A method for calculating available power transmission capacity based on decision tree

Abstract

A decision tree-based available transmission capacity (ATC) calculation method is proposed. To solve the problem that physical calculation methods are difficult to balance the calculation speed and efficiency, the algorithm in artificial intelligence is used to establish the relationship between the input margin characteristics and ATC in a data-driven manner, so as to improve the calculation speed and accuracy of ATC. To solve the problem that the decision process of existing ATC artificial intelligence calculation methods is not known, the decision path and key features in the decision tree model are defined as visual rules, which balances the visualization of the calculation process. The framework of ATC calculation method is proposed, which meets the requirements of online calculation through offline modeling and online application, and makes the method applicable.

Description

Technical Field

[0001] This invention belongs to the field of power system technology and relates to a method for calculating available transmission capacity based on decision trees. Background Technology

[0002] The development of China's electricity market is deepening, and improving the inter-provincial electricity spot market is crucial for promoting the large-scale consumption of renewable energy and ensuring the optimal allocation of resources in the short term. However, the existing inter-provincial electricity spot market considers only one trading direction, leading to problems such as congestion of some transmission channels and underutilization of others in complex electricity market environments. Constructing a two-way trading market is an important means to solve the problem of rational resource allocation across multiple regions. How to calculate the available transfer capacity (ATC) for inter-provincial transactions in real time is a key issue in building a two-way trading market. It is of great significance for guiding electricity market transactions, promoting the consumption of renewable energy, and ensuring the safe and stable operation of the system.

[0003] ATC (Advanced Transmission Capacity) refers to the remaining commercially usable transmission capacity in the actual physical transmission network, based on existing transmission contracts. This definition indicates that in a power market environment, the issue of grid transmission capacity is no longer simply about regional power exchange capacity, but rather the maximum power that can be increased between regions or between points, based on existing transmission contracts and while ensuring the safe and reliable operation of the system. It is a measure of the transmission capacity retained by the actual transmission network based on existing transmission contracts, and can be conceptually represented as...

[0004] ,

[0005] In the formula: TTC (Total Transfer Capability) is the maximum transmission capacity, reflecting the total transmission capacity on the interconnected system tie lines under the condition of meeting various system safety and reliability requirements; TRM (Transfer Reliability Margin) is the transmission reliability margin, reflecting the impact of uncertainties on the transmission capacity between interconnected systems; CBM (Capacity Benefit Margin) is the capacity benefit margin, reflecting the transmission capacity that the transmission network should retain to ensure the smooth execution of irrevocable transmission services in ETC; ETC (Existing Transmission Commitments) is the transmission capacity occupied by existing transmission agreements (including retail user services). Depending on the stability of the ETC contract, transmission contracts can be further described using terms such as "revocable" and "irrevocable," "planned" and "reserved" transmission. When the transmission volume between interconnected networks is too large, and random interference endangers the safe operation of the system, it is necessary to reduce some transmission services. This leads to transmission congestion. In practical applications, TRM and CBM are taken as a fixed percentage of TTC based on operational experience or actual conditions, while ETC is dynamically adjusted according to the transmission agreement.

[0006] Currently, many ATC calculation methods proposed by scholars both domestically and internationally can be categorized into physical calculation methods and artificial intelligence calculation methods. Physical calculation methods include continuous power flow methods, optimal power flow methods, and probabilistic calculation analysis methods. The continuous power flow method is a static stability analysis method for power systems based on the Newton-Raphson iterative method. It is used to calculate the power flow distribution and steady-state load capacity of a power system. The calculation process considers factors such as network topology, electrical parameters, load characteristics, and generator power characteristics. Its advantages are high calculation accuracy and good stability, but it requires a large amount of computational resources and time. The optimal power flow method is a steady-state analysis method for power systems based on power system optimization theory. In the calculation process, maximizing the power of transmission channels is taken as the objective function, network element constraints are used as constraints, an optimization model is established, and a nonlinear programming algorithm is used to solve the model, thereby calculating the available transmission capacity. The advantages of this method are fast calculation speed and high accuracy, but it requires complex modeling and optimization of the power system. The probabilistic ATC calculation analysis method introduces probability functions into the optimization model, which can reflect the impact of the volatility and uncertainty of new energy sources and loads on the calculation results. However, its expression of uncertainty has mathematical limitations and is difficult to apply to online calculations.

