An optimization method and system for MIQP problems based on an improved random branching strategy

Abstract

The present application discloses an optimization processing method and system for MIQP problems based on an improved random branching strategy. The method includes constructing a graph network model according to a multi-source historical data set, traversing the paths from power nodes to each load node in the target distribution network based on a graph search algorithm, and generating a candidate set of branching strategies for the reconstruction of the target distribution network according to the paths; reconstructing the target distribution network under typical distribution network operation scenarios, and calculating the expected values of different branching strategies in the candidate set of branching strategies during the reconstruction process; inputting real-time load characteristic data, real-time line characteristic data, and real-time topology characteristic data into a pre-trained decision tree model to obtain the initial weight parameters of each branching strategy during the reconstruction of the target distribution network; generating strategy weight parameters of the branching strategies based on the initial weight parameters and the expected values, and optimizing the solution process of the reconstruction of the target distribution network according to the strategy weight parameters. The method provided by the present application improves the efficiency of the reconstruction process.

Description

Technical Field

[0001] The present application relates to the technical field of power systems, and in particular, to an optimization processing method and system for MIQP problems based on an improved random branching strategy. Background Art

[0002] The distribution network is a key component of the power system, undertaking the important task of safely and efficiently distributing electric power from the transmission network to each user. With the development of the social economy, the electricity consumption demand of users is constantly changing, and the load fluctuations are becoming increasingly frequent, which puts higher requirements on the adaptability and flexibility of the distribution network.

[0003] Traditional distribution network reconstruction solvers lack effective variable selection and pruning strategies and are difficult to find a satisfactory solution within a specified time.

[0004] Therefore, how to optimize the existing solution process to improve the distribution network reconstruction efficiency has become a technical problem that needs to be urgently solved by those skilled in the art. Summary of the Invention

[0005] The present application provides an optimization processing method for MIQP problems based on an improved random branching strategy to solve the technical problem of how to optimize the existing optimization processing method for MIQP problems based on an improved random branching strategy, so as to achieve the technical effect of improving the distribution network reconstruction efficiency.

[0006] To solve the above technical problem, an embodiment of the present application provides an optimization processing method for MIQP problems based on an improved random branching strategy, which is characterized by including:

[0007] Constructing a multi-source historical data set based on the obtained historical load characteristic data, historical line characteristic data, and historical topology characteristic data of the target distribution network;

[0008] Constructing a graph network model according to the multi-source historical data set, traversing the paths from the power source nodes to each load node in the target distribution network based on a graph search algorithm, and generating a candidate set of branching strategies for the reconstruction of the target distribution network according to the paths;

[0009] Generating multiple typical distribution network operation scenarios based on the multi-source historical data set, reconstructing the target distribution network under the typical distribution network operation scenarios according to the branch and bound algorithm, and calculating the expected values of different branching strategies in the candidate set of branching strategies during the reconstruction process, where the expected values reflect the number of nodes and time information searched by each branching strategy during the reconstruction process of the target distribution network;

[0010] During the reconstruction process of the target distribution network, the obtained real-time load characteristic data, real-time line characteristic data, and real-time topology characteristic data are input into a pre-trained decision tree model to obtain the initial weight parameters of each branch strategy in the reconstruction process of the target distribution network;

[0011] Based on the initial weight parameters and the expected values, generate the strategy weight parameters of the branch strategy, and optimize the solution process of the reconstruction of the target distribution network according to the strategy weight parameters.

[0012] As one preferred solution, the method for constructing a graph network model based on the multi-source historical data set, traversing the paths from the power supply nodes to each load node in the target distribution network based on a graph search algorithm, and generating a candidate set of branch strategies for the reconstruction of the target distribution network includes:

[0013] Taking the power supply nodes and load nodes in the target distribution network as nodes in the graph network, and the lines in the target distribution network as edges in the graph network to construct the graph network model;

[0014] Analyze all the paths traversed from the power supply nodes to the load nodes, and generate different branch strategies according to the analysis results;

[0015] Screen the branch strategies, and obtain a candidate set of branch strategies for the reconstruction of the target distribution network according to the screening results.

[0016] As one preferred solution, the method for generating multiple typical distribution network operation scenarios based on the multi-source historical data set includes:

[0017] Determine the number of clusters of the multi-source historical data set according to the elbow method;

[0018] Perform clustering processing on the multi-source historical data set based on the K-means clustering algorithm and the number of clusters to obtain multiple clusters; the clustering center of each cluster represents a typical distribution network operation scenario;

[0019] Generate multiple distribution network operation scenarios based on the clustering centers.

[0020] As one preferred solution, the method for reconstructing the target distribution network under the typical distribution network operation scenario according to the branch and bound algorithm, and calculating the expected values of different branch strategies in the candidate set of branch strategies during the reconstruction process includes:

[0021] Construct a reconstruction optimization model of the target distribution network;

[0022] Input the load characteristic data, line characteristic data, and topology characteristic data under the typical distribution network operation scenario into the reconstruction optimization model, and solve according to the branch and bound algorithm;

[0023] Obtain the number of nodes searched and time information under each branch strategy in the solution process, and assign corresponding expected values to each branch strategy according to the number of nodes and the time information.

[0024] As one of the preferred solutions, the assigning corresponding expected values to each branch strategy according to the number of nodes and the time information further includes:

[0025] Obtain the occurrence probability of each distribution network operation scenario based on the multi-source historical data set, and generate the weight of each distribution network operation scenario based on the occurrence probability;

[0026] Perform a weighted sum of the initial expected values of the branch strategies under each typical distribution network operation scenario based on the weights to obtain the expected value of each branch strategy.

