New energy vehicle charging station site recommendation method, system, device, medium and product

Abstract

The application provides a new energy vehicle charging station site recommendation method, system, equipment, medium and product, and relates to the technical field of intelligent transportation, and the method comprises the following steps: acquiring multi-dimensional target feature data of a target new energy vehicle and real-time exposure station feature data; inputting the multi-dimensional target feature data and the real-time exposure station feature data into a charging station site recommendation model to obtain a charging station site recommendation result corresponding to the target new energy vehicle output by the charging station site recommendation model; the charging station site recommendation model is obtained by training a deep learning model based on training samples marked with positive and negative sample labels; wherein the training samples marked with negative sample labels are adjusted by a weight coefficient in the training process. The application provides more accurate charging station site recommendation results for users.

Description

Technical Field

[0001] The present invention relates to the field of intelligent transportation technology, and in particular to a method, system, device, medium and product for recommending charging stations for new energy vehicles. Background Art

[0002] With the increasing popularity of new energy vehicles, charging issues are gaining increasing attention. Users demand accurate, convenient, and personalized charging station recommendations. Charging station operators also need a unified, efficient station recommendation platform to improve user reach, station click-through rates, and order conversion rates.

[0003] Current intelligent recommendation solutions for new energy vehicle charging stations typically use click data as positive samples and exposed but unclicked data as negative samples for model training, lacking in-depth research into the unique behavioral characteristics of new energy vehicle owners. Unlike e-commerce recommendation scenarios, new energy vehicle users often experience repeated exposure before making a final charging decision, and each exposure may gradually approach the final charging station as the user drives purposefully. This progressive selection behavior causes the distance-based candidate set to consistently focus on the final charging station before the final charge. This leads to systematic bias when directly training the model using the exposed sample set, resulting in lower accuracy in charging station recommendations.

[0004] Therefore, there is an urgent need for a new energy vehicle charging station recommendation method, system, equipment, medium and product to solve the above problems. Summary of the Invention

[0005] In response to the problems existing in the prior art, the present invention provides a new energy vehicle charging station recommendation method, system, equipment, medium and product.

[0006] The present invention provides a method for recommending new energy vehicle charging stations, comprising:

[0007] Acquire multi-dimensional target feature data and real-time exposure site feature data of a target new energy vehicle, wherein the multi-dimensional target feature data includes behavioral feature data generated by a target user terminal of the target new energy vehicle when selecting a charging site during a historical period; the real-time exposure site feature data is behavioral feature data generated by the target user terminal when browsing target exposure sites in real time during the current driving process of the target new energy vehicle; the target exposure sites represent charging sites browsed by the target user terminal but not yet determined to be selected;

[0008] inputting the multi-dimensional target feature data and the real-time exposure site feature data into a charging site recommendation model to obtain a charging site recommendation result corresponding to the target new energy vehicle and output by the charging site recommendation model; the charging site recommendation model is obtained by training a deep learning model based on training samples marked with positive and negative sample labels; wherein the training samples marked with negative sample labels are adjusted in training contribution by a weight coefficient in the training process; the corresponding negative sample weight coefficient of the training samples marked with the negative sample labels is set in the training process and is calculated by a negative sample weighting formula, and the negative sample weighting formula is specifically:

[0009] ;

[0010] wherein, represents a negative sample weight coefficient, 、 and represents a hyperparameter, represents an exposure site total browsing time length, represents an exposure site total display number, represents an exposure site interaction number, represents exposure site time length information, represents a repeated exposure site proportion, represents a simple negative sample, represents a negative sample weight coefficient corresponding to the simple negative sample, represents a sample exposure site.

[0011] According to the new energy automobile charging site recommendation method provided by the application, the charging site recommendation model is obtained by the following steps:

[0012] According to the sample user portrait information and the sample user historical statistical features, a sample user feature is constructed, wherein the sample user portrait information is the user portrait corresponding to the sample user end in the sample new energy vehicle; the sample user historical statistical features include the site features, time features, latitude and longitude features and charging station number features of the charging site selected by the sample user end at a historical moment;

[0013] According to the charging site portrait information, supporting facility information, charging history success rate information and charging site historical attraction rate in the preset charging station distribution area, a sample site feature is constructed; wherein the charging site historical attraction rate is calculated based on the distance between the charging site selected by the sample user end at a historical moment and the sample user end;

[0014] According to the sample user terminal in the historical moment, the corresponding user real-time latitude and longitude and exposure time, the sample real-time exposure feature data is constructed;

[0015] According to a plurality of sample user historical statistical features and a plurality of sample real-time exposure feature data, a sample user historical sequence feature is constructed;

[0016] According to the exposure site browsing total time length generated by the sample user terminal at the historical moment, the exposure site display total number, the exposure site interaction times, the exposure site time length information and the repeated exposure site proportion, the sample user site selection behavior feature is constructed; wherein the exposure site time length information represents the time length between the sample exposure site and the order charging site, and the order charging site is the charging site determined by the sample user terminal in the sample order;

[0017] According to the sample user feature, the sample site feature, the sample user historical sequence feature and the sample real-time exposure feature data, the training sample is generated;

[0018] The training sample set is constructed by marking the positive sample label for the training sample in which the sample exposure site is selected as the order charging site by the sample user terminal, and marking the negative sample label for the training sample without marking the positive sample label.

[0019] Based on the training sample set, the deep learning model is trained to obtain the charging site recommendation model.

[0020] According to the new energy automobile charging site recommendation method provided by the application, the negative sample label is marked for the training sample without marking the positive sample label, which comprises:

[0021] Based on the training sample in which the sample exposure site is not selected as the order charging site by the sample user terminal, a simple negative sample is randomly determined from the unexposed site corresponding to each sample exposure site before the generation of the sample charging order of the sample user terminal, and each sample exposure site before the generation of the sample charging order of the sample user terminal is determined as a non-simple negative sample, wherein the unexposed site is a charging site not browsed by the sample user terminal;

[0022] According to the sample user site selection behavior feature in the simple negative sample and the non-simple negative sample, the corresponding negative sample weight coefficient is calculated;

[0023] Based on the training sample without marking the positive sample label, the negative sample label is constructed; and according to the negative sample weight coefficient, the corresponding weight coefficient is set for the training sample marked with the negative sample label in the training process.

