Training method of charging service recommendation model and charging service recommendation method and device

Abstract

The invention provides a training method of a charging service recommendation model, and a charging service recommendation method and apparatus. The training method of the charging service recommendation model comprises the steps of obtaining charging station information and order information according to buried points reserved in a charging service handling program; the charging station information comprises position information and supporting facility information of the charging station, and the order information comprises charging order density and order success rate; constructing attraction characteristics according to the charging station information and the order information based on an attraction model; the attraction characteristics are used for counting the frequency of accessing the charging service handling program by the user and the user conversion rate; and performing iterative training on the deep learning model by taking ordered order data in the order information as a positive sample, taking exposed and non-ordered order data in the order information as a negative sample and taking the attraction feature as an input feature to obtain a charging service recommendation model. According to the method, the charging service recommendation efficiency and the generalizability of the charging service platform are improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular to a training method for a charging service recommendation model, a charging service recommendation method and a device. Background Art

[0002] In daily business, we often encounter the dilemma that after users open the display interface of a certain type of application, the business browsing volume is high, the usage click-through rate is low, and the user conversion rate is even lower.

[0003] For example, with the widespread use of electric vehicles, users can use a variety of charging service platforms, applications or mini-programs to find different charging station locations and charging prices. These platforms usually use distance and price as different screening levels for users to choose from. Users often open multiple applications based on their own usage habits and repeatedly select to determine the appropriate charging station. The charging platforms that are not selected by users expose the problem of weak business competitiveness.

[0004] In the related art, the display content of the charging service recommendation interface is usually adjusted by manually collected user feedback data (the current effective order volume of the charging station). For example, the charging station with more cumulative revenue is displayed at the top. Since different users have different charging habits, adjusting the recommendation priority of the charging station only by relying on whether the order is effective can often only meet the charging habits of a small number of users in a specific area, resulting in low applicability of the charging service platform and difficulty in promotion. In addition, the workload of manually collecting and analyzing data is complicated, which will lead to low efficiency and error-prone charging service recommendations. Summary of the invention

[0005] The present invention provides a training method for a charging service recommendation model, a charging service recommendation method and a device, so as to solve the defects that the prior art uses manually collected user feedback data to adjust the display content of the charging service recommendation interface, resulting in large deviation, low applicability of the charging service platform and difficulty in promotion, and low efficiency of manual collection and analysis of data, resulting in low efficiency of charging service recommendation, thereby improving the charging service recommendation efficiency and the promotability of the charging service platform.

[0006] The present invention provides a training method for a charging service recommendation model, comprising: Obtain charging station information and order information according to the reserved points in the charging service handling procedure; wherein the charging station information includes the location information and supporting facilities information of the charging station, and the order information includes the charging order density and order success rate; Constructing an attraction feature according to the charging station information and the order information based on a gravity model; wherein the attraction feature is used to count the frequency of users accessing the charging service processing program and the user conversion rate; The order data that has been placed in the order information is used as a positive sample, the order data that has been exposed but not ordered in the order information is used as a negative sample, and the deep learning model is iteratively trained with the attractiveness feature as an input feature. When the maximum number of iterations is met or the deep learning model converges, the charging service recommendation model is obtained.

[0007] According to a training method for a charging service recommendation model provided by the present invention, the deep learning model is determined based on a relational database, a Torch architecture, and an ONNX architecture.

[0008] According to a method for training a charging service recommendation model provided by the present invention, after acquiring charging station information and order information according to the buried points reserved in the charging service handling procedure, the method further includes: The target features in the charging station information and the order information are cleaned to obtain cleaned charging station information and cleaned order information; wherein the target features include at least one of the distance between the user and the target charging station, the time it takes the user to arrive at the target charging station, the user's location information, and the user's usage behavior features; the target charging station is a charging station that has been exposed and has no user orders.

[0009] According to a method for training a charging service recommendation model provided by the present invention, after obtaining the charging service recommendation model, the method further includes: Inputting the test sample into the charging service recommendation model to obtain multiple recommendation results; Scoring the multiple recommendation results based on expert experience, and sorting the multiple recommendation results according to the scores to obtain a target recommendation result; The charging service recommendation model is supervised and fine-tuned using the target recommendation result as a training sample to obtain an optimized charging service recommendation model.

