Distribution network power flow control method and system based on distributed power supply

Abstract

The present invention provides a distribution network power flow control method and system based on distributed power sources, which relates to the technical field of distribution network power flow control. The method and system include dividing the distribution network into a plurality of control areas according to the load density and distributed power penetration rate of the distribution network, combining cluster analysis and fuzzy evaluation algorithm, and using a pre-built regional division model; establishing a corresponding regional power flow control model according to the adjusted control area division result, using a distributed solution algorithm to perform distributed solution on the regional power flow control model of all areas, using an iterative algorithm to perform global correction on the local optimal power flow control strategy of each control area, obtaining the global optimal power flow control strategy, and obtaining the current state and uncertain parameters of the distribution network in real time, automatically triggering the control area division and power flow control strategy solution process; realizing the rapid real-time application of the global optimal power flow control strategy, and completing the distribution network power flow control.

Description

Technical Field

[0001] The present invention relates to a power distribution network flow control technology, and in particular to a power distribution network flow control method and system based on a distributed power source. Background Art

[0002] With the access of a large number of distributed power sources, the power flow characteristics of the distribution network are becoming increasingly complex. The traditional centralized power flow control method is difficult to adapt to the high-dimensional, nonlinear, and strongly coupled characteristics of the distribution network, and faces problems such as low computational efficiency, poor robustness, and insufficient real-time performance. In the prior art, some methods adopt the idea of ​​hierarchical partitioning to divide the distribution network into multiple control areas and perform power flow optimization within the area. However, when dividing the area, it is mainly based on geographical location information, and the electrical characteristics of the distribution network and the access of distributed power sources are not fully considered, making it difficult to obtain a reasonable control area division result. When formulating power flow control strategies for each area, most of the existing methods do not consider the uncertainty of distributed power output and load demand, and the robustness needs to be improved. In addition, when performing global coordinated optimization, the existing methods use centralized optimization models and algorithms, which have low computational efficiency and are difficult to achieve real-time update and application of power flow control strategies. Summary of the invention

[0003] The present invention provides a power distribution network flow control method and system based on distributed power sources, which can solve the problems in the prior art.

[0004] A first aspect of the present invention,

[0005] A method for controlling power flow in a distribution network based on a distributed power source is provided, comprising:

[0006] According to the load density and distributed generation penetration rate of the distribution network, combined with cluster analysis and fuzzy evaluation algorithm, the distribution network is divided into several control areas using a pre-built area division model to obtain the initial control area division result, and the changes in the load density and distributed generation penetration rate of the distribution network are monitored in real time. According to the monitoring results, the initial control area division result is adaptively adjusted to obtain the adjusted control area division result;

[0007] According to the adjusted control area division results, a corresponding regional power flow control model is established. The regional power flow control model takes minimizing the regional grid loss and voltage deviation as the regional comprehensive optimization goal, expresses the uncertainty of distributed power output and load demand as a confidence interval, and converts the confidence interval into a deterministic constraint of the regional power flow control model through a robust optimization method;

[0008] A distributed solution algorithm is used to perform distributed solution on the regional power flow control model of all areas to obtain the local optimal power flow control strategy of each control area. At the same time, the power balance constraint between regions is introduced, and a global consistency correction model is established. The iterative algorithm is used to perform global correction on the local optimal power flow control strategy of each control area to obtain the global optimal power flow control strategy. The current state and uncertain parameters of the distribution network are obtained in real time, and the control area division and power flow control strategy solution process are automatically triggered to generate an updated global optimal power flow control strategy in each control cycle. At the same time, based on offline scene matching and online strategy calling technology, the rapid real-time application of the global optimal power flow control strategy is realized to complete the power flow control of the distribution network.

[0009] Preferably,

[0010] According to the load density and distributed generation penetration rate of the distribution network, combined with cluster analysis and fuzzy evaluation algorithm, the distribution network is divided into several control areas using the pre-built area division model to obtain the initial control area division results, and the changes in the load density and distributed generation penetration rate of the distribution network are monitored in real time. The initial control area division results are adaptively adjusted according to the monitoring results, and the adjusted control area division results include:

[0011] According to the current state of the distribution network, load density, distributed power penetration rate and electrical distance are selected as characteristic parameters, wherein the load density reflects the load characteristics of different regions, the distributed power penetration rate reflects the power distribution of different regions, and the electrical distance reflects the electrical connection strength between nodes;

[0012] An improved fuzzy c-means clustering algorithm is used to initially classify the distribution network nodes, the load density and the distributed generation penetration rate are used as inputs of the regional division model, and the initial classification results of the distribution network nodes are obtained by optimizing the objective function to minimize the intra-class distance and the inter-class distance;

[0013] The initial classification result is optimized by using a fuzzy evaluation method, the electrical distance is used as an additional evaluation index, the membership degree of each distribution network node to other class centers is calculated, and the distribution network nodes are reclassified according to the maximum membership principle to obtain a revised classification result;

[0014] Determine whether the revised classification result meets the preset regional division requirements, which include the load density dispersion degree and power penetration rate dispersion degree within each region and the load density balance degree and power penetration rate balance degree between regions. If so, use the current classification result as the initial control area division result; otherwise, return to the step of initial classification of distribution network nodes, and reclassify and optimize the distribution network nodes;

[0015] During the operation of the distribution network, the load changes and power output changes in each area are monitored in real time. When the load change or power output change in any area exceeds the preset threshold, the regional boundaries are dynamically adjusted according to the current operating status of the distribution network, and the regional division model parameters are updated to obtain the adjusted control area division results.

[0016] Preferably,

[0017] According to the current operating status of the distribution network, the regional boundaries are dynamically adjusted, and the regional division model parameters are updated. The adjusted control area division results include:

[0018] Acquire real-time operation data of the distribution network, the real-time operation data including load data, power supply data and topology data, and model the real-time operation data as a non-stationary time series, the time series including an input feature vector and an output target vector, wherein the input feature vector includes load data, power supply data and topology data, and the output target vector includes a balance index and an independence index;

[0019] Based on the time series, an online learning algorithm is used to minimize the cumulative loss function and update the parameters of the regional division model in real time. The cumulative loss function is obtained by summing up the loss function at each moment. The loss function at each moment includes a balance loss and an independence loss. The balance loss measures the discrete degree of the internal load density and power penetration rate of each region, and the independence loss measures the connection strength between regions.

[0020] Design an adaptive learning rate adjustment strategy to dynamically adjust the learning rate of each parameter based on the cumulative square sum of historical gradients and output the updated parameters;

[0021] The updated parameters are decoded into regional division results, and the control regional division results after dynamic adjustment of the distribution network are obtained according to the node membership information by combining soft division and hard division.

