Day-ahead schedulable potential prediction method considering charging and discharging behaviors of electric vehicle
By optimizing the BP neural network using the VB model and genetic algorithm, the problems of computational complexity and information leakage in predicting the charging and discharging behavior of electric vehicles are solved, achieving the effects of simplified calculation and reduced risk.
Patent Information
- Application Number
- CN202410617460.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-05-17
- Publication Date
- 2024-11-26
AI Technical Summary
Existing technologies for predicting the charging and discharging behavior of electric vehicles have high computational complexity, a large number of decision variables and constraints, resulting in a heavy computational burden and a high risk of leakage of user electricity consumption information.
By combining the VB model with genetic algorithms and BP neural networks, and optimizing the weight thresholds through data normalization, network training, and genetic algorithms, a classification model for electric vehicle charging and discharging behavior is constructed. The relaxed Minkowski summation method is used to express the cluster resource flexibility, simplifying the optimization solution process.
It effectively reduces the number of decision variables and constraints in the optimization process, lowers computational complexity, simplifies the computational burden of large-scale equipment, and reduces the risk of leakage of user electricity information.
Smart Images

Figure CN119025983A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of electric vehicle charging load modeling methods based on Monte Carlo method, and in particular to a method for predicting the day-ahead dispatchable potential taking into account the charging and discharging behavior of electric vehicles. Background Art
[0002] In the current market, load aggregators can participate in market-based transactions by reasonably dispatching the charging and discharging of electric vehicles. In this context, this chapter proposes a method for predicting the dispatchable potential that takes into account the charging and discharging behavior of electric vehicles. First, the feasibility of the dispatchable potential of electric vehicles is analyzed, and a dispatchable potential framework for electric vehicles is constructed; secondly, based on the survey data on the charging willingness of electric vehicle owners, a charging behavior classification model based on a genetic algorithm optimized neural network is established to determine whether electric vehicles have the ability to participate in dispatchability and obtain a dispatchable electric vehicle cluster. Finally, based on the virtual battery (VB) model of Minkowski summation, the maximum charging and discharging power and power boundary of the electric vehicle cluster are obtained, which provides a reference for load aggregators to implement different vehicle-grid interaction strategies and participate in day-ahead market transactions. Summary of the invention
[0003] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a method for predicting the day-ahead dispatchable potential taking into account the charging and discharging behavior of electric vehicles, using the VB model, a quantitative method that uses a set of standard battery parameters to describe the resource flexibility on the demand side. The VB model characterizes the resource flexibility of the cluster with an integrated power feasible domain, which can greatly reduce the number of decision variables, equations, and inequality constraints in the optimization process, solve the computational complexity problem caused by the coupling relationship between multiple decision variables, and effectively simplify the computational burden of optimizing and solving large-scale equipment. At the same time, its integrated form can reduce the risk of leakage of user electricity information under a reasonable operating framework.
[0004] In order to achieve the purpose of the present invention, the technical solution adopted by the present invention is:
[0005] The present invention discloses a method for predicting the day-ahead dispatchable potential taking into account the charging and discharging behavior of electric vehicles, comprising the following steps:
[0006] S1, Electric vehicle charging behavior classification process:
[0007] ① Based on historical charging data of electric vehicles and survey data on users’ willingness to participate in V2G;
[0008] ② Use genetic algorithm to optimize the initial weights and thresholds of BP neural network and train BP neural network;
[0009] ③ Classify electric vehicle charging behavior through trained artificial neural networks;
[0010] S2, BP neural network classification;
[0011] S3, genetic algorithm optimization weight threshold;
[0012] S4, dispatchable potential forecast.
[0013] The BP neural network classification in step S2 includes data normalization processing, network creation, network training and error definition.
[0014] The data normalization process: Due to the inconsistency of the units of the input data, some large-scale data will cause problems such as slow convergence of the neural network and long training time, and the range of the activation function of the output layer of the neural network is limited; therefore, the data needs to be normalized and mapped to the [0,1] interval;
[0015]
[0016] In the formula, y is the normalized data of x; min is the minimum value of x; max is the maximum value of x.