[0010] According to a method for training a charging service recommendation model provided by the present invention, after obtaining a plurality of recommendation results, the method further includes: A step recall strategy is adopted to filter the distance nodes corresponding to the plurality of recommendation results respectively with distance as a cutoff index to obtain a set of sites to be recalled; the distance nodes are distance representations between the user and the plurality of charging stations; Sorting each site to be recalled in the set of sites to be recalled according to the refined ranking model to obtain a sorted set of sites to be recalled; Re-shooting each site to be recalled in the sorted set of sites to be recalled according to the target indicator to obtain a re-ranked recommendation result; the target indicator includes at least one of AUC, AUC@3, RECALL@3 and MRR@3; The re-arranged recommendation results are used as training samples to perform supervised fine-tuning on the charging service recommendation model to obtain a new charging service recommendation model.

[0011] The present invention also provides a charging service recommendation method, comprising: Obtaining a user charging request, where the user input request includes charging station query information and order query information; The user charging request is processed based on the charging service recommendation model to obtain a charging service recommendation result; wherein the charging service recommendation model is trained by the training method of the charging service recommendation model.

[0012] The present invention also provides a training device for a charging service recommendation model, comprising: An information acquisition module, used to acquire charging station information and order information according to the reserved points in the charging service handling procedure; wherein the charging station information includes the location information and supporting facilities information of the charging station, and the order information includes the charging order density and order success rate; A feature construction module, configured to construct an attractiveness feature according to the charging station information and the order information based on a gravity model; wherein the attractiveness feature is used to count the frequency of users accessing the charging service processing program and the user conversion rate; A training module is used to iteratively train the deep learning model using the order data that has been placed in the order information as positive samples, the order data that has been exposed but not placed in the order information as negative samples, and the attractiveness features as input features, to obtain the charging service recommendation model when the maximum number of iterations is met or the deep learning model converges.

[0013] The present invention also provides a charging service recommendation device, comprising: A charging request acquisition module, used to acquire a user's charging request, wherein the user input request includes charging station query information and order query information; The recommendation module is used to process the user charging request based on the charging service recommendation model to obtain a charging service recommendation result; wherein the charging service recommendation model is trained by the training method of the charging service recommendation model.

[0014] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the training method for the charging service recommendation model or the charging service recommendation method as described above is implemented.

[0015] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a charging service recommendation model training method or a charging service recommendation method as described in any one of the above.

[0016] The present invention also provides a computer program product, including a computer program, wherein when the computer program is executed by a processor, the training method of the charging service recommendation model or the charging service recommendation method as described above is implemented.

[0017] The training method, charging service recommendation method and device provided by the present invention construct an attractiveness feature for counting the frequency of users visiting the charging service processing program and the user conversion rate based on the charging station information and the order information through a gravity model, and then use the order data that has been placed in the order information as positive samples, and the order data that has been exposed but not placed in the order information as negative samples, and use the attractiveness feature as an input feature to iteratively train the deep learning model to obtain the charging service recommendation model. The model can recommend the most suitable charging station to the user based on information such as the user's preferences and historical behaviors, thereby improving the efficiency of charging service recommendations and the scalability of the charging service platform. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 This is one of the flow charts of the training method of the charging service recommendation model provided by the present invention.

[0020] Figure 2 This is the second flow chart of the training method of the charging service recommendation model provided by the present invention.

[0021] Figure 3 It is a flow chart of the charging service recommendation method provided by the present invention.

[0022] Figure 4 It is a structural schematic diagram of a training device for a charging service recommendation model provided by the present invention.

[0023] Figure 5 It is a structural schematic diagram of the charging service recommendation device provided by the present invention.

[0024] Figure 6 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0025] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0026] Combine the following Figure 1-Figure 5 The present invention describes a charging service recommendation model training method, a charging service recommendation method and a device.

[0027] Figure 1 is one of the flow charts of the training method of the charging service recommendation model provided by the present invention, such as Figure 1 As shown, the training method of the charging service recommendation model includes the following steps: Step 110: Obtain charging station information and order information according to the reserved points in the charging service processing procedure; wherein the charging station information includes the location information and supporting facilities information of the charging station, and the order information includes the charging order density and order success rate.

