Charging behavior transfer learning method for long-tail users

Abstract

The invention relates to the technical field of user behavior migration management, and discloses a long-tail user-oriented charging behavior migration learning method, which comprises the following steps of: 1, acquiring management data of charging behaviors of all users and management data of all charging areas, and classifying to form a data set; 2, setting a time axis, analyzing the change degree of each charging time of the user in combination with the user data set, and generating a fluctuation index; 3, according to the user data set, analyzing the charging power of each charging behavior of the user, judging behavior attributes, carrying out statistics to generate a fast and slow charging ratio, and accurately analyzing user behavior characteristics; 4, analyzing the use state of the charging pile in each region in real time according to the region data set and the fast and slow charging proportion, and generating a supply index; and 5, setting a frequency threshold value and a fluctuation threshold value of fixed numerical values, judging the long-tail user in combination with the user data set and the fluctuation index, and generating a charging behavior migration suggestion, so that the utilization rate of personalized migration resources is high.

Description

Technical Field

[0001] The present invention relates to the technical field of user behavior migration management, and specifically to a charging behavior migration learning method for long-tail users. Background Art

[0002] As the new energy vehicle industry is developing rapidly, charging platforms, as key infrastructure, have a complex and diverse user structure, among which the long-tail user group is often easily overlooked.

[0003] The long-tail user group primarily includes niche groups who are extremely sensitive to charging times and whose vehicle usage scenarios are relatively specialized. For example, taxi drivers, busy on the job, urgently need to quickly find reliable and efficient charging stations nearby. If charging platforms can accurately deliver information about charging stations that are immediately available, ideally located, and reasonably priced, fully meeting their time-sensitive charging needs, they will undoubtedly win the long-term trust and support of these users. Then there are the electric truck drivers involved in logistics and transportation. While their routes are relatively fixed, they cover a wide swath of the city, placing high demands on the stability and compatibility of the charging facilities along their route. If charging platforms can plan appropriate routes for them and provide early warning of potential charging station failures, they can help streamline logistics operations and ultimately make these drivers loyal users.

[0004] Currently, traditional charging behavior transfer learning methods often overlook low-frequency or irregular long-tail users. These users charge randomly, in scattered locations, and infrequently. They often face issues such as occupied charging piles, large price fluctuations, and poor site matching. Charging platforms lack targeted optimization measures, which can easily lead to poor service experience and low resource utilization, further exacerbating user churn. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention provides a charging behavior transfer learning method for long-tail users, which has the advantages of accurate analysis of user behavior characteristics and high utilization of personalized migration resources. It solves the problem of poor service experience and low resource utilization for long-tail users caused by traditional charging behavior transfer learning methods.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a charging behavior transfer learning method for long-tail users, characterized in that it includes the following steps:

[0007] Step 1: Connect to the big data platform through the network to obtain the management data of all users' charging behaviors and the management data of all charging areas, and classify them into user data sets and area data sets;

[0008] Step 2: Set the time axis Z, and then combine it with the user data set to analyze the degree of change in the user's charging time each time, and generate the corresponding fluctuation index Bdz;

[0009] Step 3: Based on the user data set, analyze the charging power gl of each user's charging behavior, determine the behavior attributes, and statistically generate the fast and slow charging ratio Kmb;

[0010] Step 4: Based on the regional data set and the fast and slow charging ratio Kmb, the usage status of charging piles in each region is analyzed in real time to generate the corresponding supply index Gyz;

[0011] Step 5: Set a fixed number threshold CY and a fluctuation threshold BDY. Then, combine the user dataset and the fluctuation index Bdz to identify long-tail users and generate corresponding charging behavior migration recommendations.

[0012] Preferably, in step 1, the user data set includes the total number of charging times, charging start time, charging amount, charging duration and charging location of each user.

[0013] Preferably, in step 1, the regional data set includes the area of ​​each charging area, the total number of fast charging piles, the number of available fast charging piles, the number of available slow charging piles and the total number of slow charging piles.

