CN111159533B - Intelligent charging service recommendation method and system based on user image - Google Patents

Abstract

The invention provides an intelligent charging service recommendation method and system based on user portraits, which includes: obtaining an initial optional charging station based on the state of an electric vehicle, the current coordinates of the electric vehicle and the resource situation of the charging station; and based on the pre-obtained user portrait. The label weight determines the pushable charging stations from the initial optional charging stations; the user portrait label weight is determined based on the multi-task deep neural network training on the characteristic data and user behavior data in the user order data. Through the implementation of the above solution, the charging demand in user travel scenarios determined by the current electric vehicle status and user portrait can be accurately predicted, thereby providing users with the optimal charging station selection solution. The implementation of the above solution not only enhances the user's sense of gain, but also improves the charging efficiency. Site operating income.

Description

An intelligent charging service recommendation method and system based on user portraits

Technical field

The invention relates to the field of electric vehicle charging, and in particular to an intelligent recommendation system for electric vehicle charging stations.

Background technique

The recommendation system is a personalized information push system that recommends information, products, etc. that the user is interested in to the user based on the user's information needs, interests, etc. A good recommendation system can not only provide users with personalized services, but also establish a close relationship with users, making users dependent on recommendations. In recent years, the development of machine learning and deep learning and other fields has provided methodological guidance for recommendation systems, which have been widely used in many fields. Among them, the most typical field with good development and application prospects is the field of e-commerce.

However, in the field of charging services in the electric vehicle after-service market, the application of recommendation systems is not mature. The reason lies in the particularity of the "commodity" recommended in the charging service scenario. This "commodity" is essentially electric energy, similar to a gas station, but when it provides energy services to users, the factors that affect the user's decision-making often go beyond the attributes of the product itself, such as site location, traffic status, etc. At the same time, unlike gasoline, the properties of the electric energy provided by the charging station are much more complex, including: 1) Similar to the quality of gasoline that affects the service life of the gasoline pump, the size of the charging power will also affect the use of the power battery. lifespan; 2) Because the charging time is long and the number of charging stations is small, the queuing time at charging stations is much larger than that at gas stations; 3) Unlike refueling prices that are priced by region, charging costs are mainly determined by electricity prices, charging service fees and parking fees , now electricity prices generally adopt peak and valley electricity prices, which will fluctuate greatly at different times of the day. With the introduction of the dynamic pricing mechanism of charging service fees, the randomness of charging fees will become increasingly large; 4) Although the demand side of electricity There is no difference in quality, but from the supply side, the electricity supplied by users may come from traditional thermal power plants or renewable energy. As carbon emission rights are piloted in some cities, whether the charged energy is "green electricity" is also potential. One of the influencing factors that may affect users' decisions in the future. To sum up, the "commodity" attribute of charging services has great randomness and uncertainty. Therefore, how to improve the matching between the charging recommendation scheme in the charging station recommendation system and the actual needs of users is a key issue that technicians in this field need to solve. question.

Contents of the invention

In order to improve the matching between the charging recommendation scheme in the charging station recommendation system and the actual needs of users, the present invention adopts the following technical solutions:

A smart charging service recommendation method based on user portraits, including:

Based on the status of the electric vehicle, the current coordinates of the electric vehicle and the resource situation of the charging station, obtain an initial optional charging station;

Based on the pre-obtained user portrait tag weight, determine pushable charging stations from the initial optional charging stations;

The weight of the user portrait label is determined based on the multi-task deep neural network training on the feature data and user behavior data in the user order data.

Preferably, training user portrait label weights includes:

Based on the set user portrait tags, perform tag matching between the feature data in the user order data and the user behavior data, and generate a training log;

Process the training log through feature hashing and low-frequency filtering methods, or equal-frequency discretization methods to obtain training data;

Bring the training data into the multi-task deep neural network model for training to obtain the user portrait label weight;

The user portrait tags include: opening hours, idle rate, distance, charging power, cost and environment;

The characteristic data includes: user characteristic data, asset characteristic data and environment characteristic data.

