A method for evaluating adjustable capacity of electric vehicle cluster considering user response willingness

Abstract

The application belongs to the technical field of fuel cells, and relates to an electric vehicle cluster adjustable capacity evaluation method considering user response willingness, comprising the following steps: 1, establishing a unified energy storage model of an electric vehicle EV individual; 2, establishing a generalized energy storage GES resource general aggregation model of an electric vehicle EV cluster; 3, obtaining theoretical demand response potential based on a two-way quantum long short-term memory neural network to predict an electric vehicle EV cluster aggregation model parameter set; 4, obtaining a demand response willingness coefficient based on a TSK fuzzy system to quantize user side demand response willingness; and 5, obtaining actual demand response potential by combining data and model driven results; the application considers the heterogeneity of electric vehicles to model individuals in a unified energy storage manner, and encapsulates the cluster as a generalized energy storage resource general aggregation model, so that the theoretical demand response potential of the electric vehicle cluster is accurately characterized by a smaller number of parameters without affecting the model accuracy, and the actual demand response potential is obtained by considering the user response willingness.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of fuel cells and relates to a method for evaluating adjustable capacity of an electric vehicle cluster considering user response willingness. BACKGROUND

[0002] As an important part of the new energy system and a key carrier for achieving the "double carbon" goal, the new power system is developing rapidly. Distributed renewable energy, mainly wind and photovoltaic, is gradually replacing fossil fuels for large-scale development and utilization. While bringing clean and low carbon, its strong randomness and volatility have brought challenges to the safe and efficient operation of the power system. The "mobile energy storage" feature of electric vehicles makes them an important demand response resource. As a new energy transportation tool, its stock continues to grow with the support of national policies, and has great demand response potential, which can promote renewable energy consumption and support the reliable and stable operation of the power system while ensuring user demand.

[0003] Currently, the charging modes of electric vehicles (EVs) are divided into fast charging, slow charging, and battery replacement. Among them, the fast charging mode refers to EV charging directly with the maximum charging power, which is uncontrollable and cannot provide flexibility, belonging to conventional rigid load. The slow charging mode refers to adjusting the power size according to the actual demand of the power grid when time permits, which has the ability of load shifting and reverse power supply, belonging to mobile energy storage. The battery replacement mode refers to quickly replacing the battery to complete charging, and the replaced battery participates in dispatching according to the demand of the power grid, but the current battery replacement infrastructure is not perfect, and the acceptance of the battery replacement mode by users is low, so the demand response potential is low. Therefore, accurate evaluation of the actual adjustable capacity of the EV slow charging mode cluster can lay the foundation for smoothing the strong randomness and volatility of new energy power generation modes.

[0004] However, the overall operation characteristics of the electric vehicle cluster and the willingness of users to participate in demand response are difficult to model effectively, making it impossible to accurately evaluate the actual adjustable capacity of the electric vehicle cluster. Specifically, the following problems exist:

[0005] 1) Existing technical solutions do not consider the heterogeneity of the electric vehicle cluster when aggregating it, making it impossible to establish an accurate aggregation model;

[0006] 2) Existing technical solutions do not consider the coupling relationship between different factors and various uncertainties when predicting the parameters of the electric vehicle cluster aggregation model, making it difficult to accurately evaluate its demand response potential;

[0007] 3) Existing technical solutions do not fully consider the demand response willingness of users when evaluating the adjustable capacity of the electric vehicle cluster, making the evaluation results may not be feasible.

[0008] Therefore, an electric vehicle cluster adjustable capacity evaluation method is urgently needed to calculate the demand response potential considering the user response willingness, thereby supporting the safe and efficient operation of the power system. SUMMARY

[0009] The technical solution adopted by the present application to solve the technical problems is: an electric vehicle cluster adjustable capacity evaluation method considering user response willingness, comprising the following steps:

[0010] Step S1, establishing a unified energy storage model of an electric vehicle EV individual;

[0011] Step S2, establishing a generalized energy storage GES resource general aggregation model of an electric vehicle EV cluster;

[0012] Step S3, predicting the theoretical demand response potential based on a two-way quantum long short-term memory neural network prediction of the electric vehicle EV cluster aggregation model parameter set;

[0013] Step S4, quantifying the user side demand response willingness based on a TSK fuzzy system to obtain a demand response willingness coefficient;

[0014] Step S5, combining the data and model driven results to obtain the actual demand response potential.

