Traffic energy joint distribution optimization method based on data driving

Abstract

The invention discloses a traffic energy joint distribution optimization method based on data driving. The method comprises the following steps: S1, collecting traffic network parameters and power grid parameters according to historical data; s2, establishing a dynamic traffic-energy distribution model according to the obtained parameters of the power grid and the traffic network; s3, learning a nonlinear relationship by adopting a multilayer perceptron, and training by using historical operation data or simulation data to obtain a trained multilayer perceptron mapping function; s4, the trained multi-layer perceptron mapping function is converted into mixed integer constraints through piecewise linearization and binary variable control, and a reconstructed MIP model capable of being embedded into an optimization solver is obtained; and S5, performing global optimization by using the reconstructed MIP model, and outputting an optimal / near-optimal traffic-energy joint distribution scheme. According to the method, the charge and discharge-traffic dynamic coupling mechanism is described as fidelity as possible while solvability is kept, and the defects of fixed power linearization and traditional envelope approximation in flexible utilization and precision are overcome.

Description

Technical Field

[0001] This invention relates to the field of power-traffic coupling optimization and vehicle-to-grid (V2G) spatiotemporal coordinated scheduling, and in particular to a data-driven joint allocation optimization method for traffic energy. Background Technology

[0002] With the large-scale deployment of electric vehicles and charging stations, the coupling between the power system and the transportation system is becoming increasingly close, and V2G (Vehicle-to-Grid) provides a new opportunity to improve grid flexibility. Compared with static or semi-dynamic traffic assignment, dynamic traffic assignment (DTA) can better characterize the time-varying characteristics of travel demand and traffic flow, and therefore has more advantages in power-transportation joint scheduling. However, most existing DTA models often set the charging and discharging power of electric vehicles as a given parameter to avoid nonlinearity. Although this simplification facilitates linear solutions, it significantly reduces the flexibility that can be provided to the grid for V2G, especially in scenarios where the synergistic effects of electric vehicles need to be fully explored (such as resilient scheduling). On the other hand, once charging and discharging power is introduced as a decision variable, nonlinear coupling terms appear in DTEA, making the model non-convex and difficult to solve directly. Traditional linearization methods (such as the McCormick envelope approximation) are insufficient in both accuracy and applicability.

[0003] The "Power-Traffic Coupling Optimization Method Based on Dynamic Traffic Flow Assignment Theory" simplifies nonlinear constraints by using fixed charging and discharging power or conventional linearization approximations during modeling. Most existing Dynamic Traffic Flow Assignment (DTA) models often set the charging and discharging power of electric vehicles as a given parameter to avoid nonlinearity. While this simplification facilitates linear solutions, it significantly reduces the flexibility provided to the power grid for V2G, especially in scenarios requiring full exploitation of the synergistic effects of electric vehicles (such as resilient dispatch). On the other hand, once charging and discharging power is introduced as a decision variable, nonlinear coupling terms appear in the DTEA, making the model non-convex and difficult to solve directly. Traditional linearization methods (such as the McCormick envelope approximation) have shortcomings in both accuracy and applicability. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a data-driven joint allocation optimization method for traffic energy. While maintaining solvability, it aims to accurately characterize the dynamic coupling mechanism of charging and discharging and traffic, thus making up for the deficiencies of fixed power linearization and traditional envelope approximation in terms of flexibility and accuracy.

[0005] The objective of this invention is achieved through the following technical solution: a data-driven joint allocation optimization method for traffic energy, comprising the following steps:

[0006] S1. Collect transportation network parameters and power grid parameters based on historical data;

[0007] S2. Based on the obtained power grid and transportation network parameters, establish a dynamic transportation-energy allocation model;

[0008] S3. Use a multilayer perceptron to learn nonlinear relationships, and use historical operating data or simulation data for training to obtain a trained multilayer perceptron mapping function;

[0009] S4. Through the learning and reconstruction of the equivalent MIP, the trained multilayer perceptron mapping function is transformed into mixed integer constraints through piecewise linearization and binary variable control, resulting in a reconstructed MIP model that can be embedded in the optimization solver.

[0010] S5. Call the commercial solver Gurobi to solve the MIP model, use the reconstructed MIP model for global optimization, and output the optimal / near-optimal traffic-energy joint allocation scheme.

