Electric power-traffic network optimal scheduling method based on graph attention agent model

Abstract

The invention discloses an electric power-traffic network optimal scheduling method based on a graph attention agent model, and the method comprises the steps: constructing a second-order cone DistFlow optimal power flow model of a power distribution network by taking the electricity purchase quantity and the output of a distributed power supply as control variables; on the basis of the semi-dynamic traffic distribution model, the flexibility capability of the charging demand of the electric vehicle is brought into play through the dynamic charging service fee, and a semi-dynamic traffic distribution model considering dynamic charging service fee scheduling is constructed; on the basis of a multi-scene solving result, traffic network travel demands, traffic network basic parameters and charging service charges are used as inputs, multi-period traffic flows of all charging stations are used as outputs, a graph-attention agent model is trained, nonlinear mapping of charging prices and charging flows is learned, and the nonlinear mapping is used for rapid scheduling. According to the method, the user balance response can be approximately obtained in a multi-period and uncertain scene, and the charging load and the traffic flow are optimized.

Description

Technical Field

[0001] This invention relates to the field of power-transportation network collaborative optimization technology, and in particular to an optimal scheduling method for power-transportation networks based on a graph attention proxy model. Background Technology

[0002] With the increasing number of electric vehicles and the widespread deployment of charging stations, the coupling between transportation networks and power distribution networks is deepening. Especially when the charging demand of electric vehicles and the demand for traditional electricity loads reach their peak simultaneously, uncoordinated charging behavior puts significant pressure on the operation of the power distribution network (PDN). The essence of power-transport coupling lies in understanding the response of the spatiotemporal distribution of charging load to the dynamic pricing of charging stations. Therefore, capturing this price response relationship is key to achieving cross-network coordinated scheduling.

[0003] However, accurately modeling vehicle behavior and charging load distribution remains a challenge due to the impact of scheduling and inherent uncertainties. Currently, most traditional solver-based methods for solving coupled electric-transportation systems have limited computational efficiency and poor adaptability to dynamic environments across various scenarios. Furthermore, with the application of deep learning, most methods focus on unilateral sensory predictions from the distribution network or transportation network side, neglecting the role of scheduling mechanisms in electric vehicle charging behavior. This is insufficient to meet the requirements of optimal scheduling in coupled electric transportation systems. In addition, the involvement of multiple stakeholders in the power and transportation sectors raises concerns about information privacy. Therefore, there is an urgent need to develop an optimal scheduling method for electric-transportation networks based on a graph attention proxy model to learn the price response of electric vehicle charging demand. Summary of the Invention

[0004] The purpose of this invention is to provide an optimal scheduling method for power-transportation networks based on a graph attention proxy model. This invention considers the coordination of the power-transportation coupled system, establishes a semi-dynamic traffic flow model that takes into account dynamic charging service fee scheduling, and achieves dual-network coupling with the distribution network through charging stations. By inputting the optimization results of the coupled system under multiple scenarios into the graph attention proxy model, the relationship between electric vehicle charging behavior, scheduling methods, and the environment is learned. This invention not only learns the price response of electric vehicle charging demand but also considers the impact of scheduling costs on the charging behavior of electric vehicle users, ensuring the coordinated operation of the power-transportation network.

[0005] Technical Solution: To achieve the above-mentioned objectives, this invention proposes an optimal scheduling method for power-transportation networks based on a graph attention proxy model. This method includes the following steps:

[0006] Step 1: Obtain the distribution network parameters and operating coefficients. The network parameters include the distribution network topology and line impedance, and the operating coefficients include the upstream power grid power supply coefficient and the distributed generation coefficient.

[0007] Step 2: Obtain scenario data on load demand at each node of the power grid, distributed photovoltaic power output, and time-of-use pricing. Using the power purchase by the upper-level power grid and the power generation of distributed sources as control variables and the operation constraints of the distribution network as constraints, establish a second-order cone DistFlow optimal power flow model for the distribution network.

[0008] Step 3: Obtain the network parameters of the transportation network, including the network topology, capacity of each road segment, location and capacity of charging stations;

[0009] Step 4: Obtain scenario data on travel demand and electric vehicle penetration rate in the transportation network. Using charging service fees as the control variable and basic operational constraints of the transportation network as the constraint conditions, establish a semi-dynamic traffic allocation model that considers dynamic charging service fee scheduling.