[0014] Preferably, in step 2, the calculation process of the volatility index Bdz is as follows:

[0015] Based on the user data set, extract the management data of the i-th user and mark the total number of charging times of the i-th user in a month as n;

[0016] Divide the time axis Z into 24 time domains and label them as time domain 1, time domain 2, time domain 3, ..., time domain 24, where each time domain represents one hour;

[0017] Substitute the charging start time of the i-th user into the time axis Z. The time domain corresponding to each charging behavior is marked as {s1, s2, s3, ..., sn}, where s1 to sn represent the time domain corresponding to the first to n-th charging behavior of the i-th user;

[0018]

[0019] In the formula, represents the average time domain corresponding to the charging behavior of the i-th user, sf represents the time domain corresponding to the f-th charging behavior of the i-th user, f∈n, According to the standard deviation formula, the fluctuation index Bdz of the charging behavior of the i-th user in a month is calculated. i .

[0020] Preferably, in step 3, the charging power g1 calculation process is as follows:

[0021] According to the user data set, the charging amount of the i-th user at the f-th charging time is marked as d f , mark the charging time of the i-th user during the f-th charging as c f ;

[0022]

[0023] In the formula, gl f represents the charging power of the i-th user during the f-th charging.

[0024] Preferably, in step 3, the behavior attribute determination process is as follows:

[0025] If the charging power of the i-th user during the f-th charging is gl f ≥37.5kW, considered as fast charging behavior;

[0026] If the charging power of the i-th user during the f-th charging is gl f <37.5kW, it is judged as slow charging behavior.

[0027] Preferably, in step 3, the calculation process of the fast-slow charging ratio Kmb is as follows:

[0028] Count the total number of fast charging behaviors KC of the i-th user in a month i and the total number of slow charging behaviors MC i ;

[0029]

[0030] In the formula, Kmb i It represents the ratio of fast charging to slow charging behavior of the i-th user in one month.

[0031] Preferably, in step 4, the supply index Gyz calculation process is as follows:

[0032] According to the regional data set, the management data of the kth charging area is extracted and the area of ​​the kth charging area is marked as mj k , mark the total number of fast charging piles in the kth charging area as kz k , mark the available number of fast charging piles in the kth charging area as ky k , mark the total number of slow charging piles in the kth charging area as mz k , mark the available number of slow charging piles in the kth charging area as my k ;

[0033] If the i-th user's charging behavior in a month is fast and slow charging ratio Kmb i ≥1,

[0034]

[0035] In the formula, α1 represents the weight of the ratio of the available number of fast charging piles to the total number of fast charging piles. represents the density of fast charging piles in the kth charging area, α2 represents the weight for the density of fast charging piles, α1 and α2 are both constants, α1+α2=1, According to the weights α1 and α2, the supply index Gyz of the kth charging area for the i-th user is calculated. k ;

[0036] If the i-th user's charging behavior in a month is fast and slow charging ratio Kmb i <1,

[0037]

[0038] In the formula, β1 represents the weight of the ratio of the available number of slow charging piles to the total number of slow charging piles. represents the density of slow charging piles in the kth charging area, β2 represents the weight for the density of slow charging piles, β1 and β2 are both constants, β1+β2=1, According to the weights β1 and β2, the supply index Gyz of the kth charging area for the i-th user is calculated. k .

[0039] Preferably, in step 5, the long-tail user judgment process is as follows:

[0040] If the total number of charging times n of the i-th user in a month is lower than the number threshold CY, it means that the i-th user is a long-tail user with less charging behavior;

[0041] If the fluctuation index Bdz of the charging behavior of the i-th user i Exceeding the fluctuation threshold BDY indicates that the i-th user is a long-tail user with unstable charging habits;

[0042] If the charging locations of the i-th user do not belong to the same charging area within a month, it means that the i-th user is a long-tail user with unstable charging habits.

[0043] Preferably, in step five, for long-tail users with less charging behavior, it is recommended that users be given priority to choose a charging area adjacent to the last charging location; for long-tail users with unstable charging habits, it is recommended that users be given priority to choose a charging area with a higher supply index Gyz.

[0044] Compared with the existing technology, the present invention provides a charging behavior transfer learning method for long-tail users, which has the following beneficial effects:

[0045] 1. The present invention connects to a big data platform via a network to obtain management data on all users' charging behaviors and all charging areas, and classifies them into user datasets and area datasets. A time axis Z is set, and then combined with the user dataset, the degree of change in each user's charging time is analyzed to generate a corresponding fluctuation index Bdz. This quantifies the regularity of user charging times and provides a calculable migration basis. Based on the user dataset, the charging power gl of each user's charging behavior is analyzed, the behavior attributes are determined, and the fast and slow charging ratio Kmb is statistically generated. This accurately analyzes user behavior characteristics and precisely reflects user preferences for different charging speeds.