[0027] As one of the preferred solutions, the reconstruction optimization model of the target distribution network includes:

[0028]

[0029] Among them, represents the set of time periods, represents the set of generating units, represents the generating unit The power generation cost per unit time, represents the generating unit In the time period The power generation power, represents the set of lines, represents the line The loss cost coefficient, represents the line In the time period The power flow, represents the set of opening and closing operations, represents the line The opening operation cost, represents the line The closing operation cost, And respectively represent the closing and opening operation decision variables of the line In the time period The opening and closing operation decision variables.

[0030] As one of the preferred solutions, the inputting the obtained real-time multi-source prediction data set and the branch strategy candidate set into the pre-trained decision tree model to obtain the initial weight parameters of the target distribution network reconstruction branch strategy includes:

[0031] Construct an initial decision tree model based on Extra Trees;

[0032] Train the initial decision tree model according to the multi-source historical data set, the candidate set of branch strategies, and the expected value data corresponding to each branch strategy to obtain a trained decision tree model;

[0033] Input the obtained real-time multi-source prediction data set into the decision tree model to obtain the initial weight parameters of each branch strategy in the distribution network reconstruction process.

[0034] As one preferred solution, the initial decision tree model includes:

[0035]

[0036] Among them, Gini is the Gini impurity, which is used to evaluate the quality of the splitting point and is defined as:

[0037]

[0038] Among them, is the probability of class and is the total number of classes.

[0039] As one preferred solution, after generating the strategy weight parameters of the branch strategy based on the initial weight parameters and the expected value, and optimizing the solution process of the target distribution network reconstruction according to the strategy weight parameters, it further includes:

[0040] Execute the target reconstruction strategy determined by the solution result, and monitor the operation state data of the distribution network in real time;

[0041] Update the target reconstruction strategy in real time according to the operation state data to complete the adaptive adjustment of the target distribution network reconstruction.

[0042] Another embodiment of the present application provides an MIQP problem optimization processing system based on an improved random branch strategy, including:

[0043] An acquisition module, configured to construct a multi-source historical data set based on the obtained historical load characteristic data, historical line characteristic data, and historical topology characteristic data of the target distribution network;

[0044] A construction module, configured to construct a graph network model according to the multi-source historical data set, traverse the paths from the power supply nodes to each load node in the target distribution network based on the graph search algorithm, and generate a candidate set of branch strategies for the target distribution network reconstruction according to the paths;

[0045] A solution module, configured to generate multiple typical operation scenarios of a distribution network based on the multi-source historical data set, reconstruct the target distribution network under the typical operation scenarios of the distribution network according to the branch and bound algorithm, and calculate the expected values of different branch strategies in the branch strategy candidate set during the reconstruction process, where the expected values reflect the number of nodes searched and time information of each branch strategy during the reconstruction process of the target distribution network;

[0046] A prediction module, configured to input the acquired real-time load characteristic data, real-time line characteristic data, and real-time topology characteristic data into a pre-trained decision tree model during the reconstruction process of the target distribution network, and obtain the initial weight parameters of each branch strategy during the reconstruction process of the target distribution network;

[0047] An optimization module, configured to generate the strategy weight parameters of the branch strategy based on the initial weight parameters and the expected values, and optimize the solution process of the reconstruction of the target distribution network according to the strategy weight parameters.

[0048] Compared with the prior art, the beneficial effects of the embodiments of the present application are at least one of the following:

[0049] (1) The present application constructs a graph network model using a multi-source historical data set, and traverses the paths from power supply nodes to load nodes through a graph search algorithm, systematically generating a branch strategy candidate set for the reconstruction of the target distribution network. This method can comprehensively consider various possible path combinations in the distribution network, avoid missing important reconstruction schemes, and provide a wider selection space for finding the optimal or sub-optimal reconstruction strategy.

[0050] (2) The present application calculates the expected values of different branch strategies in the branch strategy candidate set during the reconstruction process, and the expected values reflect the number of searched nodes and time information. Through this quantitative evaluation method, the advantages and disadvantages of different branch strategies can be intuitively compared, providing an objective and reliable basis for the selection of strategies, and helping to improve the efficiency and accuracy of the reconstruction process.

[0051] (3) Through the optimized reconstruction strategy, the present application can quickly and effectively transfer loads and restore power supply when some lines or equipment fail, reduce the power outage time and scope, ensure the normal power consumption of users, and improve the power supply reliability of the distribution network. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 is a schematic flowchart of an optimization processing method for an MIQP problem based on an improved random branch strategy in one embodiment of the present application;

[0053] Figure 2 is a schematic diagram of an optimization processing system for an MIQP problem based on an improved random branch strategy in one embodiment of the present application. Detailed implementation manners

[0054] The following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part rather than all of the embodiments of the present application. The purpose of providing these embodiments is to make the disclosure of the present application more thorough and comprehensive. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without making creative efforts belong to the scope protected by the present application.

[0055] In the description of the present application, the terms "first", "second", "third", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, features defined with "first", "second", "third", etc. may explicitly or implicitly include one or more of such features. In the description of the present application, unless otherwise stated, the meaning of "a plurality" is two or more.

[0056] In the description of the present application, it should be noted that unless otherwise clearly defined and limited, the terms "installed", "connected", "connected" should be understood in a broad sense. For example, it may be a fixed connection, a detachable connection, or an integral connection; it may be a mechanical connection or an electrical connection; it may be directly connected or indirectly connected through an intermediate medium, and it may be the communication inside two components. The terms "vertical", "horizontal", "left", "right", "up", "down" and similar expressions used herein are only for the purpose of illustration and do not indicate or imply that the device or component referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus cannot be understood as a limitation to the present application. The term "and / or" used herein includes any and all combinations of one or more of the related listed items. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to specific circumstances.

