Electric vehicle schedulable potential probability assessment method and system based on user charging behavior portrait

Abstract

The invention discloses an electric vehicle schedulable potential probability assessment method and system based on a user charging behavior portrait, and the method comprises the steps: extracting user charging behavior characteristics through employing EV user real charging data and vehicle data, and constructing a multi-dimensional charging behavior portrait label system; dividing the users into different clusters based on a classification decision model to obtain classification results of the users; for the uncertainty of the EV charging behavior, the constraint condition is probabilized under the given confidence level, and a single EV energy probability feasible region and probability schedulable power under different confidence levels are obtained; carrying out approximation on the energy feasible region of the monomer EV by utilizing an improved chirnol polyhedron approximation method; minkowski summation is carried out on the approximated energy probability feasible regions of the clusters, and the schedulable power potential of the EV clusters is obtained by directly summing the probability schedulable power potentials of the clusters under different confidence coefficients by the users.

Description

Technical Field

[0001] This invention relates to the field of electric vehicle dispatchability potential assessment, and in particular to a method and system for probabilistic assessment of electric vehicle dispatchability potential based on user charging behavior profiles. Background Technology

[0002] With the acceleration of global energy transition and transportation electrification, the number of electric vehicles (EVs) is gradually increasing. Data from the International Energy Agency (IEA) shows that global EV sales reached 17 million units by 2024 and are projected to exceed 20 million units by 2025. EVs are not only electricity consumers, but their bidirectional power regulation characteristics also give them enormous potential in grid regulation. Accurately assessing their potential as "mobile energy storage units" in grid regulation is of great significance for clean energy development and the coordinated low-carbon development of electricity and transportation.

[0003] The dispatchable potential of electric vehicles (EVs) is influenced by users' charging habits. Traditional assessments of EV dispatchable potential often rely on the assumption that "user behavior is driven by objective conditions," failing to adequately consider psychological effects and leading to a disconnect between assessment results and real-world scenarios. Furthermore, when approximating the feasible region of EVs, existing methods are limited to treating dispatchable capacity at different times as physically equivalent boundaries, failing to reflect the economic value differences across time-of-use pricing. Simultaneously, EV users' travel patterns and charging / discharging needs are random, but existing methods employ deterministic boundaries that cannot quantify the probabilistic fluctuations in EV charging and discharging behavior. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method and system for probabilistically assessing the dispatchable potential of electric vehicles based on user charging behavior profiles. The assessment method and system apply user profiles, which consider user psychology, to characterize charging behavior and capture the psychological decision-making logic behind it. Simultaneously, by employing a probabilistic assessment method to evaluate the dispatchable potential of EVs and characterize boundary fluctuations, it can effectively assess the dispatchable potential of different user clusters of electric vehicles, which is beneficial for promoting clean energy and the coordinated development of electricity and low-carbon transportation.

[0005] The first aspect of this invention is to provide a method for probabilistically assessing the dispatchable potential of electric vehicles based on user charging behavior profiles, comprising:

[0006] Step 1: Extract user charging behavior features that describe user charging habits using real charging data and vehicle data of EV users in the area to be predicted. Refine and integrate the charging behavior features from different dimensions to build a multi-dimensional charging behavior profile label system.

[0007] Step 2: Using the multi-dimensional charging behavior profile tag system built in Step 1, the information gain of each profile tag is used to measure the degree of influence of each dimension profile tag in Step 1 on user classification; the classification decision model is trained and tested to obtain the user classification results.

[0008] Step 3: Based on the user charging behavior characteristics obtained in Step 1 and the user classification results obtained in Step 2, for users of the same category, obtain the feasible energy domain and schedulable power potential of a single EV unit based on the virtual battery model. Considering the uncertainty of EV charging behavior, the constraints are probabilistically converted at a given confidence level to obtain the probabilistic feasible energy domain and probabilistic schedulable power of a single EV unit under different confidence levels. Repeat the above steps to obtain the feasible energy domain of a single EV unit to obtain the feasible energy domain of EV units for all users.

[0009] Step 4: Approximate the individual EV energy probability feasible region obtained in Step 3 using the improved Chino polyhedron method. Solve to maximize the similarity between the spatial region defined by the Chino polyhedron and the EV dispatchable energy feasible region. At the same time, in order to cope with the different requirements of the power grid and the electricity market for the sustainability duration and dispatchable energy of flexibility resources at different times, consider the two dimensions of time value and energy value at different times. By setting different weights for the diameter in the approximation region in the time and energy dimensions respectively, the feasible region for different time value and energy value time periods is approximated. The improved Chino polyhedron weighted approximation is completed for the energy probability feasible region of all EVs.

[0010] Step 5: After applying the Chino polyhedral weighted approximation to the energy feasible region of all EV users, perform Minkowski summation on users belonging to the same cluster. The aggregated energy feasible region can be represented by the center and scaling factor of the Chino polyhedron.

[0011]

[0012]

[0013]

[0014]

[0015] In the formula: The cluster aggregation energy approximates the feasible region; Summing for Minkowski; Let be the approximate feasible energy region for user i in cluster j at confidence level σ. and These represent the center and scaling factor of the aggregated Chino polyhedron under different confidence levels;

[0016] The probabilistic schedulable power potential of cluster j at different confidence levels can be obtained by directly summing the power of each user:

[0017]

[0018]

[0019] In the formula: and The up-adjustment and down-adjustment power potentials of cluster j at time t under confidence level σ are respectively represented; thus, the schedulable power potential of the EV cluster is obtained.

[0020] Based on the dispatchable energy probability obtained in step four and the dispatchable power potential obtained in step five, operators are guided to use the electricity pricing mechanism to encourage some users to reduce charging power in order to reduce grid load.

[0021] In one embodiment, the EV user's real charging data in step one includes vehicle battery capacity, average charging amount per charge, vehicle charging status parameters, vehicle charging time point and / or vehicle charging location; the travel data includes average daily mileage and average daily number of trips; and the vehicle attribute data includes vehicle identification, vehicle purpose and / or vehicle energy consumption.