[0024] According to the new energy vehicle charging station recommendation method provided by the application, after the deep learning model is trained based on the training sample set and the charging station recommendation model is obtained, the method further comprises:

[0025] Based on the K-means clustering algorithm, the user classification processing is performed on the training sample set, and the training sample subset corresponding to each user type is obtained;

[0026] Based on the training sample subset, the parameter fine-tuning is performed on the charging station recommendation model, and the charging station classification recommendation model corresponding to each user type is obtained.

[0027] According to the new energy vehicle charging station recommendation method provided by the application, the multi-dimensional target feature data and the real-time exposure station feature data are input into the charging station recommendation model, and the charging station recommendation result corresponding to the target new energy vehicle output by the charging station recommendation model is obtained, which comprises:

[0028] It is judged whether the target user end exists corresponding to the user type;

[0029] If the target user end exists corresponding to the user type, according to the user type, the multi-dimensional target feature data and the real-time exposure station feature data are input into the charging station classification recommendation model corresponding to the user type, and the charging station recommendation result is obtained;

[0030] If the target user end does not exist corresponding to the user type, the multi-dimensional target feature data and the real-time exposure station feature data are input into the charging station recommendation model, and the charging station recommendation result is obtained.

[0031] The application also provides a new energy vehicle charging station recommendation system, comprising:

[0032] The feature data acquisition module is used for acquiring multi-dimensional target feature data and real-time exposure station feature data of a target new energy vehicle, wherein the multi-dimensional target feature data comprises behavior feature data generated when a target user end in the target new energy vehicle selects a charging station in a historical period; the real-time exposure station feature data is behavior feature data generated when the target user end browses a target exposure station in real time in a current driving process of the target new energy vehicle; and the target exposure station represents a charging station browsed by the target user end and not selected.

[0033] The charging station recommendation module is configured to input the multi-dimensional target feature data and the real-time exposure station feature data into a charging station recommendation model to obtain a charging station recommendation result corresponding to the target new energy vehicle output by the charging station recommendation model; the charging station recommendation model is obtained by training a deep learning model based on training samples marked with positive and negative samples; wherein the training samples marked with negative sample labels are adjusted by a weight coefficient in the training process; the corresponding negative sample weight coefficient of the training samples marked with the negative sample labels is set in the training process, and is calculated by a negative sample weighting formula, the negative sample weighting formula is specifically:

[0034] ;

[0035] wherein, the negative sample weight coefficient is represented by, 、 and the hyperparameter is represented by, the total exposure station browsing time is represented by, the total number of exposure station displays is represented by, the number of exposure station interactions is represented by, the exposure station duration information is represented by, the repeated exposure station proportion is represented by, the simple negative sample is represented by, the negative sample weight coefficient corresponding to the simple negative sample is represented by, the sample exposure station is represented by.

[0036] The application also 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 implements the new energy vehicle charging station recommendation method of any of the above.

[0037] The application also provides a non-transitory computer readable storage medium having a computer program stored thereon, wherein the computer program is executable by a processor to implement the new energy vehicle charging station recommendation method of any of the above.

[0038] The application also provides a computer program product including a computer program, wherein the computer program is executable by a processor to implement the new energy vehicle charging station recommendation method of any of the above.

[0039] The new energy vehicle charging station recommendation method, system, device, medium and product provided by the application, by acquiring multi-dimensional target feature data of the target new energy vehicle and real-time exposure station feature data, and then inputting the data into the charging station recommendation model trained by the training samples labeled with positive and negative sample labels, a more accurate charging station recommendation result is provided for the user. BRIEF DESCRIPTION OF DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the application or prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced below. Obviously, the drawings in the following description are some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.

[0041] Figure 1 The flowchart of the new energy vehicle charging station recommendation method provided by the application;

[0042] Figure 2 The structure diagram of the new energy vehicle charging station recommendation system provided by the application;

[0043] Figure 3 The structure diagram of the electronic device provided by the application. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solutions and advantages of the application more clear, the technical solutions in the application will be described clearly and completely below in combination with the drawings in the application. Obviously, the described embodiments are some embodiments of the application, not all embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor belong to the protection scope of the application.

[0045] With the popularity of new energy vehicles, the charging problem of new energy vehicles becomes more and more important. Car owners urgently need a precise, convenient and personalized charging station recommendation service, and charging station operators also need to use a unified and efficient station recommendation platform to expand user reach, improve station click rate and order conversion rate.

[0046] Under this background, combined with user historical charging behavior, surrounding station information, and not relying on user travel planning recommendation system, such system is deployed by charging station operators, by integrating user and charging station attributes, interaction behavior and interaction context and other massive data, using neural network model to predict user preference for each charging station, and generating personalized recommendation list for users accordingly.

[0047] Recommendation system technology has been widely applied in e-commerce, social media, and advertisement pushing. In the process of training recommendation system models, the commonly used frameworks include collaborative filtering, dual tower model, Wide&Deep model, and sequence recommendation model covers DIN, BERT4REC model. In addition, transfer learning is also widely used in recommendation system to further improve the recommendation effect.

[0048] It is worth noting that the recommendation system model is extremely sensitive to input samples, so the selection and enhancement of positive and negative samples become the key direction of the industry to optimize the recommendation system. In the existing intelligent recommendation charging station system, the user's current location, charging station surrounding information and supporting equipment information are mainly used for word segmentation and context coding, and the click rate is improved as the optimization target, and deep learning model is used for modeling. However, the application effect of deep learning recommendation system is highly dependent on the selection and processing of training samples. At present, the intelligent recommendation solution of new energy charging station often directly uses click data as positive samples and exposure non-click data as negative samples for model training, but this approach lacks in-depth study of the unique behavior characteristics of new energy vehicle owners.

[0049] Unlike e-commerce recommendation scenarios, new energy vehicle users often have repeated exposure before making the final charging decision, and each exposure may gradually approach the final charging station as the vehicle owner drives to the destination. This gradual selection behavior causes the precision recall-based candidate set to always recall around the final charging station before the final charging, resulting in a systematic bias when directly using the exposure sample set to train the model.

[0050] In addition, the existing charging site recommendation system has not fully considered the differentiated needs of different types of new energy vehicle owners. For example, there are significant differences between operating vehicle owners and private vehicle owners in charging station selection: operating vehicle owners pay more attention to charging efficiency and economy, while private vehicle owners may pay more attention to charging experience and convenience. If this demand difference is ignored, it will lead to a lack of targetedness in the recommendation results, which cannot meet the actual needs of different user groups, and thus reduce the accuracy of the recommendation system and user satisfaction.