[0057] In the description of the present application, it should be noted that unless otherwise defined, all technical and scientific terms used in the present application have the same meanings as those commonly understood by those skilled in the technical field to which the present application belongs. The terms used in the specification of the present application are only for the purpose of describing specific embodiments and are not intended to limit the present application. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to specific circumstances.

[0058] An embodiment of the present application provides an optimization processing method for MIQP problems based on an improved random branching strategy. Specifically, please refer to Figure 1 , Figure 1The flowchart of the optimization method for the MIQP problem based on the improved random branching strategy in one embodiment of the present application is shown, which includes S1 - S6:

[0059] S1: Construct a multi - source historical dataset based on the obtained historical load characteristic data, historical line characteristic data, and historical topology characteristic data of the target distribution network;

[0060] Among them, the mixed - integer quadratic programming (MIQP) problem is a mathematical optimization problem that combines a quadratic objective function and linear constraints. In the distribution network reconfiguration problem based on MIQP, it is necessary to collect various data related to the distribution network, which are the basis for subsequent analysis and processing. The specific data collected includes data such as substations, transformers, loads, switches, disconnectors, and topological relationships, and data cleaning and pre - processing are carried out.

[0061] Specifically, for missing load data and branch parameters, the interpolation method is used for filling, while for discrete data such as switch states, the mode is used for filling.

[0062] After dealing with the missing data, using statistical principles, such as calculating statistical quantities such as the mean and standard deviation of the data, outliers are identified by setting thresholds. For example, for load data, if the deviation of a data point from the mean exceeds 3 times the standard deviation, then this data point may be an outlier. The outliers are corrected according to the actual physical meaning. For example, if the sudden increase in load data is due to recording errors, the load values in adjacent time periods can be referred to for reasonable correction.

[0063] Since different types of data have different magnitudes and units, it will have an adverse impact on the training of machine learning models. Therefore, minimum - maximum normalization processing is performed on the load data and branch parameters. Through normalization processing, the impact of magnitude differences on the model can be eliminated, enabling the model to treat different features more fairly.

[0064] Based on the characteristics of the distribution network reconfiguration problem, features that have a significant impact on the optimization solution are selected. These features can better reflect the operating state and reconfiguration requirements of the distribution network, specifically including:

[0065] Load - related features: The load demands of each node directly reflect the power consumption situation of the distribution network, and the peak load reflects the maximum power consumption demand of the distribution network during a certain period, which is very important for evaluating the power supply capacity and load distribution of the distribution network;

[0066] Line - related features: The line impedance affects the power loss during the power transmission process, and the active power reflects the actual power transmitted by the line. These features play a key role in analyzing the operating state of the line and optimizing the power distribution of the line;

[0067] Topological features: The node degree represents the number of lines connected to the node, and the connection relationship describes the topological structure between nodes. They are crucial for understanding the network layout and connectivity of the distribution network and are important bases for distribution network reconfiguration.

[0068] In one embodiment of the present application, constructing new features can improve the learning ability and generalization performance of the machine learning model. Specifically, it may include:

[0069] Comprehensive load index: Considering multiple load-related factors, such as load changes in different time periods, load distribution, etc., a comprehensive index is constructed to more comprehensively describe the load characteristics;

[0070] Branch utilization rate: By calculating the ratio of the actual power to the rated power of the branch, the branch utilization rate is obtained. This index can reflect the usage of the branch and help optimize the load distribution and network topological structure of the branch.

[0071] S2: Construct a graph network model based on the multi-source historical dataset. Traverse the paths from the power source nodes to each load node in the target distribution network based on the graph search algorithm, and generate a candidate set of branch strategies for the target distribution network reconfiguration according to the paths;

[0072] Preferably, in one embodiment of the present application, constructing a graph network model based on the multi-source historical dataset, traversing the paths from the power source nodes to each load node in the target distribution network based on the graph search algorithm, and generating a candidate set of branch strategies for the target distribution network reconfiguration, including:

[0073] Taking the power source nodes and load nodes in the target distribution network as nodes in the graph network, and the lines in the target distribution network as edges in the graph network to construct the graph network model;

[0074] Analyze all the paths obtained by traversing from the power source nodes to the load nodes, and generate different branch strategies according to the analysis results;

[0075] Screen the branch strategies, and obtain a candidate set of branch strategies for the target distribution network reconfiguration according to the screening results.

[0076] Identify the power source nodes and load nodes in the target distribution network from the multi-source historical dataset and define them as nodes in the graph network. The power source nodes are the sources of power supply, such as substation buses; the load nodes are the places where electrical energy is consumed, such as factories and residential areas. These nodes can carry various attribute information, such as the power generation capacity and power generation cost of the power source nodes, and the historical load curve and load prediction value of the load nodes. These attributes help to more accurately analyze the operating state of the distribution network subsequently.

[0077] Regard the lines in the target distribution network as the edges in the graph network. The attributes of the edges can be extracted from multi-source historical datasets, such as the resistance, reactance, susceptance, current-carrying capacity, line length, etc. of the lines. These attributes are crucial for calculating the losses of electric energy during transmission on the lines and judging the carrying capacity of the lines. By determining the relationship between nodes and edges, the topological structure of the graph network can be constructed.