[0015] Preferably, in step S1, the general form of the electric vehicle EV individual model is represented as:

[0016]

[0017] In formula (5), and respectively represent the grid-connected battery power of the i-th electric vehicle EV at time t and time t+1; and respectively represent the battery power of the i-th electric vehicle EV at the grid-connected time and the off-grid time; and respectively represent the grid-connected and off-grid states of the i-th electric vehicle EV at time t and time t+1, which are Boolean variables, 0 representing off-grid and 1 representing grid-connected; and respectively represent the charging and discharging efficiency of the i-th electric vehicle EV; and respectively represent the charging / discharging state of the i-th electric vehicle EV during the period from t to t+1, which are Boolean variables, 0 representing that the electric vehicle EV is not charging / discharging, and 1 representing that the electric vehicle EV is charging / discharging; and respectively represent the charging / discharging power of the i-th electric vehicle EV during the period from t to t+1; Δt represents the unit time period size of the electric vehicle EV charging / discharging.

[0018] Preferably, in step S2, the individual energy storage model of the electric vehicle EV is encapsulated as a general aggregation model of the generalized energy storage GES resource based on the individual energy storage model of the electric vehicle EV, and the model is:

[0019]

[0020] In formula (12), and respectively represent the electric quantity of the generalized energy storage GES at time t and t+1; P t GES,cha represents the charging power of the generalized energy storage GES at t to t+1; P t GES,dis represents the discharging power of the generalized energy storage GES at t to t+1; represents the self-loss coefficient of the electric quantity of the generalized energy storage GES at t to t+1; represents the charging efficiency of the generalized energy storage GES at t to t+1; represents the reciprocal of the discharging efficiency of the generalized energy storage GES at t to t+1; δ t GES represents the electric quantity change value of the generalized energy storage GES caused by the on-grid and off-grid electric vehicles EV at t to t+1;

[0021] The demand response potential of the electric vehicle EV cluster is accurately characterized by a set of unified parameter sets, which are as follows:

[0022]

[0023] In formula (16), the parameter α GES is the relationship between the electric quantity at the next time and the electric quantity at the current time, that is, the equivalent value of the self-loss coefficient of the electric quantity at the corresponding time; the parameter β GES is the equivalent value of the charging efficiency of all on-grid electric vehicles EV at the corresponding time; γ GES is the reciprocal of the equivalent value of the discharging efficiency of all on-grid electric vehicles EV at the corresponding time; δ GES is the sum of the electric quantities of the off-grid and on-grid electric vehicles EV at the corresponding time; the parameters and are respectively the sum of the upper and lower limits of the charging / discharging power of all on-grid electric vehicles EV at the corresponding time; the parameters and are respectively the sum of the upper and lower limits of the electric quantity of all on-grid electric vehicles EV at the corresponding time.

[0024] Preferably, in step S3, the operating characteristics of the electric vehicle EV are divided into three categories: electric vehicle EV parameters, user travel plans, and uncertainty factors; wherein the electric vehicle EV parameters include vehicle parameters and real-time electric quantity The user travel plan includes the expected plan, i.e., the grid-connected time and the expected off-grid time of the electric vehicle (EV) user and the power requirement The uncertainty factors include external environment, weather, season, date and time-of-use electricity price; and a data-driven method is used to predict the parameters of the electric vehicle (EV) cluster aggregation model.

[0025] More preferably, the data-driven method in step S3 uses a quantum machine learning method to predict the parameters;

[0026] Step S3 includes the following sub-steps:

[0027] Step S3-1: Variational quantum circuit

[0028] Quantum encoding layer: encode the classical data into quantum state, and convert the quantum bit state from the reference state |0> to the target state through U(x) operation;

[0029] Variational layer: V(θ) operation entangles the quantum bits through multiple quantum gates with adjustable parameters and rotates to the target state;

[0030] Quantum measurement layer: measure the quantum state of part or all quantum bits on the basis of quantum computing, and output the calculation result of VQC;

[0031] Step S3-2: Quantum long short-term memory neural network

[0032] Replace the classical neural network in the LSTM basic unit with VQC, thereby extending it to the quantum field, and the main function of VQC is feature extraction and data compression; the calculation formula of QLSTM is as follows:

[0033] f t =σ(VQC1([h t-1 ,x t ]))(17)

[0034] i t =σ(VQC2([h t-1 ,x t ]))(18)

[0035] o t =σ(VQC4([h t-1 ,x t ]))(19)

[0036]

[0037] h t =VQC5(o t *tanh(c t ))(22)

[0038] y t = VQC6(o t *tanh(c t )) (23)

[0039] In formulas (17)-(23), f t , i t and o t represent the output vectors of the forget gate, the input gate and the output gate at time t, respectively; sigma represents a sigmoid activation function, which maps the corresponding variable to the interval [0, 1]; h t-1 represents the state vector of the hidden layer at time t-1; x t represents the sequence input vector at time t; represents the input unit state, i.e., the candidate state vector input to the memory unit at time t; tanh represents a hyperbolic tangent activation function; c t represents the state vector of the memory unit at time t; * represents element-wise multiplication of the corresponding matrices; y t represents the final output;

[0040] Step S3-3: optimization process;

[0041] In the optimization process, the gradient of the VQC is calculated using the parameter offset rule, and the calculation expression is:

[0042]

[0043] In formula (24), x represents an input value; theta i represents the quantum parameter of the i-th VQC circuit; f(·) represents the output with respect to the input and the quantum parameter;

[0044] The loss function is minimized by backpropagating the gradient between the VQC and the classical neural network, thereby iteratively optimizing the classical-quantum hybrid computing model;

[0045] Step S3-4: bidirectional quantum long short-term memory neural network

[0046] The Bi-QLSTM is selected as the framework for parameter prediction;

[0047] The Bi-QLSTM is composed of two QLSTMs with the same structure but opposite directions. The output result of the Bi-QLSTM contains the past and future information in the input, and the weights are shared during the training process, thereby improving the generalization ability of the model without increasing the amount of data.

[0048] Preferably, in the step S4, the demand response willingness in the electric vehicle EV operation characteristics is increased, and the demand response willingness includes: an incentive mechanism and a user psychology; the incentive mechanism is a user's benefit of participating in demand response, that is, a product of an interactive power and a time-of-use price; the user psychology is a state of charge of the electric vehicle EV and a user's recognition degree of demand response; a user demand response willingness coefficient is calculated through a model driven calculation, so as to obtain an actual adjustable capacity of the electric vehicle EV cluster.

[0049] More preferably, in the step S4, the user demand response willingness coefficient is calculated through a model driven calculation, and a TSK fuzzy system is used to quantify the user demand response willingness.

[0050] The step S4 includes the following sub-steps:

[0051] The step of the user side demand response willingness quantification model based on the TSK fuzzy system is as follows:

[0052] Step S4-1: a d-dimensional input vector x=[x1, x2, …, x d ] T A k,i ; wherein d=3, and the input vectors respectively represent a battery SOC, an incentive price and a demand response recognition degree, and the corresponding fuzzy subsets are "SOC low", "SOC medium", "SOC high", "price low", "price medium", "price high" and "degree low", "degree medium", "degree high";

[0053] Step S4-2: a Gaussian distribution is selected as a membership function of the input vector, and the battery SOC, the incentive price and the demand response recognition degree are respectively fuzzified, that is, the corresponding membership function values are taken as inputs of the TSK fuzzy rule; the Gaussian membership function can be expressed as:

[0054]

[0055] In formula (25), A k,i and σ k,i respectively represent a mean value and a standard deviation of the corresponding membership function;

[0056] Step S4-3: according to the fuzzy subsets of the input vector and the "IF THEN" fuzzy rule base, fuzzy reasoning is performed to determine a fuzzy set of an output variable; wherein K=27, and a general expression form is as follows:

[0057]

[0058] In formula (26), represents the function expression of the kth fuzzy rule; q k,i represents the parameter corresponding to the ith dimension input vector in the kth fuzzy rule; q k,0 represents the bias parameter of the kth fuzzy rule;

[0059] Step S4-4: calculating the final response willingness function according to the fuzzy set of the output variable; the demand response willingness output of the TSK fuzzy system after defuzzification is represented as:

[0060]

[0061] In formula (27), (28), (29), W EV represents the demand response willingness quantization result of the electric vehicle EV user; and respectively represent the membership function and the normalized membership function corresponding to the kth fuzzy rule.

[0062] The beneficial effects of the present application are:

[0063] 1. The present application considers the heterogeneity of electric vehicles to unify individual energy storage modeling, and encapsulates the cluster as a general aggregation model of generalized energy storage resources, so as to accurately represent the demand response potential of the electric vehicle cluster with a smaller number of parameters without affecting the model accuracy.

[0064] 2. The present application considers the characteristics of massive high-dimensional, interactive coupling, dynamic change and high nonlinearity of operation data when predicting the aggregation model parameters of the electric vehicle cluster, adopts the Bi-QLSTM model to better represent the potential structure of high-dimensional data in Hilbert space, uses the superposition, entanglement and parallel characteristics of quantum states to improve the learning and representation ability of coupled data complex relationship and accelerate the training process, thereby improving the overall performance of the model.

[0065] 3. The present application fully considers the influence of the state of charge of the battery of the electric vehicle, the incentive price and the user's demand response recognition degree on the demand response willingness when evaluating the adjustable capacity of the electric vehicle cluster, selects the TSK fuzzy system with excellent nonlinear approximation ability to accurately quantify the demand response willingness of the user. BRIEF DESCRIPTION OF DRAWINGS

[0066] Fig. 1 is a schematic diagram of a method for evaluating the adjustable capacity of an electric vehicle cluster considering the response willingness of a user according to the present application;

[0067] Fig. 2 is a general architecture diagram of a variational quantum circuit according to the present application;

[0068] Fig. 3is a variational quantum circuit architecture diagram of the QLSTM of the application;

[0069] Fig. 4 is a QLSTM neural network structure diagram of the application;

[0070] Fig. 5 is a Bi-QLSTM neural network structure diagram of the application;

[0071] Fig. 6 is a TSK fuzzy system structure diagram of the application. DETAILED DESCRIPTION

[0072] The related technologies in the present application will be described in detail below with the accompanying drawings of the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of the present application.

