulation and network complexity.

[0025] To meet the diverse regulation requirements of regional power grids and the economy, security and reliability of regional power grid operation, the present invention proposes a day-ahead regulation method for power grids based on dynamic division of distributed photovoltaic virtual clusters, which conducts dynamic division of distributed photovoltaic virtual clusters and day-ahead pre-regulation for high-penetration regional power grids to ensure the economic, safe and stable operation of regional power grids.

[0026] Specifically, Figure 1 Fig. shows a schematic flow chart of the day-ahead regulation method for power grids based on dynamic division of distributed photovoltaic virtual clusters provided by an embodiment of the present invention.

[0027] As Figure 1 shown, the method includes steps S110-S140, and the following will elaborate on steps S110-S140 and related steps in detail.

[0028] S110, obtaining characteristic data of distributed photovoltaic, where the characteristic data includes active voltage sensitivity, reactive voltage sensitivity, active regulation capacity, reactive regulation capacity, and regulation cost.

[0029] It is easy to understand that the characteristic data of distributed photovoltaic is crucial for power grid planning, operation and management. These characteristic data can help understand how photovoltaic responds to different power grid conditions and the impact of photovoltaic on power grid stability. Specifically, in this embodiment, the characteristic data of distributed photovoltaic includes but is not limited to active voltage sensitivity, reactive voltage sensitivity, active regulation capacity, reactive regulation capacity, and regulation cost.

[0030] Among them, the active voltage sensitivity refers to the degree of influence of the change in the active power output of distributed photovoltaics on the voltage at the point of common connection. The reactive voltage sensitivity refers to the influence of the reactive power provided by distributed photovoltaics on the voltage level at its point of common connection. The active regulation capacity refers to the maximum range within which distributed photovoltaics can adjust their active power output without affecting their long-term performance. The reactive regulation capacity refers to the maximum value of reactive power that distributed photovoltaics can provide or absorb without compromising their own performance.

[0031] The regulation cost involves the economic cost incurred to achieve the aforementioned regulation. It not only includes direct costs such as equipment upgrades and installation of additional energy storage systems, but may also involve indirect costs such as potential revenue losses due to power rationing. In addition, frequent use of the regulation capacity may also accelerate the aging of equipment, thereby increasing maintenance costs.

[0032] The above data can be obtained through on-site measurement and testing, and no specific limitations are imposed here.

[0033] On the basis of obtaining the characteristic data of distributed photovoltaics in step S110, further, step S120 is executed.

[0034] S120, based on the characteristic data and regulation objectives of distributed photovoltaics, dynamically divides the distributed photovoltaic virtual clusters through the improved community theory of weighted clustering.

[0035] It is easy to understand that for different regulation objectives, corresponding characteristic data should be selected as the cluster division index. For example, if the economic optimum is the regulation objective, then the reactive voltage sensitivity, reactive regulation capacity, and regulation cost are selected as the cluster division indexes in the operation mode of the economic regulation cluster; if the safe and stable operation of the distribution network is the regulation objective, then the active voltage sensitivity, reactive voltage sensitivity, active regulation capacity, and reactive regulation capacity are selected as the cluster division indexes in the operation mode of the emergency regulation cluster.

[0036] Subsequently, according to the selected cluster division indexes, the distributed photovoltaic virtual clusters can be dynamically divided through the improved community theory of weighted clustering.

[0037] Weighted clustering refers to performing clustering analysis by considering the weights of edges in a network (such as the network topology structure composed of distributed photovoltaics). Traditional clustering methods often only consider whether there is a connection between nodes, without considering the strength or importance of these connections. After introducing weights, the true situation of the relationship between nodes (such as photovoltaic nodes) can be more accurately reflected.

[0038] Improving the community theory, community detection aims to discover the community structure in a network, that is, which nodes in the network are more inclined to form tightly connected small groups. Community detection algorithms include the Girvan-Newman algorithm, the Louvain method, etc.

[0039] In this embodiment, by introducing the concept of weighted clustering into the improved community theory, a more refined and realistic community division can be achieved. For example, in a distribution network, adjusting the grid partition according to the actual load (weight) on the transmission line helps to improve the operating efficiency and stability of the system.

[0040] It should be noted that the process of dividing the distributed photovoltaic virtual clusters obtained in this embodiment is a dynamically changing and continuously optimizing process, so that two or more distributed photovoltaic virtual clusters are obtained.