[0078] Conduct a comprehensive analysis of all paths traversed from the power source node to the load node, mainly including:

[0079] Path length: The longer the path, the greater the possible power loss on the line, which will affect the operation efficiency of the distribution network. The path length can be measured by calculating the sum of the weights of each edge on the path;

[0080] Line load condition: Check the historical load data of each line on the path and the current load prediction situation, and avoid selecting lines whose loads are already close to or exceed their capacities to prevent line overloads from causing failures;

[0081] Switch status: The realization of a path often requires adjusting the opening and closing states of switches in the distribution network. Factors such as the feasibility, operation cost, and operation time of switch operations need to be considered;

[0082] Power supply reliability: Analyze the impact of the path on power supply reliability. For example, some paths may have a single-point failure risk. Once a certain line or device fails, it may cause power outages for some loads.

[0083] According to the results of the path analysis, generate different branch strategies, specifically including:

[0084] Switch operation strategy: By changing the opening and closing states of certain switches, change the topological structure of the distribution network, so as to realize different power supply paths. For example, close a certain tie switch and open some sectionalizing switches to transfer the load from one line to another;

[0085] Load transfer strategy: According to the line load conditions on the path, transfer some loads from the lines with higher loads to the lines with lower loads to achieve balanced load distribution, reduce line losses, and improve the operation efficiency of the distribution network;

[0086] Power supply distribution strategy: Adjust the output power of the power source nodes so that different power source nodes supply power to different load nodes, optimize the utilization efficiency of the power sources, and improve the stability and reliability of the power supply.

[0087] According to the set screening conditions, the generated branch strategies are evaluated and screened one by one, and those strategies that do not meet the conditions are removed. After screening, the remaining branch strategies constitute the candidate set of branch strategies for the target distribution network reconstruction. This candidate set provides multiple feasible options for subsequent distribution network reconstruction decisions. Through further evaluation and optimization, the optimal or sub-optimal reconstruction strategy can be selected from it.

[0088] S3: Generate multiple typical distribution network operation scenarios based on the multi-source historical data set. Reconstruct the target distribution network under the typical distribution network operation scenarios according to the branch and bound algorithm, and calculate the expected values of different branch strategies in the candidate set of branch strategies during the reconstruction process. The expected value reflects the number of nodes and time information searched by each branch strategy during the target distribution network reconstruction process.

[0089] Preferably, in an embodiment of the present application, generating multiple typical distribution network operation scenarios based on the multi-source historical data set includes:

[0090] Determine the number of clusters of the multi-source historical data set according to the elbow method;

[0091] Perform clustering processing on the multi-source historical data set based on the K-means clustering algorithm and the number of clusters to obtain multiple clusters; where the clustering center of each cluster represents a typical distribution network operation scenario;

[0092] Generate multiple distribution network operation scenarios based on the clustering centers.

[0093] Based on the multi-source historical data set, covering historical load characteristic data, historical line characteristic data, and historical topology characteristic data, use appropriate methods to generate multiple typical distribution network operation scenarios. This can utilize data mining techniques such as clustering analysis to classify historical data according to factors such as load patterns and topological structures to form representative operation scenarios. For example, according to the load characteristics in different seasons and different time periods, typical scenarios such as summer peak and winter trough are divided.

[0094] Preferably, in an embodiment of the present application, reconstructing the target distribution network under the typical distribution network operation scenario according to the branch and bound algorithm and calculating the expected values of different branch strategies in the candidate set of branch strategies during the reconstruction process includes:

[0095] Construct a reconstruction optimization model for the target distribution network;

[0096] Input the load characteristic data, line characteristic data, and topology characteristic data under the typical distribution network operation scenario into the reconstruction optimization model and solve it according to the branch and bound algorithm;

[0097] Obtain the number of nodes and time information searched under each branch strategy during the solution process, and assign corresponding expected values to each branch strategy according to the number of nodes and time information.

[0098] Input the load characteristic data, line characteristic data, and topological characteristic data under typical distribution network operation scenarios into the reconstruction optimization model, and use the branch and bound algorithm for solution. The specific steps are as follows:

[0099] Set the initial feasible solution and optimal solution. The initial feasible solution can be the operating state of the current distribution network, and the optimal solution is initialized to a relatively large value.

[0100] Divide the solution space of the problem into multiple sub-problems, and each sub-problem corresponds to a possible switch state combination or power generation power distribution scheme. For example, select a switch with an undetermined state, and set it to two cases: closed and open respectively, to form two sub-problems.

[0101] For each sub-problem, calculate the upper and lower bounds of its objective function. The upper bound can be obtained by relaxing the constraint conditions, etc., and the lower bound can be obtained by some heuristic algorithms or lower bound estimation methods. If the lower bound of a sub-problem is greater than the current optimal solution, then this sub-problem can be pruned and no further search is needed.

[0102] Continuously select a potential sub-problem for branching and bounding until all possible sub-problems are searched or the termination conditions are met (such as reaching the maximum search depth, finding the optimal solution, etc.).

[0103] During the solution process, record the number of nodes searched and time information under each branching strategy. The number of nodes reflects the complexity of the search process, and the time information reflects the search efficiency. Assign initial expected values to each branching strategy based on this information.

[0104] Specifically, conduct statistical analysis on the exploration costs of all recorded sample paths, calculate statistics such as the mean, median, and standard deviation to understand the distribution of exploration costs. Determine the prior probabilities of different topological adjustment strategies based on historical data or experience. Based on the prior probabilities and sample paths of the exploration costs calculate the total expected number of nodes under the given prior probabilities This value reflects the number of nodes that need to be explored during the search process of the branch and bound algorithm under different topological adjustment strategies, and further evaluates the overall performance of this strategy combination in the reconstruction of the distribution network.