[0022] In another embodiment, step one specifically includes:

[0023] The steps for extracting user charging behavior features are as follows: The acquired real charging data and vehicle data of EV users are cleaned and anomaly processed. By identifying the transition nodes between charging status and vehicle dwell status, segments are created, dividing continuous data into independent charging event segments. Each segment corresponds to a complete user charging behavior process, thereby extracting user charging behavior features. These features include: charging start time, end time, arrival time, departure time, charging duration, dwell time, idle time, charging power, battery state of charge (SOC) at charging start, SOC at charging end, charging location, and percentage of charged charge.

[0024] The steps to build a multi-dimensional charging behavior profile tag system are as follows: Extract and integrate user charging behavior characteristics from different dimensions to build five types of profile tags: user charging time, user charging space, power preference, electricity price sensitivity and range anxiety, thereby building a multi-dimensional charging behavior profile tag system.

[0025] Among them, the user charging time tag uses a clustering algorithm to directly cluster the arrival and departure times to divide users, thereby representing the user charging time tag;

[0026] User charging space labeling considers whether the user's charging location crosses urban areas, the impact of the reference date feature weight on the user's cross-city charging behavior, and quantifies the number of cross-city charging times to characterize the user's charging space label.

[0027] Power preference labels are defined by statistically analyzing the distribution of user charging power across different power ranges and determining user type based on the proportion of fast and slow charging cycles.

[0028] Electricity price sensitivity labels are calculated by analyzing users’ charging behavior under different electricity price environments. Then, based on the numerical values, a clustering algorithm is used to classify the electricity price sensitivity labels to represent users’ electricity price sensitivity labels.

[0029] The range anxiety label measures a user's range anxiety level by measuring the initial state of charge (SOC) of the EV. Then, a clustering algorithm is used to determine the degree of the user's range anxiety label based on the numerical value, thus representing the range anxiety label.

[0030] Specifically, based on observations of user charging time tags, users are categorized into: nighttime charging, daytime charging, random charging, and emergency charging.

[0031] In the power preference label, the user's power preference label is represented by slow charging user, fast charging user, or fast and slow charging balanced user;

[0032] In the electricity price sensitive label, an electricity price sensitivity index is constructed based on the proportion of a user's EV charging electricity during off-peak hours to the total charging electricity. The index is then clustered based on the value and divided into electricity price sensitive users and electricity price insensitive users.

[0033] In the range anxiety label, the user's range anxiety level is measured by the initial SOC of EV charging. The user's initial SOC value is divided into multiple intervals according to the battery level from low to high. Different weights are assigned to each interval based on the charging frequency of each interval. This is used to calculate the user's range anxiety level, and clustering is performed based on the range anxiety level to classify the user's range anxiety label.

[0034] In one embodiment, step two specifically includes:

[0035] S21: Using the multi-dimensional charging behavior profile label system constructed in step one, the information gain of each profile label is used to measure the degree of influence of each dimension profile label in step one on user classification; the root node dataset D is divided into multiple subsets based on the different values ​​of the feature with the largest information gain, and each subset corresponds to a branch of the decision tree.

[0036] S22: Repeat the feature selection step S21 for each newly generated subset to generate a new branch; iterate this process until all samples in the subset belong to the same category and then stop iterating.

[0037] S23: Remove branches with information gain lower than a preset gain threshold to form an updated classification decision tree; the classification decision tree divides user charging behavior into subdivided groups with clear feature labels, and each leaf node corresponds to a typical user profile.

[0038] In one embodiment, step S21 specifically includes:

[0039] The information gain of the profile labels from step one is used to measure the influence of each profile label on user classification in each dimension; specifically, for a dataset D with J categories, the information entropy is expressed as:

[0040]

[0041] In the formula: D is the dataset; C j Let be the set of samples belonging to class j in the dataset; This indicates the number of samples in the set.

[0042] Assume feature A m (m=1,2,...,5,) represent the five dimensions of the profile labels in step one, respectively. There are n... m Different values According to feature A m The dataset D can be divided into n m Subset Then the conditional entropy is:

[0043]

[0044] Finally, based on the information entropy and conditional entropy, the information gain g(D,A) can be obtained. m The specific calculation is shown in the following formula; the greater the information gain, the greater the contribution of the feature to the classification:

[0045]

[0046] Calculate all features A m The information gain is used to find the feature with the largest information gain. The root node dataset D is divided into multiple subsets according to the different values ​​of this feature, and each subset corresponds to a branch of the decision tree.

[0047] In one embodiment, step three specifically includes:

[0048] S31: Based on the user behavior characteristics obtained in step one and the user classification results obtained in step two, characterize the dispatchable capacity and dispatchable power boundary of different types of individual EVs; whereby the dispatchable capacity of EVs is formed by the dual constraints of charging and discharging capacity and time to form a two-dimensional energy feasible region, and the dispatchable power boundary of EVs includes the potential for upward adjustment of power and the potential for downward adjustment of power. The potential for upward adjustment of power is when the EV increases the charging power or decreases the discharging power, resulting in an increase in the equivalent grid load; the potential for downward adjustment of power is when the EV decreases the charging power or increases the discharging power, resulting in a decrease in the equivalent grid load.

[0049] S32: Based on the schedulable capacity and schedulable power boundaries obtained in step S31, calculate the probability distribution of key variables of the energy feasible region and schedulable power potential. Reconstruct the constraints at a given confidence level, expanding them to the individual EV energy probabilistic feasible region represented by the number of intervals for the lower and upper limits of energy at a given confidence level σ, and further expanding them to the individual EV probabilistic schedulable power represented by intervals for the upward and downward power potentials at a given confidence level σ; the key variables include: the charging power of user i in cluster j. Remaining battery power at the time of EV network access Arrival time at the charging location Time of leaving the charging location .