[0022] Preferably,

[0023] The calculation formula for the regional comprehensive optimization objective of minimizing regional grid loss and voltage deviation is as follows:

[0024] ;

[0025] Among them, J represents the comprehensive optimization target, α represents the weight coefficient of network loss, and N b represents the number of nodes in the region, G ij represents the conductance between nodes i and j, V i Represents the voltage amplitude of node i, V j represents the voltage amplitude at node j, θ ijrepresents the voltage phase angle difference between node i and node j, β represents the voltage deviation weight coefficient, V ref，i Represents the reference voltage of node i.

[0026] Preferably,

[0027] The distributed solution algorithm is used to perform distributed solution on the regional power flow control model of all regions, and the local optimal power flow control strategy of each control area is obtained, including:

[0028] According to the topological structure and control requirements of the distribution network, the distribution network is divided into multiple control areas, each of which is equipped with a regional controller. At the same time, according to the communication topology of the distribution network, the communication links and directions between the regional controllers are determined, and the communication topology directed graph of the control area is constructed;

[0029] Initialize the penalty factor, primal variables and dual variables of the distributed solution algorithm, wherein the primal variables include the power output and load power of each control area, and the dual variables are the boundary power of each control area;

[0030] Each regional controller solves the regional optimization sub-problem in parallel to obtain the updated value of the original variable. At the same time, each regional controller asynchronously exchanges the updated value of the dual variable according to the communication topology directed graph. If the number of edges in the communication topology directed graph changes by more than the communication topology change threshold, the distributed solution algorithm is restarted and the distributed solution algorithm is returned to the initialization step.

[0031] The update values ​​of the original variables and the update values ​​of the dual variables are encrypted using homomorphic encryption technology, and the update values ​​of the dual variables are introduced with random perturbations using differential privacy technology to obtain the encrypted update values ​​of the original variables and the perturbed update values ​​of the dual variables. Based on the encrypted update values ​​of the original variables and the perturbed update values ​​of the dual variables, a hierarchical decreasing strategy is used to update the penalty factor and the dual variables.

[0032] The updated penalty factor, the encrypted original variable update value and the perturbed dual variable update value are used to solve the next round of regional optimization subproblems until the distributed solution algorithm meets the preset convergence conditions and outputs the local optimal power flow control strategy for each control area.

[0033] Preferably,

[0034] The inter-regional power balance constraint is introduced, a global consistency correction model is established, and an iterative algorithm is used to globally correct the local optimal power flow control strategy of each control area. The global optimal power flow control strategy includes:

[0035] A global consistency correction model including inter-regional power balance constraints is established, with the local optimal power flow control strategy of each control area as the optimization variable, the minimum operation cost of the entire network as the optimization goal, and the inter-regional power exchange constraints considered;

[0036] Send the initial parameters of the global correction model to each regional controller;

[0037] Each regional controller uses an optimization algorithm to train the local optimal power flow control strategy based on local historical operation data to obtain the first local optimal power flow control strategy, and uses an encrypted communication protocol to encrypt the first local optimal power flow control strategy and upload it to the central aggregation server;

[0038] The central aggregation server performs weighted averaging on the first local optimal power flow control strategy received in the ciphertext domain to obtain the parameters of the global correction model, and sends it to each regional controller. Each regional controller uses the parameters of the received global correction model to fine-tune and update the local local optimal power flow control strategy.

[0039] Based on the fine-tuned and updated local optimal power flow control strategy of each region, the network loss and global optimization objective function value of each region are calculated to determine whether the global optimal conditions are met;

[0040] If the global optimal condition is not met, the controller of each region returns to the step of training the local optimal power flow control strategy and starts a new round of iteration. If the global optimal condition is met, the current local optimal power flow control strategy of each region is output as the global optimal power flow control strategy.

[0041] The second aspect of the present invention,

[0042] Provided is a power distribution network flow control system based on distributed power sources, comprising:

[0043] The first unit is used to divide the distribution network into several control areas according to the load density and distributed power penetration rate of the distribution network, combined with cluster analysis and fuzzy evaluation algorithm, using a pre-built area division model to obtain an initial control area division result, and to monitor the changes in the load density and distributed power penetration rate of the distribution network in real time, and to adaptively adjust the initial control area division result according to the monitoring result to obtain an adjusted control area division result;

[0044] The second unit is used to establish a corresponding regional power flow control model according to the adjusted control area division result. The regional power flow control model takes minimizing the regional grid loss and voltage deviation as the regional comprehensive optimization goal, expresses the uncertainty of the distributed power output and load demand as a confidence interval, and converts the confidence interval into a deterministic constraint of the regional power flow control model through a robust optimization method;

[0045] The third unit is used to use a distributed solution algorithm to perform distributed solution on the regional power flow control model of all regions, obtain the local optimal power flow control strategy of each control region, introduce the power balance constraint between regions, establish a global consistency correction model, and use an iterative algorithm to perform global correction on the local optimal power flow control strategy of each control region to obtain the global optimal power flow control strategy, and obtain the current state and uncertain parameters of the distribution network in real time, automatically trigger the control area division and power flow control strategy solution process, and generate an updated global optimal power flow control strategy in each control cycle; at the same time, based on offline scene matching and online strategy calling technology, the global optimal power flow control strategy can be quickly and real-time applied to complete the power flow control of the distribution network.

[0046] The third aspect of the present invention,

[0047] An electronic device is provided, comprising:

[0048] processor;

[0049] a memory for storing processor-executable instructions;

[0050] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0051] A fourth aspect of the present invention,

[0052] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0053] In the present invention, by combining the distribution network control area division method based on load density and distributed power penetration rate with cluster analysis and fuzzy evaluation algorithm, the electrical characteristics and power distribution of the distribution network can be comprehensively considered to obtain a more reasonable control area division result, laying the foundation for subsequent power flow control. In view of the uncertainty of distributed power output and load demand, a confidence interval representation method is introduced, and it is equivalently converted into the deterministic constraints of the regional power flow control model through a robust optimization method, thereby improving the robustness and reliability of the power flow control strategy. A distributed solution algorithm is used to solve the regional power flow control model, which reduces the complexity of solving the optimization problem and improves the calculation efficiency. At the same time, inter-regional power balance constraints and a global consistency correction model are introduced, and distributed decision-making and global coordination are realized through an iterative algorithm to ensure the global optimality of power flow control. An adaptive trigger mechanism is designed to automatically trigger the control area redivision and power flow control strategy solution process according to the real-time changes in the distribution network load and power supply, thereby realizing the real-time update of the power flow control strategy and improving the real-time and adaptability of the control. Based on offline scene matching and online strategy calling technology, the global optimal power flow control strategy is quickly applied in real time, which greatly shortens the time from decision-making to execution of power flow control and improves the efficiency and availability of power flow control. Compared with the traditional centralized power flow control method, the distributed power flow control method based on control area division proposed in this invention can better adapt to the high-dimensional, nonlinear, and strong coupling characteristics of the distribution network, and has the advantages of high computational efficiency, strong robustness, and good real-time performance, providing a new solution for the optimized operation of the distribution network. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 It is a flow chart of a method for controlling power flow in a distribution network based on distributed power sources according to an embodiment of the present invention;

[0055] Figure 2 It is a schematic diagram of the structure of a power distribution network flow control system based on distributed power sources according to an embodiment of the present invention. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0057] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0058] Figure 1 FIG. 1 is a flow chart of a method for controlling power flow in a distribution network based on distributed power sources according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0059] S101. According to the load density and distributed generation penetration rate of the distribution network, combined with cluster analysis and fuzzy evaluation algorithm, the distribution network is divided into several control areas using a pre-built area division model to obtain an initial control area division result, and the changes in the load density and distributed generation penetration rate of the distribution network are monitored in real time. According to the monitoring results, the initial control area division result is adaptively adjusted to obtain an adjusted control area division result.