Preferably, the training logs are processed through feature hashing and low-frequency filtering methods, including:

Convert the training log into a real number matrix group through feature hashing;

Perform low-frequency filtering processing on the discrete features of the real number matrix group, and remove features that are less than the occurrence frequency threshold to form the training data.

Preferably, the training log is processed through an equal-frequency discretization method, including:

Divide the data in the training log according to fixed frequencies to obtain several groups of equal sample sizes.

Preferably, based on the status of the electric vehicle, the current coordinates of the electric vehicle and the resources of the charging station, the initial optional charging stations include:

Calculate the safe driving range of electric vehicles in their current state;

Taking the current coordinate position of the electric vehicle as the center of the circle and the safe driving range as the radius, determine the safe driving range of the electric vehicle;

Set all available charging stations within the safe driving range of the electric vehicle as initial optional charging stations;

Wherein, the remaining driving range of the electric vehicle in the current state is determined by the safe driving range.

Preferably, the calculation formula for the remaining driving range of the electric vehicle in its current state is as follows:

Among them, S _rest is the remaining driving range, SOC ₁ is the power battery charge at the current coordinate position, SOH is the health status of the power battery, C is the rated capacity of the power battery, V is the current voltage of the power battery, and InitEC is the unit driving mileage. energy consumed.

Preferably, based on the pre-obtained user portrait tag weight, the pushable charging stations are determined from the initial optional charging stations, including:

Calculate the user charging location decision-making influence factor value of each power station in the initial optional charging station;

Determine the charging stations that can be pushed based on the pre-obtained user portrait tag weight and the user charging location decision-making influence factor value of each power station;

Label the pushable charging station with the corresponding label and push it to customers;

The influence factor value of the user charging location decision of the power station includes: time influence factor value, cost influence factor value and environmental influence factor value.

Preferably, the calculation formula for determining pushable charging stations is as follows:

Rank=α·ΔT+β·P+γ·X

Among them, Rank is the initial optional power station score, α is the time tag weight, β is the cost tag weight, γ is the environmental tag weight, △T is the time impact factor value, P is the cost impact factor value, and X is the environmental impact factor value.

Preferably, the time influence factor value calculation formula is as follows:

△T＝△T ₂ +△T ₃ +△T ₄ +△T ₅

Among them, △T is the time influence factor value, △T ₂ is the driving time of the electric vehicle from the current location to the initial optional charging station location, △T ₃ is the queuing time after arriving at the initial optional charging station, △T ₄ is the initial The optional charging station payment time, △T ₅ is the charging time of the electric vehicle at the initial optional charging station.

Preferably, the calculation formula for the charging time △T ₅ of an electric vehicle at the initial optional charging station is as follows:

Among them, △E is the charging capacity of the electric vehicle, and P _power is the charging power.

Preferably, the calculation formula for electric vehicle charging capacity △E is as follows:

ΔE={SOC ₂ -(SOC ₁ +ΔSOC)}×C

Among them, SOC ₂ is the battery charge level that the electric vehicle is expected to achieve, SOC ₁ is the battery charge level of the electric vehicle at the current location, ΔSOC is the battery charge level loss of the electric vehicle from the current location to the initial optional charging station location, and C is the electric vehicle's battery charge level. The rated capacity of a car battery.

Preferably, the cost impact factor value P is calculated as follows:

P＝[P ₁ +P ₂ ]×△E+P ₃ ×(△T ₃ +△T ₄ +△T ₅ );

Among them, P ₁ is the charging unit price of the initial optional charging station, P ₂ is the charging service fee of the initial optional charging station, △E is the charging capacity of the electric vehicle, P ₃ is the parking fee of the initial optional charging station, and △T ₃ is the initial optional charging station parking fee. The queuing time after the optional charging station, △T ₄ is the payment time at the initial optional charging station, and △T ₅ is the charging time of the electric vehicle at the initial optional charging station.