[0017] The network is created as a feedforward three-layer network. Assume that the network has 3 nodes in the input layer, q nodes in the hidden layer, 1 node in the output layer, and the weight between the input layer and the hidden layer is
[0018] w k ; b k , b are the thresholds of the hidden layer and the output layer respectively; the transfer function of the hidden layer is f1, and the transfer function of the output layer is f2;
[0019] The output of the hidden layer node is:
[0020]
[0021] Where k = 1, 2, ..., q
[0022] The output of the output layer node is:
[0023]
[0024] The network training and error definition network training is a process of constantly correcting weights and thresholds to minimize the output error of the network through training;
[0025] Assume that P sets of training data are input. Since the output layer node is 1, y p is the output of the network, and the mean square error function is used to obtain the error MSE of the P group training data, that is:
[0026]
[0027] In the formula, P represents the number of training samples, Represents the expected output value of the network, y p Represents the actual output value of the network. The cumulative error BP algorithm is used to adjust w jk , so that the global error MSE becomes smaller, that is:
[0028]
[0029] Where η is the learning rate.
[0030] The weights are modified iteratively until the error meets the requirements.
[0031] w jk =w jk +Δw jk (6)
[0032] In step S3, the genetic algorithm optimization weight threshold comprises the following steps:
[0033] a, population initialization;
[0034] Obtain BP network structure information, encode the weight from input layer to hidden layer, hidden layer threshold, weight from hidden layer to output layer, output layer threshold, and use binary encoding for each weight and threshold;
[0035] b. Construct fitness function
[0036] Considering that the network training goal of the BP neural network is to minimize the global error, the norm of the error matrix between the predicted value and the expected value of the predicted sample is used as the output of the objective function;
[0037] c. Weight threshold optimization
[0038] After a series of selection, crossover and mutation operations, it continuously evolves to seek the optimal solution. Find the optimal BP neural network initial weights and thresholds.
[0039] The dispatchable potential prediction is based on the VB model of electric vehicles, and its energy storage characteristics are expressed as
[0040] (1) Charge and discharge constraints
[0041]
[0042] (2) Constraints on power changes
[0043]
[0044] (3) Constraints on charging and discharging behavior
[0045]
[0046] (4) Battery capacity constraints
[0047]
[0048] Formula (3.9), Formula (3.10), Formula (3.11) and Formula (3.12) are the charge and discharge constraints, power change constraints and charge and discharge behavior constraints of the VB model respectively; among them, and They represent the charging and discharging power of electric vehicle n during period t, and is the maximum charge and discharge power of the general VB model, T n EV represents the grid-connected time set of electric vehicle n; s n,t and n,t-1 They represent the battery power of car n in time period t and the previous period respectively; η ch and η dis They represent the charging and discharging efficiency respectively; Δt represents the scheduling time; and is the initial charging time and the stopping charging time of electric vehicle n, X n,t represents the state of electric vehicle n at time t, X n,t =1 means that electric vehicle n is connected to the grid during period t; s n,t is the battery capacity of electric vehicle n at time t, and Indicates the upper and lower limits of the battery capacity of electric vehicle n;
[0049] For electric vehicles, the flexibility of their charging and discharging behaviors can be expressed by using the time-shifted power and energy upper and lower boundaries in the general VB model; the relaxed Minkowski summation method is used to aggregate them; Minkowski addition is an addition method applicable to Euclidean space, and the expression is shown in formula (3.5). Its physical essence is the expansion set of multiple spaces;
[0050]
[0051] Where: A and B represent two variable spaces; a and b are the elements in them respectively; denote their Minkowski sum;
[0052] For the VB model, the relaxed Minkowski summation can be used to further extend the charge and discharge boundary and power change boundary of a single electric vehicle to the charge and discharge boundary and power change boundary of the electric vehicle group, that is,
[0053]
[0054] In the formula, is the maximum charge and discharge power of the VB model, S t is the charge boundary of the VB model, ΔS t The change in VB power caused by the change in the grid-connected state of the electric vehicle;
[0055] Dispatchable Potential Prediction Based on VB Model
[0056] By combining equations (3.10), we can solve the electric quantity boundary of the VB model at time t, namely:
[0057]
[0058] In the formula, S t is the power boundary of the VB model at time t, and it follows the following constraints:
[0059]
[0060]
[0061]
[0062] Formula (3.9), Formula (3.10) and Formula (3.11) are the maximum charging power constraint, maximum discharge constraint and power constraint of the VB model respectively; are the charging power and discharging power of the VB model at time t.