[0105] Define the objective of the distribution network reconstruction as the objective function, and the strategy weight parameters of different branch strategies as the input variables of the objective function. The goal is to find a set of optimal strategy weight parameters to minimize the objective function. Use a Gaussian process regression model to model the posterior distribution of the objective function. Among them, the Gaussian process is a probability model that can model an unknown function and give the probability distribution of the function values. Learn the distribution characteristics of the objective function through the set of solutions in the historical dataset (that is, the different combinations of strategy weight parameters recorded before and their corresponding objective function values).

[0106] In each iteration, the Bayesian optimizer uses the Gaussian process to predict the unevaluated reconstruction scheme combinations (that is, different combinations of strategy weight parameters), and obtains the probability distribution of the objective function values corresponding to each combination. Then, according to a certain acquisition function (such as expected improvement, upper confidence bound, etc.), select the combination with the highest potential improvement effect for evaluation. Conduct an actual evaluation of the selected combination to obtain its corresponding objective function value. Add this data to the historical dataset and update the Gaussian process regression model to more accurately reflect the distribution of the objective function.

[0107] Repeat the above processes of prediction, selection, and update for multiple rounds of iterative optimization. As the number of iterations increases, the Bayesian optimizer gradually narrows the search range and approaches the optimal solution. When certain termination conditions are met (such as reaching the maximum number of iterations, convergence of the objective function value, etc.), the iteration stops. At this time, the combination of strategy weight parameters obtained is the optimal combination of strategy weight parameters applicable to the current distribution network operation state, corresponding to the initial expected values of each branch strategy.

[0108] Preferably, in an embodiment of the present application, when allocating corresponding expected values to each branch strategy according to the number of nodes and time information, it further includes:

[0109] Obtain the occurrence probability of each distribution network operation scenario based on the multi-source historical dataset, and generate the weight of each distribution network operation scenario based on the occurrence probability;

[0110] Perform a weighted sum of the initial expected values of the branch strategies under each typical distribution network operation scenario based on the weights to obtain the expected value of each branch strategy.

[0111] Based on multi-source historical data sets, the occurrence probability of each distribution network operation scenario is obtained through methods such as statistical analysis and machine learning. For example, according to historical data and meteorological forecasts and other information, the probability of the summer peak scenario appearing is predicted. The weight of each distribution network operation scenario is generated according to the scenario occurrence probability, and the weight is usually proportional to the occurrence probability. The initial expected values of the branch strategies under each typical distribution network operation scenario are weighted and summed based on these weights to obtain the final expected value of each branch strategy. The expected value obtained in this way comprehensively considers the influence of different scenarios and can better reflect the performance of the branch strategy in actual operation.

[0112] Preferably, in an embodiment of the present application, the reconstruction optimization model of the target distribution network includes:

[0113]

[0114] Wherein, represents the set of time periods, represents the set of generator sets, represents the generator set The power generation cost per unit time, represents the generator set in the time period The generated power, represents the set of lines, represents the line The loss cost coefficient, represents the line in the time period The power flow, represents the set of opening and closing operations, represents the line The cost of opening operation, represents the line The cost of closing operation, and respectively represent the line in the time period The closing and opening operation decision variables.

[0115] The goal of distribution network reconstruction is to minimize the total operating cost of the system. However, in the optimization process, a series of constraints must be considered to ensure the safe, stable and feasible operation of the system. The following are the common constraints for distribution network reconstruction:

[0116] Power balance constraint: The power demand of each node must be met, and the power balance relationship needs to be satisfied among all the generated power, load demand and line power flow in the power grid. The specific expression is:

[0117]

[0118] Among them, represents the power generation of the th node during the time period . is the th node's load demand during the time period . is the set of lines connected to node . is the power flow from node to node during the time period .

[0119] Generator output constraint: The power output of the generator cannot exceed its capacity upper limit or be lower than its minimum output limit, which can usually be expressed as:

[0120]

[0121] Among them, and are the minimum and maximum outputs of generator respectively, is the output of generator during the time period .

[0122] Line power flow constraint: To prevent overload, the power flow of each line must meet its maximum transmission capacity. It can be expressed as:

[0123]

[0124] Among them, is the power flow of line during the time period , is the maximum power transmission capacity of line .

[0125] Voltage magnitude constraint: The voltage magnitude of each node needs to be maintained within a certain range to ensure the stable operation of the power grid. The voltage magnitude constraint can be expressed as:

[0126]

[0127] Among them, represents the voltage magnitude of node during the time period , and are the minimum and maximum allowable values of the voltage respectively.

[0128] Switch operation constraints: In the reconstruction of the distribution network, the switch operations of nodes or lines must follow certain constraints. For line switch operations, the following constraints need to be satisfied:

[0129] If a certain line is opened during the time period then the power flow of this line must be zero:

[0130]

[0131] If a certain line is closed during the time period then the power flow of this line is not zero, and the power flow constraint of the line needs to be satisfied:

[0132]

[0133] where is the switch state of line during the time period , represents that line is disconnected, 1 represents that line is connected.

[0134] Load scheduling constraints: The load of the power grid must be scheduled according to the actual demand. Load scheduling constraints usually take into account the volatility of the load and the prediction error, and the specific form is:

[0135]

[0136] Minimum opening and closing time constraints: In order to avoid frequent switch operations, the reconstruction of the distribution network also needs to limit the operation frequency of switch equipment. Generally speaking, the minimum opening and closing time intervals of switch equipment are a specific value to prevent overly frequent operations. It can be expressed as:

[0137] .