[0028] In this step, through the reserved tracking points in the charging service processing procedure, rich charging station information and order information can be obtained; for example, in response to the user's behavior of opening the APP again to scan the code before placing an order, the user arrival identification technology is used to obtain the corresponding tracking point data, that is, the charging station information and order information.

[0029] Specifically, users can easily query the specific locations of charging piles across the country through professional charging pile query apps or the built-in navigation system of electric vehicles. These applications usually provide map functions to display the precise locations of charging piles, making it easier for users to find the nearest charging station; the official websites of charging pile operators also usually provide charging pile query services. Users can view detailed information about the charging pile, including address and other location information, by visiting the official website and entering the destination or current location.

[0030] In addition, while providing charging pile location information, charging pile query apps and official websites usually also display the supporting facilities information of the charging piles, such as charging pile type (AC pile, DC pile, etc.), charging power, number of charging ports, etc.; in addition, some charging pile query apps also provide real-time status information of charging stations, such as whether the charging pile is idle, charging costs, etc. This information is very useful for users to select charging stations and plan charging trips.

[0031] In this embodiment, the charging order density can be obtained by analyzing the number of charging orders at the charging station within a certain time period; for example, the number and time distribution of charging orders at the charging station are counted and analyzed to obtain the charging order density; by comparing the charging order densities of different charging stations, the busyness and service capabilities of each charging station can be evaluated, providing a reference for charging station planning and optimization.

[0032] In this embodiment, the order success rate is an important indicator reflecting the quality of charging services and user experience, and can be obtained by calculating the ratio of successful charging orders to all orders (including failed orders).

[0033] For example, the definition of a successful charging order usually includes a power level greater than a certain threshold (such as 0.2 kWh) or a charging time exceeding a certain time (such as 2 minutes); a failed order refers to a situation where the power level is lower than the threshold or the charging time is insufficient; by analyzing the order success rate, problems and bottlenecks in the charging service can be identified, and corresponding measures can be taken to improve and optimize it.

[0034] Step 120: construct an attraction feature based on the gravity model according to the charging station information and the order information; wherein the attraction feature is used to count the frequency of users accessing the charging service processing program and the user conversion rate.

[0035] In this step, the charging station information includes but is not limited to supporting facilities and their scores, historical order density, historical order success rate, etc.

[0036] In another modified embodiment, the site characteristics are taken as the core, the user's historical charging statistics are used as the auxiliary, and the gravity model in the geographic information system is referred to to construct the corresponding attraction characteristics.

[0037] In this embodiment, the supporting facilities (such as rest areas, catering services, etc.) of the charging station are scored by collecting user feedback data on the service quality of the supporting facilities, and the supporting facilities and their scores are obtained to represent the added value and user experience of the charging station. In addition, the geographical location of the charging station determines its accessibility and convenience, and the type (such as AC piles, DC piles) and number of charging piles determine the charging capacity and service scope of the charging station, which are important considerations for users to choose a charging station.

[0038] In this embodiment, the density of historical orders is counted based on the order information, which can reflect the busyness and usage frequency of the charging station. The success rate of historical orders is counted based on the order information, which can reflect the service quality and user experience of the charging station.

[0039] In this embodiment, the gravity model is a widely used spatial interaction model, which is a mathematical equation used to analyze and predict the form of spatial interaction. In the site selection and recommendation of charging stations, this embodiment can draw on the idea of ​​the gravity model to quantify the interaction between charging stations and users.

[0040] Specifically, attractiveness characteristics are determined by the following factors: (1) The supporting facilities score of a charging station is used as one of the important indicators to measure its attractiveness. The higher the supporting facilities score, the better the added value and user experience of the charging station, and the greater its attractiveness to users.

[0041] (2) The historical order density and success rate are used as indicators to measure the service quality and user satisfaction of the charging station. The higher the historical order density and the higher the success rate, the better the service quality of the charging station and the higher the user satisfaction.