[0041] After dynamically dividing the distributed photovoltaic virtual clusters, further, step S130 is executed.

[0042] S130, based on a pre-trained distributed photovoltaic prediction model, according to the historical output data, historical meteorological data, and meteorological prediction data of the distributed photovoltaics in the distributed photovoltaic virtual cluster, predict the future output data of each distributed photovoltaic; wherein, the distributed photovoltaic prediction model is constructed based on the attention mechanism and the long short-term memory network.

[0043] It is easy to understand that in this embodiment, a distributed photovoltaic prediction model is pre-trained to predict the future output data of the distributed photovoltaics. Specifically, the obtained historical output data, historical meteorological data, and meteorological prediction data of the distributed photovoltaics are input into the distributed photovoltaic prediction model together, and the future output data of each distributed photovoltaic can be obtained as the output.

[0044] Among them, the historical output data of the distributed photovoltaics can be directly exported through the monitoring system or log files, and the historical meteorological data and meteorological prediction data can be obtained by querying data from meteorological service agencies.

[0045] It should be noted that the prediction object of the distributed photovoltaic prediction model in this embodiment is the distributed photovoltaic virtual cluster dynamically divided in step S120, including multiple distributed photovoltaics.

[0046] It is also worth mentioning that the distributed photovoltaic prediction model provided in this embodiment is constructed based on the attention mechanism and the long short-term memory network. In particular, during the prediction process, an evaluation function is defined, which can better optimize the parameter weights, so as to obtain a more accurate prediction result.

[0047] After predicting the future output data of each distributed PV in the distributed PV virtual cluster, further, step S140 is executed.

[0048] S140, based on the future output data of each distributed PV in the distributed PV virtual cluster, according to the regulation target and the corresponding regulation target function, regulate the distributed PV virtual cluster.

[0049] It is easy to understand that according to the future output data of each distributed PV in the distributed PV virtual cluster and combined with the actual output data of each distributed PV in the distributed PV virtual cluster, the corresponding active power command and reactive power command can be determined. Subsequently, according to the active power command and the reactive power command, the distributed PV virtual cluster is regulated through the regulation target function corresponding to the current regulation target.

[0050] Among them, the regulation target functions corresponding to different regulation targets are different, which will be described in detail in the following embodiments.

[0051] It should be noted that the day-ahead regulation method of the power grid based on the dynamic division of the distributed PV virtual cluster provided in this embodiment can be applied to any type of power grid, especially to the regional-level power grid.

[0052] In this embodiment, by obtaining the characteristic data of the distributed PV, the characteristic data includes active power voltage sensitivity, reactive power voltage sensitivity, active power regulation capacity, reactive power regulation capacity, and regulation cost; based on the characteristic data of the distributed PV and the regulation target, the distributed PV virtual cluster is dynamically divided by the improved community theory of weight clustering; based on the pre-trained distributed PV prediction model, according to the historical output data, historical meteorological data, and meteorological prediction data of the distributed PV in the distributed PV virtual cluster, the future output data of each distributed PV is predicted; among them, the distributed PV prediction model is constructed based on the attention mechanism and the long short-term memory network; based on the future output data of each distributed PV in the distributed PV virtual cluster, according to the regulation target and the corresponding regulation target function, the distributed PV virtual cluster is regulated. This method can effectively classify the distributed PV by dynamically dividing the distributed PV virtual cluster through the improved community theory of weight clustering, thereby improving the regulation ability and stable operation ability of the low-voltage power grid; at the same time, by regulating the distributed PV virtual cluster according to the predicted future output data and regulation target of the distributed PV, the economy, safety, and reliability of the low-voltage power grid operation are improved.

[0053] On the basis of the above embodiments, further, the following will describe in detail the dynamic division process of the distributed virtual cluster.

[0054] Based on the characteristic data and regulation objectives of distributed photovoltaic, a dynamically partitioned distributed photovoltaic virtual cluster is obtained through an improved community theory of weight clustering, including: determining cluster partition indicators from the characteristic data of distributed photovoltaic according to the regulation objectives; calculating a similarity matrix according to the cluster partition indicators; using the modularity indicator to correct the weights of the cluster partition indicators and optimize the similarity matrix to partition the distributed photovoltaic virtual cluster.

