Power distribution network power scheduling optimization method and system

Abstract

The invention discloses a power distribution network power scheduling optimization method and system, which are used for solving the technical problem of poor calculation efficiency of a power distribution network power scheduling optimization scheme caused by great calculation complexity of joint solution in a traditional power distribution network power scheduling optimization method. The method comprises the following steps: constructing an initial classification tree model according to an obtained optimal topology scheme in a plurality of power distribution network operation scenes and network operation states in the plurality of power distribution network operation scenes; performing global optimization on a tree structure of the initial classification tree model by using a mixed integer programming method based on a preset feature selection and division mechanism to generate an optimal classification tree model; and when a current power distribution network operation state of the power distribution network is received, based on a mathematical programming method, solving an optimization problem corresponding to the power distribution network by adopting the optimal classification tree model according to the current power distribution network operation state, and generating a target power distribution network power scheduling scheme.

Description

Technical Field

[0001] The present invention relates to the technical field of power system operation optimization, and in particular to a distribution network power dispatch optimization method and system. Background Art

[0002] Optimizing power dispatch in distribution networks is a key technology for achieving safe, economical, and efficient power grid operation. Typical distribution network power dispatch optimization problems aim to determine the optimal operating strategy while satisfying the physical constraints of the distribution system (such as power flow balance, voltage limits, and line capacity). The goal is typically to minimize network losses, reduce operating costs, or improve system reliability.

[0003] During the operation of the distribution network, the network topology is not fixed, but can be flexibly adjusted through switching operations, thereby further reducing power flow congestion, increasing the absorption of renewable energy, and improving overall operational performance. This leads to the problem of coordinated topology reconstruction and scheduling optimization. Traditional methods for solving such problems mainly rely on mixed integer nonlinear programming models, attempting to optimize the topology scheme and operation strategy simultaneously under a unified framework. However, this problem essentially involves the coupling of discrete variables and continuous variables, and is extremely difficult to solve. Especially in large-scale networks or multi-scenario conditions, it often faces challenges such as high computational overhead, easy to fall into local optimality, or even failure to converge. In addition, with the large-scale access of flexible resources such as distributed power sources, electric vehicles, and energy storage systems, the uncertainty of operating status has been further aggravated, and traditional methods have gradually shown bottlenecks in efficiency and robustness in practical applications.

[0004] The topology structure in traditional distribution network power dispatch optimization methods is usually represented by a set of 0-1 variables to represent the topology s at time t. t These variables are coupled with the continuous voltage, current, and power variables, forming a MINLP (Mixed-Integer Nonlinear Programming) problem. Because the combinatorial space of topological variables grows exponentially, the combined solution faces extremely high computational complexity, resulting in poor computational efficiency in optimizing power dispatch in distribution networks. Summary of the Invention

[0005] The present invention provides a distribution network power dispatch optimization method and system for solving the technical problem that the traditional distribution network power dispatch optimization method faces great computational complexity in joint solution, resulting in poor computational efficiency of the distribution network power dispatch optimization scheme.

[0006] A first aspect of the present invention provides a method for optimizing power dispatching in a distribution network, comprising:

[0007] Obtain the optimal topology solutions and network operation status under multiple distribution network operation scenarios;

[0008] constructing an initial classification tree model according to the optimal topology solutions under the plurality of distribution network operation scenarios and the network operation states under the plurality of distribution network operation scenarios;

[0009] Based on a preset feature selection and partitioning mechanism, a mixed integer programming method is used to globally optimize the tree structure of the initial classification tree model to generate an optimal classification tree model;

[0010] When the current distribution network operating status of the distribution network is received, based on the mathematical programming method, the optimal classification tree model is used to solve the optimization problem corresponding to the distribution network according to the current distribution network operating status, and a target distribution network power scheduling plan is generated.

[0011] Optionally, constructing an initial classification tree model according to the optimal topology solutions under the plurality of distribution network operation scenarios and the network operation states under the plurality of distribution network operation scenarios includes:

[0012] Based on the optimal topology solutions under the multiple distribution network operation scenarios and the network operation states under the multiple distribution network operation scenarios, generating multiple optimal topology solution category numbers and multiple network operation state feature vectors;

[0013] An initial classification tree model is constructed with a plurality of the network operation state feature vectors as input and a plurality of the optimal topology solution category numbers as output.

[0014] Optionally, the generating of a plurality of optimal topology scheme category numbers and a plurality of network operation status feature vectors based on a plurality of optimal topology schemes under the distribution network operation scenarios and a plurality of network operation statuses under the distribution network operation scenarios includes:

[0015] Encode the optimal topology solutions under each of the distribution network operation scenarios respectively, and output the optimal topology solution category number corresponding to the optimal topology solution under each of the distribution network operation scenarios;

[0016] Feature extraction is performed on the network operation status under each of the distribution network operation scenarios, and a network operation status feature vector corresponding to the network operation status under each of the distribution network operation scenarios is output.