[0010] Step 5: Based on the second-order cone DistFlow optimal power flow model of the distribution network in Step 2 and the semi-dynamic traffic assignment model considering dynamic charging service fee scheduling in Step 4, the distribution network and the transportation network are coupled through charging stations, and a cost model of the distribution-transportation network coupled system is constructed.

[0011] Step 6: Based on the cost model of the power distribution-transportation network coupled system obtained in Step 5 under multiple scenarios, optimize the scheduling of the coupled system to obtain the charging station traffic flow. Take the transportation network travel demand, basic parameters of the transportation network, and charging service fee as inputs, and the charging station traffic flow as output. Add a bidirectional cross-attention mechanism on the basis of graph convolutional network and Transformer network to establish a proxy model based on graph-attention network for training and learning.

[0012] Step 7: Collect network parameters of the transportation network and real-time travel demand of the transportation network, and input the charging service fee into the agent model obtained in Step 6 to solve the problem, and output the traffic flow of the charging station to obtain the optimal scheduling method of the power-transportation network based on the graph-attention agent model.

[0013] Furthermore, in step 2, the optimal power flow model for the second-order cone DistFlow distribution network is as follows:

[0014] (A-1)

[0015] (A-2)

[0016] (A-3)

[0017] (A-4)

[0018] (A-5)

[0019] (A-6)

[0020] (A-7)

[0021] (A-8)

[0022] (A-9)

[0023] in, and They represent Timetable Active power and reactive power; and They represent Time Node The active and reactive power outputs of distributed power sources; , , They represent the lines respectively. Resistance, reactance, and impedance; express Timetable The current; Represents nodes Connected child nodes The set that constitutes; and They represent Timetable Active power and reactive power; and They represent Time Node The active and reactive loads; express Time Node Square of voltage amplitude at point; express Time Node Square of voltage amplitude at point; Indicates the line The upper limit of the current amplitude; and Representing nodes respectively The lower and upper limits of the square of the voltage amplitude; express Time Node The conventional power load at the location; , and , They are nodes The lower and upper limits of the active and reactive power output of the generator; For nodes The generator's ramp-climbing limitations; Represents nodes Connected charging sections gather; express Charging stations Traffic flow; This represents the average charging requirement per electric vehicle. Represents a set of time periods; Represents nodes A collection of connected generators; Indicates the economic cost of the power distribution network; The unit price of electricity for The power distribution network constantly purchases electricity from the superior power grid. and For nodes The energy production cost coefficient.

[0024] Furthermore, in step 4, the semi-dynamic traffic assignment model considering dynamic charging service fee scheduling is as follows:

[0025] (A-10)

[0026] (A-11)

[0027] (A-12)

[0028] (A-13)

[0029] (A-14)

[0030] (A-15)

[0031] (A-16)

[0032] (A-17)

[0033] (A-18)

[0034] (A-19)

[0035] (A-20)

[0036] in, Let represent the set of paths k for electric vehicles with starting point r and destination s; rs represents a starting point-destination pair. Represents the set of origin and destination points for transportation demand; This represents the traffic flow of electric vehicles choosing path k; , These represent the sets of regular road sections and charging road sections, respectively. , These represent traffic flow on regular road sections and charging road sections, respectively. , These represent the number of vehicles that can be accommodated on regular road sections and charging road sections, respectively. This represents a binary constant used to determine whether the road segment 'a' where the electric vehicle is located belongs to path 'k'. If it does, the value is 1; otherwise, it is 0. Indicates travel time for a road segment; , represents the free passage time under zero traffic flow conditions on regular road sections and charging road sections, respectively; J is the charging time model parameter; The travel time on path k; Let be the initial traffic demand for an electric vehicle to travel from the starting point r to the destination s at time t; Let t be the remaining flow rate of the electric vehicle between OD and rs; Indicates the duration of a time period; Let t be the corrected travel demand for an electric vehicle traveling from the starting point r to the destination s at time t. This represents the cost per unit of travel time. Indicates the unit price of charging; , These represent congestion pricing and charging service fees, respectively. This represents the average charging requirement per electric vehicle. This represents the toll cost that an electric vehicle with a starting point r and a destination s needs to pay when choosing route k. This represents the minimum travel cost for an electric vehicle with a starting point r and a destination s. This represents the cost of travel for users of the transportation network.