[0050] In another embodiment, step three, calculating the EV schedulable capacity and EV schedulable power boundary, includes:

[0051] The schedulable capacity of an EV is represented by a convex polyhedron constrained by a set of inequalities, forming a two-dimensional energy feasible region by the dual constraints of charge / discharge capacity and time:

[0052]

[0053] In the formula: Let be the energy feasible region for user i under cluster j. Let be the set of points within the feasible region; and These are the coefficient matrix and the constant column vector of the inequality, respectively;

[0054] The remaining battery capacity of EV user i under cluster j at time t can be calculated from the actual charging and discharging power:

[0055]

[0056] In the formula: The remaining battery level at the time of EV registration; The time of network access; The charging or discharging power at time t;

[0057] The up-adjustment and down-adjustment potentials of EV dispatchable power are as follows:

[0058] Based on the transitions between charging, discharging, and idle states of an EV, the interaction modes between the EV and the power grid are categorized into four types: Mode 1 (idle → charging) and Mode 4 (discharging → idle) correspond to upward adjustment potential; Mode 2 (charging → idle) and Mode 3 (idle → discharging) correspond to downward adjustment potential. The dispatchable power potential is related to the charging and discharging states. The upward and downward adjustment potentials of an EV at a charging location are calculated as follows:

[0059]

[0060]

[0061] In the formula: and This represents the up and down potential of the i-th user in the j-th cluster at time t, where the down potential is a negative value; , , These are the state variables of the EV at time t when it is in the idle, charging, and discharging states. The corresponding state variable is 1 only when it is in the corresponding state, and 0 otherwise.

[0062] A second aspect of the present invention is to provide a probabilistic assessment system for the dispatchable potential of electric vehicles based on user charging behavior profiles, comprising:

[0063] The module for constructing a multi-dimensional profile tagging system utilizes real charging data and vehicle data of EV users in the area to be predicted to extract user charging behavior features that can describe user charging habits. These charging behavior features are then refined and integrated from different dimensions to construct a multi-dimensional charging behavior profile tagging system.

[0064] Clustering module using classification decision model: A multi-dimensional charging behavior profile label system is constructed using the multi-dimensional profile label system module. The information gain of each profile label is used to measure the influence of each dimension profile label in the multi-dimensional profile label system module on user classification. Based on the classification decision model, training and testing are performed to obtain the user clustering results.

[0065] The module for obtaining the probabilistic feasible region and probabilistic schedulable power of a single EV is as follows: Based on the user charging behavior characteristics obtained by constructing a multi-dimensional profile tag system module and the user classification results based on the cluster division module using a classification decision model, the constraints are probabilized under a given confidence level to address the uncertainty of EV charging behavior, thereby obtaining the probabilistic feasible region and probabilistic schedulable power of a single EV at different confidence levels.

[0066] Improved Kino polyhedron weighted approximation module: The improved Kino polyhedron method is used to approximate the feasible energy probability domain of the single EV obtained in step 3. The solution maximizes the similarity between the spatial region defined by the Kino polyhedron and the feasible energy domain of EV scheduling. Considering the two dimensions of time value and energy value at different time periods, different weights are set for the diameter in the approximation domain in the time and energy dimensions to achieve approximation of the feasible domain for different time value and energy value time periods. The improved Kino polyhedron weighted approximation is completed for the feasible energy probability domain of all EVs.

[0067] The probabilistic EV schedulable power potential acquisition module: After performing a Cino polyhedral weighted approximation on the energy feasible region of all EV users using an improved Cino polyhedral weighted approximation module, Minkowski summation is performed on users belonging to the same cluster. The aggregated energy feasible region can be represented by the center and scaling factor of the Cino polyhedron. The up-adjustment and down-adjustment power potential of cluster j at time t under different confidence levels is directly summed for each user, thereby obtaining the EV cluster schedulable power potential.

[0068] A third aspect of this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described in the above embodiments.

[0069] A fourth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described above.

[0070] A fifth aspect of this application provides a computer program product, including a computer program that, when executed, is used to implement the method described above.

[0071] Compared with existing technologies, the electric vehicle dispatchability potential probability assessment method and system based on user charging behavior profiles provided by this invention have the following beneficial effects:

[0072] 1. This invention constructs a multi-dimensional profile tag system, comprehensively analyzes factors such as user charging behavior characteristics, travel patterns, electricity price sensitivity, and range anxiety, and comprehensively depicts user charging behavior characteristics. It overcomes the limitation that relying on a single dimension cannot accurately depict demand. By identifying different types of charging time, power preferences, and electricity price sensitivity, it can optimize the scheduling of charging resources and provide support for improving the utilization efficiency of the power grid and charging facilities.

[0073] 2. This invention adopts a user charging behavior profiling method based on a classification decision model. The resulting user clusters not only reflect the differences in user charging behavior in different clusters, but also the regularity of user charging behavior within the same cluster.

[0074] 3. This invention introduces the improved Kino polyhedron algorithm to approximate the probabilistic feasible region of energy for a single EV. By constructing normal vectors and scaling factors, it achieves approximation of the feasible region of energy at different confidence levels, thereby improving the accuracy of the evaluation and reducing the computational burden.

[0075] 4. In terms of the probability assessment of the schedulable potential of EV clusters, this invention uses energy feasible regions at different confidence levels to reflect the boundary fluctuations and probabilities. Attached Figure Description

[0076] Figure 1 This is a flowchart of a method for probabilistically assessing the dispatchable potential of electric vehicles based on user charging behavior profiles.

[0077] Figure 2 This is a schematic diagram of the EV energy and power boundary in step three of the electric vehicle dispatchability potential probability assessment method; (a) shows the controllable EV energy boundary; (b) shows the uncontrollable EV energy boundary; (c) shows the controllable EV power boundary; (d) shows the uncontrollable EV energy boundary;

[0078] Figure 3 The similarity between the spatial region defined by the Chino polyhedron constructed in step four of the electric vehicle dispatchability potential probability assessment method and the EV dispatchability energy feasible region;

[0079] Figure 4 This is a schematic diagram of the energy probability feasible region results of two different charging user clusters after the approximation in step four of the electric vehicle dispatchable potential probability assessment method. The solid box represents the daytime charging cluster, and the dashed box represents the nighttime charging cluster.

[0080] Figure 5a This diagram illustrates the results of the upward and downward adjustment potential of the nighttime charging user cluster after step five of the electric vehicle dispatchability potential probability assessment method. Figure 5b This diagram illustrates the potential for increasing and decreasing power output of the daytime charging cluster after step five is executed. Detailed Implementation

[0081] To make the objectives, technical solutions, beneficial effects, and significant advancements of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings provided in the examples of the present invention. Obviously, all the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort and in accordance with the content, implementation methods, and drawings of the present invention are within the scope of protection of the present invention.

[0082] It should be noted that the terms "first," "second," "third," etc., in the specification and claims of this invention are only used to distinguish different objects, and not to describe a specific order.