[0055] It is easy to understand that when the regulation objective is economic optimization, the reactive voltage sensitivity, reactive regulation capacity, and regulation cost in the characteristic data are determined as the cluster partition indicators; when the regulation objective is the safe and stable operation of the power grid, the active voltage sensitivity, reactive voltage sensitivity, active regulation capacity, and reactive regulation capacity in the characteristic data are determined as the cluster partition indicators.

[0056] Subsequently, the selected cluster partition indicators are normalized to compress their numerical ranges to , and the normalization formula is as follows in Equation (1).

[0057] (1).

[0058] In Equation (1), represents the value of the -th cluster partition indicator of the -th distributed photovoltaic after normalization, represents the value of the -th cluster partition indicator of the -th distributed photovoltaic before normalization, represents the minimum value of the -th cluster partition indicator of the -th distributed photovoltaic, represents the maximum value of the -th cluster partition indicator of the -th distributed photovoltaic.

[0059] Immediately afterwards, according to the normalized cluster partition indicators, a similarity matrix is calculated, and the specific calculation formula is as follows in Equation (2).

[0060] (2).

[0061] (3).

[0062] In Equation (2), represents the weight of the -th cluster partition indicator, represents the number of cluster partition indicators, represents the -th cluster partition indicator of the The value of a cluster division index represents the th cluster division index value of the th distributed photovoltaic before normalization. is a 0-1 function. If the communication methods between distributed photovoltaics are the same, then it takes the value of 1; if the communication methods between distributed photovoltaics are different, then it takes the value of 0. In this embodiment, it is default that the communication methods between distributed photovoltaics are the same, and it is constantly 1.

[0063] The numerical value in the similarity matrix represents the characteristic similarity between two distributed photovoltaics. The closer the numerical value is to 1, the more similar the characteristics of the two distributed photovoltaics are.

[0064] Compare the numerical value in the similarity matrix with the set threshold, and thus the distributed photovoltaics can be preliminarily divided to obtain the distributed photovoltaic virtual clusters.

[0065] Furthermore, based on the calculation of the modularity index, the weights of each cluster division index are corrected, which is also a further optimization of the similarity matrix, so as to obtain the best distributed photovoltaic virtual cluster division result. Among them, the calculation formula of the modularity index is as follows formula (4).

[0066] (4).

[0067] (5).

[0068] In formulas (4)-(5), represents the modularity index, represents the sum of the weights of all edges in the network. represents the weight of the edge between distributed photovoltaic and distributed photovoltaic . If there is no connection between distributed photovoltaic and distributed photovoltaic , then takes the value of 0; otherwise, takes the value of 1. represents the sum of the weights of the edges on distributed photovoltaic , represents the sum of the weights of the edges on distributed photovoltaic . is a 0-1 function, in , respectively represent the numbers of the distributed photovoltaic virtual clusters (communities) where distributed photovoltaic , distributed photovoltaic are located. When ​ takes the value of 1, conversely, takes the value of 0.

[0069] For a similarity matrix, it is necessary to calculate its corresponding modularity index to determine whether the modularity index meets the requirements, such as whether the modularity index is greater than a certain set value. If it is satisfied, the distributed photovoltaic virtual clusters are divided according to the current similarity matrix; if not, it is necessary to modify the weights of the clustering division indicators of each cluster, and then optimize the similarity matrix so that its corresponding modularity index can meet the set requirements.

[0070] Among them, the initial weights of the clustering division indicators of each cluster are custom-set according to the actual situation. When modifying the weights of the clustering division indicators of each cluster, the direction of modification (increasing the weight or decreasing the weight) can be determined through multiple debugging. The larger the modularity index, the better the divided distributed photovoltaic virtual clusters.

[0071] Based on the above, multiple distributed photovoltaic virtual clusters can be divided.