[0017] Optionally, the mathematical programming method is based on using the optimal classification tree model to solve the optimization problem corresponding to the distribution network according to the current distribution network operation state to generate a target distribution network power dispatching plan, including:

[0018] Extracting features of the current distribution network operation state and outputting a feature vector of the current distribution network operation state;

[0019] Using the current distribution network operation state feature vector as input to the optimal classification tree model, and outputting a target topology scheme number;

[0020] Decoding the target topology scheme number to generate a target topology scheme;

[0021] Based on the target topology solution, the mathematical programming method is used to solve the optimization problem corresponding to the distribution network, and a target distribution network power scheduling solution is output.

[0022] Optionally, the objective function corresponding to the optimal classification tree model is specifically:

[0023] ;

[0024] Where N is the number of samples, that is, the total number of distribution network operation scenarios; tr represents the structure of the classification tree, which includes splitting variables, thresholds, and leaf node categories; represents the classification error; Represents the complexity weight coefficient, which is used to control the model size; Indicates the complexity of the tree structure; is the i-th optimal topology solution; is the true topology solution corresponding to the i-th optimal topology solution.

[0025] Optionally, the optimization problem corresponding to the distribution network is specifically:

[0026] ;

[0027] Where F is the objective function of the optimization problem corresponding to the distribution network; is the decision variable at time t, representing the target distribution network power dispatching plan at time t; is the characteristic vector of the current distribution network operation status at time t; is the target topology solution at time t; is the price coefficient, i.e. the price per unit of electricity; is the power function; It is the power flow balance constraint of the power system; is an inequality constraint; 、 is the value range limit of the decision variable, that is, the value range limit of the distribution network power.

[0028] A second aspect of the present invention provides a distribution network power dispatch optimization system, comprising:

[0029] An acquisition module is used to obtain the optimal topology solution and network operation status under multiple distribution network operation scenarios;

[0030] A construction module, configured to construct an initial classification tree model according to the optimal topology solutions under the plurality of distribution network operation scenarios and the network operation states under the plurality of distribution network operation scenarios;

[0031] An optimization module is used to perform global optimization on the tree structure of the initial classification tree model based on a preset feature selection and partitioning mechanism using a mixed integer programming method to generate an optimal classification tree model;

[0032] The solution module is used to, when receiving the current distribution network operating status of the distribution network, use the optimal classification tree model based on the mathematical programming method to solve the optimization problem corresponding to the distribution network according to the current distribution network operating status, and generate a target distribution network power scheduling plan.

[0033] A third aspect of the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the distribution network power dispatch optimization method as described in any one of the above items.

[0034] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed, implements the steps of the distribution network power dispatch optimization method as described in any one of the above items.

[0035] A fifth aspect of the present invention provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions, wherein when the program instructions are executed by a computer, the computer executes the steps of the distribution network power dispatch optimization method as described in any one of the above items.

[0036] It can be seen from the above technical solutions that the present invention has the following advantages:

[0037] The above technical solution of the present invention provides a method for optimizing power dispatching of a distribution network. First, the optimal topology schemes under multiple distribution network operation scenarios and the network operation status under multiple distribution network operation scenarios are obtained; then, an initial classification tree model is constructed according to the optimal topology schemes under multiple distribution network operation scenarios and the network operation status under multiple distribution network operation scenarios; based on a preset feature selection and partitioning mechanism, a mixed integer programming method is used to perform global optimization on the tree structure of the initial classification tree model to generate an optimal classification tree model; finally, when the current distribution network operation status of the distribution network is received, the optimal classification tree model is used based on a mathematical programming method to optimize the current distribution network operation status. The optimization problem corresponding to the distribution network is solved in the row state to generate a target distribution network power dispatching scheme; based on the above scheme, the present invention uses a preset feature selection and partitioning mechanism to globally optimize the tree structure of the constructed initial classification tree model using a mixed integer programming method to obtain an optimal classification tree model, and then combines the mathematical programming method to solve the optimization problem corresponding to the distribution network to generate a target distribution network power dispatching scheme. The present invention combines the mathematical programming method and the optimal classification tree model to decouple and solve complex coupling problems, thereby quickly completing the solution of the optimization problem, greatly shortening the decision-making time, and thereby improving the computational efficiency of the distribution network power dispatching optimization scheme. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0039] Figure 1 A flowchart of a method for optimizing power dispatching in a distribution network according to the first embodiment of the present invention;

[0040] Figure 2 This is a structural block diagram of a distribution network power dispatch optimization system provided in Example 2 of the present invention. DETAILED DESCRIPTION

[0041] The embodiments of the present invention provide a distribution network power dispatch optimization method and system for solving the technical problem that the joint solution in the traditional distribution network power dispatch optimization method faces great computational complexity, resulting in poor computational efficiency of the distribution network power dispatch optimization solution.

[0042] In order to make the purpose, features, and advantages of the present invention more obvious and easy to understand, the technical solutions 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 embodiments described below 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 making creative work are within the scope of protection of the present invention.

[0043] See also Figure 1 , Figure 1 This is a flowchart of the steps of a distribution network power scheduling optimization method provided in Example 1 of the present invention.

[0044] The present invention provides a power distribution network power dispatch optimization method, comprising:

[0045] Step 101: Obtain optimal topology solutions under multiple distribution network operation scenarios and network operation states under multiple distribution network operation scenarios.