[0007] Artificial intelligence methods have been extensively studied by scholars in recent years. By training artificial intelligence models offline and applying them online, both computational speed and accuracy have been improved. Commonly used artificial intelligence models include: back-propagation neural network (BPNN), decision tree, and multiple linear regression. The Tsinghua University team proposed the concept of safety rules, using multiple linear regression to construct a mapping relationship between the limit transmission capacity and key influencing variables. The calculation results are interpretable, but only for a single limit; they cannot effectively assess multiple limits that vary with operating conditions. The Sichuan University team focused on rapid prediction of cross-sectional limits under new energy access, proposing a typical data-driven prediction model based on Lasso regression. However, the model's generalization ability under different operating scenarios needs improvement. Furthermore, the Sichuan University team also proposed a data-driven limit transmission capacity prediction model based on deep belief networks (DBN), which can adapt to high input variable dimensionality, but its prediction effect is poor for small-scale feature variable inputs and lacks interpretability. The random forest algorithm performs slightly worse in regression problems than in classification problems, failing to predict beyond the sample itself. Decision tree-based models have good interpretability and speed in the decision-making process and are currently widely used in online voltage stability assessment and voltage stability margin calculation, but their application in calculating available transmission capacity is limited.

[0008] Electricity spot trading markets typically operate in 15-minute intervals, requiring real-time calculations of available transmission capacity to ensure the rational utilization of transmission channels. Physical methods for calculating available transmission capacity struggle to balance computational efficiency and accuracy, placing high demands on system computing power for online applications. Among artificial intelligence methods for calculating available transmission capacity, black-box neural networks offer good performance but the computational process is unknown; decision trees provide a known decision-making process but suffer from low accuracy and underfitting. Therefore, existing AI methods cannot simultaneously satisfy both high computational accuracy and visualized computational processes.

[0009] In summary, the physical calculation methods commonly used in ATC calculations struggle to balance computational efficiency and accuracy. Online applications demand high system computing power, necessitating artificial intelligence (AI) methods to address the accuracy issue. Among AI methods, black-box neural networks offer good computational performance, but the computation process remains unknown; decision trees, on the other hand, can improve computational accuracy to some extent. Summary of the Invention

[0010] This invention addresses the problems of existing technologies by providing a method for calculating available transmission capacity based on decision trees.

[0011] A decision tree-based method for calculating available transmission capacity is proposed. This method extracts rules between input features and available transmission capacity offline, establishes a decision tree model, and applies the model online to calculate available transmission capacity in real time, ensuring both speed and accuracy of the calculation.

[0012] Meanwhile, the decision tree model belongs to the "white box model" of artificial intelligence, which can provide decision paths and key features, and also takes into account the visualization of the calculation process.

[0013] A method for calculating available transmission capacity based on decision trees includes the following steps: offline steps and online steps.

[0014] The offline process involves collecting historical or simulation data to establish a training sample set with electrical capacity margin and scheduling mode as input features and available transmission capacity as the target variable.

[0015] The ATC calculation model consists of a decision tree model and visualization rules. The decision tree algorithm learns the input feature information from the training sample set to obtain the decision tree model, which is used to calculate the ATC value. The visualization rules are the decision paths and node feature information in the decision tree model, and feature importance is evaluated by calculating feature weights.

[0016] The online process transforms the online running data into the input feature form of the ATC calculation model, inputs it into the established ATC calculation model, and the decision tree model provides the ATC calculation value corresponding to the current state and provides a visualization path of the current calculation process to obtain key features.

[0017] The advantages of this invention are: the ATC calculation process is divided into two parts: an online step and an offline step. The online part establishes a decision tree model by learning from training set samples, while the offline step only requires specific parameters to obtain the ATC value for each line. Unlike black-box network models in artificial intelligence, where decision steps are unknown and engineers lack decision-making references if system adjustments are needed, the decision tree model is a "white-box model" in artificial intelligence. It can provide decision paths and key features, helping decision-makers better understand and adjust the system. Furthermore, the online part is already trained and a reliable decision tree model is established when calculating the ATC value, significantly improving the offline decision-making process.