Preferably, the time tag weight α is calculated as follows:

α=α ₁ +α ₂ +α ₃ +α ₄

Among them, α ₁ is the opening time tag weight, α ₂ is the idle rate tag weight, α ₃ is the distance tag weight, and α ₄ is the charging power tag weight.

Based on the same inventive concept, the present invention also provides an intelligent charging service recommendation system based on user portraits, including:

Initial optional charging station summoning module and pushable charging station determination module;

The initial optional charging station calling module is used to obtain initial optional charging stations based on the status of the electric vehicle, the current coordinates of the electric vehicle and the resource situation of the charging station;

The pushable charging station determination module is used to determine pushable charging stations from the initial optional charging stations based on the pre-obtained user portrait tag weight;

The weight of the user portrait label is determined based on the multi-task deep neural network training on the feature data and user behavior data in the user order data.

Preferably, an intelligent charging service recommendation system based on user portraits also includes a user portrait label weight training module. The user portrait label weight training module includes:

Label matching unit, data processing unit and data training unit;

The label base matching unit is used to perform label matching on the characteristic data in the user order data and the user behavior data according to the set user portrait label, and generate a training log;

The data processing unit is used to process the training log through feature hashing and low-frequency filtering methods, or equal-frequency discretization methods to obtain training data;

The data training unit is used to bring the training data into the multi-task deep neural network model for training, and obtain the user portrait label weight;

The user portrait tags include: opening hours, idle rate, distance, charging power, cost and environment;

The characteristic data includes: user characteristic data, asset characteristic data and environment characteristic data.

Preferably, the initial optional charging station summoning module includes:

Safe driving range calculation unit, safe driving range calculation unit and initial optional charging station determination unit;

The safe driving range calculation unit is used to calculate the safe driving range of the electric vehicle in its current state;

The safe driving range calculation unit is used to determine the safe driving range of the electric vehicle with the position of the electric vehicle as the center of the circle and the safe driving range as the radius;

The initial optional charging station determination unit is used to set all available charging stations within the safe driving range as initial optional charging stations;

Wherein, the remaining driving range of the electric vehicle in the current state is determined by the safe driving range.

Preferably, the charging station determination module can be pushed, including:

User charging location decision influencing factor value calculation unit, pushable charging station determination unit and push unit;

The user charging location decision influencing factor value calculation unit is used to calculate the user charging location decision influencing factor value of each power station in the initial optional charging station;

The pushable charging station determination unit is used to determine the pushable charging station based on the pre-obtained user portrait tag weight and the user charging location decision-making influence factor value of each power station;

The push unit is used to label the pushable charging station with a corresponding label and push it to customers;

The influence factor value of the user charging location decision of the power station includes: time influence factor value, cost influence factor value and environmental influence factor value.

Compared with the closest existing technology, the technical solution provided by the present invention has the following beneficial effects:

The invention provides an intelligent charging service recommendation method and system based on user portraits, which includes: obtaining an initial optional charging station based on the state of an electric vehicle, the current coordinates of the electric vehicle and the resource situation of the charging station; and based on the pre-obtained user portrait. The label weight determines the pushable charging stations from the initial optional charging stations; the user portrait label weight is determined based on the multi-task deep neural network training on the characteristic data and user behavior data in the user order data. Through the implementation of the above solution, the charging demand in user travel scenarios determined by the current electric vehicle status and user portrait can be accurately predicted, thereby providing users with the optimal charging station selection solution. The implementation of the above solution not only enhances the user's sense of gain, but also improves the charging efficiency. Site operating income.

Description of the drawings

Figure 1 is a schematic flow chart of an intelligent charging service recommendation method based on user portraits provided by the present invention;

Figure 2 is a schematic diagram of the basic structure of an intelligent charging service recommendation system based on user portraits provided by the present invention;

Figure 3 is a detailed structural diagram of an intelligent charging service recommendation system based on user portraits provided by the present invention;

Figure 4 is a schematic technical roadmap of an intelligent charging service recommendation system based on user portraits provided in an embodiment of the present invention;

Figure 5 is a schematic diagram of a product prototype of an intelligent charging service recommendation system based on user portraits provided in an embodiment of the present invention.