[0037] Furthermore, the cost model for the power distribution-transportation network coupled system considering dynamic charging service fee scheduling in step 5 is as follows:

[0038] (A-21)

[0039] in, This represents the total cost of the coupled system.

[0040] Furthermore, in step 6, the graph-attention proxy model learns the relationship between charging flow, charging service fees, and the traffic environment to achieve optimal scheduling of the power-transportation network. Its proxy model module includes the following parts:

[0041] (1) Graph Convolutional Network Module:

[0042] (A-22)

[0043] (A-23)

[0044] (A-24)

[0045] In the formula, Traffic sections and road sections The connection between them, where adjacent represents a road segment. and road sections There is a connection relationship; Indicates the first The feature matrix of the layer; yes The set represents the connection structure of the entire graph; It is the identity matrix; Indicates the relationship between a node and itself; It is a degree matrix, representing the number of edges terminating at each node, with the diagonal elements being the degree of the nodes; Indicates the first Layer weight matrix; ReLU is a non-linear activation function.

[0046] (2) Bidirectional cross-attention module:

[0047] (A-25)

[0048] (A-26)

[0049] (A-27)

[0050] (A-28)

[0051] (A-29)

[0052] In the formula, This represents the node feature matrix output by the graph convolutional network; Represents the global feature matrix; This represents the attention value of the node feature matrix to the global feature matrix; This represents the attention value of the global feature matrix to the node feature matrix; This represents the exponential normalization function; , , They represent The corresponding weight matrix of queries, keys, and values; , , express The corresponding weight matrix of queries, keys, and values; Indicates the dimension size of the key; This represents the node feature matrix updated after the attention operation; This represents the updated global feature matrix after the attention operation; Presentation layer normalization operation; Indicates the regularization method; This indicates matrix concatenation at the node level; This represents the feature matrix obtained by concatenating the node feature matrix and the global feature matrix, and is used as the input to the Transformer module;

[0053] (3) Transformer module:

[0054] (A-30)

[0055] (A-31)

[0056] (A-32)

[0057] (A-33)

[0058] (A-34)

[0059] (A-35)

[0060] In the formula, , and These represent the query vector, key vector, and value vector, respectively. , , They represent the first A person's attention , and The corresponding weight matrix; This represents the attention mechanism of computational standards; Indicates the first One point of attention; For the number of attention heads; Indicates multi-headed attention; This represents the output weights of the multi-head attention concatenation; This represents the input feature matrix after learnable positional encoding; Indicates positional information within the sequence; This represents a feedforward neural network; This is the normalized output vector of the attention layer; and This is the weight vector; and For bias terms; It is a fully connected layer; This is the fused feature vector processed by the encoder; The output is the traffic flow at charging stations after optimal scheduling of the power-transportation network coupled system.

[0061] Beneficial effects: Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0062] Compared with traditional electric vehicle (EV) optimization scheduling schemes in power distribution-traffic coupled systems, the technical solution of this invention, based on semi-dynamic traffic flow, leverages the flexibility of EVs to guide charging load distribution through dynamic charging service fees. It also trains a graph-attention proxy model to learn the nonlinear relationship between charging prices and EV charging response, which helps quantify the flexibility of EV charging demand and facilitates vehicle scheduling based on the needs of the coupled system. Case study results demonstrate that the proposed method can utilize the flexibility of EVs, helping to alleviate traffic congestion and improve the operational safety of the power distribution network (PDN). Attached Figure Description

[0063] Figure 1 This is a flowchart of the method of the present invention;

[0064] Figure 2 This is a topology diagram of a power distribution-traffic coupling system;

[0065] Figure 3 It is the accuracy error of the traffic flow at charging stations learned by different deep learning models;

[0066] Figure 4 It refers to the charging load distribution of charging stations under the fixed charging service fee and dynamic charging service fee scheduling methods. Detailed Implementation

[0067] The present invention will be further illustrated below with reference to the accompanying drawings and specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. After reading this invention, any modifications of the invention in various equivalent forms by those skilled in the art will fall within the scope defined by the appended claims.

[0068] like Figure 1 As shown, this invention proposes an optimal scheduling method for power-transportation networks based on a graph attention proxy model. This method includes the following steps:

[0069] Step 1: Obtain the distribution network parameters and operating coefficients. The network parameters include the distribution network topology and line impedance, and the operating coefficients include the upstream power grid power supply coefficient and the distributed generation coefficient.