[0011] The beneficial effects of this invention are: This invention uses neural networks to learn nonlinearity and mixed integer constraints to replicate solvable structures, and is aimed at the DTEA problem under spatiotemporal V2G cooperation. While maintaining solvability, it describes the dynamic coupling mechanism of charging and discharging-traffic as faithfully as possible, making up for the shortcomings of fixed power linearization and traditional envelope approximation in terms of flexibility and accuracy. Attached Figure Description

[0012] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0013] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings, but the scope of protection of the present invention is not limited to the following description.

[0014] like Figure 1 As shown, a data-driven joint allocation optimization method for traffic energy includes the following steps:

[0015] S1. Collect transportation network parameters and power grid parameters based on historical data;

[0016] In step S1, the transportation network parameters include:

[0017] set of start and end pairs It contains multiple origin-destination pairs w; the origin-destination pair w refers to the journey of a vehicle from its origin to its destination in a transportation system.

[0018] Charge / discharge efficiency η c η d The lower and upper bounds of the capacity parameters for the interaction between the traffic flow and the power grid for the selected travel path k belonging to the origin-end pair w in charging station n.

[0019] Electric vehicle travel demand in charging station n that belongs to the choice of travel path k for the origin-end pair w. With battery capacity limitations The charging and discharging power limitation of electric vehicles belonging to the starting and ending points of the selected travel path k in charging station n.

[0020] The power grid parameters mentioned include spatiotemporal charging and discharging electricity prices.

[0021] S2. Based on the obtained power grid and transportation network parameters, establish a dynamic transportation-energy allocation model;

[0022] In step S2, the objective function of the optimization model is to minimize travel time and energy cost, and maximize V2G discharge benefits; the constraints include traffic flow conservation, EV travel demand constraints, and power-transport coupling constraints.

[0023] Construct the following coupled traffic-energy allocation constraints:

[0024]

[0025]

[0026] Among them, the optimization model variables are in, For charging station n, the energy flow variable of the traffic flow at time t for the selected passage path k in the origin-end pair w; and Let be the inflow and outflow of energy at charging station n at time t; Let be the charging and discharging power at charging station n at time t; These represent the vehicle flow, inflow, and outflow at charging station n at time t, respectively. It is a binary variable, representing the charging and discharging state of the electric vehicle;

[0027] Constraint (1.1) calculates the energy flow change process within charging station n at time t; Constraint (1.2) specifies the state of charge of the vehicle flow within charging station n at time t; Constraint (1.3) calculates the state of charge of the vehicle flow flowing into charging station n at time t; Constraint (1.4) specifies the state of charge of the vehicle flow flowing out of charging station n at time t; Constraint (1.5) specifies the charging and discharging power limit for the vehicle flow within charging station n at time t that belongs to the selected passage path k in the start-end pair w; Constraint (1.6) restricts any vehicle flow from being simultaneously charging and discharging at a certain time; Constraint (1.7) calculates the total charging and discharging power of the vehicle flow within charging station n at time t that belongs to the selected passage path k in the start-end pair w.

[0028] The objective function is constructed as follows:

[0029]

[0030] The objective function's first term is to minimize the charging and discharging cost of electric vehicles, the second term is to minimize the charging and discharging time cost of electric vehicles at charging stations, and the third term is to minimize the travel time cost of electric vehicles.

[0031] S3. Use a multilayer perceptron to learn nonlinear relationships, and use historical operating data or simulation data for training to obtain a trained multilayer perceptron mapping function;

[0032] Because the optimization model constructed above contains nonlinear terms. Two continuous variables Multiplying yields the result; the learning ability of neural networks is needed to fit these nonlinear terms.

[0033] Construct an MLP model, namely a multilayer perceptron model;

[0034] Construct a dataset for training the MLP. The vectors in the dataset are obtained from historical running data or simulation data. This dataset contains several data samples, each containing:

[0035] Input features:

[0036] Output result:

[0037] The MLP is trained using the MLP dataset. Specifically, the feedforward propagation mechanism of the neural network is used to learn the nonlinear relationship between the input features and the output results, thus obtaining a trained MLP model, which is a trained multilayer perceptron mapping function.

[0038] S4. Through the learning and reconstruction of the equivalent MIP, the trained multilayer perceptron mapping function is transformed into mixed integer constraints through piecewise linearization and binary variable control, resulting in a reconstructed MIP model that can be embedded in the optimization solver.