[0138] S4: During the reconstruction process of the target distribution network, input the obtained real-time load characteristic data, real-time line characteristic data, and real-time topology characteristic data into the pre-trained decision tree model to obtain the initial weight parameters of each branch strategy in the reconstruction process of the target distribution network;

[0139] Preferably, in an embodiment of the present application, input the obtained real-time multi-source prediction data set and the candidate set of branch strategies into the pre-trained decision tree model to obtain the initial weight parameters of the target distribution network reconstruction branch strategy, including:

[0140] Construct an initial decision tree model based on Extra Trees;

[0141] Train the initial decision tree model based on the multi-source historical dataset, the candidate set of branch strategies, and the expected value data corresponding to each branch strategy to obtain a trained decision tree model;

[0142] Input the obtained real-time multi-source prediction dataset into the decision tree model to obtain the initial weight parameters of each branch strategy during the distribution network reconstruction process.

[0143] Among them, Extra Trees (Extremely Randomized Trees) is an ensemble learning method and an extension of decision trees. Different from traditional decision trees, Extra Trees introduces more randomness in the process of constructing decision trees. When dividing nodes, instead of finding the optimal division among all possible division points of all features, it randomly selects features and division points and then selects the optimal one from them. This additional randomness makes the model have stronger generalization ability and can effectively avoid overfitting problems.

[0144] Organize the multi-source historical dataset, the candidate set of branch strategies, and the expected value data to make them meet the model input requirements. Preferably, in an embodiment of the present application, the initial decision tree model includes:

[0145]

[0146] Among them, Gini is the Gini impurity, which is used to evaluate the quality of the split point and is defined as:

[0147]

[0148] Among them, is the probability of class , is the total number of classes.

[0149] Optimize the hyperparameters of the Extra Trees classifier through grid search, such as the number of trees , the maximum depth , the number of features randomly selected for each node , etc., to obtain the best performance on the validation set . Randomly sample parameter combinations within the predefined parameter space and select the parameter combination that minimizes the loss function .

[0150] Comprehensively use other appropriate evaluation metrics such as accuracy, recall, and precision to evaluate the prediction performance of the Extra Trees classifier on the validation set to ensure that it can accurately predict the strategy weight parameters. The calculation formulas of each evaluation metric are as follows:

[0151] Accuracy:

[0152]

[0153] Where:

[0154] TP is True Positive, which is the number of samples correctly predicted as the positive class.

[0155] TN is True Negative, which is the number of samples correctly predicted as the negative class.

[0156] FP is False Positive, which is the number of samples incorrectly predicted as the positive class.

[0157] FN is False Negative, which is the number of positive-class samples incorrectly predicted as the negative class.

[0158] Precision:

[0159]

[0160] This represents the proportion of samples actually being the positive class among the samples predicted as the positive class.

[0161] Recall:

[0162]

[0163] This represents the proportion of samples correctly predicted as the positive class among the samples actually being the positive class.

[0164] F1-score:

[0165]

[0166] F1-score is the harmonic mean of precision and recall, used to comprehensively measure the performance of the classification model on positive-class samples. F1-score is more stable under the trade-off between precision and recall, especially in the case of class imbalance.

[0167] During the reconstruction process of the target distribution network, load characteristic data, line characteristic data, and topological characteristic data of the distribution network are collected in real time to form a real-time multi-source prediction data set. These data reflect the current actual operating state of the distribution network. The obtained real-time multi-source prediction data set is input into the trained decision tree model, and the model will predict the performance of each branch strategy according to the learned rules to obtain the expected value corresponding to each branch strategy. According to the predicted expected value, the initial weight parameters of each branch strategy in the target distribution network reconstruction process are determined. For example, the expected value can be normalized, and the normalized expected value can be used as the initial weight parameter so that the sum of the weight parameters is 1 to reflect the relative importance of each branch strategy under the current operating state of the distribution network.

[0168] S5: Generate the strategy weight parameters of the branch strategy based on the initial weight parameters and the expected value, and optimize the solution process of the target distribution network reconstruction according to the strategy weight parameters.

[0169] Preferably, in an embodiment of the present application, generating the strategy weight parameters of the branch strategy based on the initial weight parameters and the expected value, and optimizing the solution process of the target distribution network reconstruction according to the strategy weight parameters further includes:

[0170] Execute the target reconstruction strategy determined by the solution result, and monitor the operating state data of the distribution network in real time;

[0171] According to the operating state data, the target reconstruction strategy is updated in real time to complete the adaptive adjustment of the target distribution network reconstruction.

[0172] To ensure the efficiency and adaptability of the strategy weight parameters in practical applications, the system introduces a dynamic optimization mechanism, continuously updates the strategy weight parameters and performs solution in combination with the prediction results of the Extra Trees classifier. The specific steps are as follows:

[0173] When processing each new MIQP problem instance, first predict the preliminary strategy weight parameters through the Extra Trees classifier , and then combine the stored optimal strategy weights , and obtain the final strategy weight parameters through weighted fusion .

[0174]

[0175] During the actual solution process, record the performance of the strategy weight parameters (such as exploration cost , solution time etc.), and use this feedback information to continuously optimize the Extra Trees classifier and the Bayesian optimization model.

[0176] Regularly update the Extra Trees classifier and Bayesian optimization model using newly collected feedback data to ensure that the models can adapt to the dynamic changes of MIQP problem instances and newly emerging complex constraint conditions, and update the training dataset:

[0177]

[0178] Retrain the model using the updated Retrain the Extra Trees classifier and readjust the policy weight parameters through Bayesian optimization.