[0046] The network operation status in the distribution network operation scenario includes node load, voltage, power output, etc.

[0047] It should be noted that multiple distribution network operation scenarios refer to a collection of typical operating states of the distribution network under various operating conditions, including time, load levels, and distributed generation output. These scenarios can be derived through historical operational data mining, load forecast curve construction, and random perturbation generation. They aim to fully cover the operating characteristics of the distribution network under different seasons, load peaks and valleys, and renewable energy fluctuations. Each scenario corresponds to a complete set of system input characteristics (such as node load, voltage, and power output), and its optimal topology solution is obtained through offline calculation using traditional optimization models, such as the Mixed-Integer Second-Order Cone Programming (MISOCP) model. Using these data samples, the present invention constructs an input-output mapping, providing a high-quality and comprehensive knowledge base for training the Optimal Classification Tree (OCT) model, ensuring the model's good generalization and robustness in practical applications. Furthermore, a single optimal topology solution can correspond to multiple distribution network operation scenarios.

[0048] Step 102: construct an initial classification tree model based on the optimal topology solutions under multiple distribution network operation scenarios and the network operation states under multiple distribution network operation scenarios.

[0049] It should be noted that in the traditional distribution network dispatch optimization model, the topology structure is usually represented by a set of 0-1 variables to represent the topology s at time t. t, these variables are coupled with continuous voltage, current, and power variables to form a MINLP problem. Due to the exponential growth of the combinatorial space of topological variables, the joint solution faces extremely high computational complexity, making it difficult to meet the response speed requirements in real-time applications. To solve the above problems, the present invention proposes a method of converting the topology optimization problem into a multi-class classification problem. That is, the process of "searching" for the optimal topology structure at runtime is converted into a process of "predicting" an optimal topology category, significantly reducing the solution dimension and time overhead.

[0050] Specifically, step 102 may include the following sub-steps S21-S22:

[0051] Step S21: Based on the optimal topology solutions under multiple distribution network operation scenarios and the network operation status under multiple distribution network operation scenarios, generate multiple optimal topology solution category numbers and multiple network operation status feature vectors;

[0052] Furthermore, step S21 may include the following sub-steps S211-S212:

[0053] Step S211: Encode the optimal topology solutions under each distribution network operation scenario respectively, and output the optimal topology solution category number corresponding to the optimal topology solution under each distribution network operation scenario;

[0054] Step S212: extract features of the network operation status under each distribution network operation scenario, and output a network operation status feature vector corresponding to the network operation status under each distribution network operation scenario.

[0055] Step S22: construct an initial classification tree model with multiple network operation status feature vectors as input and multiple optimal topology solution category numbers as output.

[0056] It should be noted that the present invention generates a large amount of optimal topology solution data under operating scenarios offline, and uses this to train an OCT model (initial classification tree model). First, a training sample set is constructed through a large amount of simulation generation or historical operating data. The sample features are the network operating status under different scenarios (such as node load, voltage, power output, etc.), and the corresponding labels are the optimal topology solutions (expressed in the form of switch state vectors) obtained by offline solving the traditional optimization model under each scenario. On this basis, an OCT model with global optimal characteristics is constructed to learn the mapping relationship between input features and optimal topology. The model takes the operating status characteristics of the distribution network as input and the optimal topology strategy as output, thereby establishing a mapping relationship between input and output. Through this model, the system can quickly determine the topology structure suitable for the current operating status before scheduling, effectively avoiding the computational cost brought by the combination of variables.

[0057] Specifically, the present invention obtains the network operating status under distribution network operation scenarios by simulating or using historical operating data to construct a large number of operation scenario samples (i.e., network operating status under multiple distribution network operation scenarios). Each sample is composed of network status characteristics, including node load, voltage, and power output. For each sample scenario, the optimal topology solution for the distribution network operation scenario is solved offline using a mixed integer optimization model (such as MILP or MISOCP) to obtain the optimal topology solution for that scenario, which serves as the "label" for the sample.

[0058] Furthermore, the topology structure (i.e., a specific set of switch state combinations) in each optimal topology solution is mapped to a discrete category number, for example:

[0059] Topology A (switch 1 open, 2 closed, 3 open) → Category 1;

[0060] Topology B (switch 1 closed, 2 open, 3 closed) → Category 2;

[0061] It is worth mentioning that when dealing with high-dimensional topology, similar topological structures can be clustered into the same category to reduce the number of categories.

[0062] Furthermore, for the construction of the initial classification tree model, the current operating state feature vector of the distribution network, that is, the network operating state feature vector (node load, voltage forecast, distributed power output, etc.) as input, the optimal topology scheme is numbered, that is, the optimal topology scheme category number As input, build the initial classification tree model , this paper formalizes the topology optimization problem as a supervised learning task, where, is the number of identified topological classes, Represents the model obtained by training the optimal classification tree, that is, the classification tree model.

[0063] Step 103: Based on the preset feature selection and partitioning mechanism, a mixed integer programming method is used to globally optimize the tree structure of the initial classification tree model to generate an optimal classification tree model.