[0018] To address the challenge of balancing computational speed and efficiency in physical calculation methods, this paper utilizes the decision tree algorithm from artificial intelligence, employing a data-driven approach to establish the relationship between input margin features and available transmission capacity, thereby improving the calculation speed and accuracy of available transmission capacity. To address the issue of the unknowable decision-making process in existing AI-based available transmission capacity calculation methods, the decision paths and key features in the decision tree model are defined as visualization rules, thus ensuring the visualization of the calculation process. A framework for available transmission capacity calculation is proposed, and through offline modeling and online application, the requirements for online computation are met, making the method feasible for practical application. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. As shown in the figures:

[0020] Figure 1 This invention presents a framework for a decision tree-based method to calculate available transmission capacity.

[0021] Figure 2 This is a flowchart illustrating the simulation process for calculating the sample set in this invention.

[0022] Figure 3 This is a schematic diagram of the partial decision tree model structure constructed for this invention.

[0023] Figure 4 This is a partitioned topology diagram of the IEEE-39 node example used in an embodiment of the present invention.

[0024] Figure 5 This represents a partial decision path in the decision tree model constructed in this invention.

[0025] Figure 6 These are the top five features in the decision tree model constructed in this invention. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] Example 1: As Figure 1 , Figure 2 , Figure 3 , Figure 4 , Figure 5 and Figure 6 As shown, a method for calculating available transmission capacity based on decision trees is proposed. This method applies artificial intelligence data-driven steps to improve calculation speed and accuracy, and uses a decision tree algorithm to facilitate visualization of the calculation process.

[0028] A method for calculating available transmission capacity based on decision trees includes offline steps and online steps.

[0029] The offline process involves collecting historical or simulation data to establish a training sample set with electrical capacity margin and scheduling mode as input features and available transmission capacity as the target variable.

[0030] The ATC calculation model consists of a decision tree model and visualization rules. The decision tree algorithm learns the input feature information from the training sample set to obtain the decision tree model, which is used to calculate the ATC value. The visualization rules are the decision paths and node feature information in the decision tree model, and feature importance is evaluated by calculating feature weights.

[0031] The online process transforms the online running data into the input feature form of the ATC calculation model, inputs it into the established ATC calculation model, and the decision tree model provides the ATC calculation value corresponding to the current state and provides a visualization path of the current calculation process to obtain key features.

[0032] The available transmission capacity calculation samples of this invention are generated in batches by a power system simulator (Power System Simulator / Engineering, PSS / E) driven by Python language. Based on the 8760-hour load curve, the system operates in 8760 states, records the electrical quantity margin of each state, and calculates the available transmission capacity of each state using the continuous power flow method.

[0033] The input features are electrical margin and scheduling mode, and the target variable is available transmission capacity. To ensure consistency in data units, the input data needs to be normalized, meaning that data of the same type are represented by values ​​less than 1. Input Feature Set for:

[0034] (1)

[0035] In the formula, The scheduling direction vector consists of the generators and load unit adjustment power that need to be scheduled during the ATC calculation process; This is the generator active power margin vector, which consists of the active power output margins of each generator connected to the sending end of the current transmission channel; This is the generator reactive power margin vector, which consists of the reactive power output margins of each generator at the sending end connected to the current transmission channel; This is the node voltage margin vector, which consists of the voltage magnitude margins of each node at the sending and receiving ends connected to the current transmission channel. This is the node power angle margin vector, which consists of the voltage phase angle margins of each node at the sending and receiving ends connected to the current transmission channel; It is the active power margin vector of the transmission line, which consists of the active power margins of each transmission line at the sending and receiving ends connected by the transmission channel;

[0036] Scheduling direction vector The expression is:

[0037] (2)

[0038] In the formula, For generator The unit regulating power; For load The unit growth power.

[0039] Generator active power margin vector The expression is:

[0040] (3)

[0041] In the formula, For generator The effective margin; For generator The maximum active power output; For generator Initial value of active power output.