[0083] It should also be noted that the following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0084] like Figure 1 As shown, a method for probabilistically assessing the dispatchable potential of electric vehicles based on user charging behavior profiles includes...

[0085] Step 1: Preprocess the data by extracting user charging behavior characteristics using historical charging and travel data and vehicle attribute data, and constructing a multi-dimensional profile tag system.

[0086] Step 11: Obtain electric vehicle user charging data at different time points in the area to be predicted, and construct an electric vehicle user charging database. The data includes actual EV user charging data, travel data, and vehicle attribute data. The actual EV user charging data includes vehicle battery capacity, average charging amount per charge, vehicle charging status parameters, vehicle charging time point, and / or vehicle charging location. The travel data includes average daily mileage and average daily number of trips. The vehicle attribute data includes vehicle identification, vehicle purpose, and / or vehicle energy consumption.

[0087] Step 12: Clean and segment user charging data

[0088] Outlier detection and data verification are performed on the data in the running database to remove data with charging time of less than 15 minutes or more than 24 hours, charging power exceeding the vehicle's maximum power, charging start SOC exceeding 100% or ending SOC below 0%.

[0089] The cleaned data is segmented into segments by identifying the transition nodes between charging status and vehicle stationary status. Continuous data is divided into independent charging event segments, and user charging behavior features are extracted. These user charging behavior features include: charging start time, end time, arrival time, departure time, charging duration, stationary duration, idle time, charging power, battery state of charge (SOC) at the start of charging, SOC at the end of charging, charging location, and percentage of charged charge.

[0090] Step 13: Construct a multi-dimensional charging behavior profile tag system based on the user charging characteristics obtained in Step 12.

[0091] Based on charging behavior characteristics, five types of user profile tags are constructed: user charging time, user charging space, power preference, electricity price sensitivity, and range anxiety. Charging behavior characteristics are extracted and integrated from different dimensions to construct a multi-dimensional charging behavior profile tag system.

[0092] The user charging time tagging process considers the distribution of user charging time and employs the HDBSCAN clustering algorithm, which saves memory through sparse storage, to cluster the data into a two-dimensional space. Based on the clustering results of arrival and departure times, the characteristics of user charging time distribution are analyzed, and users are divided into several corresponding categories based on the final number of clusters. The categories are named according to the observed behavioral characteristics, such as nighttime charging, daytime charging, random charging, and emergency charging.

[0093] The user charging space tag takes into account the possibility that users may charge in different cities or in a fixed city. It uses date feature weights to account for the impact of certain specific dates (such as holidays, weekends, etc.) on users' cross-city charging behavior and quantifies the number of cross-city charging times.

[0094] The power preference label reflects a user's preference for different charging powers. This label directly affects the fluctuation of charging time and charging load. By statistically analyzing the distribution of charging power characteristics in different power ranges during user charging behavior, and determining the user type based on the proportion of fast and slow charging times, the user's power preference label is represented as a slow-charging user, a fast-charging user, or a balanced fast and slow charging user.

[0095] The electricity price sensitivity label is quantified based on users' charging behavior under different electricity price environments. As electricity prices fluctuate, users' choices of charging times will differ: electricity price-sensitive users tend to charge during the period with the lowest electricity price during their stay, while electricity price-insensitive users usually start charging immediately after arriving at the charging location and ignore the price difference during the charging period. Therefore, this embodiment uses the proportion of EV charging during off-peak hours to the total charging amount to construct an electricity price sensitivity index. The electricity price sensitivity value is [0,1], where a value of 1 indicates that all EV charging activities are carried out during periods of low electricity price, and the larger the value, the greater the EV user's electricity price sensitivity.

[0096] Specifically, an electricity price sensitivity index is constructed using the proportion of EV charging during off-peak electricity price periods to the total charging amount, as shown in the formula below. As shown, the electricity price sensitivity value is [0,1]. A value of 1 indicates that all EV charging activities are performed only when electricity prices are low, and a higher value indicates greater electricity price sensitivity for EV users. Based on the electricity price sensitivity, K-means clustering algorithm is used for clustering. The elbow rule is used to find the optimal number of clusters k1, and user electricity price sensitivity label levels are defined using integers between 1 and k1. The higher the level, the more sensitive the user is to electricity prices.

[0097]

[0098] In the formula: This indicates user i's sensitivity to electricity prices; Total number of charging activities for user i; This represents the amount of electricity charged by user i during the period with the lowest electricity price in the nth charging activity. , and These represent the charging amount during user i's nth charge in three time periods: peak, flat, and valley. This can be calculated from the charging time and power during each of these three periods. Peak, valley, and flat times can be defined according to the time-of-use pricing system.

[0099] Range anxiety reflects a user's sense of security regarding battery capacity; the lower the remaining SOC of the EV, the more pronounced the range anxiety. Since the remaining SOC is represented by the initial SOC during charging, the higher the initial SOC, the more pronounced the user's range anxiety. Therefore, this embodiment uses the initial SOC during EV charging to measure the user's range anxiety level. The user's initial SOC value is divided into five consecutive intervals, from low to high, in 20% increments of battery capacity.

[0100] Based on the charging frequency in each range, different weights are assigned to each range to calculate the user's range anxiety level R. i , specifically as Eq. As shown in the diagram. Based on the level of range anxiety, K-means clustering is also used for clustering. The elbow rule is used to find the optimal number of clusters k2. The user's range anxiety level is defined using integers between 1 and k2, with higher values ​​indicating greater range anxiety.

[0101]

[0102] In the formula: and Let i be the charging frequency and weight of user i in the k-th SOC interval, respectively, where the weight is given by equation [equation missing]. To determine the correlation between different charging start-up SOC ranges and range anxiety.

[0103] The charging weights for different State of Charge (SOC) zones are adjusted based on user behavior preferences. The distribution of user charging times across different SOC zones is analyzed (e.g., nighttime charging is mostly for pre-charging, while daytime low SOC charging is mostly due to urgent needs), assigning extra weight to daytime low SOC zones. Basic weights are initialized using charging urgency, and the average time interval between a user's arrival at the charging station and the start of charging is calculated for each SOC zone. Shorter time intervals indicate a more urgent user charging need.