[0072] In this embodiment, by determining the clustering division indicators from the characteristic data of distributed photovoltaics according to the regulation target, calculating the similarity matrix according to the clustering division indicators, using the modularity index to correct the weights of the clustering division indicators, optimizing the similarity matrix to divide the distributed photovoltaic virtual clusters, and then based on the pre-trained distributed photovoltaic prediction model, according to the historical output data, historical meteorological data and meteorological prediction data of the distributed photovoltaics in the distributed photovoltaic virtual clusters, predicting the future output data of each distributed photovoltaic. Thus, based on the future output data of each distributed photovoltaic in the distributed photovoltaic virtual clusters, according to the regulation target and the corresponding regulation target function, the distributed photovoltaic virtual clusters are regulated. This method dynamically divides the distributed photovoltaic virtual clusters through the improved community theory of weight clustering, can effectively classify the distributed photovoltaics, and then improve the regulation ability and stable operation ability of the low-voltage power grid; at the same time, by regulating the distributed photovoltaic virtual clusters according to the predicted future output data and regulation target of the distributed photovoltaics, the economy, safety and reliability of the low-voltage power grid operation are improved.

[0073] On the basis of the above embodiment, further, the prediction of the future output data of distributed photovoltaics will be described in detail below.

[0074] It is easy to understand that first, the historical output data, historical meteorological data and meteorological prediction data are input into the pre-trained distributed photovoltaic prediction model for output prediction to obtain the hidden state vectors of the set time series . Among them, is the length of the time series window, that is, the set time series, represents the current moment.

[0075] Let be the hidden state matrix, whose row vectors represent the states of a single variable at different time steps, and whose column vectors represent the gate neuron parameters of the LSTM (Long Short-Term Memory) in the distributed photovoltaic prediction model at the same time step.

[0076] Use one-dimensional convolution to extract the time pattern matrix from the hidden state matrix. The expression of the time pattern matrix is as shown in Equation (6) below.

[0077] (6).

[0078] In Equation (6), represents the CNN (Convolutional Neural Network) filter. Through multiple convolutional operations of the CNN on the time series, the hidden state vector is extracted. represents the th filter at the row, represents the length of the time series window, that is, the filter length, represents the current time, represents the padding length in the convolutional calculation, represents the maximum length of the weight. Generally, .

[0079] Subsequently, define the evaluation function according to the time pattern matrix, the weight matrix, and the hidden state vector, and calculate the parameter weights according to the evaluation function. Among them, the evaluation function and the calculation formulas of the parameter weights are as shown in Equations (7)-(8) below.

[0080] (7).

[0081] (8).

[0082] In Equations (7)-(8), represents the vector at the row, represents the weight matrix, represents the parameter weights.

[0083] Immediately afterwards, sum the parameter weights with the weights of the time pattern matrix to obtain the attention, as shown in Equation (9) below.

[0084] (9).

[0085] In formula (9), represents the characteristic quantity of the input data. In this embodiment, takes the value of 4.

[0086] Finally, the attention and the hidden state vector are linearly mapped and added together, and then the future output data of each distributed photovoltaic in the distributed photovoltaic virtual cluster can be obtained. Specifically, see the following formula (10).

[0087] (10).

[0088] In formula (10), , , respectively represent , and weight matrices.

[0089] In this embodiment, when the distributed photovoltaic prediction model performs output prediction, an evaluation function is defined to optimize the parameter weights, so that more accurate output prediction results can be obtained.

[0090] On the basis of the above embodiments, further, the regulation process of the distributed photovoltaic virtual cluster under different regulation objectives will be described in detail below.

[0091] Based on the future output data of each distributed photovoltaic in the distributed photovoltaic virtual cluster, according to the regulation objective and the corresponding regulation objective function, the distributed photovoltaic virtual cluster is regulated, including: determining the active power command and the reactive power command according to the future output data of each distributed photovoltaic in the distributed photovoltaic virtual cluster; when the regulation objective is economic optimization, regulating the first distributed photovoltaic virtual cluster through the first regulation objective function according to the active power command and the reactive power command. When the regulation objective is the safe and stable operation of the power grid, regulating the second distributed photovoltaic virtual cluster through the second regulation objective function according to the active power command and the reactive power command.

[0092] Among them, the first distributed photovoltaic virtual cluster is an economic regulation cluster, the second distributed photovoltaic virtual cluster is an emergency regulation cluster, both the first distributed photovoltaic virtual cluster and the second distributed photovoltaic virtual cluster are included in the distributed photovoltaic virtual cluster, and both the first distributed photovoltaic virtual cluster and the second distributed photovoltaic virtual cluster satisfy the active power constraint and the reactive power constraint.