[0051] To solve the problems in the prior art, the application provides a new energy vehicle charging station site recommendation method based on user behavior, overcomes the sample selection bias caused by the direct use of exposure-click data of a traditional recommendation system, eliminates the data bias caused by the progressive selection characteristics of new energy vehicle owners, and proposes a sample dynamic weighting mechanism based on the real charging behavior of users, effectively alleviating the candidate set distribution deviation problem caused by the distance recall mechanism. Secondly, the application solves the problem of user demand difference. A user group differentiated feature engineering system is established for the essential difference in charging demand of different user types (such as operation vehicle users and private vehicle users), the technical limitation that a single recommendation model cannot adapt to the personalized needs of different user groups is overcome, a personalized recommendation strategy based on user intelligent classification is realized, and the accuracy and practicality of the new energy charging station recommendation system are significantly improved, providing more accurate and personalized charging station recommendation services for different types of users.

[0052] Figure 1 The flowchart of the new energy vehicle charging station site recommendation method provided by the application is shown in Figure 1 The application provides a new energy vehicle charging station site recommendation method, which comprises the following steps:

[0053] In step 101, multi-dimensional target feature data of a target new energy vehicle and real-time exposure site feature data are obtained, wherein the multi-dimensional target feature data includes behavior feature data generated by a target user end in the target new energy vehicle when selecting a charging station site in a historical period; the real-time exposure site feature data is behavior feature data generated by the target user end when browsing a target exposure site in real time during the current driving process of the target new energy vehicle; and the target exposure site represents a charging station site that is browsed by the target user end and not selected.

[0054] In the application, the multi-dimensional target feature data refers to behavior feature data generated by a target user end (i.e. a vehicle owner or user) in a target new energy vehicle when selecting a charging station site in a historical period. These data may include but are not limited to the user's past charging habits, preferred site types (such as fast charging stations and slow charging stations), charging time distribution, sensitivity to charging prices, preferences for site facilities (such as whether there are rest areas, convenience stores, etc.), and the correlation between the user's historical driving route and the selection of charging station sites. By analyzing these multi-dimensional target feature data, the user's charging behavior and preferences can be understood in depth, thereby providing a strong basis for subsequent charging station site recommendation.

[0055] The real-time exposure site feature data refers to behavior feature data generated by the target user terminal when the target new energy vehicle is currently driven and the target user terminal is currently browsing the target exposure site. The real-time exposure site feature data includes the duration of browsing the site, detailed information of browsing the site (such as price, location, facilities, etc.), whether a collection or sharing operation is performed, and the interaction behavior of the user and the site (such as clicking navigation, viewing evaluation, etc.). In the present application, the target exposure site refers to a charging site that is browsed by the user but has not yet been determined to be selected.

[0056] In the present application, the real-time exposure site feature data reflects the immediate needs and points of interest of the user in the current situation. By capturing these real-time behavior features, the current preferences and decision-making processes of the user can be more accurately grasped, thereby providing more accurate and timely charging site recommendations for the user.

[0057] In step 102, the multi-dimensional target feature data and the real-time exposure site feature data are input into a charging site recommendation model to obtain a charging site recommendation result corresponding to the target new energy vehicle output by the charging site recommendation model. The charging site recommendation model is obtained by training a deep learning model based on training samples labeled with positive and negative sample labels. The training samples labeled with negative sample labels are adjusted in the training process by adjusting the contribution of the weight coefficient. The corresponding negative sample weight coefficient of the training samples labeled with the negative sample label is calculated by a negative sample weighting formula in the training process. The negative sample weighting formula is specifically as follows:

[0058] ;

[0059] wherein, represents the negative sample weight coefficient, , and represents a hyperparameter, represents the total duration of exposure site browsing, represents the total number of exposure site displays, represents the number of exposure site interactions, represents exposure site duration information, represents the proportion of repeated exposure sites, represents a simple negative sample, represents the negative sample weight coefficient corresponding to the simple negative sample, represents a sample exposure site.

[0060] In the present invention, the multi-dimensional target feature data and real-time exposure site feature data obtained in the above embodiment are input into the charging site recommendation model. The charging site recommendation model will analyze and calculate based on these data, and finally output the charging site recommendation results for the target new energy vehicle.

[0061] In the present invention, the charging site recommendation model is obtained through deep learning model training. The training samples include multi-dimensional sample feature data and sample real-time exposure feature data marked with positive and negative sample labels. Among them, the multi-dimensional sample feature data comes from user behavior records of multiple new energy vehicles, covering the charging selection behavior of different users in different situations; the sample real-time exposure feature data describes the characteristics of charging sites that users browsed but may not have selected.

[0062] Furthermore, these training samples are marked with corresponding positive and negative sample labels, and the positive and negative sample labels are used to indicate whether the user finally selected a certain charging station as the charging location, wherein the positive sample label indicates that the user selected the station, and the negative sample label indicates that the user did not select the station. In the present invention, for training samples marked with negative sample labels, their training contributions will be adjusted by weight coefficients during the training process. The purpose of weighted processing is to better reflect the importance or confidence of negative samples in model training. For example, certain selected stations that are repeatedly exposed during driving and the user exposure time is too short will be given lower weights. Preferably, the present invention weights the loss function of the training samples based on the negative sample weights, so that when training the deep learning model, more attention can be paid to those exposure records that have a greater impact on user decisions, thereby improving the accuracy of the recommendation results.

[0063] The new energy vehicle charging station recommendation method provided by the present invention obtains multi-dimensional target feature data of the target new energy vehicle and real-time exposure site feature data, and then inputs this data into a charging station recommendation model trained by training samples marked with positive and negative sample labels, providing users with more accurate charging station recommendation results.