[0063] The beneficial effects of the present invention are:
[0064] (1) The present invention uses the VB model, a quantitative method that uses a set of standard battery parameters to describe the resource flexibility of the demand side. The VB model characterizes the resource flexibility of the cluster with an integrated power feasible domain, which can greatly reduce the number of decision variables, equations, and inequality constraints in the optimization process, solve the computational complexity problem caused by the coupling relationship between multiple decision variables, and effectively simplify the computational burden of optimizing large-scale equipment. At the same time, its integrated form can reduce the risk of leakage of user power information under a reasonable operating framework. BRIEF DESCRIPTION OF THE DRAWINGS
[0065] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION
[0066] The present invention is further described below:
[0067] See also Figure 1 ,
[0068] The present invention discloses a method for predicting the day-ahead dispatchable potential taking into account the charging and discharging behavior of electric vehicles, comprising the following steps:
[0069] S1, Electric vehicle charging behavior classification process:
[0070] ④ Based on historical charging data of electric vehicles and survey data on users’ willingness to participate in V2G;
[0071] ⑤ Use genetic algorithm to optimize the initial weights and thresholds of BP neural network and train BP neural network;
[0072] ⑥ Classify electric vehicle charging behavior through trained artificial neural networks;
[0073] S2, BP neural network classification;
[0074] S3, genetic algorithm optimization weight threshold;
[0075] S4, dispatchable potential forecast.
[0076] The BP neural network classification in step S2 includes data normalization processing, network creation and network training and error definition.
[0077] The data normalization process: Due to the inconsistency of the units of the input data, some large-scale data will cause problems such as slow convergence of the neural network and long training time, and the range of the activation function of the output layer of the neural network is limited; therefore, the data needs to be normalized and mapped to the [0,1] interval;
[0078]
[0079] In the formula, y is the normalized data of x; min is the minimum value of x; max is the maximum value of x.
[0080] The network is created as a feed-forward three-layer network. Assume that the network has 3 nodes in the input layer, q nodes in the hidden layer, 1 node in the output layer, and the weight between the input layer and the hidden layer is w k ; b k , b are the thresholds of the hidden layer and the output layer respectively; the transfer function of the hidden layer is f1, and the transfer function of the output layer is f2;
[0081] The output of the hidden layer node is:
[0082]
[0083] Where k = 1, 2, ..., q
[0084] The output of the output layer node is:
[0085]
[0086] The network training and error definition network training is a process of constantly correcting weights and thresholds to minimize the output error of the network through training;
[0087] Assume that P sets of training data are input. Since the output layer node is 1, y p is the output of the network, and the mean square error function is used to obtain the error MSE of the P group training data, that is:
[0088]
[0089] In the formula, P represents the number of training samples, Represents the expected output value of the network, y p Represents the actual output value of the network. The cumulative error BP algorithm is used to adjust w jk , so that the global error MSE becomes smaller, that is:
[0090]
[0091] Where η is the learning rate.
[0092] The weights are modified iteratively until the error meets the requirements.
[0093] w jk =w jk +Δw jk (6)
[0094] In step S3, the genetic algorithm optimization weight threshold comprises the following steps:
[0095] a, population initialization;
[0096] Obtain BP network structure information, encode the weight from input layer to hidden layer, hidden layer threshold, weight from hidden layer to output layer, output layer threshold, and use binary encoding for each weight and threshold;
[0097] b. Construct fitness function
[0098] Considering that the network training goal of the BP neural network is to minimize the global error, the norm of the error matrix between the predicted value and the expected value of the predicted sample is used as the output of the objective function;
[0099] d. Weight threshold optimization
[0100] After a series of selection, crossover and mutation operations, it continuously evolves to seek the optimal solution. Find the optimal BP neural network initial weights and thresholds.