[0042] Generator reactive margin vector The expression is:

[0043] (4)

[0044] In the formula, For generator reactive power margin; For generator The maximum reactive power output; For generator Initial value of reactive power output.

[0045] Node voltage margin vector The expression is:

[0046] (5)

[0047] In the formula, busbar Voltage margin; busbar The initial voltage value. As the power flow in the transmission channel increases, the system bus voltage will show a decreasing trend, and the lower limit of the per-unit value of the voltage amplitude constraint is 0.9 pu.

[0048] Nodal work angle margin vector The expression is:

[0049] (6)

[0050] In the formula, busbar Phase margin; busbar The initial value of the phase angle. As the power flow in the transmission channel increases, the phase angle of the busbars at both ends of the system line will gradually increase, and the limit of the busbar phase angle is not more than 180 degrees.

[0051] Line active power margin vector The expression is:

[0052] (7)

[0053] In the formula, For the line The effective margin; For the line The maximum active power; For the line The initial value of active power.

[0054] Target variable Available transmission capacity Treating it as continuous data, its expression is:

[0055] (8)

[0056] Decision tree algorithms fall under the category of supervised learning in machine learning. They utilize training data with input features, selecting sample features as the root and internal nodes of the decision tree, and iteratively generating multiple trees. In a decision tree model, the root and internal nodes contain the input features, each branch represents the decision conditions, and the leaf nodes represent the gain of the corresponding path. Each leaf node corresponds to a decision path, and the final calculation result is the value of the decision tree gain.

[0057] By learning from the decision results of different feature groups, the overall computational performance of the model is gradually improved. The optimization objective for generating the decision tree is:

[0058] (9)

[0059] (10)

[0060] In the formula, is the loss function; in this invention, we choose the root mean square error (MSE); n is the total number of samples in this tree. It is the number of samples in the left subtree; Probability of choosing the left subtree; Probability of choosing the right subtree;

[0061] When selecting the optimal feature for node splitting, the Classification and Regression Tree (CART) algorithm is used to calculate the gain of the feature on node splitting. The feature that produces the largest gain is selected as the optimal feature. The calculation formula is:

[0062] (13)

[0063] In the formula, , These represent the left and right subtrees after the split, respectively. This indicates the node after the split, which is the left leaf node. This indicates the node that reaches the right leaf node after the split. Represents the left subtree ; Represents the right subtree ; Represents the left subtree ; Represents the right subtree .

[0064] A method for calculating available transmission capacity based on decision trees is proposed, wherein the decision tree algorithm is used to learn the mapping rules between input features and ATC (Automatic Transmission Capacity), and a decision tree model and visualization rules are established. The model establishment includes the following steps:

[0065] Step 1: Transform the online running data into the form of the input feature sample set given by equations (1)-(8).

[0066] Step 2: Initialize the decision tree model parameters, including: maximum tree depth, loss function, solution method, minimum weight of leaf nodes, and initialize the initial ATC values. .

[0067] Step 3: Build a decision tree using the input feature sample set.

[0068] The steps to build a decision tree:

[0069] Step 1) Input the feature space consisting of the sample set, and traverse the features. Utilizing features Scanning cut points The goal is to minimize the difference between the output values ​​of the two segmented feature subsets and the target value, and to calculate the feature gain to select the optimal splitting feature. and split point Choose the optimal combination that satisfies the objective. .

[0070] Step 2) Utilize the optimal Combining partitioned subset space , And calculate the corresponding subset output value.

[0071] Step 3) Repeat steps 1) and 2) to... , The two subsets are further divided. When the stopping condition for subset division is met, the score of the optimal leaf node is calculated. This process continues until the stopping condition for all subsets is met.

[0072] Step 4: Determine whether the depth of the currently established decision tree has reached the set value. If it has not reached the set value, repeat step 3; if it has reached the set value, output the gain of the established decision tree as the ATC calculation value of the decision tree model.

[0073] Example 2: Figure 1 , Figure 2 , Figure 3 , Figure 4 , Figure 5 and Figure 6 As shown, a method for calculating available transmission capacity based on decision trees is proposed. This method applies the concept of artificial intelligence data-driven thinking to improve calculation speed and accuracy, and uses a decision tree algorithm to make the calculation process more visual.