[0104]

[0105] In the formula: This is the sensitivity coefficient for daytime periods, typically set to 0.2-0.8; and The number of times user i is charged during the day and night in interval k; Based on the weights, it can be expressed by the formula Sure.

[0106]

[0107] In the formula: This represents the response time of user i in interval k, i.e., the time interval between arriving at the charging station and starting charging. When This means that the user should charge immediately upon arriving at the charging location. ε is a very small positive number.

[0108] Step 2: Divide users into different clusters using a classification decision model; integrate and analyze the constructed multi-dimensional profile tag system, and train and test the classification decision model to obtain the clustering results of users.

[0109] S21: Feature Selection

[0110] The information gain of the profile labels from step one is used to measure the influence of each profile label on user classification for each dimension. Specifically, for a dataset D with J categories, the information entropy is expressed as:

[0111] (5)

[0112] In the formula: D is the dataset; C j Let be the set of samples belonging to class j in the dataset; This indicates the number of samples in the set.

[0113] Assume feature A m (m=1,2,...,5, representing the five dimensions of the profile labels in step one) There are n m Different values According to feature A m The dataset D can be divided into n m Subset Then the conditional entropy is:

[0114] (6)

[0115] Finally, based on the information entropy and conditional entropy, the information gain g(D,A) can be obtained. mThe specific calculation is shown in equation (7). The greater the information gain, the greater the contribution of the feature to the classification.

[0116] (7)

[0117] Calculate all features A m The information gain is used to find the feature with the largest information gain. The root node dataset D is divided into multiple subsets according to the different values ​​of this feature, and each subset corresponds to a branch of the decision tree.

[0118] S22: Building a Decision Tree

[0119] For each newly generated subset, repeat the feature selection step S21. That is, treat each subset obtained in step S21 as a new dataset, and again select the feature with the highest information gain from the remaining features to split the dataset, generating a new branch. Iterate this process until the stopping condition is met, that is, all samples in the subset belong to the same category.

[0120] S23: Decision Tree Pruning

[0121] If user profiles are overly segmented, treating random combinations of user characteristics as general patterns will lead to poor classification results for new data. Decision tree pruning removes users whose information gain is below a preset gain threshold. Branches prone to overfitting, the gain threshold Set to 0.05. Pruning the decision tree improves the accuracy and generalization ability of profiling user charging behavior. The classification decision tree can divide user charging behavior into subgroups with clear feature labels, with each leaf node corresponding to a typical user profile.

[0122] Step 3: To address the uncertainty in EV charging behavior, the constraints are probabilistically transformed at a given confidence level to obtain the probabilistic feasible region and probabilistic schedulable power of a single EV.

[0123] S31: Based on the user behavior characteristics obtained in step one and the user classification results obtained in step two, and based on the differences in charging time, power, and energy data of different users, characterize the dispatchable capacity and dispatchable power boundary of different types of single EVs; whereby the dispatchable capacity of EVs is formed by the dual constraints of charging and discharging capacity and time to form a two-dimensional energy feasible domain, and the dispatchable power of EVs is divided into upward power potential and downward power potential. The upward power potential is when the EV increases the charging power or decreases the discharging power, and the equivalent grid load increases; the downward power potential is when the EV decreases the charging power or increases the discharging power, and the equivalent grid load decreases.

[0124] The schedulable capacity of an EV is represented by a convex polyhedron constrained by a set of inequalities, forming a two-dimensional energy feasible region by the dual constraints of charge / discharge capacity and time:

[0125] (8)

[0126] In the formula: Let be the energy feasible region for user i under cluster j. Let be the set of points within the feasible region; and These are the coefficient matrix and constant column vector of the inequality, respectively.

[0127] The remaining battery capacity of EV user i under cluster j at time t can be calculated from the actual charging and discharging power:

[0128] (9)

[0129] In the formula: The remaining battery level at the time of EV registration; The time of network access; Let t be the charging or discharging power at time t.

[0130] To ensure battery health and accommodate potential emergency vehicle use during charging, a baseline battery level typically needs to be set. The specific details are shown in equation (10).

[0131] (10)

[0132] In the formula: This indicates the minimum SOC set to protect the health of the EV battery; This represents the total number of trips made by this EV in the dataset; This represents the SOC consumed during the b-th trip of the EV; This refers to the rated capacity of the EV battery, expressed in kWh.

[0133] Figure 2 (a), (b), (c), and (d) illustrate schematic diagrams of the energy and power boundaries of controllable and uncontrollable EVs, respectively. This invention only discusses controllable EVs; for... Figure 2 (b) EVs that require continuous charging due to insufficient expected parking time, and whose energy and power boundaries lack flexibility, are not within the scope of this discussion. Figure 2 As shown in (a), the upper boundary of EV energy is a broken line abc, which is represented by the minimum value of the grid-connected charging and the maintenance of full charge to off-grid at each moment during the grid-connected time; the lower boundary of energy is represented by a broken line afed, which is represented by the maximum value of the grid-connected discharge, the maintenance of the reference charge, and the charging before off-grid. The upper and lower boundaries of energy can be represented by equations (11) and (12). The upper and lower boundaries of the power of differentiated EV user i under cluster j can be calculated by the charging and discharging power and efficiency, as represented by equations (13) and (14).

[0134] (11)

[0135] (12)

[0136] (13)

[0137] (14)

[0138] In the formula: The charging power for user i in cluster j is determined by the user's power preference label; The user's desired battery level can be represented by the average battery level at the end of charging. The discharge power is a negative value, and different values ​​are set according to the vehicle model. , These are the upper (+) and lower (-) boundaries of EV energy and power at time t, respectively; and These are the arrival and departure times from the charging location, respectively. and These are the charging and discharging efficiencies, respectively.

[0139] The dispatchable power of a single EV is divided into upward and downward power potential. Upward power potential refers to the EV increasing its charging power or decreasing its discharging power, resulting in an increase in the equivalent grid load; downward power potential refers to the EV decreasing its charging power or increasing its discharging power, resulting in a decrease in the equivalent grid load. This embodiment categorizes the EV-grid interaction modes into four types based on the EV's transitions between charging, discharging, and idle states: Mode 1 (idle → charging) and Mode 4 (discharging → idle) correspond to upward potential; Mode 2 (charging → idle) and Mode 3 (idle → discharging) correspond to downward potential. The dispatchable power potential is related to the charging and discharging state. The upward and downward potential of the EV at the charging location are calculated as follows:

[0140] (15)

[0141] (16)

[0142] In the formula: and This represents the up and down potential of the i-th user in the j-th cluster at time t, where the down potential is a negative value; , , These are the state variables of the EV at time t when it is in the idle, charging, and discharging states. The corresponding state variable is 1 only when it is in the corresponding state, and 0 otherwise.