[0060] Among them, the cluster analysis algorithm is used to classify nodes with similar load density and distributed power penetration characteristics into one category to form the initial control area division result. The fuzzy evaluation algorithm is used to evaluate the rationality of the initial control area division result. By establishing a fuzzy evaluation index system including multiple indicators such as load density and distributed power penetration rate, the comprehensive evaluation value of each initial control area is calculated, and the initial control area division result is optimized and adjusted according to the comprehensive evaluation value to obtain a more reasonable control area division result.

[0061] In an optional embodiment,

[0062] According to the load density and distributed generation penetration rate of the distribution network, combined with cluster analysis and fuzzy evaluation algorithm, the distribution network is divided into several control areas using the pre-built area division model to obtain the initial control area division results, and the changes in the load density and distributed generation penetration rate of the distribution network are monitored in real time. The initial control area division results are adaptively adjusted according to the monitoring results, and the adjusted control area division results include:

[0063] According to the current state of the distribution network, load density, distributed power penetration rate and electrical distance are selected as characteristic parameters, wherein the load density reflects the load characteristics of different regions, the distributed power penetration rate reflects the power distribution of different regions, and the electrical distance reflects the electrical connection strength between nodes;

[0064] An improved fuzzy c-means clustering algorithm is used to initially classify the distribution network nodes, the load density and the distributed generation penetration rate are used as inputs of the regional division model, and the initial classification results of the distribution network nodes are obtained by optimizing the objective function to minimize the intra-class distance and the inter-class distance;

[0065] The initial classification result is optimized by using a fuzzy evaluation method, the electrical distance is used as an additional evaluation index, the membership degree of each distribution network node to other class centers is calculated, and the distribution network nodes are reclassified according to the maximum membership principle to obtain a revised classification result;

[0066] Determine whether the revised classification result meets the preset regional division requirements, which include the load density dispersion degree and power penetration rate dispersion degree within each region and the load density balance degree and power penetration rate balance degree between regions. If so, use the current classification result as the initial control area division result; otherwise, return to the step of initial classification of distribution network nodes, and reclassify and optimize the distribution network nodes;

[0067] During the operation of the distribution network, the load changes and power output changes in each area are monitored in real time. When the load change or power output change in any area exceeds the preset threshold, the regional boundaries are dynamically adjusted according to the current operating status of the distribution network, and the regional division model parameters are updated to obtain the adjusted control area division results.

[0068] For example, first, according to the current operating status of the distribution network, load density, distributed power penetration rate and electrical distance are selected as characteristic parameters, among which load density reflects the load characteristics of different regions, distributed power penetration rate reflects the power distribution situation in different regions, and electrical distance characterizes the electrical connection strength between distribution network nodes.

[0069] In the initial area division stage, the improved fuzzy c-means clustering algorithm is used to classify the distribution network nodes. The load density and distributed generation penetration rate are used as the input features of the clustering model. The initial classification results of the distribution network nodes are obtained by optimizing the objective function to minimize the ratio of the sum of squared distances within the class to the sum of squared distances between classes. The sum of squared distances within the class reflects the discrete degree of the characteristic parameters of the nodes in the same class, and the sum of squared distances between classes reflects the differences between different classes. Minimizing the ratio of the two can obtain compact and highly separated clustering results.

[0070] In order to further optimize the initial classification results, a fuzzy evaluation method is introduced, and the electrical distance is used as an additional evaluation index to calculate the membership of each distribution network node with other class centers. The membership degree indicates the similarity between a node and a certain category. The larger the value, the more likely the node belongs to that category. The distribution network nodes are reclassified according to the maximum membership principle, that is, the nodes are divided into the class with the largest membership, and the revised classification results are obtained.

[0071] Determine whether the revised classification results meet the preset regional division requirements, mainly including the degree of dispersion of load density and power penetration within each region, as well as the degree of balance of load density and power penetration between regions. Among them, the degree of dispersion within the region reflects the consistency of the characteristic parameters of the nodes in the same region. The smaller the degree of dispersion, the more similar the node characteristics in the region. The degree of balance between regions reflects whether the load and power distribution between different regions is balanced. The higher the degree of balance, the more uniform the load and power distribution in each region. If the revised classification results meet the preset requirements, the current classification results will be used as the initial control area division plan, otherwise return to the initial clustering step, adjust the clustering parameters, and reclassify and optimize the distribution network nodes.

[0072] Considering the dynamic change characteristics of the load and power supply of the distribution network, in order to adapt to the actual operation needs, during the operation of the distribution network, it is necessary to monitor the load changes and power output changes in each area in real time. When the load change amplitude or power output change amplitude of any area exceeds the preset threshold, it is considered that the current regional division scheme is no longer applicable, and the regional boundaries need to be dynamically adjusted according to the current operating status of the distribution network. During the adjustment, the characteristic parameters in the regional division model are updated, the node membership is recalculated, and the nodes are re-divided according to the maximum membership principle to obtain the adjusted control area division result to adapt to the real-time operating status changes of the distribution network.

[0073] In this embodiment, multiple characteristic parameters such as load density, distributed power penetration rate and electrical distance are comprehensively considered, which can fully reflect the load characteristics, power distribution and electrical connection strength between nodes in different areas of the distribution network, making the regional division more comprehensive and reasonable. The improved fuzzy c-means clustering algorithm is used for initial regional division. By optimizing the ratio of intra-class distance and inter-class distance, a compact and highly separated clustering result is obtained, which lays a good foundation for subsequent regional division. The fuzzy evaluation method is introduced to optimize the initial clustering results, and the electrical distance is used as an additional evaluation index. The nodes are reclassified in combination with the maximum membership principle, which improves the accuracy and interpretability of regional division. By setting quantitative evaluation indicators for regional division, including the discrete degree of load density within the region, the discrete degree of power penetration rate, and the balanced degree of load density and power penetration rate between regions, a clear optimization goal and evaluation criteria are provided for regional division. In view of the dynamic change characteristics of the load and power supply of the distribution network, a dynamic adjustment mechanism for regional boundaries is proposed to monitor the changes in the operating parameters of each region in real time. When the preset threshold is exceeded, the regional redivision is triggered, and the dynamic adaptation of the control regional division to the actual operating state of the distribution network is realized.