[0064] The optimal classification tree model consists of the model objective (objective function), the classification tree model structure, and the pre-set feature selection and partitioning mechanism. The goal is to find an OCT tree that minimizes the total cost after all input samples are entered. The tree structure must satisfy the constraints specified in the "model structure" and "feature selection and partitioning mechanism."

[0065] It should be noted that after completing the classification transformation of the topology optimization problem and constructing the training dataset, to ensure the optimality, interpretability, and inference efficiency of the classification results, this paper uses an optimal classification tree model as the core component of the topology prediction model. This model not only quickly determines the optimal topology strategy for the current operating state but also provides clear decision logic in the form of a tree structure, making it suitable for the deployment and operation of intelligent scheduling systems.

[0066] Furthermore, regarding the model objective (i.e., the objective function corresponding to the optimal classification tree model), the design goal of the optimal classification tree model is to minimize the training error within a limited model complexity while ensuring a clear model structure for fast online execution. Unlike traditional greedy tree construction, OCT uses a global optimization strategy to design the tree structure, avoiding overfitting or model instability caused by local optimality. The objective function corresponding to the optimal classification tree model is specifically:

[0067] ;

[0068] Where N is the number of samples, that is, the total number of distribution network operation scenarios; tr represents the structure of the classification tree, which includes splitting variables, thresholds, and leaf node categories; represents the classification error; Represents the complexity weight coefficient, which is used to control the model size; Indicates the complexity of the tree structure, such as the number of nodes, depth, number of leaf nodes, etc. is the i-th optimal topology solution; is the true topology solution corresponding to the i-th optimal topology solution.

[0069] For the model structure of the classification tree model, the optimal classification tree model consists of three parts: internal nodes, leaf nodes and path structure. Internal nodes (split nodes): Each node contains a conditional judgment rule in the form of: , represents the characteristic variable of the j-th operating state The size of the threshold is determined and the samples are divided into the left / right subtree.

[0070] Leaf node: Each leaf node outputs a topology strategy category , corresponding to a fixed switch combination or network structure.

[0071] Path structure: A decision path is formed from the root node to the leaf node, and multiple feature splitting rules are applied in sequence, and finally a classification result is given at a leaf node.

[0072] Furthermore, for the preset feature selection and partitioning mechanism, in order to build a globally optimal classification tree model, this method needs to make the following judgments: ① At each new node, you must choose to split or stop. ② After the node stops splitting, you must choose a label to assign to this new leaf node. ③ If you choose to branch, you must choose the variable to branch. ④ When classifying the training points according to the tree being built, all samples must choose to assign a leaf node to satisfy the structure of the tree. Specifically, the following variables are introduced:

[0073] ;

[0074] in, Indicates the Whether the internal nodes are divided on feature j, Indicates the Internal nodes are divided on the running feature vector j, otherwise the running feature vector j does not participate in the division; Indicates the Whether the node is activated, Indicates the The node is activated, otherwise it is not activated; Indicates whether sample i is assigned to a leaf node , Indicates that sample i is assigned to a leaf node Otherwise, it is not assigned to the leaf node ; represents the partition threshold of the node; Represents a leaf node The predicted category of the output, K is the total number of leaf nodes. Based on this, the tree construction model constraints can be written as follows:

[0075] ;

[0076] in, Indicates that each node can only be divided on one feature. If , indicating that the node has not been divided. 、 .

[0077] Based on the above foundation, we can get the constraint , Make sure there is only parent node Activate, the child nodes can be activated, among which, For the parent node.

[0078] The sample path constraint is defined, indicating that samples are assigned to leaf nodes only if all splitting conditions on the path are satisfied. The Big M method is used to control the constraints on samples going to the left and right subtrees.

[0079] Furthermore, we define the formula , which defines the sample path constraint and represents the sample Assigned to leaf nodes All splitting conditions on the path must be satisfied. The Big M method is used to control the constraints on samples going to the left and right subtrees.

[0080] Further, define the constraints , this constraint means that a sample can only be assigned to one leaf node.

[0081] Ultimately, based on the above process, all splitting conditions and sample paths of the tree are encoded into a mixed integer model. Through training, the classification tree will select the most discriminative features for node splitting, ultimately constructing the optimal strategy tree (i.e., the optimal classification tree model).

[0082] Step 104: When the current distribution network operating status of the distribution network is received, based on a mathematical programming method, an optimal classification tree model is used to solve the optimization problem corresponding to the distribution network according to the current distribution network operating status, and a target distribution network power dispatching plan is generated.

[0083] Mathematical programming is a technique for solving decision-making problems through mathematical modeling and optimization algorithms. Its core goal is to find the optimal solution to an objective function (such as minimizing cost or maximizing efficiency) under given constraints. Mathematical programming methods include convex optimization methods and numerical optimization methods.