[0073] As shown in Figs. 1-6 , the embodiment provides an electric vehicle cluster adjustable capacity evaluation method considering user response willingness.

[0074] 1. Unified energy storage model of electric vehicle individual

[0075] In the present application, the charging mode of EV is slow charging.

[0076] Considering a cluster containing N EVs, due to the difference in grid-connected period of different EVs, the definition domain of different individuals is heterogeneous, and the specific form can be expressed as:

[0077]

[0078] Three kinds of changes of battery capacity of grid-connected EVs are considered. Firstly, at the grid-connected time of EV, the change of grid-connected battery capacity can be expressed as:

[0079]

[0080] wherein, and respectively represent the grid-connected battery capacity of the i-th EV before and after the grid-connected time; dt represents the short time period size of EV grid connection and off-grid; represents the battery capacity of the i-th EV at the grid-connected time.

[0081] Then, during the normal grid-connected period of EV, the change of grid-connected battery capacity can be expressed as:

[0082]

[0083] wherein, and and and and and and and and

[0084] Finally, at the off-grid time, the on-grid battery state of charge can be expressed as:

[0085]

[0086] where, and and

[0087] Therefore, the general form of the EV individual model can be expressed as:

[0088]

[0089]

[0090] where, and and and and and and and and

[0091] 2. Generalized Energy Storage Resource Aggregation Model for Electric Vehicle Cluster

[0092] Based on the unified energy storage model of EV individuals, the cluster is packaged as a generalized energy storage (GES) resource aggregation model. The model is as follows:​​

[0093]

[0094] wherein, and respectively represent the state of charge of the GES at time t and t+1; P t GES,cha represents the charging power of the GES during the period t to t+1; P t GES,dis represents the discharging power of the GES during the period t to t+1; represents the self-loss coefficient of the state of charge of the GES during the period t to t+1; represents the charging efficiency of the GES during the period t to t+1; represents the inverse of the discharging efficiency of the GES during the period t to t+1; represents the change value of the state of charge of the GES during the period t to t+1 caused by the EVs and off-grid; and respectively represent the upper / lower limit of the charging power of the GES during the period t to t+1; and respectively represent the upper / lower limit of the discharging power of the GES during the period t to t+1; and represent the upper / lower limit of the state of charge of the GES during the period t to t+1.

[0095] Therefore, the demand response potential of the EV cluster can be accurately characterized by a set of unified parameters, the aggregation method can characterize the demand response potential of the EVs with fewer parameters without affecting the accuracy of the model, and the EV cluster can be called and controlled in a unified manner. The parameter set is as follows:

[0096]

[0097] According to the individual model of the EV and the cluster aggregation model, the parameter a GES is the relationship between the state of charge at the next time and the state of charge at the current time, that is, the equivalent value of the self-loss coefficient of the state of charge at the corresponding time; the parameter b GES is the equivalent value of the charging efficiency of all grid-connected EVs at the corresponding time; g GES is the inverse of the equivalent value of the discharging efficiency of all grid-connected EVs at the corresponding time; d GES is the sum of the state of charge of the off-grid and grid-connected EVs at the corresponding time; the parameters and respectively represent the sum of the upper / lower limit of the charging / discharging power of all grid-connected EVs at the corresponding time; the parameters and respectively represent the sum of the upper / lower limit of the state of charge of all grid-connected EVs at the corresponding time.

[0098] 3. Bidirectional quantum long short-term memory driven evaluation method for adjustable capacity of electric vehicle cluster under consideration of user response willingness

[0099] Precise evaluation of the adjustable capacity of the EV cluster is very important for the power dispatch department to develop a reasonable dispatch plan according to the actual operation of the power grid, therefore, the present application further proposes a bidirectional quantum long short-term memory (Bi-QLSTM) driven evaluation method for the adjustable capacity of the electric vehicle cluster under consideration of the user response willingness, which comprehensively evaluates the actual demand response potential of the EV cluster according to the operation characteristics and user willingness of the EV cluster and through model-data hybrid driving.

[0100] The theoretical demand response potential of the EV cluster can be accurately characterized according to a set of unified parameters (formula 16) of the GES resource general aggregation model (formula 12). According to the physical meaning of the parameter set, it is closely related to the operation characteristics of the EV. The present application divides the operation characteristics of the EV into three categories: EV parameters, user travel plan and uncertainty factors. Among them, the EV parameters include vehicle parameters and real-time power The user travel plan includes the expected plan, i.e. the expected on-grid time and the expected off-grid time of the EV user and the power requirement At the same time, various uncertainty factors are considered, such as the actual plan, i.e. the actual off-grid time of the EV user, the external environment, i.e. the weather, the season, the date and the time-of-use electricity price, etc. Since the operation characteristics of the EV cluster are complex and difficult to model accurately, the present application adopts a data-driven method to predict the parameters of the aggregation model.