[0074] In an optional embodiment,

[0075] According to the current operating status of the distribution network, the regional boundaries are dynamically adjusted, and the regional division model parameters are updated. The adjusted control area division results include:

[0076] Acquire real-time operation data of the distribution network, the real-time operation data including load data, power supply data and topology data, and model the real-time operation data as a non-stationary time series, the time series including an input feature vector and an output target vector, wherein the input feature vector includes load data, power supply data and topology data, and the output target vector includes a balance index and an independence index;

[0077] Based on the time series, an online learning algorithm is used to minimize the cumulative loss function and update the parameters of the regional division model in real time. The cumulative loss function is obtained by summing up the loss function at each moment. The loss function at each moment includes a balance loss and an independence loss. The balance loss measures the discrete degree of the internal load density and power penetration rate of each region, and the independence loss measures the connection strength between regions.

[0078] Design an adaptive learning rate adjustment strategy to dynamically adjust the learning rate of each parameter based on the cumulative square sum of historical gradients and output the updated parameters;

[0079] The updated parameters are decoded into regional division results, and the control regional division results after dynamic adjustment of the distribution network are obtained according to the node membership information by combining soft division and hard division.

[0080] Exemplarily, first obtain the real-time operation data of the distribution network, including load data, power supply data and topology data. The load data represents the power demand of each node, the power supply data represents the output of the distributed power source, and the topology data represents the connection relationship between the nodes. The acquired real-time operation data is modeled as a non-stationary time series, and each moment of the time series includes an input feature vector and an output target vector. The input feature vector is composed of load data, power supply data and topology data, which reflects the current operating status of the distribution network. The output target vector represents the performance indicators of the regional division, including the balance index and the independence index. The balance index measures the matching degree between the load demand and power supply in each control area, and the independence index measures the electrical connection strength between different control areas.

[0081] Based on the time series obtained by modeling, the online learning algorithm is used to update the parameters of the regional division model in real time. The goal of the online learning algorithm is to minimize the cumulative loss function so that the regional division model can continuously adapt to changes in the distribution of real-time operation data. The cumulative loss function is obtained by summing up the loss function at each moment, which represents the overall performance of the regional division model over the entire time period. The loss function at each moment includes two parts: balance loss and independence loss. Balance loss measures the degree of discreteness of the load density and power penetration rate within each region. The smaller the discreteness, the more balanced the load demand and power supply within the region. Independence loss measures the connection strength between regions. The smaller the connection strength, the lower the degree of electrical coupling between regions and the higher the independence of operation of each region.

[0082] In the parameter update process, in order to further improve the algorithm's adaptability and convergence speed, an adaptive learning rate adjustment strategy is designed. This strategy dynamically adjusts the learning rate of each parameter based on the cumulative square sum of historical gradients. For parameters that are updated frequently and have large gradient changes, the learning rate is small to avoid oscillation caused by too fast parameter updates. For parameters that are updated less frequently and have slow gradient changes, the learning rate is large to accelerate the convergence speed of the parameters. By adaptively adjusting the learning rate, the stability and convergence speed of the algorithm can be balanced, and the efficiency of online learning can be improved.

[0083] After the parameters of the regional division model are updated, the updated parameters need to be decoded into the actual control area division results of the distribution network. A combination of soft division and hard division is adopted to take into account the continuity and discreteness of the division results. Soft division refers to dividing each node into multiple control areas and assigning different membership weights. The membership weight indicates the degree of correlation between the node and each control area. The larger the weight, the stronger the correlation between the node and the area. Hard division refers to the hard regional allocation of nodes according to the membership weight, and the nodes are divided into the control area with the highest membership. By combining soft division and hard division, a clear boundary division result can be obtained while ensuring the continuity of the control area. Finally, the dynamically adjusted control area division scheme of the distribution network is obtained.

[0084] In this embodiment, by modeling the real-time operation data of the distribution network as a time series and using an online learning algorithm to continuously update the parameters of the regional division model, the division of the control area can be dynamically adjusted according to the real-time operation status of the distribution network. This dynamic division can better adapt to the actual operation of the distribution network and improve the efficiency and performance of regional control. In the process of regional division, by introducing balance loss and independence loss in the loss function, the load density and power penetration rate within each region can be more balanced, and the connection strength between regions is low, thereby improving the rationality and effectiveness of regional division. An adaptive learning rate adjustment strategy is designed, and the learning rate of each parameter is dynamically adjusted according to the cumulative square sum of historical gradients, which can accelerate the convergence of the model and improve the efficiency of parameter updating. By combining soft division and hard division, the regional division result is obtained according to the membership information of the node, so that the regional boundary is smoother and the discontinuity problem caused by hard division is avoided. Since the regional division is based on the real-time operation data of the distribution network, it can avoid leaking the user's privacy information and protect data privacy.

[0085] S102. According to the adjusted control area division results, a corresponding regional power flow control model is established. The regional power flow control model takes minimizing the regional grid loss and voltage deviation as the comprehensive optimization goal at the regional level, expresses the uncertainty of distributed power output and load demand as confidence intervals, and converts the confidence intervals into deterministic constraints of the regional power flow control model through a robust optimization method.

[0086] Among them, network loss reflects the loss of electric energy during transmission and is an important indicator for evaluating the economic operation of distribution networks; voltage deviation reflects the degree of deviation between the voltage of each node and the nominal voltage and is an important indicator for evaluating the power quality of distribution networks. By weightedly combining network loss and voltage deviation and constructing a comprehensive optimization target at the regional level, it is possible to achieve optimal control of regional power flow while taking into account both economic efficiency and power quality.

[0087] Considering the large uncertainty of the output and load demand of distributed power sources in the distribution network, in order to improve the robustness of the regional power flow control model, this paper adopts the confidence interval method to model the uncertainty of the output and load demand of distributed power sources. According to the predicted value and prediction error distribution of the output and load demand of distributed power sources, a confidence interval with a certain confidence level (such as 90% or 95%) is constructed to limit the uncertainty of the output and load demand of distributed power sources within the confidence interval. Then, using the duality theory of robust optimization, the confidence interval constraint is equivalently transformed into the deterministic constraint condition of the regional power flow control model, thereby obtaining a robust optimization power flow control model that includes the uncertainty of distributed power output and load demand.

[0088] Specifically, we first select the minimization of regional network loss and voltage deviation as the optimization goal. Considering the different dimensions of network loss and voltage deviation, we introduce weighting coefficients to obtain the comprehensive optimization goal at the regional level. Furthermore, according to the information such as the regional power grid topology, line parameters, load and power distribution, a regional power flow equation group is established, including node power balance equations and branch power flow equations. According to the prediction error distribution of distributed power output and load demand, the confidence interval method or scenario method is used to generate confidence intervals or discrete scenarios of uncertain parameters, and the regional uncertainty set is constructed. The duality theory of robust optimization is introduced to transform the optimization problem with uncertain parameters into an equivalent deterministic optimization problem, and a regional power flow control model is obtained.