[0070] Step 2: Obtain scenario data on load demand at each node of the power grid, distributed photovoltaic power output, and time-of-use pricing. Using the power purchase by the upper-level power grid and the power generation of distributed sources as control variables and the operation constraints of the distribution network as constraints, establish a second-order cone DistFlow optimal power flow model for the distribution network.

[0071] Step 3: Obtain the network parameters of the transportation network, including the network topology, capacity of each road segment, location and capacity of charging stations;

[0072] Step 4: Obtain scenario data on travel demand and electric vehicle penetration rate in the transportation network. Using charging service fees as the control variable and basic operational constraints of the transportation network as the constraint conditions, establish a semi-dynamic traffic allocation model that considers dynamic charging service fee scheduling.

[0073] Step 5: Based on the second-order cone DistFlow optimal power flow model of the distribution network in Step 2 and the semi-dynamic traffic assignment model considering dynamic charging service fee scheduling in Step 4, the distribution network and the transportation network are coupled through charging stations, and a cost model of the distribution-transportation network coupled system is constructed.

[0074] Step 6: Based on the cost model of the power distribution-transportation network coupled system obtained in Step 5 under multiple scenarios, optimize the scheduling of the coupled system to obtain the charging station traffic flow. Take the transportation network travel demand, basic parameters of the transportation network, and charging service fee as inputs, and the charging station traffic flow as output. Add a bidirectional cross-attention mechanism on the basis of graph convolutional network and Transformer network to establish a proxy model based on graph-attention network for training and learning.

[0075] Step 7: Collect network parameters of the transportation network and real-time travel demand of the transportation network, and input the charging service fee into the agent model obtained in Step 6 to solve the problem, and output the traffic flow of the charging station to obtain the optimal scheduling method of the power-transportation network based on the graph-attention agent model.

[0076] Furthermore, in step 2, the optimal power flow model for the second-order cone DistFlow distribution network is as follows:

[0077] (A-1)

[0078] (A-2)

[0079] (A-3)

[0080] (A-4)

[0081] (A-5)

[0082] (A-6)

[0083] (A-7)

[0084] (A-8)

[0085] (A-9)

[0086] in, and They represent Timetable Active power and reactive power; and They represent Time Node The active and reactive power outputs of distributed power sources; , , They represent the lines respectively. Resistance, reactance, and impedance; express Timetable The current; Represents nodes Connected child nodes The set that constitutes; and They represent Timetable Active power and reactive power; and They represent Time Node The active and reactive loads; express Time Node Square of voltage amplitude at point; express Time Node Square of voltage amplitude at point; Indicates the line The upper limit of the current amplitude; and Representing nodes respectively The lower and upper limits of the square of the voltage amplitude; express Time Node The conventional power load at the location; , and , They are nodes The lower and upper limits of the active and reactive power output of the generator; For nodes The generator's ramp-climbing limitations; Represents nodes Connected charging sections gather; express Charging stations Traffic flow; This represents the average charging requirement per electric vehicle. Represents a set of time periods; Represents nodes A collection of connected generators; Indicates the economic cost of the power distribution network; The unit price of electricity for The power distribution network constantly purchases electricity from the superior power grid. and For nodes The energy production cost coefficient.

[0087] Furthermore, in step 4, the semi-dynamic traffic assignment model considering dynamic charging service fee scheduling is as follows:

[0088] (A-10)

[0089] (A-11)

[0090] (A-12)

[0091] (A-13)

[0092] (A-14)

[0093] (A-15)

[0094] (A-16)

[0095] (A-17)

[0096] (A-18)

[0097] (A-19)

[0098] (A-20)