[0046] 2. The present invention uses regional data sets and the fast-slow charging ratio Kmb to analyze the usage status of charging piles in each area in real time, generate a corresponding supply index Gyz, and specifically judge the real-time supply capacity of the area. It sets a fixed number threshold CY and a fluctuation threshold BDY, and then combines the user data set and the fluctuation index Bdz to judge long-tail users with less charging behavior and unstable charging habits. For long-tail users with less charging behavior, it is recommended that users choose charging areas adjacent to the last charging location, which is in line with the long-tail users' dependence on habitual paths, reduces exploration costs, and improves acceptance. For long-tail users with unstable charging habits, it is recommended that users choose charging areas with a higher supply index Gyz, which maximizes the success rate of queue-free charging, meets the random needs of long-tail users, and has high utilization rate of personalized migration resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a step diagram of the method of the present invention. DETAILED DESCRIPTION

[0048] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0049] Traditional charging behavior transfer learning methods often ignore low-frequency or irregular long-tail users. These users have random charging behaviors, scattered locations, and low frequency. They often face problems such as occupied charging piles, large price fluctuations, and low site matching. Charging platforms lack targeted optimization measures, which can easily lead to poor service experience and low resource utilization, further exacerbating user churn. Therefore, a charging behavior transfer learning method for long-tail users is provided. Please refer to Figure 1 , including the following steps:

[0050] Step 1: Connect to the big data platform through the network to obtain the management data of all users' charging behaviors and the management data of all charging areas, and classify them into user data sets and area data sets;

[0051] The user data set includes the total number of charging times, charging start time, charging amount, charging duration and charging location of each user;

[0052] The regional dataset includes the area of ​​each charging area, the total number of fast charging piles, the number of available fast charging piles, the number of available slow charging piles, and the total number of slow charging piles;

[0053] Step 2: Set the time axis Z, and then combine it with the user data set to analyze the degree of change in the user's charging time each time, and generate the corresponding fluctuation index Bdz;

[0054] The calculation process of the volatility index Bdz is as follows:

[0055] Based on the user data set, extract the management data of the i-th user and mark the total number of charging times of the i-th user in a month as n;

[0056] Divide the time axis Z into 24 time domains and label them as time domain 1, time domain 2, time domain 3, ..., time domain 24, where each time domain represents one hour;

[0057] Substitute the charging start time of the i-th user into the time axis Z. The time domain corresponding to each charging behavior is marked as {s1, s2, s3, ..., sn}, where s1 to sn represent the time domain corresponding to the first to n-th charging behavior of the i-th user;

[0058]

[0059] In the formula, represents the average time domain corresponding to the charging behavior of the i-th user, sf represents the time domain corresponding to the f-th charging behavior of the i-th user, f∈n, According to the standard deviation formula, the fluctuation index Bdz of the charging behavior of the i-th user in a month is calculated. i ,quantify the regularity of users’ charging time and provide a calculable migration basis;

[0060] Step 3: Based on the user data set, analyze the charging power gl of each user's charging behavior, determine the behavior attributes, and statistically generate the fast and slow charging ratio Kmb;

[0061] The calculation process of charging power gl is as follows:

[0062] According to the user data set, the charging amount of the i-th user at the f-th charging time is marked as df, and the charging time of the i-th user at the f-th charging time is marked as c f ;

[0063]

[0064] In the formula, glf represents the charging power of the i-th user during the f-th charging;

[0065] The behavior attribute judgment process is as follows:

[0066] If the charging power of the i-th user during the f-th charging is gl f ≥37.5kW, considered as fast charging behavior;

[0067] If the charging power of the i-th user during the f-th charging is gl f <37.5kW, judged as slow charging behavior;

[0068] The calculation process of the fast and slow charging ratio Kmb is as follows:

[0069] Count the total number of fast charging behaviors KC of the i-th user in a month i and the total number of slow charging behaviors MC i ;

[0070]

[0071] In the formula, Kmb i represents the ratio of fast and slow charging behavior of the i-th user in a month, accurately reflecting the user's preference for different charging speeds;