[0039] Suppose a multilayer perceptron contains an f-th layer network that needs to learn W. f b f Network parameters, among which, This is the output of the previous layer. As the input to the first layer of the network, o tIt is the output result of the neural network; where formula (2.1) constructs the input layer of the neural network; formula (2.2) is the affine transformation of the neural network; formula (2.3) uses the activation function to change negative numbers to 0 and positive numbers to keep their original values; formula (2.4) is the output layer of the neural network;

[0040]

[0041] Since the formula (2.3) contains the operator max, which is a non-linear operator that the gurobi optimization solver has difficulty handling, the formula is linearized.

[0042] The mapping function of the trained multilayer perceptron is transformed into a mixed integer constraint through piecewise linearization and binary variable control, and the "neural network prediction value" is transformed into a MIP form that can be embedded in the optimization solver.

[0043] In formula (3.1), an auxiliary variable is introduced. Reconstruct formula (2.2), and use the Big M method to linearize the constraints in formulas (3.2)-(3.3), where M is a sufficiently large parameter, and constraint (3.4) is an auxiliary variable. Provide a definition;

[0044]

[0045] S5. Call the commercial solver Gurobi to solve the MIP model, use the reconstructed MIP model for global optimization, and output the optimal / near-optimal traffic-energy joint allocation scheme.

[0046] In step S5, the reconstructed MIP model is first used to characterize the optimization problem. and The correlation between them is then determined, and the strategy is solved based on the dynamic traffic-energy allocation model, objective function, and constraints.

[0047] The strategies obtained include: traffic allocation strategies for electric vehicles participating in vehicle-to-grid interaction, including: the traffic flow of electric vehicles belonging to the origin-end pair w within charging station n at time t, via the selected path k. At time t, within charging station n, the selected travel path k for the electric vehicle waiting flow belongs to the origin-end pair w.

[0048] Electric vehicle charging and discharging strategies that participate in vehicle-to-grid interaction include: The charging and discharging power at charging station n at time t.

[0049] The method also includes strategy distribution and execution steps: pushing electric vehicle traffic allocation strategies to traffic guidance screens and in-vehicle navigation systems; and distributing electric vehicle charging and discharging strategies to charging pile controllers.

[0050] The foregoing description illustrates and describes a preferred embodiment of the present invention. However, as previously stated, it should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the inventive concept described herein through the foregoing teachings or techniques or knowledge in related fields. Any modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A data-driven joint allocation optimization method for traffic energy, characterized in that: Includes the following steps: S1. Collect transportation network parameters and power grid parameters based on historical data; S2. Based on the obtained power grid and transportation network parameters, establish a dynamic transportation-energy allocation model; S3. Use a multilayer perceptron to learn nonlinear relationships, and use historical operating data or simulation data for training to obtain a trained multilayer perceptron mapping function; S4. Through the learning and reconstruction of the equivalent MIP, the trained multilayer perceptron mapping function is transformed into mixed integer constraints through piecewise linearization and binary variable control, resulting in a reconstructed MIP model that can be embedded in the optimization solver. S5. Call the commercial solver Gurobi to solve the MIP model, use the reconstructed MIP model for global optimization, and output the optimal / near-optimal traffic-energy joint allocation scheme.

2. The data-driven joint allocation optimization method for traffic energy according to claim 1, characterized in that: In step S1, the transportation network parameters include: set of start and end pairs Contains multiple start and end pairs w; Charge / discharge efficiency η c η d The lower and upper bounds of the capacity parameters for the interaction between the traffic flow and the power grid for the selected travel path k belonging to the origin-end pair w in charging station n. Electric vehicle travel demand in charging station n that belongs to the choice of travel path k for the origin-end pair w. With battery capacity limitations The charging and discharging power limitation of electric vehicles belonging to the starting and ending points of the selected travel path k in charging station n. The power grid parameters mentioned include spatiotemporal charging and discharging electricity prices.