[0099] in, Let represent the set of paths k for electric vehicles with starting point r and destination s; rs represents a starting point-destination pair. Represents the set of origin and destination points for transportation demand; This represents the traffic flow of electric vehicles choosing path k; , These represent the sets of regular road sections and charging road sections, respectively. , These represent traffic flow on regular road sections and charging road sections, respectively. , These represent the number of vehicles that can be accommodated on regular road sections and charging road sections, respectively. This represents a binary constant used to determine whether the road segment 'a' where the electric vehicle is located belongs to path 'k'. If it does, the value is 1; otherwise, it is 0. Indicates travel time for a road segment; , represents the free passage time under zero traffic flow conditions on regular road sections and charging road sections, respectively; J is the charging time model parameter; The travel time on path k; Let be the initial traffic demand for an electric vehicle to travel from the starting point r to the destination s at time t; Let t be the remaining flow rate of the electric vehicle between OD and rs; Indicates the duration of a time period; Let t be the corrected travel demand for an electric vehicle traveling from the starting point r to the destination s at time t. This represents the cost per unit of travel time. Indicates the unit price of charging; , These represent congestion pricing and charging service fees, respectively. This represents the average charging requirement per electric vehicle. This represents the toll cost that an electric vehicle with a starting point r and a destination s needs to pay when choosing route k. This represents the minimum travel cost for an electric vehicle with a starting point r and a destination s. This represents the cost of travel for users of the transportation network.

[0100] Furthermore, the cost model for the power distribution-transportation network coupled system considering dynamic charging service fee scheduling in step 5 is as follows:

[0101] (A-21)

[0102] in, This represents the total cost of the coupled system.

[0103] Furthermore, in step 6, the graph-attention proxy model learns the relationship between charging flow, charging service fees, and the traffic environment to achieve optimal scheduling of the power-transportation network. Its proxy model module includes the following parts:

[0104] (1) Graph Convolutional Network Module:

[0105] (A-22)

[0106] (A-23)

[0107] (A-24)

[0108] In the formula, Traffic sections and road sections The connection between them, where adjacent represents a road segment. and road sections There is a connection relationship; Indicates the first The feature matrix of the layer; yes The set represents the connection structure of the entire graph; It is the identity matrix; Indicates the relationship between a node and itself; It is a degree matrix, representing the number of edges terminating at each node, with the diagonal elements being the degree of the nodes; Indicates the first Layer weight matrix; ReLU is a non-linear activation function.

[0109] (2) Bidirectional cross-attention module:

[0110] (A-25)

[0111] (A-26)

[0112] (A-27)

[0113] (A-28)

[0114] (A-29)

[0115] In the formula, This represents the node feature matrix output by the graph convolutional network; Represents the global feature matrix; This represents the attention value of the node feature matrix to the global feature matrix; This represents the attention value of the global feature matrix to the node feature matrix; This represents the exponential normalization function; , , They represent The corresponding weight matrix of queries, keys, and values; , , express The corresponding weight matrix of queries, keys, and values; Indicates the dimension size of the key; This represents the node feature matrix updated after the attention operation; This represents the updated global feature matrix after the attention operation; Presentation layer normalization operation; Indicates the regularization method; This indicates matrix concatenation at the node level; This represents the feature matrix obtained by concatenating the node feature matrix and the global feature matrix, and is used as the input to the Transformer module;

[0116] (3) Transformer module:

[0117] (A-30)

[0118] (A-31)

[0119] (A-32)

[0120] (A-33)

[0121] (A-34)

[0122] (A-35)

[0123] In the formula, , and These represent the query vector, key vector, and value vector, respectively. , , They represent the first A person's attention , and The corresponding weight matrix; This represents the attention mechanism of computational standards; Indicates the first One point of attention; For the number of attention heads; Indicates multi-headed attention; This represents the output weights of the multi-head attention concatenation; This represents the input feature matrix after learnable positional encoding; Indicates positional information within the sequence; This represents a feedforward neural network; This is the normalized output vector of the attention layer; and This is the weight vector; and For bias terms; It is a fully connected layer; This is the fused feature vector processed by the encoder; The output is the traffic flow at charging stations after optimal scheduling of the power-transportation network coupled system.

[0124] Case Analysis

[0125] The superiority of the power-transportation network optimal scheduling method based on the graph attention proxy model described in this invention is illustrated below through a numerical example. This invention uses a coupled system of a 13-node transportation network and a 33-node distribution network as the example; the network topology diagram is shown below. Figure 2 To compare the superiority of the proposed method, a power-transportation network coordinated scheduling method was performed using a fixed charging service fee and the proposed dynamic charging service fee scheduling method. Simultaneously, backpropagation (BP) networks, gated recurrent networks (GRU) networks, graph convolutional networks (GCN) networks, Transformer networks, and the graph-attention proxy model proposed in this invention were used for training and learning.