[0072] Step 4: Based on the regional data set and the fast and slow charging ratio Kmb, the usage status of charging piles in each region is analyzed in real time to generate the corresponding supply index Gyz;

[0073] The calculation process of the supply index Gyz is as follows:

[0074] According to the regional data set, the management data of the kth charging area is extracted and the area of ​​the kth charging area is marked as mj k , mark the total number of fast charging piles in the kth charging area as kz k , mark the available number of fast charging piles in the kth charging area as ky k , mark the total number of slow charging piles in the kth charging area as mz k , mark the available number of slow charging piles in the kth charging area as my k ;

[0075] If the i-th user's charging behavior in a month is fast and slow charging ratio Kmb i ≥1, indicating that users prefer to choose fast charging piles.

[0076]

[0077] In the formula, α1 represents the weight of the ratio of the available number of fast charging piles to the total number of fast charging piles. represents the density of fast charging piles in the kth charging area, α2 represents the weight for the density of fast charging piles, α1 and α2 are both constants, α1+α2=1, According to the weights α1 and α2, the supply index Gyz of the kth charging area for the i-th user is calculated. k ;

[0078] If the i-th user's charging behavior in a month is fast and slow charging ratio Kmb i <1, indicating that users prefer slow charging piles.

[0079]

[0080] In the formula, β1 represents the weight of the ratio of the available number of slow charging piles to the total number of slow charging piles. represents the density of slow charging piles in the kth charging area, β2 represents the weight for the density of slow charging piles, β1 and β2 are both constants, β1+β2=1, According to the weights β1 and β2, the supply index Gyz of the kth charging area for the i-th user is calculated. k , targeted judgment of regional real-time supply capabilities;

[0081] Step 5: Set a fixed number threshold CY and a fluctuation threshold BDY. Then, combine the user dataset and the fluctuation index Bdz to identify long-tail users and generate corresponding charging behavior migration recommendations.

[0082] The long-tail user judgment process is as follows:

[0083] If the total number of charging times n of the i-th user in a month is lower than the number threshold CY, it means that the i-th user is a long-tail user with less charging behavior, and the user's charging experience should be optimized in a timely manner;

[0084] If the fluctuation index Bdz of the charging behavior of the i-th user i If the fluctuation threshold BDY is exceeded, it means that the i-th user is a long-tail user with unstable charging habits, and the charging success rate should be improved in time;

[0085] If the charging locations of the i-th user are not in the same charging area within a month, it means that the i-th user has unstable charging habits and the difficulty of finding charging piles should be reduced.

[0086] Accurately identify long-tail users and fill service gaps in data-sparse scenarios. For long-tail users with less charging behavior, it is recommended that users be given priority to choose charging areas adjacent to the last charging location. This is in line with the long-tail users' dependence on habitual paths, reduces exploration costs, and improves acceptance. For long-tail users with unstable charging habits, it is recommended that users be given priority to choose charging areas with a higher supply index Gyz, maximizing the success rate of queue-free charging, meeting the random needs of long-tail users, and achieving high utilization of personalized migration resources.

[0087] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A charging behavior transfer learning method for long-tail users, characterized by: The following steps are involved: Step 1: Connect to the big data platform through the network to obtain the management data of all users' charging behaviors and the management data of all charging areas, and classify them into user data sets and regional data sets; Step 2: Set the time axis Z, and then combine it with the user data set to analyze the degree of change in the user's charging time each time, and generate the corresponding fluctuation index Bdz; Step 3: Based on the user data set, analyze the charging power gl of each user's charging behavior, determine the behavior attributes, and statistically generate the fast and slow charging ratio Kmb; Step 4: Based on the regional data set and the fast and slow charging ratio Kmb, the usage status of charging piles in each region is analyzed in real time to generate the corresponding supply index Gyz; Step 5: Set a fixed number threshold CY and a fluctuation threshold BDY. Then, combine the user dataset and the fluctuation index Bdz to identify long-tail users and generate corresponding charging behavior migration recommendations.

2. The charging behavior transfer learning method for long-tail users according to claim 1 is characterized by: In step 1, the user data set includes the total number of charging times, charging start time, charging amount, charging time and charging location of each user.

3. The charging behavior transfer learning method for long-tail users according to claim 2 is characterized by: In step 1, the regional data set includes the area of ​​each charging area, the total number of fast charging piles, the number of available fast charging piles, the number of available slow charging piles, and the total number of slow charging piles.