[0089] In an optional embodiment,

[0090] The calculation formula for the regional comprehensive optimization objective of minimizing regional grid loss and voltage deviation is as follows:

[0091] ;

[0092] Among them, J represents the comprehensive optimization target, α represents the weight coefficient of network loss, and N b represents the number of nodes in the region, G ij represents the conductance between nodes i and j, V i Represents the voltage amplitude of node i, V j represents the voltage amplitude at node j, θ ij represents the voltage phase angle difference between node i and node j, β represents the voltage deviation weight coefficient, V ref，i Represents the reference voltage of node i.

[0093] In this embodiment, by establishing a regional power flow control model, minimizing the regional network loss and voltage deviation as the comprehensive optimization goal, the power flow in the region can be optimized and controlled, and the economic operation efficiency and power quality of the distribution network can be improved. Considering the large uncertainty in the output of distributed power sources and load demand in the distribution network, the confidence interval method is used to model these uncertainties, and the confidence interval is equivalently converted into the deterministic constraints of the model through the robust optimization method, thereby improving the robustness of the regional power flow control model, so that it can adapt to the uncertain fluctuations of distributed power sources and loads. By constructing a comprehensive optimization goal by weighted combination of network loss and voltage deviation, the unified treatment of the two goals of economy (network loss) and power quality (voltage deviation) is achieved, so that the regional power flow control can be optimized while taking into account both economy and power quality. The dual theory of robust optimization is introduced to convert the optimization problem with uncertain parameters into a deterministic optimization problem, avoiding the large amount of computational burden brought by the scenario method or Monte Carlo method, and improving the efficiency of model solution. The robustness and efficiency of power flow control are improved, and the distributed characteristics of the distribution network are adapted, which is of great significance to improving the operation efficiency and reliability of the distribution network.

[0094] S103. A distributed solution algorithm is used to perform distributed solution on the regional power flow control model of all regions to obtain the local optimal power flow control strategy of each control region. At the same time, the power balance constraint between regions is introduced, and a global consistency correction model is established. An iterative algorithm is used to perform global correction on the local optimal power flow control strategy of each control region to obtain the global optimal power flow control strategy, and the current state and uncertain parameters of the distribution network are obtained in real time. The control area division and power flow control strategy solution process are automatically triggered to generate an updated global optimal power flow control strategy in each control cycle. At the same time, based on offline scene matching and online strategy calling technology, the global optimal power flow control strategy is quickly and real-time applied to complete the power flow control of the distribution network.

[0095] Among them, the distributed solution algorithm can choose the improved alternating direction method of multipliers (ADMM) for distributed solution. ADMM is a distributed optimization algorithm based on Lagrange multiplier method and variable partitioning technology. It decomposes the original problem into several sub-problems and solves each sub-problem alternately in an iterative manner to finally obtain the optimal solution to the original problem. By treating the regional-level robust optimization power flow control model of each control area as a sub-problem, the ADMM algorithm is used to solve the sub-problems of all regions in parallel to obtain the local optimal power flow control strategy of each control area, including distributed power generation output scheduling, adjustable load control and reactive power compensation.

[0096] Since each control area independently solves its own regional power flow control model, it may lead to unbalanced power flow between regions. In order to ensure the balance of the overall power flow of the distribution network, the inter-regional power balance constraint is introduced on the basis of the ADMM algorithm, and a global consistency correction model is established. The global consistency correction model takes the power flow balance of the inter-regional tie line as the constraint condition, and the local optimal power flow control strategy of each control area as the decision variable to construct a global optimization problem. Then, an iterative algorithm is used to solve the global optimization problem. In each iteration, the decision variables of other areas are fixed, the decision variables of the current area are optimized, and the global optimal power flow control strategy is obtained by iteration until convergence. The global optimal power flow control strategy not only satisfies the power balance constraint between regions, but also takes into account the local optimality of each control area, realizing the coordinated optimization of global and local.

[0097] A rolling optimization control framework based on a real-time trigger mechanism is adopted. By acquiring the current state information of the distribution network (such as node voltage, branch power flow, etc.) and uncertain parameter information (such as distributed power output forecast value, load demand forecast value and its confidence interval, etc.) in real time, the control area division and power flow control strategy solution process are automatically triggered, and an updated global optimal power flow control strategy is generated in each control cycle. At the same time, a combination of offline scene matching and online strategy calling is adopted to pre-generate the global optimal power flow control strategy for a variety of typical scenarios offline, and construct a scenario-strategy mapping table. During online operation, the real-time state of the distribution network is matched with the offline scene, and the global optimal power flow control strategy closest to the current scene is quickly retrieved and directly applied to the power flow control of the distribution network, thereby greatly improving the real-time and adaptability of the control strategy.

[0098] In an optional embodiment,

[0099] The distributed solution algorithm is used to perform distributed solution on the regional power flow control model of all regions, and the local optimal power flow control strategy of each control area is obtained, including:

[0100] According to the topological structure and control requirements of the distribution network, the distribution network is divided into multiple control areas, each of which is equipped with a regional controller. At the same time, according to the communication topology of the distribution network, the communication links and directions between the regional controllers are determined, and the communication topology directed graph of the control area is constructed;

[0101] Initialize the penalty factor, primal variables and dual variables of the distributed solution algorithm, wherein the primal variables include the power output and load power of each control area, and the dual variables are the boundary power of each control area;

[0102] Each regional controller solves the regional optimization sub-problem in parallel to obtain the updated value of the original variable. At the same time, each regional controller asynchronously exchanges the updated value of the dual variable according to the communication topology directed graph. If the number of edges in the communication topology directed graph changes by more than the communication topology change threshold, the distributed solution algorithm is restarted and the distributed solution algorithm is returned to the initialization step.

[0103] The update values ​​of the original variables and the update values ​​of the dual variables are encrypted using homomorphic encryption technology, and the update values ​​of the dual variables are introduced with random perturbations using differential privacy technology to obtain the encrypted update values ​​of the original variables and the perturbed update values ​​of the dual variables. Based on the encrypted update values ​​of the original variables and the perturbed update values ​​of the dual variables, a hierarchical decreasing strategy is used to update the penalty factor and the dual variables.

[0104] The updated penalty factor, the encrypted original variable update value and the perturbed dual variable update value are used to solve the next round of regional optimization subproblems until the distributed solution algorithm meets the preset convergence conditions and outputs the local optimal power flow control strategy for each control area.