[0074] The method was applied to an embodiment of the IEEE-9 node computational example to verify its effectiveness. Each bus node is numbered, with G at the node indicating a generator node. Line segment connections between buses represent the existence of a transmission channel. The IEEE-9 node computational example was divided into three regions: Region 1, Region 2, and Region 3.

[0075] Figure 4 This section describes the partitioning of the IEEE-9 node example. It calculates the available transmission capacity from region 3 to region 2. The sample set contains 43 input features and 8760 samples. The initial decision tree model parameters are set as follows: maximum tree depth 385, loss function is the mean squared error function, solution method is tree model solving, and minimum leaf node weight is 10.

[0076] Generator node 1, generator node 2, generator node 3, transformer node 4, load node 5, transformer node 6, load node 7, transformer node 8, and load node 9. In region 1, generator node 1 is connected to transformer node 4. Transformer node 4 is connected to load node 5 and load node 9 respectively. Load node 5 is connected to transformer node 6 in region 3. Transformer node 6 is connected to load node 7 and generator node 3 respectively. Load node 9 is connected to transformer node 8. In region 2, transformer node 8 is connected to generator node 2 and load node 7 respectively.

[0077] The example uses the IEEE-9 three-machine nine-node example. Figure 4 The nodes directly connected to G represent generator nodes, which are PV nodes and can be used as power selling nodes. The remaining nodes are PQ nodes without generators. Load nodes can increase load to purchase electricity. This example is divided into three regions, each containing one generator node and two non-generator nodes.

[0078] The path from the root node to any leaf node constitutes a decision tree. Figure 4 This refers to a portion of the decision path in the decision tree model when inputting samples 1 and 2 from the sample set. When the decision tree model calculates the ATC, it sequentially traverses a decision tree path based on the current input features. Each path generates a corresponding ATC decision value.

[0079] Depend on Figure 4 It can be seen that when sample 1 in the input sample set passes through a decision path, it satisfies the following conditions: the active power margin of generator node 3 is greater than 209.472MW, the active power margin of the node 6-node 7 line is less than or equal to 109.818MW, the voltage of node 8 is greater than 110.867V, and the path finally generates a calculated gain of 38 at the leaf node.

[0080] When inputting sample 2, the active power margin of generator node 3 is greater than 206.472MW, the reactive power margin of bus node 3 is less than or equal to 31.364MW, the bus voltage of node 7 is greater than 344.281, and the path finally generates a calculated gain of 7 at the leaf node.

[0081] The decision path of a decision tree can provide a visual representation of the rules and computation process for model calculation, and features with higher weights can be used as key features.

[0082] Figure 5 These are the top five features with the highest weights in the decision tree model constructed using this invention on the IEEE-9 node example. Feature weights are the number of times a feature appears in a decision tree node; the more times it appears, the more decision paths are constructed using the feature, the better the feature's decision-making effect, and the more critical the feature. Figure 5As can be seen, the top five key characteristics include line active power margin, generator active and reactive power margin, and bus phase angle margin. These key characteristics reflect important factors affecting ATC. During operation, it is necessary to pay close attention to these variables to ensure that they are within the safety margin range.

[0083] In this embodiment of the invention, the average computation time of the ATC calculation method for a single sample is 230 milliseconds, and the average computation accuracy of the test set is 95.32%. This indicates that the method uses the constructed model to evaluate the computation time, which is in the millisecond range, meeting the requirements for online computation. The computation accuracy is higher than 95%. Since ATC takes into account a certain safety margin when it is defined, this result can guarantee a certain level of security.

[0084] The steps for constructing the ATC computational model, including the overall optimization objective, feature gain calculation formula, and optimal leaf node score expression for the decision tree model, are as follows:

[0085] Decision trees generate new decision trees by fitting the residuals of the previous calculation results, thereby gradually improving the overall computational performance of the model. The optimization objective for generating new decision trees is:

[0086]

[0087]

[0088] In the formula, is the loss function; in this invention, we choose the root mean square error (MSE); n is the total number of samples in this tree. It is the number of samples in the left subtree; Probability of choosing the left subtree; Choose the probability for the right subtree.