Detailed ways

The present invention will be further described in detail below in conjunction with the accompanying drawings.

Example 1:

The recommendation system provided in this embodiment is bound to the user to implement data collection and interaction. The user can specify at least one electric vehicle to use. The flow of the recommendation method is shown in Figure 1, including: based on the status of the electric vehicle, the electric vehicle Based on the current coordinates of the car and the resources of the charging station, the initial optional charging stations are obtained; based on the pre-obtained user portrait label weights, the pushable charging stations are determined from the initial optional charging stations; the user portrait label weights are based on the multi-task deep neural network Train and determine the feature data and user behavior data in the user order data.

Based on the multi-task deep neural network, the feature data and user behavior data in the user order data are trained to generate user portrait label weights, including:

Based on the user charging location decision-making influence factor value received through the mobile APP when the user generates charging needs, determine the label system of the user profile, including opening time, idle rate, distance, charging power and cost;

Based on the user order data of the State Grid Smart Internet of Vehicles platform, the sensitivity of various users to the elasticity coefficient of charging prices and charging pile information, vehicle pile distance, road conditions and other information was analyzed, and the training of five user portrait label weights was completed, such as As shown in Figure 4.

The recommended process includes stages such as training log generation, training data generation, model training, and online scoring. When the recommendation system recommends users with charging needs, it will record the user characteristic data, asset characteristic data, and environmental characteristic data at that time, and collect user behavior feedback for this recommendation.

Tag matching: The recommendation background log will record the user characteristics, asset characteristics and environmental characteristics corresponding to the current sample. The label log will capture the user's behavioral feedback on the recommended items. The system splices the two pieces of data together according to the unique ID to generate the original training log.

The input requirements of most machine learning algorithms are real-number matrices, so the original data needs to be converted into a real-number matrix first. This process is called feature hashing (compressing the original high-dimensional feature vector into a lower-dimensional feature vector, and trying not to Loss of expressiveness of original features).

Too many extremely sparse discrete features will cause overfitting problems during the training process (making the assumptions too strict in order to obtain consistent hypotheses), and at the same time increase the number of parameter storage. To avoid this problem, low-frequency filtering is performed on discrete features, and features smaller than the occurrence frequency threshold are discarded.

Through the analysis of the training data, it can be found that the value distribution of features in different dimensions and the feature values in the same dimension are very different. For example, data on characteristics such as distance and price obeys a long-tail distribution, which is reflected in the fact that the eigenvalues of most samples are relatively small, and a small number of samples have very large eigenvalues. Therefore, in practice, normalization is performed based on the position of the feature value in the cumulative distribution function, that is, the features are divided into buckets with equal frequency to ensure that the sample size in each bucket is basically equal. This method ensures that features with different distributions can be mapped to approximately uniform distribution, thus ensuring the distinction of features between samples and the stability of their values.

A multi-task deep neural network model is used to train user label weights, even if the input layer has n_in neurons and the output layer has n_out neurons. Plus some hidden layers containing several neurons. At this time, it is necessary to find the appropriate linear coefficient matrix W and bias vector b corresponding to all hidden layers and output layers, so that the output calculated from all training sample inputs is as equal to or very close to the sample output as possible.

After obtaining the user portrait label weight matrix that satisfies the evaluation effect function, the system saves the dynamic data as static data.