[0105] For example, first, according to the topological structure and control requirements of the distribution network, the distribution network is divided into multiple control areas, and each control area is equipped with a regional controller. The regional controller is responsible for collecting the distribution network operation data in the area and communicating and coordinating with other regional controllers. At the same time, according to the communication topology of the distribution network, the communication links and directions between the regional controllers are determined, and the communication topology directed graph of the control area is constructed. The communication topology directed graph reflects the communication relationship and data transmission direction between the regional controllers.

[0106] Next, the penalty factor, primal variables, and dual variables of the distributed solution algorithm are initialized. The penalty factor is a weight coefficient used to balance the optimization objective and constraints. By adjusting the size of the penalty factor, the solution accuracy and convergence speed of the optimization problem can be controlled. The primal variables include the power output and load power of each control area, which represent the decision variables of the distribution network flow optimization control. The dual variable is the boundary power of each control area, which represents the power exchange between adjacent control areas.

[0107] During the iterative process of the distributed solution algorithm, each regional controller solves the regional optimization subproblem in parallel to obtain the updated value of the original variable. The regional optimization subproblem is to decompose the distribution network power flow optimization control problem into multiple independent subproblems. Each subproblem corresponds to a control area and is solved by the controller of the area. By solving the regional optimization subproblems in parallel, the computational efficiency of the algorithm can be significantly improved. After solving the regional optimization subproblem, each regional controller asynchronously exchanges the updated value of the dual variable according to the communication topology directed graph. The asynchronous communication mechanism can reduce the impact of communication delay on the convergence speed of the algorithm. If the number of edges in the communication topology directed graph changes by more than the preset communication topology change threshold, the distributed solution algorithm is triggered to restart and return to the distributed solution algorithm initialization step. The communication topology change threshold is used to determine whether the communication topology has changed significantly. When the communication link is disconnected or added, the algorithm performance may be degraded and the algorithm parameters need to be reinitialized.

[0108] In order to protect the privacy of distribution network operation data, this paper uses homomorphic encryption and differential privacy technology to encrypt and perturb the original variables and dual variables. Homomorphic encryption is an encryption technology that allows calculations on ciphertext, which can achieve secure calculations between multiple parties without leaking the original data. Differential privacy is a technology that protects data privacy by introducing random noise, which can prevent the leakage of sensitive information. By encrypting the updated values ​​of the original variables and the updated values ​​of the dual variables through homomorphic encryption technology, the data exchanged between regional controllers can be protected from being stolen or tampered with. By introducing random perturbations to the updated values ​​of the dual variables through differential privacy technology, malicious attackers can be prevented from inferring the statistical characteristics of the original data. The encrypted updated values ​​of the original variables and the perturbed updated values ​​of the dual variables are used for subsequent penalty factors and dual variable updates.

[0109] When updating the penalty factor and dual variable, this paper adopts a hierarchical reduction strategy, dividing the levels according to the electrical distance of the control area. The closer the distance is, the higher the level. The electrical distance indicates the degree of electrical connection between two control areas, which can be obtained through power flow calculation or sensitivity analysis. The basic idea of ​​the hierarchical reduction strategy is to use different penalty factors and dual variable update rates for control areas of different levels. The higher the level of the control area, the smaller the penalty factor and dual variable update rate. This strategy can accelerate the convergence of the algorithm and avoid the optimization incoordination problem between different control areas.

[0110] Finally, the next round of regional optimization subproblems is solved using the updated penalty factor, the updated primal variables, and the updated dual variables until the distributed solution algorithm meets the preset convergence conditions, and the local optimal power flow control strategy of each control area is output. The preset convergence conditions can be that the update amplitude of the primal variables and the dual variables is less than a given threshold, or the number of algorithm iterations reaches the maximum value. The local optimal power flow control strategy represents the optimal power output and load power scheduling scheme obtained by each regional controller under the current control area division and communication topology.

[0111] In this embodiment, the distribution network is divided into multiple control areas, each area is equipped with a controller, and each regional controller solves the regional optimization sub-problem in parallel, realizing the distributed solution of the power flow optimization problem, introducing the concept of the communication topology directed graph, and setting the communication topology change threshold. When the communication topology changes significantly, it can trigger the distributed algorithm to restart, thereby improving the robustness of the algorithm to communication failures. The original variable update value is encrypted using homomorphic encryption technology, and the dual variable update value is introduced using differential privacy technology, which effectively protects the data privacy of each control area and avoids the risk of privacy data leakage. The layered decreasing penalty factor update strategy is adopted, which can accelerate the convergence speed of the algorithm to a certain extent and improve the solution efficiency. Considering the communication topology constraints of the distribution network, each regional controller asynchronously exchanges the dual variable update value according to the communication topology directed graph, which can adapt to the distributed communication characteristics of the distribution network. By introducing communication topology, homomorphic encryption, differential privacy and other technologies, the goals of distributed solution, communication robustness, data privacy protection and other goals are achieved, providing a strong guarantee for the safe and reliable operation of the distribution network.

[0112] In an optional embodiment,

[0113] The inter-regional power balance constraint is introduced, a global consistency correction model is established, and an iterative algorithm is used to globally correct the local optimal power flow control strategy of each control area. The global optimal power flow control strategy includes:

[0114] A global consistency correction model including inter-regional power balance constraints is established, with the local optimal power flow control strategy of each control area as the optimization variable, the minimum operation cost of the entire network as the optimization goal, and the inter-regional power exchange constraints considered;

[0115] Send the initial parameters of the global correction model to each regional controller;

[0116] Each regional controller uses an optimization algorithm to train the local optimal power flow control strategy based on local historical operation data to obtain the first local optimal power flow control strategy, and uses an encrypted communication protocol to encrypt the first local optimal power flow control strategy and upload it to the central aggregation server;

[0117] The central aggregation server performs weighted averaging on the first local optimal power flow control strategy received in the ciphertext domain to obtain the parameters of the global correction model, and sends it to each regional controller. Each regional controller uses the parameters of the received global correction model to fine-tune and update the local local optimal power flow control strategy.

[0118] Based on the fine-tuned and updated local optimal power flow control strategy of each region, the network loss and global optimization objective function value of each region are calculated to determine whether the global optimal conditions are met;

[0119] If the global optimal condition is not met, the controller of each region returns to the step of training the local optimal power flow control strategy and starts a new round of iteration. If the global optimal condition is met, the current local optimal power flow control strategy of each region is output as the global optimal power flow control strategy.

[0120] Exemplarily, a global consistency correction model is first established, with the local optimal power flow control strategy of each control area as the optimization variable and the minimum operation cost of the entire network as the optimization goal. At the same time, power exchange constraints between regions are introduced to ensure power balance between regions.

[0121] In the initial stage of the algorithm, the central aggregation server distributes the initial parameters of the global correction model to each regional controller. Each regional controller uses local historical operation data to train the local optimal power flow control strategy through the optimization algorithm to obtain the first version of the local optimal power flow control strategy. To protect local data privacy, the regional controller uses an encrypted communication protocol to encrypt the first version of the local optimal power flow control strategy and upload it to the central aggregation server.