[0064] Based on the above embodiment, the charging station recommendation model is trained by the following steps:

[0065] Constructing sample user features based on sample user portrait information and sample user historical statistical features, wherein the sample user portrait information is a user portrait corresponding to a sample user terminal in a sample new energy vehicle; the sample user historical statistical features include site features, time features, latitude and longitude features, and charging station number features corresponding to the charging station determined and selected by the sample user terminal at a historical moment;

[0066] According to preset charging station site image information, supporting facility information, charging history success rate information and charging station site historical attraction rate in a charging station site delivery area, sample site features are constructed; wherein the charging station site historical attraction rate is calculated based on the distance between the charging station site selected by the sample user end at a historical moment and the sample user end;

[0067] According to the sample user end in historical moment, the corresponding user real-time latitude and longitude and exposure time, sample real-time exposure feature data are constructed;

[0068] According to a plurality of sample user historical statistical features and a plurality of sample real-time exposure feature data, sample user historical sequence features are constructed;

[0069] According to the exposure site browsing total time, exposure site display total number, exposure site interaction times, exposure site time length information and repeated exposure site proportion generated by the sample user end at a historical moment, sample user site selection behavior features are constructed; wherein the exposure site time length information represents the time length between the sample exposure site and the order charging station, and the order charging station is the charging station determined by the sample user end in the sample order;

[0070] According to the sample user features, the sample site features, the sample user historical sequence features and the sample real-time exposure feature data, the training sample is generated;

[0071] The training sample is marked as a positive sample label if the sample exposure site is selected as the order charging station by the sample user end, and the training sample is marked as a negative sample label if the positive sample label is not marked, so as to construct a training sample set.

[0072] Based on the training sample set, the deep learning model is trained to obtain the charging station recommendation model.

[0073] In the present application, the historical order data and historical exposure data of the user are collected; then each piece of information is aggregated and matched according to the user, order ID and historical exposure site ID, sample feature data are generated, and the features related to the user are cross-processed in advance to construct high-order interaction features with other features. For features with missing values, the median or mode of historical data is used for filling; for classification features, label encoding is performed; for numerical features, normalization processing is uniformly performed.

[0074] Specifically, in the present application, the sample user portrait information is derived from the user end in the sample new energy vehicle; the sample user historical statistical features include the site features (such as site type, location, etc.) corresponding to the charging station selected by the sample user end at the historical moment, the time features (such as the time period for selecting charging), the latitude and longitude features (the specific geographic location of the charging station), and the charging station number features (the specific charging station number used).

[0075] In the present application, the sample site features are constructed according to the charging station portrait information, the supporting facility information, the charging history success rate information, and the charging station historical attraction rate in the preset charging station deployment area, wherein the charging station historical attraction rate is calculated based on the distance between the charging station selected by the sample user end at the historical moment and the sample user end, reflecting the attraction of the site to the user.

[0076] Further, the sample real-time exposure feature data is constructed according to the corresponding user real-time latitude and longitude and exposure time when the sample user end browses the sample exposure site at the historical moment, which reflects the time information and the point of interest of the user when browsing the charging station. Then, the sample user historical sequence features are constructed according to the multiple sample user historical statistical features and the multiple sample real-time exposure feature data, which are input into the deep learning model in sequence, facilitating the deep learning model to capture the information of each fine-grained behavior of the user in the past, such as the charging selection behavior of the user at different times and places.

[0077] Further, the sample user site selection behavior features are constructed according to the exposure site browsing total duration, the exposure site display total number, the exposure site interaction times, the exposure site duration information, and the repeated exposure site proportion generated by the sample user end at the historical moment. Among them, the exposure site duration information represents the duration between the sample exposure site and the order charging station, reflecting the degree of attention and decision-making process of the user to the exposure site.

[0078] After constructing the feature data of the above-mentioned dimensions, the sample user features, the sample site features, the sample user historical sequence features, and the sample real-time exposure feature data cover multiple dimensions such as user, site, and user historical behavior. Then, the training samples are generated according to these multi-dimensional sample feature data.

[0079] Further, the training samples that determine the sample exposure site to be selected as the order charging station by the sample user end are marked as positive sample labels; the training samples that are not marked as positive sample labels are marked as negative sample labels, so as to construct the training sample set, wherein the training samples marked with negative sample labels will be weighted according to the weighting coefficient in the training process to improve the training efficiency of the model for important samples.

[0080] Finally, the deep learning model is trained based on the training sample set to obtain a charging station recommendation model. In the present application, a deep learning model based on the Two-Stream architecture is used, and a feature enhancement technique is used to achieve accurate recommendation. Specifically, the training sample set is input into the deep learning model, and in the feature processing stage, the model first preprocesses all input features through a feature segmentation layer, and then introduces a feature enhancement layer (such as the SENet attention mechanism, the MMOE multi-task learning module) for feature optimization. In the present application, the main body of the deep learning model adopts an extensible Two-Stream architecture, supporting multiple network implementations such as xDeepFM, DCN, and FinalMLP. In the feature fusion stage, the model uses a high-order linear fusion model to perform high-order feature interaction on the dual-stream output. In the training process of the present application, Binary Cross Entropy is used as the loss function, the sample loss function is weighted using a weighting coefficient, and the model convergence is optimized through adaptive learning rate adjustment.

[0081] On the basis of the above-mentioned embodiments, the marking negative sample labels on the training samples without marking the positive sample labels comprises:

[0082] Based on the training samples in which the sample exposure station is not selected as the order charging station by the sample user end, a simple negative sample is randomly determined from the unexposed station corresponding to the location of each sample exposure station before the generation of the sample charging order of the sample user end, and each sample exposure station before the generation of the sample charging order of the sample user end is determined as a non-simple negative sample, wherein the unexposed station is a charging station that has not been browsed by the sample user end;

[0083] According to the sample user station selection behavior features in the simple negative samples and the non-simple negative samples, respectively, the corresponding negative sample weight coefficients are calculated;

[0084] Based on the training samples without marking the positive sample labels, the negative sample labels are constructed, and according to the negative sample weight coefficients, the corresponding weight coefficients of the training samples marked with the negative sample labels are set in the training process.

[0085] The existing recommendation model often uses user clicks / order items as positive samples and exposed non-clicks / exposed non-order items as negative samples. Due to the repeated exposure of new energy vehicle owners at the same station and systematic bias in exposure behavior, the latest exposure data or all exposure data of the user cannot be directly used as negative samples for the station recommendation model.

[0086] In the present application, the purpose of the weighting operation is to better reflect the user behavior habits, especially the key time points and places that can stimulate the user to make charging consumption. Through the weighting operation, the deep learning model can pay more attention to the exposure records that have greater influence on the user's decision, thereby improving the accuracy of the prediction.

[0087] Specifically, taking a sample user's order as an example, all exposure records occurring within the distance threshold D and the time threshold T before the matching order occurs are recorded as . Within the same , the last determined order site (i.e., the finally selected charging site) is taken as the only positive sample, and other negative samples are weighted according to the negative sample weight coefficient calculated according to the sample user site selection behavior characteristics.