[0101] The dispatchable potential prediction is based on the VB model of electric vehicles, and its energy storage characteristics are expressed as
[0102] (1) Charge and discharge constraints
[0103]
[0104] (2) Constraints on power changes
[0105]
[0106] (3) Constraints on charging and discharging behavior
[0107]
[0108] (4) Battery capacity constraints
[0109]
[0110] Formula (3.9), Formula (3.10), Formula (3.11) and Formula (3.12) are the charge and discharge constraints, power change constraints and charge and discharge behavior constraints of the VB model respectively; among them, and They represent the charging and discharging power of electric vehicle n during period t, and is the maximum charge and discharge power of the general VB model, T n EV represents the grid-connected time set of electric vehicle n; s n,t and n,t-1 They represent the battery power of car n in time period t and the previous period respectively; η ch and η dis They represent the charging and discharging efficiency respectively; Δt represents the scheduling time; and is the initial charging time and the stopping charging time of electric vehicle n, X n,t represents the state of electric vehicle n at time t, X n,t =1 means that electric vehicle n is connected to the grid during period t; s n,t is the battery capacity of electric vehicle n at time t, and Indicates the upper and lower limits of the battery capacity of electric vehicle n;
[0111] For electric vehicles, the flexibility of their charging and discharging behaviors can be expressed by using the time-shifted power and energy upper and lower boundaries in the general VB model; the relaxed Minkowski summation method is used to aggregate them; Minkowski addition is an addition method applicable to Euclidean space, and the expression is shown in formula (3.5). Its physical essence is the expansion set of multiple spaces;
[0112]
[0113] Where: A and B represent two variable spaces; a and b are the elements in them respectively; denote their Minkowski sum;
[0114] For the VB model, the relaxed Minkowski summation can be used to further extend the charge and discharge boundary and power change boundary of a single electric vehicle to the charge and discharge boundary and power change boundary of the electric vehicle group, that is,
[0115]
[0116] In the formula, is the maximum charge and discharge power of the VB model, S t is the charge boundary of the VB model, ΔS t The change in VB power caused by the change in the grid-connected state of the electric vehicle;
[0117] Dispatchable Potential Prediction Based on VB Model
[0118] By combining equations (3.10), we can solve the electric quantity boundary of the VB model at time t, namely:
[0119]
[0120] In the formula, S t is the power boundary of the VB model at time t, and it follows the following constraints:
[0121]
[0122]
[0123]
[0124] Formula (3.9), Formula (3.10) and Formula (3.11) are the maximum charging power constraint, maximum discharge constraint and power constraint of the VB model respectively; are the charging power and discharging power of the VB model at time t.
[0125] Analyzing equations (3.12) and (3.13), we can see that the main factors affecting the equivalent power of the VB model are the charging time and the number of charging vehicles per unit time. Among them, the charging time is limited by the car owner's car usage habits and can be regarded as a relatively fixed value; the number of charging vehicles per unit time is closely related to the owner's own behavior and can be appropriately guided by economic incentives, which also provides a strong opportunity for the operation space of load aggregators.
[0126] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent transformations made using the contents of the present invention's specification and drawings or directly or indirectly applied in related technical fields are also included in the patent protection scope of the present invention.
Claims
1. A method for predicting the day-ahead dispatchable potential taking into account the charging and discharging behavior of electric vehicles. Features: The steps include: S1, Electric vehicle charging behavior classification process: ① Based on historical charging data of electric vehicles and survey data on users’ willingness to participate in V2G; ② Use genetic algorithm to optimize the initial weights and thresholds of BP neural network and train BP neural network; ③ Classify electric vehicle charging behavior through trained artificial neural networks; S2, BP neural network classification; S3, genetic algorithm optimization weight threshold; S4, dispatchable potential prediction.
2. A method for predicting the day-ahead dispatchable potential taking into account the charging and discharging behavior of electric vehicles according to claim 1, Features: The BP neural network classification in step S2 includes data normalization processing, network creation, network training and error definition.
3. A method for predicting the day-ahead dispatchable potential taking into account the charging and discharging behavior of electric vehicles according to claim 2, Features: The data normalization process: Due to the inconsistency of the units of the input data, some large-scale data will cause problems such as slow convergence of the neural network and long training time, and the range of the activation function of the output layer of the neural network is limited; therefore, the data needs to be normalized and mapped to the [0,1] interval; In the formula, y is the normalized data of x; min is the minimum value of x; max is the maximum value of x.