4. The charging behavior transfer learning method for long-tail users according to claim 3 is characterized by: In step 2, the calculation process of the volatility index Bdz is as follows: Based on the user data set, extract the management data of the i-th user and mark the total number of charging times of the i-th user in a month as n; Divide the time axis Z into 24 time domains and label them as time domain 1, time domain 2, time domain 3, ..., time domain 24, where each time domain represents one hour; Substitute the charging start time of the i-th user into the time axis Z. The time domain corresponding to each charging behavior is marked as {s1, s2, s3, ..., sn}, where s1 to sn represent the time domain corresponding to the first to n-th charging behavior of the i-th user; In the formula, represents the average time domain corresponding to the charging behavior of the i-th user, sf represents the time domain corresponding to the f-th charging behavior of the i-th user, f∈n, According to the standard deviation formula, the fluctuation index Bdz of the charging behavior of the i-th user in a month is calculated. i .

5. The charging behavior transfer learning method for long-tail users according to claim 4 is characterized by: In step 3, the charging power g1 calculation process is as follows: According to the user data set, the charging amount of the i-th user at the f-th charging time is marked as d f , mark the charging time of the i-th user during the f-th charging as c f ; In the formula, gl f represents the charging power of the i-th user during the f-th charging.

6. The charging behavior transfer learning method for long-tail users according to claim 5 is characterized by: In step 3, the behavior attribute judgment process is as follows: If the charging power of the i-th user during the f-th charging is gl f ≥37.5kW, considered as fast charging behavior; If the charging power of the i-th user during the f-th charging is gl f <37.5kW, it is considered slow charging behavior.

7. The charging behavior transfer learning method for long-tail users according to claim 6 is characterized by: In step 3, the calculation process of the fast and slow charging ratio Kmb is as follows: Count the total number of fast charging behaviors KC of the i-th user in a month i and the total number of slow charging behaviors MC i ; In the formula, Kmb i It represents the ratio of fast charging to slow charging behavior of the i-th user in one month.

8. The charging behavior transfer learning method for long-tail users according to claim 7 is characterized by: In step 4, the supply index Gyz calculation process is as follows: According to the regional data set, the management data of the kth charging area is extracted and the area of ​​the kth charging area is marked as mj k , mark the total number of fast charging piles in the kth charging area as kz k , mark the available number of fast charging piles in the kth charging area as ky k , mark the total number of slow charging piles in the kth charging area as mz k , mark the available number of slow charging piles in the kth charging area as my k ; If the i-th user's charging behavior in a month is fast and slow charging ratio Kmb i ≥1, In the formula, α1 represents the weight of the ratio of the available number of fast charging piles to the total number of fast charging piles. represents the density of fast charging piles in the kth charging area, α2 represents the weight for the density of fast charging piles, α1 and α2 are both constants, α1+α2=1, According to the weights α1 and α2, the supply index Gyz of the kth charging area for the i-th user is calculated. k ; If the i-th user's charging behavior in a month is fast and slow charging ratio Kmb i <1, In the formula, β1 represents the weight of the ratio of the available number of slow charging piles to the total number of slow charging piles. represents the density of slow charging piles in the kth charging area, β2 represents the weight for the density of slow charging piles, β1 and β2 are both constants, β1+β2=1, According to the weights β1 and β2, the supply index Gyz of the kth charging area for the i-th user is calculated. k .

9. The charging behavior transfer learning method for long-tail users according to claim 8 is characterized by: In step 5, the long-tail user judgment process is as follows: If the total number of charging times n of the i-th user in a month is lower than the number threshold CY, it means that the i-th user is a long-tail user with less charging behavior; If the fluctuation index Bdz of the charging behavior of the i-th user i Exceeding the fluctuation threshold BDY indicates that the i-th user is a long-tail user with unstable charging habits; If the charging locations of the i-th user do not belong to the same charging area within a month, it means that the i-th user is a long-tail user with unstable charging habits.

10. The charging behavior transfer learning method for long-tail users according to claim 9 is characterized by: In step five, for long-tail users with less charging behavior, it is recommended that users be given priority to choose a charging area adjacent to the last charging location. For long-tail users with unstable charging habits, it is recommended that users be given priority to choose a charging area with a higher supply index Gyz.

Images