When a user generates a charging demand, the initial optional charging station is first determined based on the current coordinate information of the electric vehicle:

Based on the state of charge of the electric vehicle battery, road conditions, environment, air conditioning usage status, etc., the remaining driving range of the electric vehicle is calculated. The remaining driving range calculation formula is as follows:

Among them, S _rest is the remaining driving range, SOC ₁ is the power battery charge at the current coordinate position, SOH is the health status of the power battery, C is the rated capacity of the power battery, V is the current voltage of the power battery, and InitEC is the unit driving mileage. energy consumed;

Determine the current safe driving range of electric vehicles based on 80% of the remaining driving range of electric vehicles;

Draw a circle with the current position of the electric vehicle as the center and the safe driving range as the radius to determine the safe driving range of the electric vehicle;

Search all available charging stations within the safe driving range and return them to the recommendation system as initial optional charging stations.

Secondly, based on the saved user portrait tag weight matrix, the initial optional charging stations are sorted and pushed to the user, including:

Based on the user portrait tag weight, environmental characteristics and initial optional charging station status at the current moment, all charging stations in the initial optional charging stations are sorted and scored. The calculation formula is as follows:

Rank=α·ΔT+β·P+γ·X

Rank is the initial optional power station score, α is the time tag weight, β is the cost tag weight, γ is the environmental tag weight, △T is the time impact factor value, P is the cost impact factor value, and X is the environmental impact factor value;

Among them, the calculation formula of the time influence factor value △T is as follows:

△T＝△T ₂ +△T ₃ +△T ₄ +△T ₅

△T ₂ is the driving time of the electric vehicle from the current location to the initial optional charging station location, △T ₃ is the queuing time after arriving at the initial optional charging station, △T ₄ is the payment time to the initial optional charging station, △T ₅ Charging time for electric vehicles at initial optional charging stations;

Among them, the calculation formula of the cost impact factor value P is as follows:

P＝[P ₁ +P ₂ ]×△E+P ₃ ×(△T ₃ +△T ₄ +△T ₅ );

P ₁ is the charging unit price of the initial optional charging station, P ₂ is the charging service fee of the initial optional charging station, △E is the charging capacity of the electric vehicle, P ₃ is the parking fee of the initial optional charging station, and △T ₃ is the initial optional charging station upon arrival. The queuing time after the charging station, △T ₄ is the payment time at the initial optional charging station, △T ₅ is the charging time of the electric vehicle at the initial optional charging station;

The charging decision influencing factor value is obtained as follows:

△T ₂ =t ₃ -t ₂ : Call the Amap API interface, input Loc ₂ , t ₂ and Loc ₃ , and return t ₃ ;

△T ₃ : It is necessary to obtain the statistical model of the number of vehicles connected to each site based on historical order data. For example: the fitting number of electric vehicles arriving at charging station A on a certain working day approximately obeys the Poisson distribution P(λ) , and then the queuing waiting time can be predicted based on the number of arrivals between the current site's idle status and △T ₂ .

△T ₄ : The scan code payment response time was optimized by other teams and was set to 1s in the first phase of the project;

△T ₅ The charging time △T ₅ of an electric vehicle at the initial optional charging station is calculated as follows:

△E is the charging capacity of the electric vehicle, and P _power is the charging power.

The calculation formula of electric vehicle charging capacity △E is as follows:

ΔE={SOC ₂ -(SOC ₁ +ΔSOC)}×C

SOC ₂ is the expected battery charge of the electric vehicle, SOC ₁ is the battery charge of the electric vehicle at the current location, ΔSOC is the battery charge loss of the electric vehicle from the current location to the initial optional charging station location, C is the electric vehicle battery rated capacity.

P ₁ : Call asset data and update regularly according to prices set by the National Development and Reform Commission;

_P2 : According to the "Pricing Catalog" issued by the Beijing Municipal Development and Reform Commission in 2018, the pricing power for electric vehicle charging service fees has been fully developed since April 2018. Other provinces and cities in the charging market cultivation stage are also gradually liberalizing charging services. Fee pricing power;

P ₃ : Retrieve asset data;

Finally, the top three charging solutions are labeled with different labels (such as best price, shortest time, etc.) or key information is highlighted (such as charging cost, idle rate, etc.) and pushed to users through the front end.