[0084] It should be noted that, based on the topological structure output by the optimal classification tree model, the physical model is used to solve the scheduling problem. This stage uses a fixed topology as input and solves traditional continuous optimization variables, including: power flow balance equations (AC or DC power flow models); node voltage upper and lower bounds; branch power flow capacity constraints; power source output and load adjustability bounds; and objective functions (such as minimizing network losses, operating costs, and voltage deviation). These optimization problems can be solved using convex or numerical optimization methods, which have a mature engineering foundation and rigorous physical interpretation. Since the topology is known, this model is a standard scheduling problem, and the solution is stable and efficient.

[0085] Specifically, step 104 may include the following sub-steps S41-S44:

[0086] Step S41: extract features of the current distribution network operation state and output a feature vector of the current distribution network operation state;

[0087] Step S42: using the current distribution network operation state feature vector as the input of the optimal classification tree model, and outputting the target topology scheme number;

[0088] Step S43: Decode the target topology solution number to generate a target topology solution;

[0089] Step S44: Based on the target topology solution, a mathematical programming method is used to solve the optimization problem corresponding to the distribution network, and a target distribution network power dispatching solution is output.

[0090] It should be noted that once the optimal tree structure is determined, the splitting rules for each node are explicit. Inference only needs to start from the root node and determine the sample characteristics layer by layer. Previously, online operations required a combinatorial search of topological variables (solving a mixed-integer optimization problem). Now, topological solutions can be directly output through the classification model at runtime, significantly reducing decision-making time. It is worth noting that this invention uses an optimal classification tree, written as a mixed-integer optimization form, which has global optimization and greater interpretability than ordinary tree models.

[0091] Furthermore, after completing the training of the optimal classification tree model, the present invention rapidly predicts a set of optimal or suboptimal topologies with high performance by inputting the current distribution network operating status at runtime. Next, based on this recommended topology, the present invention further constructs and solves a distribution network scheduling optimization model under physical constraints, thus completing the entire closed-loop process from "strategy selection" to "operational optimization."

[0092] After obtaining the optimal topology, the present invention uses the following typical scheduling optimization model as the solution object, that is, the optimization problem corresponding to the distribution network, which is written as follows:

[0093] ;

[0094] Where F is the objective function of the optimization problem corresponding to the distribution network; is the decision variable at time t, representing the target distribution network power dispatching plan at time t; is the characteristic vector of the current distribution network operation status at time t; is the target topology solution at time t, which represents the defined optimal topology strategy; is the price coefficient, i.e. the price per unit of electricity; is the power function, which means that in the topology , running status and scheduling decisions The performance indicators related to scheduling under the joint action of the above factors can be directly understood as the output power of different generators; It is the power flow balance constraint of the power system; is an inequality constraint; 、 is the value range limit of the decision variable, that is, the value range limit of the distribution network power.

[0095] Specifically, the optimization problem corresponding to the distribution network is defined as follows:

[0096] Objective function: F is the objective function of the optimization problem, which means that in all time periods Minimize the weighted sum of dispatch costs (or other performance objectives such as network loss, emission, etc.) within the given constraints. That is, find a set of optimization variables and topology , so that the above total objective function is minimized.

[0097] Equality constraints: The power flow balance constraint of the power system, which ensures that the power revenue and expenditure at each node are balanced, is a fundamental physical law of system operation. Ensuring that the power grid adheres to power conservation under the optimization scheme is usually a nonlinear equation, which determines the non-convexity of the problem.

[0098] Inequality constraints: These are inequality constraints that prevent the network from entering unsafe areas and ensure the feasibility of the optimization solution. They mainly include node voltage limits, branch current limits, and equipment capacity limits.

[0099] Variable bound constraints: 、 It is the value range limit of the decision variable, which limits the optimization variable Prevent unachievable scheduling results within the limits allowed by physical equipment.

[0100] On this basis, the model can replace the AC power flow equation in the dispatching model with the linearized power flow model DistFlow model, simplify the dispatching optimization problem into a convex optimization problem, and use the solver for fast calculation, and finally output the target distribution network power distribution plan (that is, the target distribution network power dispatching plan).

[0101] Furthermore, in order to verify the effectiveness and versatility of the method proposed in this invention, a variety of representative comparison methods are set as performance benchmarks, covering classic joint optimization modeling strategies and mainstream topology prediction and classification models. Specifically, they include:

[0102] MISOCP (Mixed-Integer Second-Order Cone Programming): Models the problem as a mixed-integer second-order cone optimization problem, jointly solving for the optimal topology and scheduling solution in each runtime scenario. This method uses the Gurobi optimizer for a global solution, resulting in the theoretically optimal solution.

[0103] CART (Classification and Regression Tree): A traditional classification tree model based on greedy splitting, using the Gini index as the partitioning criterion and trained using the scikit-learn framework.

[0104] RF (Random Forest): A random forest classification model composed of multiple CART trees. It uses bootstrap resampling and feature subset sampling to improve model stability and accuracy. The forest is constructed with 100 subtrees and trained using the random forest classifier provided by scikit-learn.

[0105] OCT (Optimal Classification Tree): This core classification model uses mixed integer programming to find the globally optimal tree structure within given tree depth and complexity constraints. The optimization goal is to minimize the weighted sum of classification error and tree complexity.