[0042] (3) Based on the geographical location of the charging station, calculate the distance or time cost between it and the user, and use it as an indicator to measure the accessibility and convenience of the charging station; the closer the distance or the lower the time cost, the better the accessibility and convenience of the charging station, and the more attractive it is to users.

[0043] By constructing attractiveness features based on the above factors and training a deep learning model, the deep learning model can more accurately predict the user's selection behavior between different charging stations, thereby providing users with more accurate charging station recommendation services to meet their diverse needs.

[0044] Step 130: Use the order data that has been placed in the order information as positive samples, use the order data that has been exposed but not placed in the order information as negative samples, and use the attractiveness feature as the input feature to iteratively train the deep learning model, and obtain the charging service recommendation model when the maximum number of iterations is met or the deep learning model converges.

[0045] In this step, deep learning models include but are not limited to multi-layer perceptrons, convolutional neural networks, recurrent neural networks and their variants (such as LSTM), etc.

[0046] In this embodiment, the positive samples include the charging order data that has been placed, which is used to indicate that the user finally selected a charging station for charging; the negative samples include the order data that has been exposed but not placed, which is used to indicate that the user saw the recommendation of a charging station but did not finally place an order for charging.

[0047] In this embodiment, after obtaining positive samples and negative samples, the data can be cleaned to remove missing values, outliers, etc.; the numerical features can also be standardized or normalized, and the categorical features can be encoded (such as one-hot encoding, label encoding, etc.).

[0048] Specifically, after obtaining the charging station information and order information according to the buried points reserved in the charging business processing procedure, the target features in the charging station information and the order information are cleaned to obtain the cleaned charging station information and the cleaned order information; wherein the target features include at least one of the distance between the user and the target charging station, the time it takes the user to arrive at the target charging station, the user's location information, and the user's usage behavior characteristics; the target charging station is a charging station that has been exposed and has no user orders, which improves the training quality, reduces the interference of the target features on the model training, and can improve the model prediction performance.

[0049] In this embodiment, the attractiveness feature is calculated on each positive sample and negative sample to obtain the corresponding feature map, and the feature map corresponding to the positive sample and the feature map corresponding to the negative sample are input into the LSTM (Long Short-Term Memory) for iterative training; cross-validation can be used to evaluate the performance of the model, and hyperparameters (such as learning rate, batch size, number of network layers, etc.) can be adjusted to optimize the model. During the training process, indicators such as the loss function and accuracy of the model can be monitored to ensure that the model is training in the right direction. When the preset maximum number of iterations is reached, the training is stopped; alternatively, when the loss function of the model no longer decreases significantly in multiple consecutive iterations, and indicators such as accuracy tend to be stable, it can be considered that the model has converged, and the training is stopped to obtain a charging service recommendation model.

[0050] In this embodiment, the trained charging service recommendation model is evaluated using the test set data to verify its generalization ability.

[0051] Specifically, the performance of the model can be evaluated using indicators such as accuracy, recall, F1 score, AUC, etc. Finally, the trained model can be deployed to an online environment to recommend charging stations in real time. Then, based on the user's real-time request and contextual information (such as current location, time, etc.), the recommendation results that best meet the user's needs are screened out from the candidate charging stations.

[0052] The training method of the charging service recommendation model provided by the present invention constructs an attractiveness feature for counting the frequency of users visiting the charging service processing program and the user conversion rate according to the charging station information and the order information through a gravity model, and then uses the order data that has been placed in the order information as a positive sample, and the order data that has been exposed but not placed in the order information as a negative sample, and uses the attractiveness feature as an input feature to iteratively train the deep learning model to obtain the charging service recommendation model. The model can recommend the most suitable charging station to the user based on information such as the user's preferences and historical behaviors, thereby improving the efficiency of charging service recommendations and the scalability of the charging service platform.

[0053] In some embodiments, the deep learning model is determined based on a relational database, a Torch architecture, and an ONNX architecture.

[0054] In this embodiment, in the complete process of a deep learning project, data processing, model training, and model deployment are three core links. This embodiment uses SQL to implement data processing, Torch as a model training framework, and ONNX for model deployment.