[0101] In addition, the adjustable capacity obtained by the data-driven link through the prediction of the aggregation model parameters does not consider the user's demand response willingness, and the result is the theoretical adjustable capacity of the EV cluster. In order to accurately depict the influence of the user's demand response willingness on the adjustable capacity, the present application considers the influence of two categories of factors: incentive mechanism and user psychology. Among them, the incentive mechanism is the benefit of the user participating in the demand response, i.e. the product of the interactive power and the time-of-use electricity price; the user psychology is the state of charge of the EV and the degree of recognition of the demand response by the user. Since the factors affecting the user's demand response willingness are relatively clear, the present application calculates the user's demand response willingness coefficient through model-driven calculation, so as to obtain the actual adjustable capacity of the EV cluster.

[0102] 3.1. Aggregation model parameter prediction model based on bidirectional quantum long short-term memory

[0103] EV operation characteristics are typical time series data, and the demand response potential of EV clusters is closely related to the nonlinear characteristics in time. Deep learning models can fully exploit the underlying features of data through the nonlinear transformation of hidden layers, extract abstract and easily distinguishable high-level features, and thus have excellent feature representation, data fitting and generalization capabilities. Among them, the Long Short-Term Memory (LSTM) neural network is an improved model of the Recurrent Neural Network (RNN), which introduces a gate control unit on the basis of RNN self-feedback, filters the redundant and misleading features in the time series through its unique memory and forgetting mode, and effectively controls the gradient flow, thereby realizing accurate prediction of time series of any length. However, in practical application, the operation data of EV clusters presents the characteristics of massive high-dimensional, interactive coupling, dynamic change and high nonlinearity, and the LSTM model may be difficult to represent and learn the complex relationships in the data, resulting in unstable training results and affecting the performance of the model.

[0104] As a new emerging computing paradigm, quantum computing has strong parallel computing ability and is considered as an effective way to improve the speed and performance of classical machine learning. Therefore, the data-driven link of the present application adopts a quantum machine learning method for parameter prediction. The hardware technology in the current quantum computing era is in the noisy medium-scale quantum stage, which is characterized by limited number of qubits and short decoherence time. Variational quantum algorithms have been proven to have natural noise resistance and even benefit from noise, making them particularly suitable for recent quantum devices. Therefore, a quantum-classical hybrid computing framework is built through a variational quantum circuit (VQC) to realize quantum long short-term memory (QLSTM).

[0105] The core of the quantum-classical hybrid computing framework is to convert the computing task into an optimization problem. This algorithm combines the computing power of quantum computers with the optimization methods of classical computers. Data processing is performed on the quantum computer through the VQC, and the loss function is calculated through quantum measurement. The adjustable parameters θ are optimized on the classical computer, and the computing task is completed when the loss function is minimized / maximized.

[0106] QLSTM can better represent the potential structure of high-dimensional data in Hilbert space and improve the learning and representation capabilities of complex data relationships while reducing the number of training parameters and accelerating the training process by utilizing the superposition, entanglement and parallel characteristics of quantum states, thereby improving the overall performance of the model. Therefore, the QLSTM model can more accurately predict the adjustable capacity of the EV cluster and provide a basis for the scheduling and management of EVs.

[0107] 1) Variational quantum circuit

[0108] Quantum circuit is a kind of quantum computing model, which is composed of multiple quantum bits, quantum gates and quantum measurement modules. Its computing process is to store the input data in the form of quantum information by multiple quantum bits, then operate the state of quantum bits through a series of quantum gates, and finally convert it into classical information through quantum measurement link and output the calculation result. VQC is a special quantum circuit, which has variable parameters and can be continuously iterated and optimized. Its architecture is shown in Fig. 2 、 3 , which is mainly composed of three parts.

[0109] Quantum encoding layer: encode classical data into quantum state, and convert the state of quantum bits from reference state |0> to target state through U(x) operation. Usually one quantum bit encodes one classical input feature.

[0110] Variational layer: V(θ) operation entangles and rotates the quantum bits to the target state through multiple quantum gates with adjustable parameters. This part is the actual learnable part of VQC, which is continuously optimized in the iteration process.

[0111] Quantum measurement layer: measure the quantum state of part or all quantum bits based on quantum computing, and output the calculation result of VQC. The result of this part measurement is the expected value of each quantum bit.