[0093] It is easy to understand that according to the future output data of each distributed photovoltaic in the distributed photovoltaic virtual cluster and combining with the real-time output value of each distributed photovoltaic in each distributed photovoltaic virtual cluster, the active power command can be determined and reactive power instruction After receiving the active power instruction and the reactive power instruction, the regulation values of each distributed photovoltaic are in a proportional relationship with the real-time output values of each distributed photovoltaic.

[0094] The regulation objectives include economic optimization and secure and stable operation of the distribution network. Correspondingly, the distributed photovoltaic virtual clusters are also divided into economic regulation clusters and emergency regulation clusters.

[0095] When the regulation objective is economic optimization, the node voltage is controlled within the voltage warning area, and active and reactive power coordination optimization is performed on the distributed photovoltaic virtual cluster. Specifically, the first distributed photovoltaic virtual cluster (i.e., the economic regulation cluster) is regulated through the first regulation objective function, and the first regulation objective function is as shown in Equation (11) below.

[0096] (11).

[0097] In Equation (11), represents the first regulation objective function,[[]] represents the regulation weight of the node voltage change,[[]] represents the distributed photovoltaic virtual cluster number,[[]] represents the number of economic regulation clusters,[[]] represents due to the distributed photovoltaic virtual cluster the main node voltage change caused by the output change,[[]] represents the reference value of the main node voltage change obtained after optimization,[[]] represents the reactive power regulation weight,[[]] represents the reactive power instruction,[[]] represents the distributed photovoltaic virtual cluster actual reactive power output value.

[0098] When the regulation objective is the secure and stable operation of the distribution network, the node voltage is controlled within the voltage qualified area, and active and reactive power coordination optimization is performed on the distributed photovoltaic virtual cluster. Specifically, the second distributed photovoltaic virtual cluster (i.e., the emergency regulation cluster) is regulated through the second regulation objective function, and the second regulation objective function is as shown in Equation (12) below.

[0099] (12).

[0100] In Equation (12), represents the second regulation objective function,[[]] represents the distributed photovoltaic virtual cluster number,[[]] represents the number of emergency regulation clusters,[[]] represents the active power instruction,[[]] represents the distributed photovoltaic virtual cluster The actual active power output value.

[0101] In addition, both the economic regulation cluster and the emergency regulation cluster should satisfy the active power constraint and the reactive power constraint, as specifically shown in the following formula (13).

[0102] (13).

[0103] In formula (13), represents the reactive power command, represents the distributed photovoltaic virtual cluster of the actual reactive power output value, represents the active power command, represents the distributed photovoltaic virtual cluster of the actual active power output value, represents the maximum allowable active power deviation, represents the maximum allowable active power deviation, represents the maximum allowable reactive power deviation.

[0104] In this embodiment, when the regulation target is economic optimization, according to the active power command and the reactive power command, the first distributed photovoltaic virtual cluster is regulated through the first regulation objective function, and when the regulation target is the safe and stable operation of the power grid, according to the active power command and the reactive power command, the second distributed photovoltaic virtual cluster is regulated through the second regulation objective function, improving the economy, security and reliability of the low-voltage power grid operation.

[0105] It is worth mentioning that the grid day-ahead regulation method based on the dynamic division of the distributed photovoltaic virtual cluster provided in this embodiment can divide the distributed photovoltaic into virtual clusters based on the community theory with weight correction, and dynamically update with the real-time data of the distributed photovoltaic and the regional-level power grid. At the same time, it is regulated separately according to the different regulation target requirements of the regional-level power grid, meeting the economy, security and reliability of the power grid operation.

[0106] Corresponding to the grid day-ahead regulation method based on the dynamic division of the distributed photovoltaic virtual cluster described in the above embodiments, the embodiment of the present invention also provides a grid day-ahead regulation device based on the dynamic division of the distributed photovoltaic virtual cluster.

[0107] Specifically, Figure 2 shows the structural schematic diagram of the grid day-ahead regulation device based on the dynamic division of the distributed photovoltaic virtual cluster provided in the embodiment of the present invention.