4. A method for predicting the day-ahead dispatchable potential taking into account the charging and discharging behavior of electric vehicles according to claim 2, Features: The network is created as a feed-forward three-layer network. Assume that the network has 3 nodes in the input layer, q nodes in the hidden layer, 1 node in the output layer, and the weight between the input layer and the hidden layer is w k ; b k , b are the thresholds of the hidden layer and the output layer respectively; the transfer function of the hidden layer is f 1 , the transfer function of the output layer is f 2 ; The output of the hidden layer node is: Where k = 1, 2, ..., q The output of the output layer node is: 。 5. A method for predicting the day-ahead dispatchable potential taking into account the charging and discharging behavior of electric vehicles according to claim 2, Features: The network training and error definition network training is a process of constantly correcting weights and thresholds to minimize the output error of the network through training; Assume that P sets of training data are input. Since the output layer node is 1, y p is the output of the network, and the mean square error function is used to obtain the error MSE of the P group training data, that is: In the formula, P represents the number of training samples, Represents the expected output value of the network, y p Represents the actual output value of the network. The cumulative error BP algorithm is used to adjust w jk , so that the global error MSE becomes smaller, that is: Where η is the learning rate. The weights are modified iteratively until the error meets the requirements. In jk =in jk +Δw jk (6)。 6. A method for predicting the day-ahead dispatchable potential taking into account the charging and discharging behavior of electric vehicles according to claim 1, Features: In step S3, the genetic algorithm optimization weight threshold comprises the following steps: a, population initialization; Obtain BP network structure information, encode the weight from input layer to hidden layer, hidden layer threshold, weight from hidden layer to output layer, output layer threshold, and use binary encoding for each weight and threshold; b. Construct fitness function Considering that the network training goal of the BP neural network is to minimize the global error, the norm of the error matrix between the predicted value and the expected value of the predicted sample is used as the output of the objective function; c. Weight threshold optimization After a series of selection, crossover and mutation operations, it continuously evolves to seek the optimal solution. Find the optimal BP neural network initial weights and thresholds.
7. A method for predicting the day-ahead dispatchable potential taking into account the charging and discharging behavior of electric vehicles according to claim 1, Features: The dispatchable potential prediction is based on the VB model of electric vehicles, and its energy storage characteristics are expressed as (1) Charge and discharge constraints (2) Constraints on power changes (3) Constraints on charging and discharging behavior (4) Battery capacity constraints Formula (3.9), Formula (3.10), Formula (3.11) and Formula (3.12) are the charge and discharge constraints, power change constraints and charge and discharge behavior constraints of the VB model respectively; among them, and They represent the charging and discharging power of electric vehicle n during period t, and is the maximum charge and discharge power of the general VB model, T n EV represents the grid-connected time set of electric vehicle n; s n,t and n,t-1 They represent the battery power of car n in time period t and the previous period respectively; η ch and η dis They represent the charging and discharging efficiency respectively; Δt represents the scheduling time; and is the initial charging time and the stopping charging time of electric vehicle n, X n,t represents the state of electric vehicle n at time t, X n,t =1 means that electric vehicle n is connected to the grid during period t; s nt is the battery capacity of electric vehicle n at time t, and Indicates the upper and lower limits of the battery capacity of electric vehicle n; For electric vehicles, the flexibility of their charging and discharging behaviors can be expressed by using the time-shifted power and energy upper and lower boundaries in the general VB model; the relaxed Minkowski summation method is used to aggregate them; Minkowski addition is an addition method applicable to Euclidean space, and the expression is shown in formula (3.5). Its physical essence is the expansion set of multiple spaces; Where: A and B represent two variable spaces; a and b are the elements in them respectively; denote their Minkowski sum; For the VB model, the relaxed Minkowski summation can be used to further extend the charge and discharge boundary and power change boundary of a single electric vehicle to the charge and discharge boundary and power change boundary of the electric vehicle group, that is, Where P t ch,max , P t dis,max is the maximum charge and discharge power of the VB model, S t is the charge boundary of the VB model, ΔS t The change in VB power caused by the change in the grid-connected state of the electric vehicle; Dispatchable Potential Prediction Based on VB Model By combining equations (3.10), we can solve the electric quantity boundary of the VB model at time t, namely: In the formula, S t is the power boundary of the VB model at time t, and it follows the following constraints: 0,,P t ch ,,P t ch,max (3.8) 0,,P t dis ,,P t dis,max (3.9) Formula (3.9), Formula (3.10) and Formula (3.11) are the maximum charging power constraint, maximum discharge constraint and power constraint of the VB model respectively; P t ch , P t dis are the charging power and discharging power of the VB model at time t.