[0143] S32: Probabilistic assessment of the dispatchable potential of a single EV considering the uncertainty of user charging behavior

[0144] Based on the energy feasible region and schedulable power results obtained in step S31, the probability distribution of key variables of energy feasible region and schedulable power potential is calculated, and the constraints are reconstructed under a given confidence level to expand them into an energy feasible region expressed in probability.

[0145] Specifically, regarding the charging power of cluster user i within the feasible region and schedulable power... Remaining battery power at the time of EV network access Arrival time at the charging location Time of leaving the charging location Using kernel density estimation and historical charging data, the probability density functions of the aforementioned variables are obtained.

[0146] Taking charging power as an example, for Its probability density function is:

[0147] (17)

[0148] In the formula: Let M be the probability density function; M be the number of data samples of EV user i under cluster j; and the bandwidth h be set according to Silverman's rule based on the sample standard deviation s and the sample size n. ; Let m be the m-th data point; K(·) is the kernel function. In this embodiment, a Gaussian kernel function that can handle most data distributions is selected.

[0149] The cumulative distribution function F is calculated from the probability density function:

[0150] (18)

[0151] For a one-sided confidence interval, the number of intervals with lower and upper limits of power at a given confidence level σ is used, i.e. This indicates the original mean was calculated. :

[0152] (19)

[0153] Calculate the cumulative distribution function of the above variables sequentially, and under a given confidence level σ, replace the mean of the original variables with the interval number. (Using the remaining capacity...) For example, it must meet the following requirements:

[0154] (20)

[0155] (twenty one)

[0156] In the formula: Pr(·) is the probability function; and These represent the lower and upper limits of the remaining capacity confidence interval at a confidence level of σ, respectively.

[0157] Using the same method, the determined values ​​of EV energy and power boundaries in equations (10)-(14) are replaced by interval numbers to obtain the probabilistic feasible region of single EV energy represented by multiple sets of interval numbers. For the upper and lower adjustable potentials in equations (15) and (16), time is discretized. Under the confidence level σ, the upper and lower adjustment potentials are represented by interval numbers composed of the lower upper limit of their respective confidence intervals. The calculation method of the upper and lower limits of the confidence intervals is shown in equations (22) and (23).

[0158] (twenty two)

[0159] (twenty three)

[0160] In the formula: and with formula , In , The meanings are the same, respectively t k The potential for power increase and power decrease for user i in cluster j at time t; k The time period number, T represents the total time period; , and , t at confidence level σ k The lower and upper limits of the confidence intervals for the up-adjustment and down-adjustment potentials of user i in cluster j at time point.

[0161] Based on the above steps, obtain the probabilistic feasible region of individual EV energy and the probabilistic schedulable power under different confidence levels; repeat the above steps to obtain the feasible region of individual EV energy to obtain the feasible region of EV energy for all users.

[0162] Step 4: Approximate the feasible region of EV energy using the improved Chino polyhedron expression.

[0163] The Chino polyhedron, which approximates the energy feasible region under different confidence levels, can be defined by a center point, a generating vector matrix, and a corresponding scaling factor:

[0164] (twenty four)

[0165] (25)

[0166] (26)

[0167] In the formula: It is the center point of the Chino polyhedron; To generate a vector matrix; express The corresponding scaling factor vector matrix consists of Q scaling factors. composition; and These represent the lower and upper limits of the scaling factor, respectively.

[0168] Generate vector matrix Q generated vectors extending in all directions Composition, namely:

[0169] (27)

[0170] Using the improved Cino polyhedron expression to approximate the EV energy feasible region is essentially a problem of finding the optimal inner approximation, i.e., maximizing the similarity between the spatial region bounded by the Cino polyhedron and the EV scheduling energy feasible region. For example... As shown, in this embodiment, for the energy feasible region under confidence level σ, arbitrary... Given S normal vectors, calculate the diameters of regions Z and W along the direction of the normal vectors. The similarity is measured by the ratio of the two diameters. Furthermore, by assigning different weights to the diameters in the time and energy dimensions within the approximate domain, we can approximate the feasible domain for different time and energy value periods.

[0171] (28)

[0172] (29)

[0173] (30)

[0174] In the formula: and The normal vectors of the Chino polyhedron and the original convex polyhedron are respectively... Diameter in the direction; , representing the energy value and time value weights of the feasible region time period, respectively, and their magnitudes are determined by the peak-valley-flat state of each time period; For the chino polyhedron and the convex polyhedron at the normal vector Weighted similarity in direction; (·) 2 This represents the square of each element in the matrix. This refers to element-wise multiplication of a matrix. , , , They represent and According to the normal vector The direction is decomposed into the length of the energy axis and the time axis.

[0175] According to equations (8) and (24)-(30), the optimization problem of approximating the feasible region using the improved Chino polyhedron can be determined by the following equation:

[0176] (31)

[0177] In the formula: and These are the coefficient matrix and constant column vector of the inequality, respectively.

[0178] Figure 4 The results show the energy probability feasible region results for two different charging user clusters after performing step four. The nighttime cluster, containing 38 electric private car users, has a dispatchable period concentrated between 23:30 and 10:30 the next day when calculated using the mean, with an energy range of 0.7-2.2 MWh. After performing steps one through four, the feasible region boundary gradually widens at 90%, 95%, and 99% confidence levels. The daytime cluster, containing 36 users, has a dispatchable period concentrated between 10:15 and 16:00, a significantly narrower time range than the nighttime cluster. Furthermore, with almost the same number of vehicles, the nighttime cluster has a larger dispatchable energy range because vehicles are more idle at night, resulting in more abundant electricity available for grid regulation.

[0179] Step 5: After performing the Chino polyhedral weighted approximation based on the energy feasible region of all EV users, perform the Minkowski summation on users belonging to the same cluster to obtain the schedulable power potential of the EV cluster.