[0106] To ensure fair comparisons between different methods, CART, RF, and OCT used identical training data and employed cross-validation to determine model hyperparameters. MISOCP independently solved each test scenario, serving as a reference for the optimal joint topology-scheduling solution. All of these methods were executed on the same computing platform and operating environment to ensure comparable results.

[0107] All experiments were run on the same platform: an Intel Core i7-11700 CPU, 32GB of RAM, Windows 11 operating system, and MATLAB R2021a. The maximum OCT tree depth was 10, and the minimum number of leaf node samples was 20. Each set of experimental results is the average of 50 repeated runs to eliminate the effects of occasional perturbations.

[0108] The training and solution times of the proposed method, as well as the prediction accuracy under different test systems, are shown in Tables 1 and 2. It can be observed that the joint modeling MISOCP method achieves high-precision and fast solutions in small-scale systems, with a solution time of approximately 27 seconds for a 33-node system. However, as the system scales up, the computational burden of this method increases rapidly. In a 301-node system, a single solution takes as long as 6678 seconds, presenting a significant scalability bottleneck. In comparison, the learning-based topology prediction methods (CART, RF, and the proposed OCT) demonstrate remarkable consistency in inference efficiency. Regardless of network size, the total computation time after completing topology prediction and invoking the scheduling model remains under 40 seconds for all three methods, demonstrating their suitability for rapid deployment in engineering scenarios. While CART, due to its greedy construction mechanism, consumes the least training time, its prediction accuracy after training suffers from significant disadvantages. While RF is reasonably fast during the inference phase, its training time increases rapidly with network size (for example, exceeding 3,300 seconds in a 301-node system), and its high model complexity hinders system interpretation. It is noteworthy that the proposed OCT method, despite its relatively long training time (for example, 4,197 seconds in a 301-node system), offers the ability to train once and reuse over time. During the scheduling solution phase, its computation time is only slightly higher than CART (38.66 seconds vs. 34.89 seconds), essentially on par with RF, while also offering greater interpretability and robustness.

[0109] Table 2 further demonstrates the prediction accuracy of different methods across three types of systems. It can be seen that the accuracy of the CART model decreases rapidly with system scale, from 57.66% for 33 nodes to only 31.55% for 301 nodes, failing to meet the accuracy requirements for scheduling decisions. The RF model has slightly higher overall prediction accuracy, but still exhibits significant degradation in large-scale systems. In the 301-node system, its average accuracy is 68.01%, approximately 21 percentage points lower than the OCT model. In contrast, the proposed optimal classification tree (OCT) model demonstrated high consistency and accuracy across all tested systems. Its prediction accuracy reached 95.31%, 90.16%, and 88.91% for the 33-node, 118-node, and 301-node systems, respectively, demonstrating that the model maintains stable generalization capabilities as system scale increases. Combined with its transparent structure and ease of verification and engineering deployment, the OCT model not only significantly outperforms other methods in accuracy but also demonstrates excellent practical application value.

[0110]

[0111]

[0112] As a comparison of technical effects, existing technologies can be used as a reference. The dispatch optimization problem of distribution networks plays a central role in modern power systems, especially in the context of a high proportion of distributed generation and rapidly changing loads. Dispatch optimization not only needs to meet traditional physical constraints such as power balance and voltage constraints, but also needs to adapt to the flexible adjustment of network topology to address operational risks, reduce network losses, and improve system efficiency. However, when considering reconfigurable network topology, the dispatch optimization problem will introduce a large number of binary combination variables, which are highly coupled with continuous control variables. This transforms the original problem into a non-convex, mixed-integer, nonlinear optimization problem, making its solution exponentially more difficult. Existing traditional methods (such as mixed-integer programming and heuristic algorithms) have difficulty effectively solving such problems in large-scale networks and real-time control scenarios.

[0113] To address the above-mentioned issues, the present invention provides a distribution network power dispatch optimization method that achieves efficient coordination between distribution network topology optimization and dispatch control. The core concept of this method is to rapidly identify the optimal topology through a data-driven approach, then perform physically feasible dispatch optimization within this topology. This decouples the complex coupled problem and improves computational efficiency and interpretability. Specifically, the present invention decouples the previously unified complex problem into two phases. In the first phase, an optimal classification tree (OCT) is used to learn from a large amount of historical operating data to rapidly identify the optimal topology under current operating conditions. In the second phase, based on the selected topology, a physics-based power flow model and dispatch optimizer are employed to solve the optimal power distribution scheme that satisfies operational constraints. The core advantage of this method lies in leveraging the efficiency and interpretability of the decision tree model to rapidly classify and select topology reconstruction schemes, effectively avoiding the large-scale enumeration of combinatorial topology variables during the solution phase. At the same time, the dispatch optimization component retains the physical mechanism of the power flow equation, ensuring that the optimization results meet engineering feasibility. Compared with traditional mixed-integer models, this method significantly reduces solution time, improves model response speed, and exhibits good generalizability and adaptability. The present invention also focuses on integrating a data-driven optimal classification tree model with a physics-based grid operation model, enabling efficient scheduling and control of distribution systems in complex operating environments. Furthermore, the present invention balances model interpretability with satisfying physical constraints.