[0112] 2) Quantum long short-term memory neural network

[0113] The classical neural network in the LSTM basic unit is replaced by VQC in the present application, so as to extend it to the quantum field. The main function of VQC is feature extraction and data compression. The basic structure of QLSTM is shown in Fig. 4 , and the calculation formula is as follows:

[0114] f t =σ(VQC1([h t-1 ,x t ])) (17)

[0115] i t =σ(VQC2([h t-1 ,x t ])) (18)

[0116] o t =σ(VQC4([h t-1 ,x t ])) (19)

[0117]

[0118] ht = VQC5(o t tanh(c t )) (22)

[0119] y t = VQC6(o t tanh(c t )) (23)

[0120] where f t , i t and o t denote the forget gate, input gate and output gate at time t, respectively; σ denotes the sigmoid activation function, which maps the corresponding variable to the interval [0, 1]; h t-1 denotes the state vector of the hidden layer at time t-1; x t denotes the sequence input vector at time t; denotes the input unit state, i.e., the candidate state vector input to the memory unit at time t; tanh denotes the hyperbolic tangent activation function; c t denotes the state vector of the memory unit at time t; * denotes element-wise multiplication of the corresponding matrices; y t denotes the final output.

[0121] 3) Optimization process

[0122] In the optimization process, the present application uses a parameter offset rule to calculate the gradient of VQC, and the calculation expression is:

[0123]

[0124] where x denotes the input value; θ i denotes the quantum parameter of the i-th VQC circuit; f(·) denotes the output with respect to the input and quantum parameter.

[0125] Therefore, the loss function can be minimized by backpropagating the gradient between VQC and the classical neural network, thereby iteratively optimizing the classical-quantum hybrid computing model.

[0126] 4) Bidirectional quantum long short-term memory neural network

[0127] The running characteristics of the EV cluster not only depend on historical data, but also are related to future data, having the property of bidirectional time series. In order to better learn and represent the complex relationships in the data sequence, the present application selects Bi-QLSTM as the framework for parameter prediction.

[0128] Bi-QLSTM is composed of two QLSTMs with the same structure and opposite directions, and its structure is as follows: Fig. 5The basic idea of this method is to process the input through QLSTM in forward and backward order respectively, and then splice the results of the two-way processing to obtain the final output. The output result of Bi-QLSTM contains the information of the past and future of the input, and the weight is shared during the training process, which improves the generalization ability of the model without increasing the data volume.

[0129] 3.2. User side demand response willingness quantification model based on TSK fuzzy system

[0130] TSK fuzzy system is widely used in system identification, pattern recognition and data processing due to its excellent nonlinear approximation ability. The present application considers the influence of EV battery state of charge (SOC), incentive price and user demand response recognition degree on demand response willingness, selects this method to quantify the demand response willingness of users, and evaluates the actual adjustable capacity of EV cluster. According to Fig. 6 It can be seen that the steps of the user side demand response willingness quantification model based on TSK fuzzy system are as follows:

[0131] Step 1: First, given a d-dimensional input vector x = [x1, x2, …, x d ] T and the fuzzy subset A k,i corresponding to the i-dimensional input vector in the kth fuzzy rule. In the present application, d = 3, and the input vectors represent battery SOC, incentive price and demand response recognition degree, respectively, and their corresponding fuzzy subsets are "low SOC", "medium SOC", "high SOC", "low price", "medium price", "high price" and "low degree", "medium degree", "high degree".

[0132] Step 2: Select Gaussian distribution as the membership function of the input vector, and fuzzy the battery SOC, incentive price and demand response recognition degree, i.e. take the corresponding membership function value as the input of TSK fuzzy rule. The Gaussian membership function can be expressed as:

[0133]

[0134] where, represents the Gaussian distribution membership function of the i-dimensional input vector in the kth fuzzy rule; c k,i and σ k,i respectively represent the mean and standard deviation of the corresponding membership function.

[0135] Step3: According to the fuzzy subset of the input vector and the "IF THEN" fuzzy rule base, fuzzy reasoning is carried out to determine the fuzzy set of the output variable. According to the dimension of the input vector and the number of fuzzy subsets, the number of TSK fuzzy system reasoning rules in the present application is K = 27, and the general expression form is:

[0136]

[0137] Among them, The function expression of the kth fuzzy rule is represented as q k,i The parameter corresponding to the ith dimension input vector in the kth fuzzy rule is represented as q k,0 The bias parameter of the kth fuzzy rule is represented as q

[0138] Step4: The final response willingness function is calculated according to the fuzzy set of the output variable. The demand response willingness output of the TSK fuzzy system after de-fuzzification can be expressed as:

[0139]

[0140] Among them, W EV The quantitative result of the demand response willingness of the EV user is represented as q And The membership function and the normalized membership function corresponding to the kth fuzzy rule are represented as q

[0141] The embodiment considers the heterogeneity of electric vehicles to model individuals uniformly, and establishes a general aggregation model of generalized energy storage resources accordingly;

[0142] The embodiment establishes an electric vehicle aggregation model parameter prediction model based on the Bi-QLSTM model, which better represents and learns the complex relationship in the data and accelerates the training efficiency by using the superposition, entanglement and parallel characteristics of quantum states, and improves the overall performance of the model;

[0143] The embodiment establishes a user-side demand response willingness quantification model based on the TSK fuzzy system, which realizes accurate quantification of the demand response willingness of users.