[0108] As Figure 2As shown in the figure, the device includes: a distributed photovoltaic characteristic data acquisition module 210, configured to acquire the characteristic data of distributed photovoltaics, where the characteristic data includes active voltage sensitivity, reactive voltage sensitivity, active regulation capacity, reactive regulation capacity, and regulation cost; a distributed photovoltaic virtual cluster division module 220, configured to dynamically divide distributed photovoltaic virtual clusters based on the characteristic data of distributed photovoltaics and the regulation target through an improved community theory of weight clustering; a distributed photovoltaic output data prediction module 230, configured to predict the future output data of each distributed photovoltaic based on a pre-trained distributed photovoltaic prediction model, according to the historical output data, historical meteorological data, and meteorological prediction data of the distributed photovoltaics in the distributed photovoltaic virtual cluster; wherein, the distributed photovoltaic prediction model is constructed based on an attention mechanism and a long short-term memory network; a distributed photovoltaic virtual cluster regulation module 240, configured to regulate the distributed photovoltaic virtual cluster based on the future output data of each distributed photovoltaic in the distributed photovoltaic virtual cluster, according to the regulation target and the corresponding regulation target function.

[0109] In this embodiment, the characteristic data of distributed photovoltaics is acquired by the distributed photovoltaic characteristic data acquisition module 210, and the characteristic data includes active voltage sensitivity, reactive voltage sensitivity, active regulation capacity, reactive regulation capacity, and regulation cost; the distributed photovoltaic virtual cluster division module 220 dynamically divides distributed photovoltaic virtual clusters based on the characteristic data of distributed photovoltaics and the regulation target through an improved community theory of weight clustering; the distributed photovoltaic output data prediction module 230 predicts the future output data of each distributed photovoltaic based on a pre-trained distributed photovoltaic prediction model, according to the historical output data, historical meteorological data, and meteorological prediction data of the distributed photovoltaics in the distributed photovoltaic virtual cluster; wherein, the distributed photovoltaic prediction model is constructed based on an attention mechanism and a long short-term memory network; the distributed photovoltaic virtual cluster regulation module 240 regulates the distributed photovoltaic virtual cluster based on the future output data of each distributed photovoltaic in the distributed photovoltaic virtual cluster, according to the regulation target and the corresponding regulation target function. The device can effectively classify distributed photovoltaics by dynamically dividing distributed photovoltaic virtual clusters through an improved community theory of weight clustering, thereby improving the regulation ability and stable operation ability of the low-voltage power grid; at the same time, by regulating the distributed photovoltaic virtual cluster according to the predicted future output data and regulation target of distributed photovoltaics, the economy, safety, and reliability of the low-voltage power grid operation are improved.

[0110] It should be noted that the grid day-ahead regulation device based on the dynamic division of distributed photovoltaic virtual clusters provided in the embodiments of the present invention can be correspondingly referred to the grid day-ahead regulation method based on the dynamic division of distributed photovoltaic virtual clusters described in the above embodiments, and will not be elaborated here.

[0111] Figure 3Illustrates a schematic diagram of the physical structure of an electronic device, as Figure 3 shown. The electronic device may include: a processor 310, a communications interface 320, a memory 330, and a communication bus 340. Among them, the processor 310, the communications interface 320, and the memory 330 communicate with each other through the communication bus 340. The processor 310 may call the logical instructions in the memory 330 to execute a grid day-ahead regulation method based on the dynamic partitioning of a distributed photovoltaic virtual cluster. The method includes: obtaining characteristic data of the distributed photovoltaic, where the characteristic data includes active voltage sensitivity, reactive voltage sensitivity, active regulation capacity, reactive regulation capacity, and regulation cost; dynamically partitioning the distributed photovoltaic virtual cluster through an improved community theory of weight clustering based on the characteristic data of the distributed photovoltaic and the regulation target; predicting the future output data of each distributed photovoltaic according to the historical output data, historical meteorological data, and meteorological prediction data of the distributed photovoltaic within the distributed photovoltaic virtual cluster based on a pre-trained distributed photovoltaic prediction model; where the distributed photovoltaic prediction model is constructed based on an attention mechanism and a long short-term memory network; regulating the distributed photovoltaic virtual cluster based on the future output data of each distributed photovoltaic within the distributed photovoltaic virtual cluster according to the regulation target and the corresponding regulation target function.

[0112] In addition, when the logical instructions in the above-mentioned memory 330 are implemented in the form of software functional units and sold or used as an independent product, they may be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, may be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: various media such as a USB flash drive, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk, or an optical disc that can store program codes.