[0126] Figure 3 The accuracy errors of different deep learning models learning charging station traffic flow are shown. It can be found that the graph-attention proxy model proposed in this invention extracts the network topology through a GCN layer and then trains it using the Transformer's attention mechanism. Overall, it exhibits the best performance in terms of error accuracy.

[0127] Table 1 summarizes the evaluation parameter results for different models. All models were trained using the same data. The graph-attention proxy model proposed in this invention demonstrates superior performance across all evaluation parameter metrics, with average errors improved by 57.93%, 59.35%, 43.22%, and 21.83% compared to backpropagation networks, gated recurrent networks, graph convolutional networks, and Transformer networks, respectively. This confirms that the proposed graph-attention proxy model possesses high accuracy and excellent generalization ability under various uncertain traffic network scenarios.

[0128] Table 1 Comparison of evaluation parameters for different models

[0129]

[0130] Figure 4 This paper illustrates the charging load distribution of charging stations under both fixed charging service fees and the dynamic charging service fee scheduling method proposed in this invention. Under the fixed charging service fee condition, the load of some charging stations is significantly concentrated during peak hours, which may lead to excessive node voltage or local overload. Under the dynamic pricing condition, the spatiotemporal distribution of the load is more balanced, resulting in a lower peak load and a smoother overall load curve. This helps alleviate the operational pressure on the distribution network and enhances the stability of the coupled system.

[0131] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A power-traffic network optimal scheduling method based on a graph attention agent model, characterized in that, The method comprises the following steps: Step 1, obtaining power distribution network parameters and operation coefficients, the network parameters including power distribution network topology, line impedance, the operation coefficients including upper power grid power supply coefficient, distributed power supply coefficient; Step 2, obtaining the scenario data of power grid node load demand, distributed photovoltaic output and time-of-use electricity price, taking the upper power grid power purchase quantity and the distributed power supply power generation quantity as control variables, and taking the power distribution network operation constraint as a constraint condition, a power distribution network second-order cone DistFlow optimal power flow model is established; Step 3, obtaining traffic network parameters, the network parameters including traffic network topology, each road section capacity, charging station position and capacity; Step 4, obtaining the scenario data of traffic network travel demand and electric vehicle penetration rate, taking the charging service fee as a control variable, and taking the basic traffic network operation constraint as a constraint condition, a semi-dynamic traffic distribution model considering dynamic charging service fee scheduling is established; Step 5, based on the power distribution network second-order cone DistFlow optimal power flow model in step 2 and the semi-dynamic traffic distribution model considering dynamic charging service fee scheduling in step 4, the power distribution network and the traffic network are coupled through the charging station to construct a power distribution-traffic network coupled system cost model; Step 6, based on solving the power distribution-traffic network coupled system cost model in step 5 under multiple scenarios, the charging station traffic flow obtained by optimizing and scheduling the coupled system is taken as the input, and the charging station traffic flow is taken as the output, a graph-attention network-based proxy model is established by adding a bidirectional cross-attention mechanism on the basis of a graph convolution network and a Transformer network, and the proxy model is trained and learned; Step 7, collecting the traffic network parameters and real-time traffic network travel demand and charging service fee and inputting them into the proxy model obtained in step 6 to solve, and outputting the charging station traffic flow to obtain a power-traffic network optimal scheduling method based on a graph-attention proxy model.

2. The power-traffic network optimal scheduling method based on a graph attention agent model according to claim 1, characterized in that, In step 2, the power distribution network second-order cone DistFlow optimal power flow model is as follows: (A-1) (A-2) (A-3) (A-4) (A-5) (A-6) (A-7) (A-8) (A-9) in, and They represent Timetable Active power and reactive power; and They represent Time Node The active and reactive power outputs of distributed power sources; , , They represent the lines respectively. Resistance, reactance, and impedance; express Timetable The current; Represents nodes Connected child nodes The set that constitutes; and They represent Timetable Active power and reactive power; and They represent Time Node The active and reactive loads; express Time Node Square of voltage amplitude at point; express Time Node Square of voltage amplitude at point; Indicates the line The upper limit of the current amplitude; and Representing nodes respectively The lower and upper limits of the square of the voltage amplitude; express Time Node The conventional power load at the location; , and , They are nodes The lower and upper limits of the active and reactive power output of the generator; For nodes The generator's ramp-climbing limitations; Represents nodes Connected charging sections gather; express Charging stations Traffic flow; This represents the average charging requirement per electric vehicle. represents a set of time periods; represents a set of generators connected to the node represents a set of generators connected to the node represents the economic cost of the distribution network; is the unit price of electricity, is the is the amount of electricity purchased by the distribution network from the upper-level network at the time instant and is the energy production cost coefficient of the node .