[0088] In the present application, based on the training samples of the sample exposure sites not selected as the order charging sites by the sample user end, from the charging sites around each sample exposure site of the sample user's sample charging order before it is generated, a simple negative sample (EasyNegative) is randomly determined, and the sample exposure site that has been exposed is determined as a non-simple negative sample. Among them, the simple negative sample refers to extracting N unexposed charging sites within the distance threshold D around each exposure occurrence location as the negative sample, which is used to help the deep learning model to quickly fit; the non-simple negative sample is other unselected order charging sites and exposure sites that have been browsed by the sample user end.

[0089] On the basis of the above embodiment, the negative sample weight coefficient is calculated by a negative sample weighting formula, and the negative sample weighting formula is specifically:

[0090] ;

[0091] Among them, represents the negative sample weight coefficient, , and represents a hyperparameter, represents the total exposure site browsing time, represents the total number of exposure site displays, represents the number of exposure site interactions, represents the exposure site duration information, represents the proportion of repeated exposure sites, represents the simple negative sample, represents the negative sample weight coefficient corresponding to the simple negative sample, represents the sample exposure site.

[0092] In the present application, for non-simple negative samples, the weighting process is as follows:

[0093] ;

[0094] Wherein, duration represents the total duration of the exposure site browsing, and a longer dwell time may mean that the user is more interested in the exposure content. impression_item_count represents the total number of exposure sites displayed, that is, the number of exposure sites in the exposure site browsing page displayed by the sample user end, and more exposure sites may mean that the user has browsed more content, increasing the opportunity for interaction. interaction represents the number of exposure site interactions, that is, the number of user interactions, such as site clicks, and a large number of interactions indicate that the user has a higher level of engagement with the content. timediff represents the exposure site duration information, that is, the duration between the time when the site is exposed and the time when the order is generated, and the closer the time when the site is exposed to the time when the order is generated, the more new information the exposure provides. duplication represents the proportion of repeated exposure sites, that is, the proportion of repeated sites compared to the first exposure. In the present application, the higher the repetition rate, the less new information the exposure provides, and the more likely it is that the user consciously selects the site and then repeats the image site, which is a strong interference to the model.

[0095] For simple negative samples, their weights can be fixed as In the weighting process, the hyperparameters (such as , and , etc.) involved need to be determined through training parameter adjustment and evaluated on the validation set. In the present application, hyperparameter adjustment can consider using methods such as grid search, random search and Bayesian optimization to find the optimal parameter combination and improve the prediction accuracy of the model.

[0096] Finally, based on the unlabeled positive sample label, sample user features, sample site features, sample user historical sequence features and sample real-time exposure feature data, negative sample labels are constructed, and in the training process, the training contribution of training samples labeled with negative sample labels is adjusted through the negative sample weight coefficient, thereby helping the deep learning model to learn how to distinguish between charging sites that the user may select and those that the user may not select.

[0097] On the basis of the above embodiment, after the deep learning model is trained based on the training sample set to obtain the charging site recommendation model, the method further comprises:

[0098] Based on the K-means clustering algorithm, the user classification processing is performed on the training sample set to obtain the training sample subset corresponding to each user type;

[0099] Based on the training sample subset, the charging station site recommendation model is parameter fine-tuned to obtain a charging station site classification recommendation model corresponding to each user type.

[0100] In the present application, the K-means clustering algorithm is used for user classification processing of the training sample set. As an unsupervised learning algorithm, K-means can divide data points (i.e. sample users) into K different clusters (i.e. user types), so that the data points in the same cluster are as similar as possible.

[0101] Specifically, the present application first constructs a multi-dimensional feature space containing site distribution features, time series features and user behavior habit features through a data preprocessing process. Then, the Principal Component Analysis (PCA) algorithm is applied to reduce the dimension of high-dimensional features, and the principal components with cumulative explained variance of more than 95% are selected as feature inputs, effectively reducing the data dimension and retaining key feature information. Further, an improved K-means clustering algorithm is used for user grouping, wherein the clustering effect is optimized by the Silhouette Coefficient and the Elbow Method, and the optimal clustering number K value is automatically determined, so as to dynamically identify and continuously update the classification of price-sensitive users, distance-sensitive users, battery power-sensitive users and operation vehicle users. Different user feedbacks for different classifications are distinct, and it is difficult to optimize them simultaneously when using the same deep learning recommendation algorithm for prediction. The classification results of these users will be used as the basis for subsequent classification fine-tuning models.

[0102] Further, based on the training sample set corresponding to different user categories determined in the above embodiment, a hierarchical optimization strategy based on low-rank adaptation (such as LoRA) is adopted, and all parameters in the charging station site recommendation model except the feature enhancement layer and the feature fusion layer are frozen to realize fine adjustment of the model. Then, fine-tuning tasks are performed for different user groups, and the feature fusion layer and the feature enhancement are progressively unfrozen during the fine-tuning process. In the fine-tuning process, binary cross-entropy is used as the loss function to maintain consistency with the charging station site recommendation model. Finally, the charging station site recommendation model fine-tuned based on the training sample set of different user groups forms the charging station site classification recommendation model corresponding to the user group.

[0103] On the basis of the above embodiment, the input of the multi-dimensional target feature data and the real-time exposure site feature data into the charging station site recommendation model to obtain the charging station site recommendation result corresponding to the target new energy vehicle output by the charging station site recommendation model comprises:

[0104] determining whether the target user terminal has a corresponding user type;

[0105] If the target user terminal has a corresponding user type, input the multi-dimensional target feature data and the real-time exposure site feature data into the charging site classification recommendation model corresponding to the user type according to the user type, to obtain the charging site recommendation result;

[0106] If the target user terminal does not have a corresponding user type, input the multi-dimensional target feature data and the real-time exposure site feature data into the charging site recommendation model to obtain the charging site recommendation result.

[0107] In the present application, first, the user ID of the new energy vehicle and the exposure site ID (exposure site determined based on the user terminal browsing process) are obtained as indexes; then, according to these indexes, the multi-dimensional features related to the user ID and the site ID are extracted from the database in real time, which include but are not limited to:

[0108] User features: may include the user's charging habits, preferred site type, sensitivity to price, etc.

[0109] Site features: including site location, facilities and price information, etc. These information helps the model to evaluate the attractiveness of the site.