[0180] After applying an improved Cino polyhedral weighted approximation to the energy probability feasible region for all EV users, a Minkowski summation is performed on users belonging to the same cluster, as shown in Equation (32). The aggregated energy feasible region can be represented by the center and scaling factor of the Cino polyhedron, as shown in Equation (33). The center and scaling factor of the aggregated Cino polyhedron can be directly added together by the center and scaling factor of each approximated Cino polyhedron, as shown in Equations (34) and (35).

[0181] (32)

[0182] (33)

[0183] (34)

[0184] (35)

[0185] In the formula: The cluster aggregation energy approximates the feasible region; Summing for Minkowski; Let be the approximate feasible energy region for user i in cluster j at confidence level σ. and These represent the center and scaling factor of the aggregated Cino polyhedron under different confidence levels.

[0186] The schedulable power probability potential of cluster j at different confidence levels can be obtained by directly summing the power of each user:

[0187] (36)

[0188] (37)

[0189] In the formula: and The table shows the up-adjustment and down-adjustment power potential of cluster j at time t under confidence level σ.

[0190] After performing step five, as follows Figure 5a As shown, the available power for nighttime charging remains basically unchanged between 1:00 and 8:00. This is because users are all charging during this period and few EVs leave the network. However, between 8:00 and 10:30, vehicles leave and join the network one after another, causing fluctuations in the available power. A power trough occurs around 9:45, which is when the number of users on the network in this cluster is the lowest. Figure 5b The daytime charging clusters shown exhibit two "peaks" at 12:00 and 15:00, with both the power adjustment and the charging speed reaching their peaks. This is because private passenger vehicles and commercial vehicles enter a charging state due to range requirements, creating rigid charging demand. Consequently, the power adjustment reaches its lowest point at these times, while the power adjustment reaches its peak. Charging operators can use electricity pricing mechanisms to guide some users to reduce their charging power, thereby reducing the grid load.

[0191] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention. Non-essential improvements, adjustments or substitutions made by those skilled in the art based on the content of this specification are all within the scope of protection claimed by the present invention.

Claims

1. A method for probabilistically evaluating the dispatchable potential of electric vehicles based on user charging behavior profiles, characterized in that, include: Step 1: Extract user charging behavior features that describe user charging habits using real charging data and vehicle data of EV users in the area to be predicted. Refine and integrate the charging behavior features from different dimensions to build a multi-dimensional charging behavior profile label system. Step 2: Using the multi-dimensional charging behavior profile tag system built in Step 1, the information gain of each profile tag is used to measure the degree of influence of each dimension profile tag in Step 1 on user classification. The classification decision model is used for training and testing to obtain the user's classification results; Step 3: Based on the user charging behavior characteristics obtained in Step 1 and the user classification results obtained in Step 2, for users of the same category, obtain the feasible energy domain and schedulable power potential of a single EV unit based on the virtual battery model. Considering the uncertainty of EV charging behavior, the constraints are probabilistically converted at a given confidence level to obtain the probabilistic feasible energy domain and probabilistic schedulable power of a single EV unit under different confidence levels. Repeat the above steps to obtain the feasible energy domain of a single EV unit to obtain the feasible energy domain of EV units for all users. Step 4: Approximate the individual EV energy probability feasible region obtained in Step 3 using the improved Chino polyhedron method. Solve to maximize the similarity between the spatial region defined by the Chino polyhedron and the EV dispatchable energy feasible region. At the same time, in order to cope with the different requirements of the power grid and the electricity market for the sustainability duration and dispatchable energy of flexibility resources at different times, consider the two dimensions of time value and energy value at different times. By setting different weights for the diameter in the approximation region in the time and energy dimensions respectively, the feasible region for different time value and energy value time periods is approximated. The improved Chino polyhedron weighted approximation is completed for the energy probability feasible region of all EVs. Step 5: After applying the Chino polyhedral weighted approximation to the energy feasible region of all EV users, perform Minkowski summation on users belonging to the same cluster. The aggregated energy feasible region can be represented by the center and scaling factor of the Chino polyhedron. ； ； ； ； In the formula: The cluster aggregation energy approximates the feasible region; Summing for Minkowski; Let be the approximate feasible energy region for user i in cluster j at confidence level σ. and These represent the center and scaling factor of the aggregated Chino polyhedron under different confidence levels; The probabilistic schedulable power potential of cluster j at different confidence levels can be obtained by directly summing the power of each user: ； ； In the formula: and The up-adjustment and down-adjustment power potentials of cluster j at time t under confidence level σ are respectively represented; thus, the schedulable power potential of the EV cluster is obtained. Based on the dispatchable energy probability obtained in step four and the dispatchable power potential obtained in step five, operators are guided to use the electricity pricing mechanism to encourage some users to reduce charging power in order to reduce grid load.

2. The method for assessing the dispatchable potential probability of electric vehicles based on user charging behavior profiles according to claim 1, characterized in that, Step one specifically includes: The steps for extracting user charging behavior features are as follows: The acquired real charging data and vehicle data of EV users are cleaned and anomaly processed. By identifying the transition nodes between charging status and vehicle dwell status, segments are created, dividing continuous data into independent charging event segments. Each segment corresponds to a complete user charging behavior process, thereby extracting user charging behavior features. These features include: charging start time, end time, arrival time, departure time, charging duration, dwell time, idle time, charging power, battery state of charge (SOC) at charging start, SOC at charging end, charging location, and percentage of charged charge. The steps to build a multi-dimensional charging behavior profile tag system are as follows: Extract and integrate user charging behavior characteristics from different dimensions to build five types of profile tags: user charging time, user charging space, power preference, electricity price sensitivity and range anxiety, thereby building a multi-dimensional charging behavior profile tag system. Among them, the user charging time tag uses a clustering algorithm to directly cluster the arrival and departure times to divide users, thereby representing the user charging time tag; User charging space labeling considers whether the user's charging location crosses urban areas, the impact of the reference date feature weight on the user's cross-city charging behavior, and quantifies the number of cross-city charging times to characterize the user's charging space label. Power preference labels are defined by statistically analyzing the distribution of user charging power across different power ranges and determining user type based on the proportion of fast and slow charging cycles. Electricity price sensitivity labels are calculated by analyzing users’ charging behavior under different electricity price environments. Then, based on the numerical values, a clustering algorithm is used to classify the electricity price sensitivity labels to represent users’ electricity price sensitivity labels. The range anxiety label measures a user's range anxiety level by measuring the initial state of charge (SOC) of the EV. Then, a clustering algorithm is used to determine the degree of the user's range anxiety label based on the numerical value, thus representing the range anxiety label.