3. The optimal scheduling method of power-transportation network based on graph attention agent model according to claim 2, characterized in that, In step 4, the semi-dynamic traffic distribution model considering dynamic charging service fee scheduling is as follows: (A-10) (A-11) (A-12) (A-13) (A-14) (A-15) (A-16) (A-17) (A-18) (A-19) (A-20) wherein, denotes the set of paths k for the electric vehicle with origin r and destination s; rs denotes an origin-destination pair; denotes the set of origin-destination pairs; denotes the traffic flow of electric vehicles choosing path k; , denote the set of regular links and charging links, respectively; , denote the traffic flow of regular links and charging links, respectively; , denote the number of vehicles that regular links and charging links can accommodate, respectively; denotes a binary constant that judges whether link a belongs to path k or not, taking 1 if it belongs to path k and 0 otherwise; denotes the travel time of a link; , denote the free travel time of regular links and charging links, respectively, under zero traffic flow; J is the charging time model parameter; is the travel time on path k; is the initial traffic demand of electric vehicles traveling from origin r to destination s at time t; is the residual traffic demand of electric vehicles between O-D pair rs at time t; denotes the duration of a time period; is the revised traffic demand of electric vehicles traveling from origin r to destination s at time t; denotes the unit travel time cost; denotes the unit charging price; , denote the congestion pricing and charging service fee, respectively; denotes the average charging demand per electric vehicle; denotes the travel cost of electric vehicles choosing path k with origin r and destination s; denotes the minimum travel cost of electric vehicles with origin r and destination s; denotes the travel cost of network users.

4. The optimal scheduling method of power-transportation network based on graph attention agent model according to claim 3, characterized in that, In step 5, the power distribution-traffic network coupled system cost model considering dynamic charging service fee scheduling is as follows: (A-21) wherein, represents the total cost of the coupled system.

5. The optimal scheduling method of power-transportation network based on graph attention agent model according to claim 3, characterized in that, In step 6, the graph-attention proxy model learns the relationship between the charging flow and the charging service fee and the traffic environment, realizes the learning of the power-traffic network optimal scheduling, and the proxy model module comprises the following parts: (1) a graph convolution network module; (A-22) (A-23) (A-24) wherein a table and a road segment adjacent to the road segment and a road segment are connected; denotes the feature matrix of the layer; is the set of denotes the connection structure of the whole graph; is the identity matrix; denotes the node's relation to itself; is the degree matrix, which denotes the number of edges terminating at each node, with diagonal elements being the degree of the node; denotes the weight matrix of the layer; is the nonlinear activation function ReLU; (2) a bidirectional cross-attention module; (A-25) (A-26) (A-27) (A-28) (A-29) In the formula, denotes the node feature matrix output by the graph convolution network; denotes the global feature matrix; denotes the attention value of the node feature matrix to the global feature matrix; denotes the attention value of the global feature matrix to the node feature matrix; denotes the exponential normalization function; , , respectively denote the weight matrix corresponding to the query, key, and value; , , denote the weight matrix corresponding to the query, key, and value; denotes the dimension size of the key; denotes the updated node feature matrix after the attention operation; denotes the updated global feature matrix after the attention operation; denotes the layer normalization operation; denotes the regularization method; denotes the matrix splicing in the node dimension; denotes the feature matrix obtained after the node feature matrix and the global feature matrix are spliced, which is used as the input of the Transformer module; (3) a Transformer module: (A-30) (A-31) (A-32) (A-33) (A-34) (A-35) In the formula, , and respectively represent a query vector, a key vector, and a value vector; , , respectively represent the weight matrix corresponding to , , and the first attention head; represent the calculation standard attention mechanism; represent the first attention head; , is the number of attention heads; represent multi-head attention; represent the output weight of multi-head attention splicing; represent the input feature matrix after learning position coding; represent the position information in the sequence; represent the feedforward neural network; is the normalized output vector of the attention layer; and are weight vectors; and are bias terms; is a fully connected layer; is a fusion feature vector processed by an encoder; is the output power-traffic network coupled system optimal scheduling of the charging station traffic flow.

Images