Analysis of user order data obtained during the early operation of eCharging shows that when making decisions about charging locations, users will receive status information of five charging stations through the App, including: charging station opening hours, idle rate, charging power, and fees. and the distance from the user’s current location to the charging station. In the decision-making process, users will estimate driving time, queuing time, payment time, and charging time to judge whether the opening hours of the charging station are appropriate, estimate the possible waiting time in line through the idle rate, and estimate the distance from the current location to the charging station and the map Use the intersection information on the road to estimate the driving time on the road, estimate the charging time through the charging power of the charging station, and finally estimate the charging cost through the charging price (including electricity price and charging service fee) and parking fee, and determine the user profile in this system The label system includes opening hours, idle rate, distance, charging power and cost.

The schematic diagram of the product prototype of the intelligent charging service recommendation system based on user portraits is shown in Figure 5.

Example 2:

The basic structure diagram of an intelligent charging service recommendation system based on user portraits is shown in Figure 2, including:

Initial optional charging station summoning module and pushable charging station determination module;

An initial optional charging station call module is used to obtain initial optional charging stations based on the status of the electric vehicle, the current coordinates of the electric vehicle and the resource situation of the charging station;

A pushable charging station determination module is used to determine pushable charging stations from the initial optional charging stations based on the pre-obtained user portrait tag weight;

The weight of the user portrait label is determined based on the multi-task deep neural network training on the feature data and user behavior data in the user order data.

The detailed structural diagram of an intelligent charging service recommendation system based on user portraits is shown in Figure 3. It also includes a household portrait label weight training module. The household portrait label weight training module includes:

Label matching unit, data processing unit and data training unit;

A label base matching unit, used to perform label matching on the characteristic data in the user order data and the user behavior data according to the set user portrait label, and generate a training log;

A data processing unit, used to process the training log through feature hashing and low-frequency filtering methods, or equal-frequency discretization methods to obtain training data;

A data training unit, used to bring the training data into the multi-task deep neural network model for training, and obtain the user portrait label weight;

User portrait tags include: opening hours, idle rate, distance, charging power, cost and environment;

Characteristic data includes: user characteristic data, asset characteristic data and environment characteristic data.

Among them, the data processing unit includes a feature hashing subunit and a low-frequency filtering subunit;

The feature hash subunit is used to convert the training log into a real matrix group through feature hash;

The low-frequency filtering subunit is used to perform low-frequency filtering processing on the discrete features of the real number matrix group, remove features that are less than the occurrence frequency threshold, and form the training data.

Among them, the initial optional charging station calling module includes: a safe driving mileage calculation unit, a safe driving range calculation unit and an initial optional charging station determination unit;

A safe driving range calculation unit is used to calculate the safe driving range of the electric vehicle in its current state;

A safe driving range calculation unit is used to determine the safe driving range of the electric vehicle with the position of the electric vehicle as the center of the circle and the safe driving range as the radius;

An initial optional charging station determination unit is used to set all available charging stations within the safe driving range as initial optional charging stations;

The remaining driving range of an electric vehicle in its current state is determined by the safe driving range.

The pushable charging station determination module includes: a user charging location decision influencing factor value calculation unit, a pushable charging station determination unit and a push unit;

A user charging location decision influencing factor value calculation unit, configured to calculate the user charging location decision influencing factor value of each power station in the initial optional charging station;

The pushable charging station determination unit is used to determine the pushable charging station based on the pre-obtained user portrait tag weight and the user charging location decision-making influence factor value of each power station;

A push unit, used to label the pushable charging station with a corresponding label and push it to customers;

The influence factor values of the user charging location decision of the power station include: time influence factor value, cost influence factor value and environmental influence factor value.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and do not limit the scope of protection. Although the present application has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: After reading this application, those skilled in the art can still make various changes, modifications or equivalent substitutions to the specific implementation methods of the application, but these changes, modifications or equivalent substitutions are within the scope of the claims of the application that are pending approval.