[0122] The central aggregation server performs weighted average processing on the first version of the local optimal power flow control strategy received from each region in the ciphertext domain to obtain the parameters of the global correction model and sends the results to each regional controller. The regional controller uses the received global correction model parameters to fine-tune and update the local optimal power flow control strategy.

[0123] In each round of iteration, based on the fine-tuned and updated local optimal power flow control strategy of each region, the network loss and global optimization objective function value of each region are calculated respectively, and it is judged whether the global optimal condition is met. If the global optimal condition is not met, the controller of each region is returned to perform the next round of training iteration of the local optimal power flow control strategy. If the global optimal condition is met, the current local optimal power flow control strategy of each region is output as the global optimal power flow control strategy of the whole network.

[0124] Among them, the local optimal power flow control strategy refers to a control strategy that minimizes the operating cost in each control area by optimizing the output of various types of generators in the dispatching area and adjusting the controllable load while meeting the constraints of safe operation of the power grid in the area.

[0125] The global optimal power flow control strategy is a control strategy that minimizes the operation cost of the entire network by coordinating the local optimal power flow control strategies of each region under the premise of satisfying constraints such as the power flow balance of the regional interconnection lines.

[0126] In this embodiment, by introducing inter-regional power balance constraints, it is ensured that the flow control schemes of each region can achieve inter-regional interconnection line flow balance and meet the physical constraints of the entire network. Using a distributed federated learning framework, each regional controller performs local strategy training and updates locally, and the central aggregation server only performs encrypted global parameter aggregation, which effectively protects the data privacy of each region. Through iterative optimization, each regional controller indirectly collaborates to optimize the global objective function, and finally converges to the global optimal flow control strategy for the entire network, realizing the unification of distributed decision-making and global optimization. The introduction of parallel computing reduces the computational complexity of the optimization problem and improves the scalability of the algorithm. Through technologies such as encrypted communication and differential privacy, sensitive information of each region is protected during information interaction, meeting the needs of power grid data security and privacy protection.

[0127] Figure 2 FIG. 1 is a schematic diagram of a structure of a power distribution network flow control system based on a distributed power source according to an embodiment of the present invention. Figure 2 As shown, the system comprises:

[0128] The first unit is used to divide the distribution network into several control areas according to the load density and distributed power penetration rate of the distribution network, combined with cluster analysis and fuzzy evaluation algorithm, using a pre-built area division model to obtain an initial control area division result, and to monitor the changes in the load density and distributed power penetration rate of the distribution network in real time, and to adaptively adjust the initial control area division result according to the monitoring result to obtain an adjusted control area division result;

[0129] The second unit is used to establish a corresponding regional power flow control model according to the adjusted control area division result. The regional power flow control model takes minimizing the regional grid loss and voltage deviation as the regional comprehensive optimization goal, expresses the uncertainty of the distributed power output and load demand as a confidence interval, and converts the confidence interval into a deterministic constraint of the regional power flow control model through a robust optimization method;

[0130] The third unit is used to use a distributed solution algorithm to perform distributed solution on the regional power flow control model of all regions, obtain the local optimal power flow control strategy of each control region, introduce the power balance constraint between regions, establish a global consistency correction model, and use an iterative algorithm to perform global correction on the local optimal power flow control strategy of each control region to obtain the global optimal power flow control strategy, and obtain the current state and uncertain parameters of the distribution network in real time, automatically trigger the control area division and power flow control strategy solution process, and generate an updated global optimal power flow control strategy in each control cycle; at the same time, based on offline scene matching and online strategy calling technology, the global optimal power flow control strategy can be quickly and real-time applied to complete the power flow control of the distribution network.

[0131] According to a third aspect of the embodiments of the present invention,

[0132] An electronic device is provided, comprising:

[0133] processor;

[0134] a memory for storing processor-executable instructions;

[0135] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0136] A fourth aspect of the embodiments of the present invention is:

[0137] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0138] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0139] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A power flow control method for a distribution network based on distributed power sources, characterized in that: include: According to the load density and distributed generation penetration rate of the distribution network, combined with cluster analysis and fuzzy evaluation algorithm, the distribution network is divided into several control areas using a pre-built area division model to obtain the initial control area division result, and the changes in the load density and distributed generation penetration rate of the distribution network are monitored in real time. According to the monitoring results, the initial control area division result is adaptively adjusted to obtain the adjusted control area division result; According to the adjusted control area division results, a corresponding regional power flow control model is established. The regional power flow control model takes minimizing the regional grid loss and voltage deviation as the regional comprehensive optimization goal, expresses the uncertainty of distributed power output and load demand as a confidence interval, and converts the confidence interval into a deterministic constraint of the regional power flow control model through a robust optimization method; A distributed solution algorithm is used to perform distributed solution on the regional power flow control model of all regions to obtain the local optimal power flow control strategy of each control region. At the same time, the power balance constraint between regions is introduced to establish a global consistency correction model. An iterative algorithm is used to perform global correction on the local optimal power flow control strategy of each control region to obtain the global optimal power flow control strategy, and the current state and uncertain parameters of the distribution network are obtained in real time. The control area division and power flow control strategy solution process are automatically triggered to generate an updated global optimal power flow control strategy in each control cycle. At the same time, based on offline scene matching and online strategy calling technology, the global optimal power flow control strategy is applied quickly and in real time to complete the power flow control of the distribution network, wherein the uncertain parameters include the output prediction value of the distributed power source and the corresponding confidence interval, the load demand prediction value and the corresponding confidence interval; According to the load density and distributed generation penetration rate of the distribution network, combined with cluster analysis and fuzzy evaluation algorithm, the distribution network is divided into several control areas using the pre-built area division model to obtain the initial control area division results, and the changes in the load density and distributed generation penetration rate of the distribution network are monitored in real time. The initial control area division results are adaptively adjusted according to the monitoring results, and the adjusted control area division results include: According to the current state of the distribution network, load density, distributed power penetration rate and electrical distance are selected as characteristic parameters, wherein the load density reflects the load characteristics of different regions, the distributed power penetration rate reflects the power distribution of different regions, and the electrical distance reflects the electrical connection strength between nodes; An improved fuzzy c-means clustering algorithm is used to initially classify the distribution network nodes, the load density and the distributed generation penetration rate are used as inputs of the regional division model, and the initial classification results of the distribution network nodes are obtained by optimizing the objective function to minimize the intra-class distance and the inter-class distance; The initial classification result is optimized by using a fuzzy evaluation method, the electrical distance is used as an additional evaluation index, the membership degree of each distribution network node to other class centers is calculated, and the distribution network nodes are reclassified according to the maximum membership principle to obtain a revised classification result; Determine whether the revised classification result meets the preset regional division requirements, which include the load density dispersion degree and power penetration rate dispersion degree within each region and the load density balance degree and power penetration rate balance degree between regions. If so, use the current classification result as the initial control area division result; otherwise, return to the step of initial classification of distribution network nodes, and reclassify and optimize the distribution network nodes; During the operation of the distribution network, the load changes and power output changes in each area are monitored in real time. When the load change or power output change in any area exceeds the preset threshold, the regional boundaries are dynamically adjusted according to the current operating status of the distribution network, and the regional division model parameters are updated to obtain the adjusted control area division results.