[0110] User historical order behavior sequence: records the user's past charging site features and charging behavior (such as charging capacity, charging duration), which can reflect the user's charging mode.

[0111] User historical exposure behavior sequence: records the user's exposure habits, such as exposure time period, exposure latitude and longitude sequence, which can reflect the user's potential interest and decision-making process.

[0112] In addition, the real-time exposure feature data of the request will also be extracted, such as the time information and latitude and longitude when the user browses the exposure site, etc.

[0113] Then, according to the user ID, it is determined whether the user has a stored clustering result. If the user does not have a corresponding clustering result, the main model (i.e. charging site recommendation model) will be called for inference. If the user ID has a corresponding clustering result, the classification model corresponding to the user type (i.e. charging site classification recommendation model) will be called for inference. Finally, the main model or the classification model will eventually output the scores of each charging site, which reflect the recommendation priority of each charging site, and then the charging sites are sorted according to these scores, so as to provide the optimal charging site recommendation scheme for the user. Among them, the higher the score of the site, the higher the position in the recommendation list, and the easier it is to be selected by the user.

[0114] The application can trace and identify the key time points and places stimulating the user to generate charging consumption by deeply analyzing the historical behavior trajectory of the new energy vehicle user end. Based on these key nodes, the corresponding exposure samples are dynamically weighted, thereby strengthening the key exposure samples in the user behavior trajectory, effectively alleviating the sample selection bias problem caused by the user exposure behavior habit, and significantly improving the generalization ability of the model. At the same time, the key cost features affecting the user behavior are identified, including charging price, time cost, distance cost and battery endurance cost, etc. In the model input stage, the key features are crossed, and a feature selection layer is used to strengthen the influence of the key cost features on the final result. These reinforced features can more accurately reflect the user's decision preference, thereby improving the prediction accuracy of the model. In addition, the application uses hierarchical fine-tuning technology to generate personalized recommendation models for different user groups according to the differences in behavior characteristics of different types of users, effectively learning the feedback differences of different user groups, and significantly improving the user group conversion rate which is difficult to improve in a single model.

[0115] The new energy vehicle charging station recommendation system provided by the application is described below, and the new energy vehicle charging station recommendation system described below can be correspondingly referred to the new energy vehicle charging station recommendation method described above.

[0116] Figure 2 The structure diagram of the new energy vehicle charging station recommendation system provided by the application is as follows, Figure 2As shown, the application provides a new energy vehicle charging station recommendation system, comprising a feature data acquisition module 201 and a charging station recommendation module 202, wherein the feature data acquisition module 201 is used to obtain multi-dimensional target feature data of a target new energy vehicle and real-time exposure station feature data, wherein the multi-dimensional target feature data includes behavior feature data generated by a target user end in the target new energy vehicle when selecting a charging station in a historical period; the real-time exposure station feature data is behavior feature data generated by the target user end when browsing a target exposure station in real time in the current driving process of the target new energy vehicle; the target exposure station represents a charging station that is browsed by the target user end and not selected; the charging station recommendation module 202 is used to input the multi-dimensional target feature data and the real-time exposure station feature data into a charging station recommendation model to obtain a charging station recommendation result corresponding to the target new energy vehicle output by the charging station recommendation model; the charging station recommendation model is obtained by training a deep learning model based on training samples labeled with positive and negative sample labels; wherein the training samples labeled with negative sample labels are adjusted by weight coefficients in the training process; the corresponding negative sample weight coefficient of the training sample labeled with the negative sample label is set in the training process, and is calculated by a negative sample weighting formula, the negative sample weighting formula is specifically:

[0117] ;

[0118] Wherein, represents the negative sample weight coefficient, , and represent hyperparameters, represents the total exposure station browsing time, represents the total number of exposure station displays, represents the number of exposure station interactions, represents the exposure station duration information, represents the repeated exposure station proportion, represents a simple negative sample, represents the negative sample weight coefficient corresponding to the simple negative sample, represents a sample exposure station.

[0119] The new energy vehicle charging station recommendation system provided by the application obtains multi-dimensional target feature data of a target new energy vehicle and real-time exposure station feature data, and then inputs these data into a charging station recommendation model trained by training samples labeled with positive and negative sample labels, to provide more accurate charging station recommendation results for users.

[0120] The system provided in the embodiment of the present invention is used to execute the above-mentioned method embodiments. Please refer to the above-mentioned embodiments for the specific process and detailed content, which will not be repeated here.

[0121] Figure 3 A schematic diagram of the structure of the electronic device provided by the present invention, such as Figure 3 As shown, the electronic device may include: a processor (Processor) 301, a communication interface (Communications Interface) 302, a memory (Memory) 303 and a communication bus 304, wherein the processor 301, the communication interface 302 and the memory 303 communicate with each other via the communication bus 304. The processor 301 may call the logic instructions in the memory 303 to execute a method for recommending charging sites for new energy vehicles, the method comprising: obtaining multi-dimensional target feature data and real-time exposure site feature data of a target new energy vehicle, wherein the multi-dimensional target feature data includes the behavior feature data generated by the target user terminal in the target new energy vehicle when selecting a charging site in a historical period; the real-time exposure site feature data is the behavior feature data generated by the target user terminal when browsing the target exposure site in real time during the current driving process of the target new energy vehicle; the target exposure site represents a charging site that has been browsed by the target user terminal but has not yet been determined to be selected. ; Input the multi-dimensional target feature data and the real-time exposure site feature data into the charging site recommendation model to obtain the charging site recommendation result corresponding to the target new energy vehicle output by the charging site recommendation model; the charging site recommendation model is obtained by training the deep learning model based on the training samples marked with positive and negative sample labels; wherein, the training samples marked with negative sample labels are adjusted for training contribution through weight coefficients during the training process; during the training process, the corresponding negative sample weight coefficient is set for the training samples marked with the negative sample labels, which is calculated by the negative sample weighting formula, and the negative sample weighting formula is specifically as follows:

[0122] ;

[0123] in, represents the negative sample weight coefficient, 、 and represents the hyperparameter, Indicates the total browsing time of the exposure site. Indicates the total number of exposure sites displayed. Indicates the number of interactions with the exposure site. Indicates the exposure site duration information, Indicates the proportion of repeated exposure sites, represents a simple negative sample, a negative sample weight coefficient corresponding to a simple negative sample, representing a sample exposure site.