Claims (16)

1. An intelligent charging service recommendation method based on user portraits, which is characterized by including: Based on the status of the electric vehicle, the current coordinates of the electric vehicle and the resources of the charging station, obtain the initial charging station; Based on the pre-obtained user portrait tag weight, determine the push charging station from the initial charging stations; The weight of the user portrait label is determined based on the multi-task deep neural network training on the feature data and user behavior data in the user order data; Determining push charging stations from the initial charging stations based on the pre-obtained user portrait tag weights includes: Calculate the influence factor value of user charging location decision-making for each power station in the initial charging station; Determine the push charging station based on the pre-obtained user portrait tag weight and the user charging location decision-making influence factor value of each power station; Label the push charging station with the corresponding label and push it to customers; The power station's user charging location decision-making impact factor values include: time impact factor value, cost impact factor value and environmental impact factor value; The user portrait tag weight includes the time tag weight, the cost tag weight and the environment tag weight; the time tag weight includes the opening time tag weight, the idle rate tag weight, the distance tag weight, and the charging power tag weight. 2. The method of claim 1, wherein the user portrait label weight includes: Based on the set user portrait tags, perform tag matching between the feature data in the user order data and the user behavior data, and generate a training log; Process the training log through feature hashing and low-frequency filtering methods, or equal-frequency discretization methods to obtain training data; Bring the training data into the multi-task deep neural network model for training to obtain the user portrait label weight; The user portrait tags include: opening hours, idle rate, distance, charging power, cost and environment; The characteristic data includes: user characteristic data, asset characteristic data and environment characteristic data. 3. The method of claim 2, wherein the training log is processed by feature hashing and low-frequency filtering methods, including: Convert the training log into a real number matrix group through feature hashing; Perform low-frequency filtering processing on the discrete features of the real number matrix group, and remove features that are less than the occurrence frequency threshold to form the training data. 4. The method of claim 2, wherein the training log is processed by an equal-frequency discretization method, including: Divide the data in the training log according to fixed frequencies to obtain several groups of training data with equal sample sizes. 5. The method of claim 1, wherein obtaining the initial charging station based on the state of the electric vehicle, the current coordinates of the electric vehicle and the resource situation of the charging station includes: Calculate the safe driving range of electric vehicles in their current state; Taking the current coordinate position of the electric vehicle as the center of the circle and the safe driving range as the radius, determine the safe driving range of the electric vehicle; Set all available charging stations within the safe driving range of the electric vehicle as initial charging stations; Wherein, the safe driving range of the electric vehicle in its current state is determined by the remaining driving range. 6. The method of claim 5, wherein the remaining driving range calculation formula of the electric vehicle in its current state is as follows: Among them, S _rest is the remaining driving range, SOC ₁ is the power battery charge at the current coordinate position, SOH is the health status of the power battery, C is the rated capacity of the power battery, V is the current voltage of the power battery, and InitEC is the unit driving mileage. energy consumed. 7. The method of claim 1, wherein the calculation formula for determining the push charging station is as follows: Rank=α·ΔT+β·P+γ·X Among them, Rank is the initial power station score, α is the time tag weight, β is the cost tag weight, γ is the environmental tag weight, △T is the time impact factor value, P is the cost impact factor value, and X is the environmental impact factor value. 8. The method of claim 7, wherein the time influence factor value calculation formula is as follows: △T＝△T ₂ +△T ₃ +△T ₄ +△T ₅ Among them, △T is the time influence factor value, △T ₂ is the driving time of the electric vehicle from the current location to the initial charging station location, △T ₃ is the queuing time after arriving at the initial charging station, and △T ₄ is the initial charging station payment time. , △T ₅ is the charging time of the electric vehicle at the initial charging station. 9. The method of claim 8, wherein the calculation formula for the charging time ΔT ₅ of the electric vehicle at the initial charging station is as follows: Among them, △E is the charging capacity of the electric vehicle, and P _power is the charging power. 