[0055] (1) This embodiment uses SQL query statements to extract the required charging station information and order information from a relational database (such as MySQL, PostgreSQL, etc.). This information may include selecting specific tables and columns and applying screening conditions to filter data; SQL statements are used to process missing values, abnormal values ​​or duplicate data to ensure data quality. This may include using CASE statements, JOIN operations, subqueries and other technologies to clean and transform data; then, according to the requirements of the model, SQL string functions, date functions or window functions are used to perform feature engineering and encoding classification variable conversion on the data to obtain the corresponding positive and negative sample data and feature maps; finally, the processed data is exported to a format suitable for model training to facilitate data transmission and analysis.

[0056] (2) This example uses Torch's DataLoader and Dataset classes to load and process training data. If the data has been exported as a CSV file, it can be read using the pandas library and converted into a Torch tensor using a custom Dataset class. The deep learning model is defined using Torch's nn.Module class. This includes defining the layer structure, activation function, and loss function of the deep learning model. The optimizer (such as SGD, Adam, etc.) and training data are then used to update the model's weights. For example, steps such as forward propagation, loss calculation, backpropagation, and weight update are performed. Finally, the performance of the model is evaluated on the validation set to monitor overfitting and adjust hyperparameters.

[0057] (3) After training the charging service recommendation model, this can be done by calling Torch's torch.onnx.export function to convert the Torch model and its input tensors into ONNX format files; then use ONNXRuntime, TensorRT or other ONNX-compatible inference engines to optimize the ONNX model to improve inference speed and efficiency; finally, deploy the optimized ONNX model to the target platform, such as a web server, mobile application or edge device; for example, by writing inference code, passing input data to the model, and processing the model's output.

[0058] The training method of the charging service recommendation model provided by the present invention builds a deep learning model through a relational database, Torch architecture and ONNX architecture, which can achieve efficient data storage and management, flexible model construction and adjustment, and cross-framework model migration and deployment, thereby improving the data processing efficiency, model training efficiency and model deployment efficiency of the deep learning model.

[0059] In some embodiments, after obtaining the charging service recommendation model, the method further includes: (1) Input the test sample into the charging service recommendation model to obtain multiple recommendation results.

[0060] In this embodiment, the method of obtaining the test sample is the same as that of the above-mentioned charging station information and order information, and will not be repeated in this embodiment; in addition, the above-mentioned charging station information and order information can be divided into a training set and a test set according to a certain division ratio, and the test sample belongs to the test set.

[0061] In this embodiment, the test sample includes an attractiveness feature, and the test sample is input into a trained charging service recommendation model to obtain multiple possible charging service recommendation results; these recommendation results may include nearby charging stations, recommended charging time periods, estimated charging costs, etc.

[0062] (2) Scoring multiple recommendation results based on expert experience, and sorting the multiple recommendation results according to the scores to obtain the target recommendation results.

[0063] In this embodiment, multiple recommendation results output by the model are scored based on expert experience, that is, each recommendation result is subjectively evaluated based on factors such as professional knowledge, market conditions, and user feedback. The scoring criteria include accuracy, practicality, user satisfaction, and other aspects of the recommendation results. Based on the scoring results, multiple recommendation results are sorted according to their scores. Among the sorted results, the recommendation result with the highest score is the target recommendation result, which means that the expert believes that it is the best charging service recommendation under the current test sample.

[0064] (3) The target recommendation results are used as training samples to supervise and fine-tune the charging service recommendation model to obtain the optimized charging service recommendation model.

[0065] In this embodiment, the supervised fine-tuning method includes using a variety of machine learning algorithms and techniques, such as gradient descent, stochastic gradient descent, Adam optimizer, etc. By continuously iterating and adjusting model parameters, the performance of the model can be gradually improved, so that the recommendation results output by it are more in line with expert experience and user needs.

[0066] In this embodiment, the target recommendation result is used as a training sample to adjust the network parameters of the charging service recommendation model so that the model can more accurately output high-quality recommendations similar to the target recommendation result in future recommendations.

[0067] In this embodiment, an expert scoring method is used to recommend sites that meet the business scoring requirements based on the site charging pile completion rate, supporting facilities utilization rate, order success rate, and actual business needs, thereby effectively improving the click-through rate and conversion rate of high-quality sites.