3. The data-driven joint allocation optimization method for traffic energy according to claim 1, characterized in that: In step S2, the objective function of the optimization model is to minimize travel time and energy cost, and maximize V2G discharge benefits; the constraints include traffic flow conservation, EV travel demand constraints, and power-transport coupling constraints. Construct the following coupled traffic-energy allocation constraints: Among them, the optimization model variables are in, For charging station n, the energy flow variable of the traffic flow at time t for the selected passage path k in the origin-end pair w; and Let be the inflow and outflow of energy at charging station n at time t; Let be the charging and discharging power at charging station n at time t; These represent the vehicle flow, inflow, and outflow at charging station n at time t, respectively. It is a binary variable, representing the charging and discharging state of the electric vehicle; Constraint (1.1) calculates the energy flow change process within charging station n at time t; Constraint (1.2) specifies the state of charge of the vehicle flow within charging station n at time t; Constraint (1.3) calculates the state of charge of the vehicle flow flowing into charging station n at time t; Constraint (1.4) specifies the state of charge of the vehicle flow flowing out of charging station n at time t; Constraint (1.5) specifies the charging and discharging power limit for the vehicle flow within charging station n at time t that belongs to the selected passage path k in the start-end pair w; Constraint (1.6) restricts any vehicle flow from being simultaneously charging and discharging at a certain time; Constraint (1.7) calculates the total charging and discharging power of the vehicle flow within charging station n at time t that belongs to the selected passage path k in the start-end pair w. The objective function is constructed as follows: The objective function's first term is to minimize the charging and discharging cost of electric vehicles, the second term is to minimize the charging and discharging time cost of electric vehicles at charging stations, and the third term is to minimize the travel time cost of electric vehicles.

4. The data-driven joint allocation optimization method for traffic energy according to claim 1, characterized in that: Step S3 includes: Because the optimization model constructed above contains nonlinear terms. Two continuous variables Multiplying yields the result; the learning ability of neural networks is needed to fit these nonlinear terms. Construct an MLP model, namely a multilayer perceptron model; Construct a dataset for training the MLP. The vectors in the dataset are obtained from historical running data or simulation data. This dataset contains several data samples, each containing: Input features: Output result: The MLP is trained using the MLP dataset. Specifically, the feedforward propagation mechanism of the neural network is used to learn the nonlinear relationship between the input features and the output results, thus obtaining a trained MLP model, which is a trained multilayer perceptron mapping function.

5. The data-driven joint allocation optimization method for traffic energy according to claim 1, characterized in that: Step S4 includes: Suppose a multilayer perceptron contains an f-th layer network that needs to learn W. f b f Network parameters, among which, This is the output of the previous layer. As the input to the first layer of the network, o t It is the output result of the neural network; where formula (2.1) constructs the input layer of the neural network; formula (2.2) is the affine transformation of the neural network; formula (2.3) uses the activation function to change negative numbers to 0 and positive numbers to keep their original values; formula (2.4) is the output layer of the neural network; Since the formula (2.3) contains the operator max, which is a non-linear operator that the gurobi optimization solver has difficulty handling, the formula is linearized. The mapping function of the trained multilayer perceptron is transformed into a mixed integer constraint through piecewise linearization and binary variable control, and the "neural network prediction" is transformed into a MIP form that can be embedded in the optimization solver. In formula (3.1), an auxiliary variable is introduced. Reconstruct formula (2.2), and use the Big M method to linearize the constraints in formulas (3.2)-(3.3), where M is a sufficiently large parameter, and constraint (3.4) is an auxiliary variable. Provide a definition; 6. The data-driven joint allocation optimization method for traffic energy according to claim 1, characterized in that: In step S5, the reconstructed MIP model is first used to characterize the optimization problem. and The correlation between them is then determined, and the strategy is solved based on the dynamic traffic-energy allocation model, objective function, and constraints. The strategies obtained include: traffic allocation strategies for electric vehicles participating in vehicle-to-grid interaction, including: the traffic flow of electric vehicles belonging to the origin-end pair w within charging station n at time t, via the selected path k. At time t, within charging station n, the selected travel path k for the electric vehicle waiting flow belongs to the origin-end pair w. Electric vehicle charging and discharging strategies that participate in vehicle-to-grid interaction include: The charging and discharging power at charging station n at time t.

7. The data-driven joint allocation optimization method for traffic energy according to claim 1, characterized in that: The method also includes strategy distribution and execution steps: pushing electric vehicle traffic allocation strategies to traffic guidance screens and in-vehicle navigation systems; and distributing electric vehicle charging and discharging strategies to charging pile controllers.

Images