[0144] In summary, the present application considers the heterogeneity of electric vehicles to model individuals uniformly, and encapsulates it as a general aggregation model of generalized energy storage resources, so as to accurately represent the demand response potential of the electric vehicle cluster with a smaller number of parameters without affecting the model accuracy. Therefore, the present application has a wide application prospect in the aspects of strong randomness and volatility of smooth new energy power generation mode.

[0145] It should be pointed out that the above is only the preferred embodiment of the present application, and does not limit the present application in any form. Any simple modification, equivalent change and modification of the above embodiment according to the technical essence of the present application still belongs to the scope of the technical solution of the present application.

Claims

1.A method for evaluating adjustable capacity of an electric vehicle cluster considering user response willingness, characterized in that, The method comprises the following steps: Step S1, establishing a unified energy storage model of an electric vehicle EV individual; Step S2, establishing a generalized energy storage GES resource general aggregation model of an electric vehicle EV cluster; Step S3, predicting a theoretical demand response potential based on a bidirectional quantum long short-term memory neural network prediction electric vehicle EV cluster aggregation model parameter set; Step S4, quantifying a user side demand response willingness based on a TSK fuzzy system to obtain a demand response willingness coefficient; Step S5, combining data and model driven results to obtain an actual demand response potential; In the step S3, a quantum machine learning method is used for parameter prediction; The step S3 comprises the following sub-steps: Step S3-1: a variational quantum circuit; A quantum encoding layer: encoding classical data into a quantum state, and converting the quantum bit state from a reference state |0> to a target state through a U(x) operation; A variational layer: a V(θ) operation entangles quantum bits through a plurality of quantum gates with adjustable parameters and rotates to a target state; A quantum measurement layer: measuring the quantum state of part or all quantum bits on the basis of quantum computing, and outputting the calculation result of the VQC; Step S3-2: a quantum long short-term memory neural network The classical neural network in the LSTM basic unit is replaced by a VQC, which is extended to the quantum field, and the main function of the VQC is feature extraction and data compression; the calculation formula of the QLSTM is as follows: f t = σ(VQC1([h t-1 ,x t ])) (17) i t = σ(VQC2([h t-1 ,x t ]))(18) o t = σ(VQC4([h t-1 ,x t ]))(19) h t = VQC5(o t tanh(c t ))(22) y t =VQC6(o t *fishy(c) t )) (23) In formulas (17)-(23), f t , i t , and o t represent the output vectors of the forget gate, the input gate, and the output gate at time t, respectively; σ represents a sigmoid activation function that maps the corresponding variable to the interval [0, 1]; h t-1 represents the state vector of the hidden layer at time t-1; x t represents the sequence input vector at time t; represents the input unit state, i.e., the candidate state vector input to the memory unit at time t; tanh represents a hyperbolic tangent activation function; c t represents the state vector of the memory unit at time t; * represents element-wise multiplication of the corresponding matrices; y t represents the final output; Step S3-3: an optimization process; In the optimization process, the gradient of the VQC is calculated by using a parameter offset rule, and the calculation expression is as follows: In Equation (24), x represents an input value; θ i represents a quantum parameter of the i-th VQC circuit; f(·) represents an output with respect to an input and a quantum parameter; The loss function is minimized by back-propagating the gradient between the VQC and the classical neural network, so as to iteratively optimize the classical-quantum hybrid computing model; Step S3-4: a bidirectional quantum long short-term memory neural network The Bi-QLSTM is selected as the framework for parameter prediction; The Bi-QLSTM is composed of two QLSTMs with the same structure and opposite directions, and the output result of the Bi-QLSTM contains the past and future information in the input, and the weight values are shared in the training process, thereby improving the generalization ability of the model without increasing the amount of data; In the step S4, the model driven calculation user demand response willingness coefficient adopts a TSK fuzzy system to quantify the demand response willingness of the user; The step S4 comprises the following sub-steps: The step of the user side demand response willingness quantification model based on the TSK fuzzy system is as follows: Step S4-1: Given a d-dimensional input vector x = [x1, x2,..., xd]T d ] T Aikis the fuzzy subset corresponding to the ith-dimensional input vector in the kth fuzzy rule k,i ; where d = 3, and the input vectors represent battery SOC, incentive price, and demand response acceptance level, respectively, and their corresponding fuzzy subsets are "SOC low", "SOC medium", "SOC high", "price low", "price medium", "price high", and "level low", "level medium", "level high", respectively. Step S4-2: a Gaussian distribution is selected as the membership function of the input vector, and the battery SOC, the incentive price and the demand response recognition degree are fuzzified respectively, that is, the membership function values corresponding to the three are taken as the inputs of the TSK fuzzy rule; the Gaussian membership function is expressed as: In formula (25), represents the Gaussian type distribution membership