[0114] In an embodiment of the present invention, the present invention provides a method for optimizing power dispatching of a distribution network. First, the optimal topology schemes under multiple distribution network operation scenarios and the network operation status under multiple distribution network operation scenarios are obtained; then, an initial classification tree model is constructed based on the optimal topology schemes under multiple distribution network operation scenarios and the network operation status under multiple distribution network operation scenarios; based on a preset feature selection and partitioning mechanism, a mixed integer programming method is used to perform global optimization on the tree structure of the initial classification tree model to generate an optimal classification tree model; finally, when the current distribution network operation status of the distribution network is received, the optimal classification tree model is used based on the mathematical programming method to optimize the current distribution network operation status. The operating state solves the optimization problem corresponding to the distribution network and generates a target distribution network power dispatching scheme; based on the above scheme, the present invention uses a preset feature selection and partitioning mechanism, and uses a mixed integer programming method to globally optimize the tree structure of the constructed initial classification tree model to obtain the optimal classification tree model, and then combines the mathematical programming method to solve the optimization problem corresponding to the distribution network and generate a target distribution network power dispatching scheme. The present invention combines the mathematical programming method and the optimal classification tree model to decouple and solve complex coupling problems, thereby quickly completing the solution of the optimization problem, greatly shortening the decision-making time, and thereby improving the computational efficiency of the distribution network power dispatching optimization scheme.

[0115] See also Figure 2 , Figure 2 This is a structural block diagram of a distribution network power dispatch optimization system provided in Example 2 of the present invention.

[0116] The present invention provides a power distribution network power dispatch optimization system, comprising:

[0117] An acquisition module 201 is configured to acquire an optimal topology solution and a network operation status under multiple distribution network operation scenarios;

[0118] A construction module 202 is configured to construct an initial classification tree model based on the optimal topology solutions under multiple distribution network operation scenarios and the network operation states under multiple distribution network operation scenarios;

[0119] The optimization module 203 is used to perform global optimization on the tree structure of the initial classification tree model based on a preset feature selection and partitioning mechanism using a mixed integer programming method to generate an optimal classification tree model;

[0120] The solution module 204 is used to, when receiving the current distribution network operating status of the distribution network, use the optimal classification tree model based on the mathematical programming method to solve the optimization problem corresponding to the distribution network according to the current distribution network operating status and generate a target distribution network power scheduling plan.

[0121] Furthermore, the construction module 202 includes:

[0122] The first submodule is configured to generate a plurality of optimal topology scheme category numbers and a plurality of network operation status feature vectors based on the optimal topology schemes under a plurality of distribution network operation scenarios and the network operation status under a plurality of distribution network operation scenarios;

[0123] The second submodule is used to construct an initial classification tree model with multiple network operation status feature vectors as input and multiple optimal topology solution category numbers as output.

[0124] Furthermore, the first submodule is specifically configured to:

[0125] Encode the optimal topology scheme under each distribution network operation scenario respectively, and output the optimal topology scheme category number corresponding to the optimal topology scheme under each distribution network operation scenario;

[0126] The network operation status under each distribution network operation scenario is subjected to feature extraction, and the network operation status feature vector corresponding to the network operation status under each distribution network operation scenario is output.

[0127] Furthermore, the solution module 204 is specifically configured to:

[0128] Extract features of the current distribution network operation status and output the current distribution network operation status feature vector;

[0129] The current distribution network operation state feature vector is used as the input of the optimal classification tree model, and the target topology scheme number is output;

[0130] Decoding the target topology scheme number to generate the target topology scheme;

[0131] Based on the target topology scheme, the mathematical programming method is used to solve the optimization problem corresponding to the distribution network, and the target distribution network power dispatching scheme is output.

[0132] Furthermore, the objective function corresponding to the optimal classification tree model is:

[0133] ;

[0134] Where N is the number of samples, that is, the total number of distribution network operation scenarios; tr represents the structure of the classification tree, which includes splitting variables, thresholds, and leaf node categories; represents the classification error; Represents the complexity weight coefficient, which is used to control the model size; Indicates the complexity of the tree structure; is the i-th optimal topology solution; is the true topology solution corresponding to the i-th optimal topology solution.

[0135] Furthermore, the optimization problem corresponding to the distribution network is specifically:

[0136] ;

[0137] Where F is the objective function of the optimization problem corresponding to the distribution network; is the decision variable at time t, representing the target distribution network power dispatching plan at time t; is the characteristic vector of the current distribution network operation status at time t; is the target topology solution at time t; is the price coefficient, i.e. the price per unit of electricity; is the power function; It is the power flow balance constraint of the power system; is an inequality constraint; 、 is the value range limit of the decision variable, that is, the value range limit of the distribution network power.

[0138] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices, modules and sub-modules can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0139] An embodiment of the present invention further provides a computer device comprising a memory and a processor, wherein a computer program is stored in the memory; when the computer program is executed by the processor, the processor executes the steps of the distribution network power dispatch optimization method as described in any of the above embodiments.