[0179] Adaptive weight adjustment: According to system performance metrics (such as exploration cost , solution time , etc.), dynamically adjust the weight fusion factor and the prediction weights of the Extra Trees classifier to achieve the best balance of policy weight parameters. The adjustment formula is as follows:

[0180]

[0181] where is a function adaptively adjusted according to system performance metrics. For example, the following linear adjustment strategy can be adopted:

[0182]

[0183] where is the baseline exploration cost, is the current exploration cost, is a small constant to prevent division by zero.

[0184] Configure the updated policy weight parameters into the heuristic algorithm and pruning strategy of the solver to guide the solution process. Run the solver and execute the branch and bound algorithm. During the search process, adjust the heuristic search and pruning strategies according to the policy weight parameters to improve the solution efficiency and quality.

[0185] Through the above steps, the system can effectively integrate the Extra Trees classifier and Bayesian optimization model into the storage and update process of policy weight parameters, not only improving the prediction accuracy of policy weights, but also ensuring that the branch and bound algorithm always maintains high efficiency and adaptability in the solution of the distribution network reconfiguration problem through a dynamic optimization mechanism. This method significantly improves the solution efficiency and quality of the distribution network reconfiguration problem, ensuring that the optimal or near-optimal solution can be found under various complex constraint conditions.

[0186] Preferably, another embodiment of the present application provides an MIQP problem optimization processing system based on an improved random branching strategy. Specifically, please refer toFigure 2 , Figure 2 is shown as a schematic diagram of an MIQP problem optimization processing system based on an improved random branching strategy in one embodiment of the present application, which includes:

[0187] An acquisition module 11, configured to construct a multi-source historical data set based on the acquired historical load characteristic data, historical line characteristic data, and historical topology characteristic data of the target distribution network;

[0188] A construction module 12, configured to construct a graph network model according to the multi-source historical data set, traverse the paths from the power supply nodes to each load node in the target distribution network based on the graph search algorithm, and generate a candidate set of branching strategies for the reconstruction of the target distribution network according to the paths;

[0189] A solution module 13, configured to generate multiple typical distribution network operation scenarios based on the multi-source historical data set, reconstruct the target distribution network under the typical distribution network operation scenarios according to the branch and bound algorithm, and calculate the expected values of different branching strategies in the candidate set of branching strategies during the reconstruction process, where the expected value reflects the number of nodes and time information searched by each branching strategy during the reconstruction of the target distribution network;

[0190] A prediction module 14, configured to input the acquired real-time load characteristic data, real-time line characteristic data, and real-time topology characteristic data into a pre-trained decision tree model during the reconstruction process of the target distribution network to obtain the initial weight parameters of each branching strategy during the reconstruction process of the target distribution network;

[0191] An optimization module 15, configured to perform a fusion process on the initial weight parameters and the expected values, and execute the target reconstruction strategy determined by the fusion process result.

[0192] Compared with the prior art, the beneficial effects of the embodiments of the present application are at least one of the following:

[0193] (1) The present application constructs a graph network model using a multi-source historical data set, and traverses the paths from the power supply nodes to the load nodes through the graph search algorithm, systematically generating a candidate set of branching strategies for the reconstruction of the target distribution network. This method can comprehensively consider various possible path combinations in the distribution network, avoid missing important reconstruction schemes, and provide a wider selection space for finding the optimal or sub-optimal reconstruction strategy.

[0194] (2) The present application calculates the expected values of different branching strategies in the candidate set of branching strategies during the reconstruction process, and the expected value reflects the number of searched nodes and time information. Through this quantitative evaluation method, the advantages and disadvantages of different branching strategies can be intuitively compared, providing an objective and reliable basis for the selection of strategies, and helping to improve the efficiency and accuracy of the reconstruction process.

[0195] (3) Through the optimized reconstruction strategy, this application can quickly and effectively transfer the load and restore power supply when some lines or equipment fail, reduce the power outage time and scope, ensure the normal power consumption of users, and improve the power supply reliability of the distribution network.

[0196] The above embodiments only represent several implementation manners of this application, and their descriptions are relatively specific and detailed. However, it should not be construed as a limitation of the patent scope of this application. It should be noted that for those of ordinary skill in the art, without departing from the concept of this application, several deformations and improvements can still be made, and these all belong to the protection scope of this application. Therefore, the protection scope of the patent of this application shall be subject to the appended claims.

Claims

1. An optimization method for MIQP problems based on an improved random branching strategy, characterized in that Including: Construct a multi-source historical data set based on the obtained historical load characteristic data, historical line characteristic data, and historical topology characteristic data of the target distribution network; Construct a graph network model according to the multi-source historical data set, traverse the paths from the power source nodes to each load node in the target distribution network based on the graph search algorithm, and generate a candidate set of branch strategies for the reconstruction of the target distribution network according to the paths; Generate multiple typical distribution network operation scenarios based on the multi-source historical data set, reconstruct the target distribution network under the typical distribution network operation scenarios according to the branch and bound algorithm, and calculate the expected values of different branch strategies in the candidate set of branch strategies during the reconstruction process. The expected value reflects the number of nodes and time information searched by each branch strategy during the reconstruction of the target distribution network; During the reconstruction process of the target distribution network, input the obtained real-time load characteristic data, real-time line characteristic data, and real-time topology characteristic data into a pre-trained decision tree model to obtain the initial weight parameters of each branch strategy during the reconstruction process of the target distribution network; Generate the strategy weight parameters of the branch strategy based on the initial weight parameters and the expected value, and optimize the solution process of the reconstruction of the target distribution network according to the strategy weight parameters.