[0089] When selecting the optimal feature for node splitting, the Classification and Regression Tree (CART) algorithm is used to calculate the gain of the feature on node splitting. The feature that produces the largest gain is selected as the optimal feature. The calculation formula is:

[0090]

[0091] In the formula, , These represent the left and right subtrees after the split, respectively. This indicates the node after the split, which is the left leaf node. This indicates the node that reaches the right leaf node after the split. Represents the left subtree ; Represents the right subtree ; Represents the left subtree ; Represents the right subtree .

[0092] Example 3: As Figure 1 , Figure 2 , Figure 3 , Figure 4 , Figure 5 and Figure 6 As shown, a method for calculating available transmission capacity based on decision trees is presented. This method applies artificial intelligence methods to improve calculation speed and accuracy, and uses a limit gradient boosting algorithm based on decision tree concepts to facilitate visualization of the calculation process.

[0093] A decision tree-based method for calculating available transmission capacity treats the calculation of available transmission capacity as a regression problem in machine learning. It adopts a data-driven approach and uses existing operational or simulation data to construct a training sample set consisting of electrical capacity margin, scheduling mode, and available transmission capacity.

[0094] The mapping rules between input features (electrical capacity margin, scheduling mode) and target variable (available transmission capacity) in the training sample set are then learned to construct a decision tree model. When a new operating state requires calculation of available transmission capacity, the operating state data is simply converted into input features for the decision tree model, which then calculates the corresponding available transmission capacity. The decision tree model iteratively builds multiple decision trees based on the mapping rules between input features and available transmission capacity. Finally, the results of each decision tree are summed to obtain the available transmission capacity. Therefore, it can provide a visualized decision-making process and give corresponding feature weights to help operators understand the model's decision-making process.

[0095] The overall framework of the available transmission capacity calculation method based on the limiting gradient boosting algorithm is divided into two parts: an offline stage and an online stage. The offline stage collects historical data or simulation operation data to establish a training sample set with electrical capacity margin and scheduling mode as input features, and available transmission capacity as the target variable. The available transmission capacity calculation model consists of a decision tree model and visualization rules. The decision tree model is obtained by learning the feature information from the training sample set and is used to calculate the available transmission capacity value. The visualization rules are the decision paths and node feature information in the decision tree model, and the importance of features is evaluated by calculating feature weights.

[0096] In the online phase, the online operation data is converted into the input feature form of the available transmission capacity calculation model and input into the established available transmission capacity calculation model. The decision tree model provides the calculated value of available transmission capacity corresponding to the current state and provides a visualization path of the current calculation process to obtain key features.

[0097] The available transmission capacity calculation samples of this invention are generated in batches by a power system simulator (Power System Simulator / Engineering, PSS / E) driven by Python. Based on the 8760-hour load curve, 8760 operating states of the system are generated, the electrical quantity margin of each state is recorded, and the available transmission capacity of each state is calculated using the continuous power flow method. Figure 2 The method for generating computational sample simulations is demonstrated.

[0098] Decision trees are a type of supervised learning algorithm frequently used in classification and regression analysis. They simulate the human decision-making process by classifying or predicting data through a series of conditional judgments. Decision trees are intuitive and easy to interpret, making them suitable for handling classification and regression problems. They can also be used to handle nonlinear relationships and multi-class problems. The construction process of a decision tree includes feature selection, tree generation, and tree pruning. Feature selection involves choosing the optimal attribute as a node; common methods include information gain, information gain ratio, and the Gini index. Tree generation involves recursively building the decision tree until a stopping condition is met.

[0099] Decision trees learn from the decision results of different feature groups, gradually improving the overall computational performance of the model. The optimization objective for generating decision trees is:

[0100]

[0101]

[0102] In the formula, is the loss function; in this invention, we choose the root mean square error (MSE); n is the total number of samples in this tree. It is the number of samples in the left subtree; Probability of choosing the left subtree; Choose the probability for the right subtree.

[0103] In summary, the decision tree-based ATC calculation method provided by this invention can meet the requirements of calculation speed and accuracy by applying the data-driven approach of artificial intelligence. The algorithm based on the decision tree approach can provide decision paths and key features, while also taking into account the visualization of the calculation process.