10. The method according to claim 9, wherein the calculation formula of the electric vehicle charging quantity ΔE is as follows: ΔE={SOC ₂ -(SOC ₁ +ΔSOC)}×C Among them, SOC ₂ is the expected battery charge level of the electric vehicle, SOC ₁ is the battery charge level of the electric vehicle at the current location, ΔSOC is the battery charge loss of the electric vehicle from the current location to the initial charging station location, and C is the electric vehicle battery. rated capacity. 11. The method of claim 7, wherein the calculation formula of the cost impact factor value P is as follows: P＝[P ₁ +P ₂ ]×△E+P ₃ ×(△T ₃ +△T ₄ +△T ₅ ); Among them, P ₁ is the charging unit price at the initial charging station, P ₂ is the charging service fee at the initial charging station, △E is the charging capacity of the electric vehicle, P ₃ is the parking fee at the initial charging station, and △T ₃ is the queuing time after arriving at the initial charging station. , △T ₄ is the initial charging station payment time, △T ₅ is the charging time of the electric vehicle at the initial charging station. 12. The method of claim 7, wherein the time tag weight α is calculated as follows: α=α ₁ +α ₂ +α ₃ +α ₄ Among them, α ₁ is the opening time tag weight, α ₂ is the idle rate tag weight, α ₃ is the distance tag weight, and α ₄ is the charging power tag weight. 13. An intelligent charging service recommendation system based on user portraits, used to implement an intelligent charging service recommendation method based on user portraits as claimed in claim 1, characterized in that it includes: Initial optional charging station summoning module and pushable charging station determination module; The initial optional charging station calling module is used to obtain initial optional charging stations based on the status of the electric vehicle, the current coordinates of the electric vehicle and the resource situation of the charging station; The pushable charging station determination module is used to determine pushable charging stations from the initial optional charging stations based on the pre-obtained user portrait tag weight; The weight of the user portrait label is determined based on the multi-task deep neural network training on the feature data and user behavior data in the user order data. 14. The system according to claim 13, further comprising a user portrait label weight training module, and the user portrait label weight training module includes: Label matching unit, data processing unit and data training unit; The label base matching unit is used to perform label matching on the characteristic data in the user order data and the user behavior data according to the set user portrait label, and generate a training log; The data processing unit is used to process the training log through feature hashing and low-frequency filtering methods, or equal-frequency discretization methods to obtain training data; The data training unit is used to bring the training data into the multi-task deep neural network model for training, and obtain the user portrait label weight; The user portrait tags include: opening hours, idle rate, distance, charging power, cost and environment; The characteristic data includes: user characteristic data, asset characteristic data and environment characteristic data. 15. The system of claim 13, wherein the initial optional charging station summoning module includes: Safe driving range calculation unit, safe driving range calculation unit and initial optional charging station determination unit; The safe driving range calculation unit is used to calculate the safe driving range of the electric vehicle in its current state; The safe driving range calculation unit is used to determine the safe driving range of the electric vehicle with the position of the electric vehicle as the center of the circle and the safe driving range as the radius; The initial optional charging station determination unit is used to set all available charging stations within the safe driving range as initial optional charging stations; Wherein, the remaining driving range of the electric vehicle in the current state is determined by the safe driving range. 16. The system of claim 13, wherein the pushable charging station determination module includes: User charging location decision influencing factor value calculation unit, pushable charging station determination unit and push unit; The user charging location decision influencing factor value calculation unit is used to calculate the user charging location decision influencing factor value of each power station in the initial optional charging station; The pushable charging station determination unit is used to determine the pushable charging station based on the pre-obtained user portrait tag weight and the user charging location decision-making influence factor value of each power station; The push unit is used to label the pushable charging station with a corresponding label and push it to customers; The influence factor value of the user charging location decision of the power station includes: time influence factor value, cost influence factor value and environmental influence factor value.