[0124] In addition, the logical instructions in the memory 303 described above can be implemented in the form of a software function unit and sold or used as an independent product, and can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the parts that contribute to the prior art or parts of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various program code storage media.

[0125] On the other hand, the present application also provides a computer program product, which comprises a computer program stored on a non-transitory computer readable storage medium, and the computer program comprises program instructions, when the program instructions are executed by a computer, the computer can execute the new energy vehicle charging station recommendation method provided by the above-mentioned method, the method comprises: obtaining multi-dimensional target feature data of a target new energy vehicle and real-time exposure site feature data, wherein the multi-dimensional target feature data comprises behavior feature data generated by a target user end in the target new energy vehicle when selecting a charging station in a historical period; the real-time exposure site feature data is behavior feature data generated by the target user end when browsing a target exposure site in real time in the current driving process of the target new energy vehicle; the target exposure site represents a charging station that is browsed by the target user end and not determined to be selected; inputting the multi-dimensional target feature data and the real-time exposure site feature data into a charging station recommendation model to obtain a charging station recommendation result corresponding to the target new energy vehicle output by the charging station recommendation model; the charging station recommendation model is obtained by training a deep learning model based on training samples labeled with positive and negative sample labels; wherein the training samples labeled with negative sample labels are adjusted by weight coefficients in the training process; the corresponding negative sample weight coefficient of the training samples labeled with the negative sample labels is set in the training process, and is calculated by a negative sample weighting formula, the negative sample weighting formula is specifically:

[0126] ;

[0127] wherein, denotes a negative sample weight coefficient, , and denotes a hyperparameter, denotes total exposure site browsing time, denotes total exposure site display quantity, denotes exposure site interaction times, denotes exposure site duration information, denotes repeated exposure site proportion, denotes a simple negative sample, denotes a negative sample weight coefficient corresponding to a simple negative sample, denotes a sample exposure site.

[0128] In another aspect, the application also provides a non-transitory computer readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the new energy vehicle charging station recommendation method provided by the above embodiments. The method comprises: obtaining multi-dimensional target feature data of a target new energy vehicle and real-time exposure site feature data, wherein the multi-dimensional target feature data comprises behavior feature data generated by a target user end in the target new energy vehicle when selecting a charging station in a historical period; the real-time exposure feature data is behavior feature data generated by the target user end when browsing a target exposure site in real time during the current driving process of the target new energy vehicle; the target exposure site represents a charging station that is browsed by the target user end and not determined to be selected; inputting the multi-dimensional target feature data and the real-time exposure site feature data into a charging station recommendation model to obtain a charging station recommendation result corresponding to the target new energy vehicle output by the charging station recommendation model; the charging station recommendation model is obtained by training a deep learning model based on training samples labeled with positive and negative sample labels; wherein the training samples labeled with negative sample labels are adjusted in training contribution by a weight coefficient in the training process; the corresponding negative sample weight coefficient of the training samples labeled with the negative sample labels is set in the training process, and is calculated by a negative sample weighting formula, the negative sample weighting formula is specifically:

[0129] ;

[0130] wherein, denotes a negative sample weight coefficient, , and denotes a hyperparameter, denotes total exposure site browsing time, denotes total exposure site display quantity, denotes exposure site interaction times, representing exposure site duration information, representing a repeated exposure site proportion, representing a simple negative sample, representing a negative sample weight coefficient corresponding to the simple negative sample, representing a sample exposure site.

[0131] The device embodiments described above are merely illustrative, wherein the units described as separate components can or can not be physically separated, and the components displayed as units can or can not be physical units, i.e., can be located in one place, or can be distributed to multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiment scheme according to actual needs. Those skilled in the art can understand and implement it without creative labor.

[0132] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be realized by means of software and the necessary general hardware platform, and of course can also be realized by hardware. Based on such understanding, the above technical solutions can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in each embodiment or some parts of the embodiments.

[0133] Finally, it should be noted that: the above examples are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing examples, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing examples, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for recommending new energy vehicle charging stations, characterized in that: include: Acquire multi-dimensional target feature data and real-time exposure site feature data of a target new energy vehicle, wherein the multi-dimensional target feature data includes behavioral feature data generated by a target user terminal of the target new energy vehicle when selecting a charging site during a historical period; the real-time exposure site feature data is behavioral feature data generated by the target user terminal when browsing target exposure sites in real time during the current driving process of the target new energy vehicle; the target exposure sites represent charging sites browsed by the target user terminal but not yet determined to be selected; The multi-dimensional target feature data and the real-time exposure site feature data are input into a charging site recommendation model to obtain a charging site recommendation result corresponding to the target new energy vehicle output by the charging site recommendation model; the charging site recommendation model is obtained by training a deep learning model based on training samples marked with positive and negative sample labels; wherein, the training samples marked with negative sample labels have their training contribution adjusted by weight coefficients during the training process; during the training process, the corresponding negative sample weight coefficients are set for the training samples marked with the negative sample labels, which are calculated by a negative sample weighting formula, and the negative sample weighting formula is specifically as follows: ； in, represents the negative sample weight coefficient, and represents the hyperparameter, Indicates the total browsing time of the exposure site. Indicates the total number of exposure sites displayed. Indicates the number of interactions with the exposure site. Indicates the exposure site duration information, Indicates the proportion of repeated exposure sites, represents a simple negative sample, Represents the negative sample weight coefficient corresponding to the simple negative sample, represents the sample exposure site; The charging station recommendation model is trained by the following steps: Constructing sample user features based on sample user portrait information and sample user historical statistical features, wherein the sample user portrait information is a user portrait corresponding to a sample user terminal in a sample new energy vehicle; the sample user historical statistical features include site features, time features, latitude and longitude features, and charging station number features corresponding to the charging station determined and selected by the sample user terminal at a historical moment; Constructing sample site features based on charging site profile information, supporting facilities information, historical charging success rate information, and historical charging site attraction rate within a preset charging station deployment area; wherein the historical charging site attraction rate is calculated based on the distance between the charging site determined and selected by the sample user terminal at a historical moment and the sample user terminal; Constructing sample real-time exposure feature data based on the real-time latitude and longitude and exposure time of the sample user terminal when browsing the sample exposure site at a historical moment; Constructing a historical sequence feature of the sample user based on the historical statistical features of the plurality of sample users and the real-time exposure feature data of the plurality of samples; Based on the total browsing time of the exposure sites generated by the sample user terminal at historical moments, the total number of exposure site displays, the number of exposure site interactions, the exposure site duration information, and the proportion of repeated exposure sites, the sample user site selection behavior characteristics are constructed; wherein the exposure site duration information represents the time between the sample exposure site and the ordered charging site, and the ordered charging site is the charging site determined by the sample user terminal in the sample order; generating the training sample according to the sample user characteristics, the sample site characteristics, the sample user historical sequence characteristics, and the sample real-time exposure characteristic data; Marking the training samples for which it is determined that the sample exposure site is selected by the sample user terminal as the order charging site with a positive sample label, and marking the training samples that are not marked with the positive sample label with a negative sample label, so as to construct a training sample set; Based on the training sample set, the deep learning model is trained to obtain the charging station recommendation model.