2. The method according to claim 1, characterized in that According to the current operating status of the distribution network, the regional boundaries are dynamically adjusted, and the regional division model parameters are updated. The adjusted control area division results include: Acquire real-time operation data of the distribution network, the real-time operation data including load data, power supply data and topology data, and model the real-time operation data as a non-stationary time series, the time series including an input feature vector and an output target vector, wherein the input feature vector includes load data, power supply data and topology data, and the output target vector includes a balance index and an independence index; Based on the time series, an online learning algorithm is used to minimize the cumulative loss function and update the parameters of the regional division model in real time. The cumulative loss function is obtained by summing up the loss function at each moment. The loss function at each moment includes a balance loss and an independence loss. The balance loss measures the discrete degree of the internal load density and power penetration rate of each region, and the independence loss measures the connection strength between regions. Design an adaptive learning rate adjustment strategy to dynamically adjust the learning rate of each parameter based on the cumulative square sum of historical gradients and output the updated parameters; The updated parameters are decoded into regional division results, and the control regional division results after dynamic adjustment of the distribution network are obtained according to the node membership information by combining soft division and hard division.

3. The method according to claim 1, characterized in that: The calculation formula for the regional comprehensive optimization objective of minimizing regional grid loss and voltage deviation is as follows: ； Among them, J represents the comprehensive optimization target, α represents the weight coefficient of network loss, and N b represents the number of nodes in the region, G ij represents the conductance between nodes i and j, V i Represents the voltage amplitude of node i, V j represents the voltage amplitude at node j, θ ij represents the voltage phase angle difference between node i and node j, β represents the voltage deviation weight coefficient, V ref，i Represents the reference voltage of node i.

4. The method according to claim 1, characterized in that The distributed solution algorithm is used to perform distributed solution on the regional power flow control model of all regions, and the local optimal power flow control strategy of each control area is obtained, including: According to the topological structure and control requirements of the distribution network, the distribution network is divided into multiple control areas, each of which is equipped with a regional controller. At the same time, according to the communication topology of the distribution network, the communication links and directions between the regional controllers are determined, and the communication topology directed graph of the control area is constructed; Initialize the penalty factor, primal variables and dual variables of the distributed solution algorithm, wherein the primal variables include the power output and load power of each control area, and the dual variables are the boundary power of each control area; Each regional controller solves the regional optimization sub-problem in parallel to obtain the updated value of the original variable. At the same time, each regional controller asynchronously exchanges the updated value of the dual variable according to the communication topology directed graph. If the number of edges in the communication topology directed graph changes by more than the communication topology change threshold, the distributed solution algorithm is restarted and the distributed solution algorithm is returned to the initialization step. The update values ​​of the original variables and the update values ​​of the dual variables are encrypted using homomorphic encryption technology, and the update values ​​of the dual variables are introduced with random perturbations using differential privacy technology to obtain the encrypted update values ​​of the original variables and the perturbed update values ​​of the dual variables. Based on the encrypted update values ​​of the original variables and the perturbed update values ​​of the dual variables, a hierarchical decreasing strategy is used to update the penalty factor and the dual variables. The updated penalty factor, the encrypted original variable update value and the perturbed dual variable update value are used to solve the next round of regional optimization subproblems until the distributed solution algorithm meets the preset convergence conditions and outputs the local optimal power flow control strategy for each control area.

5. The method according to claim 1, characterized in that The inter-regional power balance constraint is introduced, a global consistency correction model is established, and an iterative algorithm is used to globally correct the local optimal power flow control strategy of each control area. The global optimal power flow control strategy includes: A global consistency correction model including inter-regional power balance constraints is established, with the local optimal power flow control strategy of each control area as the optimization variable, the minimum operation cost of the entire network as the optimization goal, and the inter-regional power exchange constraints considered; Send the initial parameters of the global correction model to each regional controller; Each regional controller uses an optimization algorithm to train the local optimal power flow control strategy based on local historical operation data to obtain the first local optimal power flow control strategy, and uses an encrypted communication protocol to encrypt the first local optimal power flow control strategy and upload it to the central aggregation server; The central aggregation server performs weighted averaging on the first local optimal power flow control strategy received in the ciphertext domain to obtain the parameters of the global correction model, and sends it to each regional controller. Each regional controller uses the parameters of the received global correction model to fine-tune and update the local local optimal power flow control strategy. Based on the fine-tuned and updated local optimal power flow control strategy of each region, the network loss and global optimization objective function value of each region are calculated to determine whether the global optimal conditions are met; If the global optimal condition is not met, the controller of each region returns to the step of training the local optimal power flow control strategy and starts a new round of iteration. If the global optimal condition is met, the current local optimal power flow control strategy of each region is output as the global optimal power flow control strategy.

6. A power flow control system for a distribution network based on distributed power sources, used to implement the method described in any one of claims 1 to 5, characterized in that: include: The first unit is used to divide the distribution network into several control areas according to the load density and distributed power penetration rate of the distribution network, combined with cluster analysis and fuzzy evaluation algorithm, using a pre-built area division model to obtain an initial control area division result, and to monitor the changes in the load density and distributed power penetration rate of the distribution network in real time, and to adaptively adjust the initial control area division result according to the monitoring result to obtain an adjusted control area division result; The second unit is used to establish a corresponding regional power flow control model according to the adjusted control area division result. The regional power flow control model takes minimizing the regional grid loss and voltage deviation as the regional comprehensive optimization goal, expresses the uncertainty of the distributed power output and load demand as a confidence interval, and converts the confidence interval into a deterministic constraint of the regional power flow control model through a robust optimization method; The third unit is used to use a distributed solution algorithm to perform distributed solution on the regional power flow control model of all regions, obtain the local optimal power flow control strategy of each control region, introduce the power balance constraint between regions, establish a global consistency correction model, and use an iterative algorithm to perform global correction on the local optimal power flow control strategy of each control region to obtain the global optimal power flow control strategy, and obtain the current state and uncertain parameters of the distribution network in real time, automatically trigger the control area division and power flow control strategy solution process, and generate an updated global optimal power flow control strategy in each control cycle; at the same time, based on offline scene matching and online strategy calling technology, the global optimal power flow control strategy is applied quickly and in real time to complete the power flow control of the distribution network, wherein the uncertain parameters include the output prediction value of distributed power sources and the corresponding confidence interval, the load demand prediction value and the corresponding confidence interval.

7. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method described in any one of claims 1 to 5.

8. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images