[0113] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the grid day-ahead regulation method based on the dynamic partitioning of distributed photovoltaic virtual clusters provided by the above-mentioned various methods. The method includes: obtaining characteristic data of distributed photovoltaics, where the characteristic data includes active power voltage sensitivity, reactive power voltage sensitivity, active power regulation capacity, reactive power regulation capacity, and regulation cost; based on the characteristic data of distributed photovoltaics and the regulation target, dynamically partitioning to obtain distributed photovoltaic virtual clusters through an improved community theory of weighted clustering; based on a pre-trained distributed photovoltaic prediction model, predicting the future output data of each distributed photovoltaic according to the historical output data, historical meteorological data, and meteorological prediction data of the distributed photovoltaics within the distributed photovoltaic virtual cluster; wherein, the distributed photovoltaic prediction model is constructed based on an attention mechanism and a long short-term memory network; based on the future output data of each distributed photovoltaic within the distributed photovoltaic virtual cluster, regulating the distributed photovoltaic virtual cluster according to the regulation target and the corresponding regulation target function.

[0114] On another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it realizes the grid day-ahead regulation method based on the dynamic partitioning of distributed photovoltaic virtual clusters provided by the above-mentioned various methods. The method includes: obtaining characteristic data of distributed photovoltaics, where the characteristic data includes active power voltage sensitivity, reactive power voltage sensitivity, active power regulation capacity, reactive power regulation capacity, and regulation cost; based on the characteristic data of distributed photovoltaics and the regulation target, dynamically partitioning to obtain distributed photovoltaic virtual clusters through an improved community theory of weighted clustering; based on a pre-trained distributed photovoltaic prediction model, predicting the future output data of each distributed photovoltaic according to the historical output data, historical meteorological data, and meteorological prediction data of the distributed photovoltaics within the distributed photovoltaic virtual cluster; wherein, the distributed photovoltaic prediction model is constructed based on an attention mechanism and a long short-term memory network; based on the future output data of each distributed photovoltaic within the distributed photovoltaic virtual cluster, regulating the distributed photovoltaic virtual cluster according to the regulation target and the corresponding regulation target function.

[0115] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0116] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0117] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features. And these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for day-ahead control of a power grid based on dynamic division of distributed photovoltaic virtual clusters, characterized in that: include: Acquiring characteristic data of distributed photovoltaics, the characteristic data including active voltage sensitivity, reactive voltage sensitivity, active regulation capacity, reactive regulation capacity, and regulation cost; Based on the characteristic data and control objectives of distributed photovoltaics, the distributed photovoltaic virtual cluster is obtained through dynamic partitioning of the improved community theory of weighted clustering. Based on the pre-trained distributed photovoltaic prediction model, the future output data of each distributed photovoltaic is predicted according to the historical output data, historical meteorological data and meteorological forecast data of the distributed photovoltaic virtual cluster; wherein the distributed photovoltaic prediction model is constructed based on the attention mechanism and the long short-term memory network; Based on the future output data of each distributed photovoltaic in the distributed photovoltaic virtual cluster, the distributed photovoltaic virtual cluster is regulated according to the regulation target and the corresponding regulation target function.

2. The method for day-ahead control of a power grid based on dynamic division of distributed photovoltaic virtual clusters according to claim 1 is characterized in that: The method of obtaining a dynamically divided distributed photovoltaic virtual cluster based on the characteristic data and control objectives of distributed photovoltaics through the improved community theory of weighted clustering includes: According to the control target, determining a cluster division index from the characteristic data of distributed photovoltaics; Calculating a similarity matrix according to the clustering index; The modularity index is used to correct the weight of the cluster division index, and the similarity matrix is ​​optimized to divide the distributed photovoltaic virtual cluster.

3. The method for day-ahead control of a power grid based on dynamic division of distributed photovoltaic virtual clusters according to claim 2 is characterized in that: According to the control target, cluster division indicators are determined from the characteristic data of distributed photovoltaics, including: When the control target is economically optimal, the reactive voltage sensitivity, reactive regulation capacity and regulation cost in the characteristic data are determined as cluster division indicators; When the control target is safe and stable operation of the power grid, the active voltage sensitivity, reactive voltage sensitivity, active regulation capacity and reactive regulation capacity in the characteristic data are determined as cluster division indicators.