Claims

1. A method for calculating available transmission capacity based on decision trees, characterized in that, This includes offline and online steps. The offline process involves collecting historical or simulation data to establish a training sample set with electrical capacity margin and scheduling mode as input features, and available transmission capacity as the target variable. The ATC calculation model consists of a decision tree model and visualization rules. The decision tree algorithm learns from the input features in the training sample set to obtain the decision tree model, which is used to calculate the ATC value. The visualization rules are the decision paths and node feature information in the decision tree model; feature weights are calculated to evaluate feature importance. The online process transforms the online runtime data into the input feature form of the ATC calculation model, inputs it into the established ATC calculation model, and the decision tree model provides the ATC calculation value corresponding to the current state, as well as a visualization path of the current calculation process, thus obtaining key features. The input features are electrical margin and scheduling mode, and the target variable is available transmission capacity. To ensure consistency of data units, the input data needs to be normalized, meaning that data of the same type are represented by values ​​less than 1. (Input feature set) for: （1） In the formula, The scheduling direction vector consists of the generators and load unit adjustment power that need to be scheduled during the ATC calculation process; This is the generator active power margin vector, which consists of the active power output margins of each generator connected to the sending end of the current transmission channel; This is the generator reactive power margin vector, which consists of the reactive power output margins of each generator at the sending end connected to the current transmission channel; This is the node voltage margin vector, which consists of the voltage magnitude margins of each node at the sending and receiving ends connected to the current transmission channel. This is the node power angle margin vector, which consists of the voltage phase angle margins of each node at the sending and receiving ends connected to the current transmission channel; It is the active power margin vector of the transmission line, which consists of the active power margins of each transmission line at the sending and receiving ends connected by the transmission channel; Scheduling direction vector The expression is: （2） In the formula, For generator The unit regulating power; For load unit growth power, Generator active power margin vector The expression is: （3） In the formula, For generator The effective margin; For generator The maximum active power output; For generator Initial value of active power output Generator reactive margin vector The expression is: （4） In the formula, For generator reactive power margin; For generator The maximum reactive power output; For generator Initial value of reactive power output Node voltage margin vector The expression is: （5） In the formula, busbar Voltage margin; busbar The initial voltage value will decrease as the power flow in the transmission channel increases. The lower limit of the per-unit value for voltage amplitude constraint is 0.9 pu. Nodal work angle margin vector The expression is: （6） In the formula, busbar Phase margin; busbar The initial phase angle is given. As the power flow in the transmission channel increases, the phase angle of the busbars at both ends of the system line will gradually increase. The limit for the busbar phase angle is not to exceed 180 degrees. Line active power margin vector The expression is: （7） In the formula, For the line The effective margin; For the line The maximum active power; For the line The initial value of active power, Target variable Available transmission capacity Treating it as continuous data, its expression is: （8）, The decision tree algorithm is used to learn the mapping rules between input features and the ATC calculation model, and a decision tree model and visualization rules are established. The model establishment includes the following steps: Step 1: Transform the online running data into the form of the input feature sample set given by expressions (1) to (8). Step 2: Initialize the decision tree model parameters, including: maximum tree depth, loss function, solution method, minimum weight of leaf nodes, and initialize the initial ATC values. , Step 3: Build a decision tree using the input feature sample set.

2. The method for calculating available transmission capacity based on decision trees according to claim 1, characterized in that, The available transmission capacity calculation samples are generated in batches by a power system simulator (PSS / E) driven by Python. The system operating status is generated based on the 8760-hour load curve, the electrical quantity margin of each status is recorded, and the available transmission capacity of each status is calculated using the continuous power flow method.

3. The method for calculating available transmission capacity based on decision trees according to claim 1, characterized in that, Decision tree algorithms belong to the category of supervised learning in machine learning. They use data with input features for training, select the features of the samples as the root node and internal nodes of the decision tree, and continuously iterate to generate multiple trees. In the decision tree model, the root node and internal nodes are the input features, each branch is the decision condition, and the leaf nodes are the gain of the corresponding path. Each leaf node corresponds to a decision path, and the final calculation result is the value of the decision tree gain.

Images