3. The method for probabilistic assessment of the dispatchable potential of electric vehicles based on user charging behavior profiles according to claim 1, characterized in that, Based on observations of user charging time tags, users were categorized into: nighttime charging, daytime charging, random charging, and emergency charging. In the power preference label, the user's power preference label is represented by slow charging user, fast charging user, or fast and slow charging balanced user; In the electricity price sensitive label, an electricity price sensitivity index is constructed based on the proportion of a user's EV charging electricity during off-peak hours to the total charging electricity. The index is then clustered based on the value and divided into electricity price sensitive users and electricity price insensitive users. In the range anxiety label, the user's range anxiety level is measured by the initial SOC of EV charging. The user's initial SOC value is divided into multiple intervals according to the battery level from low to high. Different weights are assigned to each interval based on the charging frequency of each interval. This is used to calculate the user's range anxiety level, and clustering is performed based on the range anxiety level to classify the user's range anxiety label.

4. The method for assessing the dispatchable potential probability of electric vehicles based on user charging behavior profiles according to claim 1, characterized in that, Step two specifically includes: S21: Using the multi-dimensional charging behavior profile label system constructed in step one, the information gain of each profile label is used to measure the degree of influence of each dimension profile label in step one on user classification; the root node dataset D is divided into multiple subsets based on the different values ​​of the feature with the largest information gain, and each subset corresponds to a branch of the decision tree. S22: Repeat the feature selection step S21 for each newly generated subset to generate a new branch; iterate this process until all samples in the subset belong to the same category and then stop iterating. S23: Remove branches with information gain lower than a preset gain threshold to form an updated classification decision tree; the classification decision tree divides user charging behavior into subdivided groups with clear feature labels, and each leaf node corresponds to a typical user profile.

5. The method for probabilistic assessment of the dispatchable potential of electric vehicles based on user charging behavior profiles according to claim 1, characterized in that, Step three specifically includes: S31: Based on the user behavior characteristics obtained in step one and the user classification results obtained in step two, characterize the dispatchable capacity and dispatchable power boundary of different types of individual EVs; whereby the dispatchable capacity of EVs is formed by the dual constraints of charging and discharging capacity and time to form a two-dimensional energy feasible region, and the dispatchable power boundary of EVs includes the potential for upward adjustment of power and the potential for downward adjustment of power. The potential for upward adjustment of power is when the EV increases the charging power or decreases the discharging power, resulting in an increase in the equivalent grid load; the potential for downward adjustment of power is when the EV decreases the charging power or increases the discharging power, resulting in a decrease in the equivalent grid load. S32: Based on the schedulable capacity and schedulable power boundaries obtained in step S31, calculate the probability distribution of key variables of the energy feasible region and schedulable power potential. Reconstruct the constraints at a given confidence level, expanding them to the individual EV energy probabilistic feasible region represented by the number of intervals for the lower and upper limits of energy at a given confidence level σ, and further expanding them to the individual EV probabilistic schedulable power represented by intervals for the upward and downward power potentials at a given confidence level σ; the key variables include: the charging power of user i in cluster j. Remaining battery power at the time of EV network access Arrival time at the charging location Time of leaving the charging location .

6. A system for probabilistically assessing the dispatchable potential of electric vehicles based on user charging behavior profiles, characterized in that, include: The module for constructing a multi-dimensional profile tagging system utilizes real charging data and vehicle data of EV users in the area to be predicted to extract user charging behavior features that can describe user charging habits. These charging behavior features are then refined and integrated from different dimensions to construct a multi-dimensional charging behavior profile tagging system. Clustering module is divided using a classification decision model: a multi-dimensional charging behavior profile tag system is constructed using the multi-dimensional profile tag system construction module, and the information gain of each profile tag is used to measure the degree of influence of each dimension profile tag on user classification in the multi-dimensional profile tag system construction module. The user clustering results are obtained by training and testing based on the classification decision model; The module for obtaining the probabilistic feasible region and probabilistic schedulable power of a single EV is as follows: Based on the user charging behavior characteristics obtained by constructing a multi-dimensional profile tag system module and the user classification results based on the cluster division module using a classification decision model, the constraints are probabilized under a given confidence level to address the uncertainty of EV charging behavior, thereby obtaining the probabilistic feasible region and probabilistic schedulable power of a single EV at different confidence levels. Improved Kino polyhedron weighted approximation module: The improved Kino polyhedron method is used to approximate the feasible energy probability domain of the single EV obtained in step 3. The solution maximizes the similarity between the spatial region defined by the Kino polyhedron and the feasible energy domain of EV scheduling. Considering the two dimensions of time value and energy value at different time periods, different weights are set for the diameter in the approximation domain in the time and energy dimensions to achieve approximation of the feasible domain for different time value and energy value time periods. The improved Kino polyhedron weighted approximation is completed for the feasible energy probability domain of all EVs. The probabilistic EV schedulable power potential acquisition module: After performing a Cino polyhedral weighted approximation on the energy feasible region of all EV users using an improved Cino polyhedral weighted approximation module, Minkowski summation is performed on users belonging to the same cluster. The aggregated energy feasible region can be represented by the center and scaling factor of the Cino polyhedron. The up-adjustment and down-adjustment power potential of cluster j at time t under different confidence levels is directly summed for each user, thereby obtaining the EV cluster schedulable power potential.

7. An electronic device, comprising: The device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, the processor executing the program to implement the electric vehicle dispatchability potential probability assessment method based on user charging behavior profiles as described in any one of claims 1-5.

8. A computer-readable storage medium storing a computer program that, when executed by a processor, implements the method for probabilistic assessment of the dispatchable potential of electric vehicles based on user charging behavior profiles according to any one of claims 1-5.

9. A computer program product comprising a computer program, which, when executed, is used to implement the electric vehicle dispatchability potential probability assessment method based on user charging behavior profiles according to any one of claims 1-5.

Images