4. The method for day-ahead control of a power grid based on dynamic division of distributed photovoltaic virtual clusters according to claim 1, characterized in that: The distributed photovoltaic prediction model based on pre-training predicts the future output data of each distributed photovoltaic according to the historical output data, historical meteorological data and meteorological forecast data of the distributed photovoltaic virtual cluster, including: Inputting the historical output data, historical meteorological data and meteorological forecast data into a pre-trained distributed photovoltaic prediction model to obtain a hidden state vector of a set time series; Extracting a time pattern matrix from the hidden state vector, and defining an evaluation function according to the time pattern matrix, the weight matrix and the hidden state vector; According to the evaluation function, the parameter weight is calculated; Summing the parameter weight and the weight of the temporal pattern matrix to obtain attention; The attention and the hidden state vector are linearly mapped and then added to obtain the output future output data of each distributed photovoltaic in the distributed photovoltaic virtual cluster.

5. The method for day-ahead control of a power grid based on dynamic division of distributed photovoltaic virtual clusters according to claim 1, characterized in that: The method of regulating the distributed photovoltaic virtual cluster based on the future output data of each distributed photovoltaic in the distributed photovoltaic virtual cluster according to the regulation target and the corresponding regulation target function includes: Determine active power instructions and reactive power instructions based on the future output data of each distributed photovoltaic in the distributed photovoltaic virtual cluster; When the regulation target is economically optimal, the first distributed photovoltaic virtual cluster is regulated by a first regulation target function according to the active power command and the reactive power command. When the control target is safe and stable operation of the power grid, the second distributed photovoltaic virtual cluster is controlled by a second control target function according to the active power command and the reactive power command; The first distributed photovoltaic virtual cluster is an economic regulation cluster, the second distributed photovoltaic virtual cluster is an emergency regulation cluster, the first distributed photovoltaic virtual cluster and the second distributed photovoltaic virtual cluster are both included in a distributed photovoltaic virtual cluster, and the first distributed photovoltaic virtual cluster and the second distributed photovoltaic virtual cluster both meet active power constraints and reactive power constraints.

6. The method for day-ahead control of a power grid based on dynamic division of distributed photovoltaic virtual clusters according to claim 5, characterized in that: The first regulation objective function is defined as follows: ； The second regulation objective function is defined as follows: ； The active power constraint and reactive power constraint are defined as follows: ； in, represents the first regulation objective function, represents the second regulation objective function, represents the node voltage change control weight, Indicates the distributed photovoltaic virtual cluster number, represents the number of economic regulation clusters, Indicates the number of emergency control clusters, Indicates that due to the distributed photovoltaic virtual cluster The voltage change of the main nodes caused by the output change, It represents the reference value of the voltage change of the main nodes obtained after optimization. represents the reactive power control weight, Indicates reactive power instruction. Represents a distributed photovoltaic virtual cluster The actual reactive power output value, Indicates active command. Represents a distributed photovoltaic virtual cluster The actual active output value of Indicates the maximum allowable active power deviation, Indicates the maximum allowed reactive power deviation.

7. A day-ahead control device for a power grid based on dynamic division of distributed photovoltaic virtual clusters, characterized in that: include: A distributed photovoltaic characteristic data acquisition module is used to acquire characteristic data of distributed photovoltaics, wherein the characteristic data includes active voltage sensitivity, reactive voltage sensitivity, active regulation capacity, reactive regulation capacity and regulation cost; Distributed photovoltaic virtual cluster division module is used to dynamically divide distributed photovoltaic virtual clusters based on the characteristic data and control objectives of distributed photovoltaics through the improved community theory of weighted clustering; A distributed photovoltaic output data prediction module is used to predict the future output data of each distributed photovoltaic based on the pre-trained distributed photovoltaic prediction model, according to the historical output data, historical meteorological data and meteorological forecast data of the distributed photovoltaic in the distributed photovoltaic virtual cluster; wherein the distributed photovoltaic prediction model is constructed based on the attention mechanism and the long short-term memory network; The distributed photovoltaic virtual cluster control module is used to control the distributed photovoltaic virtual cluster based on the future output data of each distributed photovoltaic in the distributed photovoltaic virtual cluster, according to the control target and the corresponding control target function.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the day-ahead control method for a power grid based on dynamic division of distributed photovoltaic virtual clusters according to any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the day-ahead control method for a power grid based on dynamic division of distributed photovoltaic virtual clusters according to any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the day-ahead control method for a power grid based on dynamic division of distributed photovoltaic virtual clusters according to any one of claims 1 to 6 is implemented.

Images