function of the ith dimension input vector in the kth fuzzy rule; c k,i and σ k,i respectively represent the mean and standard deviation of the corresponding membership function; Step S4-3: according to the fuzzy subsets of the input vector and the "IF THEN" fuzzy rule base, the fuzzy set of the output variable is determined; wherein K=27, and the general expression form is as follows: In formula (26), a function expression representing the kth fuzzy rule; q k,i a parameter corresponding to the ith dimension of the input vector in the kth fuzzy rule; q k,0 a bias parameter of the kth fuzzy rule; Step S4-4: the final response willingness function is calculated according to the fuzzy set of the output variable; the demand response willingness output of the TSK fuzzy system after defuzzification is expressed as: In formulas (27), (28), and (29), W EV It represents the quantitative results of the demand response willingness of electric vehicle (EV) users; and represent the membership function and normalized membership function corresponding to the kth fuzzy rule respectively. 2.The method of claim 1, wherein, In the step S1, a general form of the electric vehicle EV individual model is represented as: in formula (5), and respectively represent the grid-connected battery power of the i-th electric vehicle EV at time t and time t+1; and respectively represent the battery power of the i-th electric vehicle EV at the grid-connected time and the off-grid time; and respectively represent the grid-connected and off-grid states of the i-th electric vehicle EV at time t and time t+1, and are Boolean variables, 0 representing off-grid and 1 representing grid-connected; and respectively represent the charging and discharging efficiencies of the i-th electric vehicle EV; and respectively represent the charging / discharging states of the i-th electric vehicle EV in the period from t to t+1, and are Boolean variables, 0 representing that the electric vehicle EV is not charging / discharging and 1 representing that the electric vehicle EV is charging / discharging; and respectively represent the charging / discharging powers of the i-th electric vehicle EV in the period from t to t+1; Δt represents the unit period size of the charging / discharging of the electric vehicle EV. 3.The method of claim 1, wherein, In the step S2, on the basis of the electric vehicle EV individual unified energy storage model, the cluster is packaged as a general aggregation model of generalized energy storage GES resource, and the model is: In formula (12), and respectively represent the electric quantity of the generalized energy storage GES at time t and time t+1; P t GES,cha represents the charging power of the generalized energy storage GES during the period from t to t+1; P t GES,dis represents the discharging power of the generalized energy storage GES during the period from t to t+1; represents the self-loss coefficient of the electric quantity of the generalized energy storage GES during the period from t to t+1; represents the charging efficiency of the generalized energy storage GES during the period from t to t+1; represents the reciprocal of the discharging efficiency of the generalized energy storage GES during the period from t to t+1; represents the electric quantity change value of the generalized energy storage GES caused by the on-grid and off-grid of the electric vehicle EV during the period from t to t+1; The demand response potential of the electric vehicle EV cluster is accurately characterized by a set of unified parameters as follows: In formula (16), parameter α GES is the relationship between the power at the next time and the power at the current time, that is, the equivalent value of the self-loss coefficient of the power at the corresponding time; parameter β GES is the equivalent value of the charging efficiency of all grid-connected electric vehicles EVs at the corresponding time; γ GES is the inverse of the equivalent value of the discharging efficiency of all grid-connected electric vehicles EVs at the corresponding time; δ GES is the sum of the powers of all grid-connected and off-grid electric vehicles EVs at the corresponding time; parameters and are respectively the sum of the upper / lower limits of the charging / discharging power of all grid-connected electric vehicles EVs at the corresponding time; parameters and are respectively the sum of the upper / lower limits of the power of all grid-connected electric vehicles EVs at the corresponding time. 4.The method of claim 1, wherein, In the step S3, the electric vehicle EV operation characteristics are classified into three categories: electric vehicle EV parameters, user travel plan and uncertainty factors; wherein the electric vehicle EV parameters include vehicle parameters and real-time power The user travel plan includes expected plan, i.e. the expected on-grid time and off-grid time of the electric vehicle EV user and power requirement The uncertainty factors include external environment, weather, season, date and time-of-use electricity price; and a data-driven method is used to predict the parameters of the electric vehicle EV cluster aggregation model. 5.The method of claim 1, wherein, In the step S4, the demand response willingness is added in the electric vehicle EV operation characteristics, and the demand response willingness includes: incentive mechanism and user psychology. The incentive mechanism is the benefit of the user participating in the demand response, that is, the product of the interactive power and the time-of-use price. The user psychology is the state of charge of the electric vehicle EV and the recognition degree of the user to the demand response. The demand response willingness coefficient of the user is calculated through the model driven calculation, so as to obtain the actual adjustable capacity of the electric vehicle EV cluster.

Images