2. The optimization processing method for the MIQP problem based on the improved random branching strategy according to claim 1, wherein, The constructing a graph network model according to the multi-source historical data set, traversing the paths from the power source nodes to each load node in the target distribution network based on the graph search algorithm, and generating a candidate set of branch strategies for the reconstruction of the target distribution network includes: Use the power source nodes and load nodes in the target distribution network as nodes in the graph network, and the lines in the target distribution network as edges in the graph network to construct the graph network model; Analyze all the paths traversed from the power source nodes to the load nodes, and generate different branch strategies according to the analysis results; Screen the branch strategies, and obtain the candidate set of branch strategies for the reconstruction of the target distribution network according to the screening results.

3. The optimization processing method for the MIQP problem based on the improved random branching strategy according to claim 1, characterized in that, The generating multiple typical distribution network operation scenarios based on the multi-source historical data set includes: Determine the number of clusters of the multi-source historical data set according to the elbow method; Perform clustering processing on the multi-source historical data set based on the K-means clustering algorithm and the number of clusters to obtain multiple clusters; where the clustering center of each cluster represents a typical distribution network operation scenario; Generate multiple distribution network operation scenarios based on the clustering centers.

4. The optimization processing method for the MIQP problem based on the improved random branching strategy according to claim 1, wherein, The reconstructing the target distribution network under the typical distribution network operation scenarios according to the branch and bound algorithm, and calculating the expected values of different branch strategies in the candidate set of branch strategies during the reconstruction process includes: Construct a reconstruction optimization model of the target distribution network; Input the load characteristic data, line characteristic data, and topology characteristic data under the typical distribution network operation scenario into the reconstruction optimization model, and solve it according to the branch and bound algorithm; Obtain the number of nodes and time information searched under each branch strategy during the solution process, and assign corresponding expected values to each branch strategy according to the number of nodes and the time information.

5. The optimization method for MIQP problems based on the improved random branching strategy according to claim 4, characterized in that Assigning corresponding expected values to each of the branch strategies according to the number of nodes and the time information further includes: Obtaining the occurrence probability of each distribution network operation scenario based on the multi-source historical data set, and generating the weight of each distribution network operation scenario based on the occurrence probability; Performing a weighted sum of the initial expected values of the branch strategies under each typical distribution network operation scenario based on the weights to obtain the expected values of each branch strategy.

6. The optimization processing method for the MIQP problem based on the improved random branching strategy according to claim 4, characterized in that The reconstruction optimization model of the target distribution network includes: Among them, represents a set of time periods, represents a set of generating units, represents the generating unit 's power generation cost per unit time, represents the generating unit in the time period 's power generation power, represents a set of lines, represents the line 's loss cost coefficient, represents the line in the time period 's power flow, represents a set of opening and closing operation sets, represents the line 's opening operation cost, represents the line 's closing operation cost, and respectively represent the line in the time period 's closing and opening operation decision variables.

7. The optimization processing method for the MIQP problem based on the improved random branch strategy according to claim 1, characterized in that, Inputting the obtained real-time multi-source prediction data set and the branch strategy candidate set into a pre-trained decision tree model to obtain the initial weight parameters of the target distribution network reconstruction branch strategy, including: Constructing an initial decision tree model based on Extra Trees; Training the initial decision tree model according to the multi-source historical data set, the branch strategy candidate set, and the expected value data corresponding to each branch strategy to obtain a trained decision tree model; Inputting the obtained real-time multi-source prediction data set into the decision tree model to obtain the initial weight parameters of each branch strategy in the distribution network reconstruction process.

8. The optimization method for MIQP problems based on the improved random branching strategy according to claim 7, characterized in that, The initial decision tree model includes: Where Gini is the Gini impurity, which is used to evaluate the quality of the split point and is defined as: where, is the probability of , is the total number of categories.

9. The optimization processing method for MIQP problems based on the improved random branching strategy as described in claim 1, wherein After generating the strategy weight parameters of the branch strategy based on the initial weight parameters and the expected value, and optimizing the solution process of the target distribution network reconstruction according to the strategy weight parameters, it further includes: Executing the target reconstruction strategy determined by the solution result, and real-time monitoring the operation state data of the distribution network; Updating the target reconstruction strategy in real time according to the operation state data to complete the adaptive adjustment of the target distribution network reconstruction.

10. An optimization processing system for MIQP problems based on an improved random branching strategy, characterized in that, It includes: An acquisition module, configured to construct a multi-source historical data set based on the obtained historical load characteristic data, historical line characteristic data, and historical topology characteristic data of the target distribution network; A construction module, configured to construct a graph network model according to the multi-source historical data set, traverse the paths from the power source nodes to each load node in the target distribution network based on a graph search algorithm, and generate a branch strategy candidate set for the target distribution network reconstruction according to the paths; A solution module, configured to generate multiple typical distribution network operation scenarios based on the multi-source historical data set, reconstruct the target distribution network under the typical distribution network operation scenarios according to the branch and bound algorithm, and calculate the expected values of different branch strategies in the branch strategy candidate set during the reconstruction process, where the expected values reflect the number of nodes and time information searched by each branch strategy during the target distribution network reconstruction process; A prediction module, configured to input the obtained real-time load characteristic data, real-time line characteristic data, and real-time topology characteristic data into a pre-trained decision tree model during the reconstruction process of the target distribution network to obtain the initial weight parameters of each branch strategy in the target distribution network reconstruction process; An optimization module is used to generate the policy weight parameters of the branch policy based on the initial weight parameters and the expected values, and optimize the solution process of the target distribution network reconstruction according to the policy weight parameters.

Images