2. The method for recommending new energy vehicle charging stations according to claim 1, characterized in that: The step of labeling the training samples that are not labeled with the positive sample labels with negative sample labels includes: Based on the training samples in which the sample exposure sites are not selected by the sample user terminal as the order charging sites, a simple negative sample is randomly determined from the unexposed sites corresponding to the locations of the sample exposure sites before the sample charging order of the sample user terminal is generated, and each sample exposure site of the sample user terminal before the sample charging order is generated is determined as a non-simple negative sample, wherein the unexposed site is a charging site that has not been browsed by the sample user terminal; Calculating corresponding negative sample weight coefficients according to the sample user site selection behavior characteristics in the simple negative sample and the non-simple negative sample respectively; Based on the training samples not marked with the positive sample labels, the negative sample labels are constructed; and according to the negative sample weight coefficients, corresponding weight coefficients are set for the training samples marked with the negative sample labels during the training process.

3. The method for recommending new energy vehicle charging stations according to claim 1, characterized in that: After training the deep learning model based on the training sample set to obtain the charging station recommendation model, the method further includes: Based on the K-means clustering algorithm, user classification processing is performed on the training sample set to obtain training sample subsets corresponding to each user type; Based on the training sample subset, the parameters of the charging site recommendation model are fine-tuned to obtain a charging site classification recommendation model corresponding to each of the user types.

4. The method for recommending new energy vehicle charging stations according to claim 3, characterized in that: The step of inputting the multi-dimensional target feature data and the real-time exposure site feature data into a charging site recommendation model to obtain a charging site recommendation result corresponding to the target new energy vehicle output by the charging site recommendation model includes: Determine whether the target user terminal corresponds to the user type; If the target user terminal has a corresponding user type, input the multi-dimensional target feature data and the real-time exposure site feature data into the charging site classification recommendation model corresponding to the user type according to the user type to obtain the charging site recommendation result; If the target user terminal does not have the corresponding user type, the multi-dimensional target feature data and the real-time exposure site feature data are input into the charging site recommendation model to obtain the charging site recommendation result.

5. A new energy vehicle charging station recommendation system, characterized in that: include: A feature data acquisition module is configured to obtain multi-dimensional target feature data and real-time exposure site feature data of a target new energy vehicle, wherein the multi-dimensional target feature data includes behavioral feature data generated by a target user terminal of the target new energy vehicle when selecting a charging site during a historical period; the real-time exposure site feature data is behavioral feature data generated by the target user terminal when browsing target exposure sites in real time during the current driving process of the target new energy vehicle; the target exposure sites represent charging sites that have been browsed by the target user terminal but have not yet been determined to be selected; A charging site recommendation module is configured to input the multi-dimensional target feature data and the real-time exposure site feature data into a charging site recommendation model to obtain a charging site recommendation result corresponding to the target new energy vehicle output by the charging site recommendation model; the charging site recommendation model is obtained by training a deep learning model based on training samples marked with positive and negative sample labels; wherein, the training contribution of the training samples marked with negative sample labels is adjusted by a weight coefficient during the training process; during the training process, the corresponding negative sample weight coefficient is set for the training samples marked with the negative sample label, which is calculated by a negative sample weighting formula, and the negative sample weighting formula is specifically as follows: ； in, represents the negative sample weight coefficient, and represents the hyperparameter, Indicates the total browsing time of the exposure site. Indicates the total number of exposure sites displayed. Indicates the number of interactions with the exposure site. Indicates the exposure site duration information, Indicates the proportion of repeated exposure sites, represents a simple negative sample, Represents the negative sample weight coefficient corresponding to the simple negative sample, represents the sample exposure site; The charging station recommendation model is trained by the following steps: Constructing sample user features based on sample user portrait information and sample user historical statistical features, wherein the sample user portrait information is a user portrait corresponding to a sample user terminal in a sample new energy vehicle; the sample user historical statistical features include site features, time features, latitude and longitude features, and charging station number features corresponding to the charging station determined and selected by the sample user terminal at a historical moment; Constructing sample site features based on charging site profile information, supporting facilities information, historical charging success rate information, and historical charging site attraction rate within a preset charging station deployment area; wherein the historical charging site attraction rate is calculated based on the distance between the charging site determined and selected by the sample user terminal at a historical moment and the sample user terminal; Constructing sample real-time exposure feature data based on the real-time latitude and longitude and exposure time of the sample user terminal when browsing the sample exposure site at a historical moment; Constructing a historical sequence feature of the sample user based on the historical statistical features of the plurality of sample users and the real-time exposure feature data of the plurality of samples; Based on the total browsing time of the exposure sites generated by the sample user terminal at historical moments, the total number of exposure site displays, the number of exposure site interactions, the exposure site duration information, and the proportion of repeated exposure sites, the sample user site selection behavior characteristics are constructed; wherein the exposure site duration information represents the time between the sample exposure site and the ordered charging site, and the ordered charging site is the charging site determined by the sample user terminal in the sample order; generating the training sample according to the sample user characteristics, the sample site characteristics, the sample user historical sequence characteristics, and the sample real-time exposure characteristic data; Marking the training samples for which it is determined that the sample exposure site is selected by the sample user terminal as the order charging site with a positive sample label, and marking the training samples that are not marked with the positive sample label with a negative sample label, so as to construct a training sample set; Based on the training sample set, the deep learning model is trained to obtain the charging station recommendation model.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the method for recommending new energy vehicle charging stations according to any one of claims 1 to 4 is implemented.

7. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for recommending new energy vehicle charging stations according to any one of claims 1 to 4 is implemented.

8. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the method for recommending new energy vehicle charging stations according to any one of claims 1 to 4 is implemented.

Images