[0140] An embodiment of the present invention further provides a computer-readable storage medium having a computer program / instruction stored thereon. When the computer program / instruction is executed by a processor, the steps of the distribution network power dispatch optimization method as described in any of the above embodiments are implemented.

[0141] An embodiment of the present invention further provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the distribution network power dispatch optimization method as described in any of the above embodiments.

[0142] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of units is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0143] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0144] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments can still be modified, or some of the technical features thereof can be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A power distribution network power dispatch optimization method, characterized in that: include: Obtain the optimal topology solution and network operation status under multiple distribution network operation scenarios; constructing an initial classification tree model according to the optimal topology solutions under the plurality of distribution network operation scenarios and the network operation states under the plurality of distribution network operation scenarios; Based on a preset feature selection and partitioning mechanism, a mixed integer programming method is used to globally optimize the tree structure of the initial classification tree model to generate an optimal classification tree model; When the current distribution network operating status of the distribution network is received, based on the mathematical programming method, the optimal classification tree model is used to solve the optimization problem corresponding to the distribution network according to the current distribution network operating status, and a target distribution network power scheduling plan is generated.

2. The power distribution network optimization method according to claim 1, characterized in that: The constructing of an initial classification tree model according to the optimal topology solutions under the plurality of distribution network operation scenarios and the network operation states under the plurality of distribution network operation scenarios comprises: Based on the optimal topology solutions under the multiple distribution network operation scenarios and the network operation states under the multiple distribution network operation scenarios, generating multiple optimal topology solution category numbers and multiple network operation state feature vectors; An initial classification tree model is constructed with a plurality of the network operation state feature vectors as input and a plurality of the optimal topology solution category numbers as output.

3. The power distribution network optimization method according to claim 2, characterized in that: The generating of a plurality of optimal topology scheme category numbers and a plurality of network operation status feature vectors based on the plurality of optimal topology schemes under the distribution network operation scenarios and the plurality of network operation statuses under the distribution network operation scenarios comprises: Encode the optimal topology solutions under each of the distribution network operation scenarios respectively, and output the optimal topology solution category number corresponding to the optimal topology solution under each of the distribution network operation scenarios; Feature extraction is performed on the network operation status under each of the distribution network operation scenarios, and a network operation status feature vector corresponding to the network operation status under each of the distribution network operation scenarios is output.

4. The power distribution network optimization method according to claim 1, characterized in that: The method based on mathematical programming adopts the optimal classification tree model to solve the optimization problem corresponding to the distribution network according to the current distribution network operation state, and generates a target distribution network power dispatching plan, including: Extracting features of the current distribution network operation state and outputting a feature vector of the current distribution network operation state; Using the current distribution network operation state feature vector as input to the optimal classification tree model, and outputting a target topology scheme number; Decoding the target topology scheme number to generate a target topology scheme; Based on the target topology solution, the mathematical programming method is used to solve the optimization problem corresponding to the distribution network, and a target distribution network power scheduling solution is output.

5. The power distribution network power dispatch optimization method according to claim 1, characterized in that: The objective function corresponding to the optimal classification tree model is specifically: ； Where N is the number of samples, that is, the total number of distribution network operation scenarios; tr represents the structure of the classification tree, which includes splitting variables, thresholds, and leaf node categories; represents the classification error; Represents the complexity weight coefficient, which is used to control the model size; Indicates the complexity of the tree structure; is the i-th optimal topology solution; is the true topology solution corresponding to the i-th optimal topology solution.

6. The power distribution network power dispatch optimization method according to claim 1, characterized in that: The optimization problem corresponding to the distribution network is specifically: ； Where F is the objective function of the optimization problem corresponding to the distribution network; is the decision variable at time t, representing the target distribution network power dispatching plan at time t; is the characteristic vector of the current distribution network operation status at time t; is the target topology solution at time t; is the price coefficient, i.e. the price per unit of electricity; is the power function; It is the power flow balance constraint of the power system; is an inequality constraint; 、 is the value range limit of the decision variable, that is, the value range limit of the distribution network power.

7. A power distribution network optimization system, characterized in that: include: An acquisition module is used to obtain the optimal topology solution and network operation status under multiple distribution network operation scenarios; A construction module, configured to construct an initial classification tree model according to the optimal topology solutions under the plurality of distribution network operation scenarios and the network operation states under the plurality of distribution network operation scenarios; An optimization module is used to perform global optimization on the tree structure of the initial classification tree model based on a preset feature selection and partitioning mechanism using a mixed integer programming method to generate an optimal classification tree model; The solution module is used to, when receiving the current distribution network operating status of the distribution network, use the optimal classification tree model based on the mathematical programming method to solve the optimization problem corresponding to the distribution network according to the current distribution network operating status, and generate a target distribution network power scheduling plan.

8. A computer device, characterized in that: The method comprises a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the processor executes the steps of the distribution network power dispatch optimization method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed, the distribution network power dispatch optimization method according to any one of claims 1 to 6 is implemented.

10. A computer program product, characterized in that The computer program product includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions, wherein when the program instructions are executed by a computer, the computer is caused to execute the distribution network power dispatch optimization method according to any one of claims 1 to 6.

Images