[0068] The training method of the charging service recommendation model provided by the present invention obtains multiple recommendation results by inputting a test sample into the charging service recommendation model, and then scores the multiple recommendation results based on expert experience, and sorts the multiple recommendation results according to the scores to obtain a target recommendation result. Finally, the charging service recommendation model is supervised and fine-tuned with the target recommendation result as a training sample to obtain an optimized charging service recommendation model. The prediction performance of the charging service recommendation model is further improved by combining the expert's professional knowledge.

[0069] In some embodiments, after obtaining multiple recommendation results, the training method of the charging service recommendation model further includes: (1) A step-by-step recall strategy is used to filter the distance nodes corresponding to multiple recommendation results with distance as the cutoff indicator to obtain a set of sites to be recalled; the distance node represents the distance between the user and multiple charging stations.

[0070] In this embodiment, the step recall strategy uses the distance between the user and multiple charging stations as a cutoff indicator. Specifically, the charging stations in the recommended results are screened according to a preset distance threshold (which may be one or more, forming a step-like shape). Charging stations with a closer distance will be retained first to form a collection of stations to be recalled. By filtering by distance, it can be ensured that the charging stations in the recommended results are actually feasible for the user, while reducing the amount of calculation for subsequent processing.

[0071] (2) Sort each site to be recalled in the site set to be recalled according to the refined ranking model to obtain a sorted site set to be recalled.

[0072] In this embodiment, the refined ranking model is used to measure the importance of multiple complex features (such as user historical behavior, service quality of charging stations, price, etc.) to provide more accurate ranking results.

[0073] In this embodiment, a trained refined ranking model is used to score and rank each site in the set of sites to be recalled, thereby obtaining a ranked set of sites to be recalled. The accuracy and relevance of the recommendation results are further improved through ranking by the refined ranking model.

[0074] (3) Re-ranking each site in the sorted set of sites to be recalled according to the target indicator to obtain a re-ranked recommendation result; the target indicator includes at least one of AUC, AUC@3, RECALL@3 and MRR@3.

[0075] In this embodiment, according to target indicators such as AUC (Area Under Curve), AUC@3 (AUC value of the first three recommended results), Recall@3 (recall rate of correct results in the first three recommended results) and MRR@3 (average reciprocal ranking of correct results in the first three recommended results), the sorted set of sites to be recalled is rearranged to further optimize the recommendation results, thereby reflecting the accuracy and diversity of the recommendation results. The sorting of multiple recommendation results can be made more in line with user needs and business goals through rearrangement.

[0076] (4) The re-ranked recommendation results are used as training samples to supervise and fine-tune the charging service recommendation model to obtain a new charging service recommendation model.

[0077] In this embodiment, the rearranged recommendation results are used as training samples to contain more accurate user preferences and charging station information. The training samples are used to perform supervised fine-tuning on the original charging service recommendation model, and the model parameters are adjusted to better fit the data. Through supervised fine-tuning, the recommendation performance of the model is further improved so that it can provide more accurate recommendation results in the future.

[0078] The training method of the charging service recommendation model provided by the present invention adopts a step recall strategy to screen multiple recommendation results with distance as a truncation indicator, sorts each to-be-recalled site in the to-be-recalled site set according to the refined ranking model, re-ranks each to-be-recalled site in the sorted to-be-recalled site set according to the target indicator, and finally supervises and fine-tunes the charging service recommendation model using the re-ranked recommendation results as training samples to obtain a new charging service recommendation model, and further improves the prediction performance of the charging service recommendation model, so as to provide users with more accurate and personalized recommendation services.

[0079] Figure 2 This is a second flow chart of the training method of the charging service recommendation model provided by the present invention. Figure 2 In the illustrated embodiment, corresponding buried point data is obtained by pre-setting buried points in the APP and mini-program related to the charging service, and the business data is obtained from the business library. After the buried point data and business data (such as order data) are cleaned, the attractive features that affect user clicks and conversions are extracted from the above data through the data processing platform Maxcompute; the attractive features are analyzed and positive samples and negative samples are constructed, and the deep learning model (corresponding model) is iteratively trained using the positive samples and negative samples. After the training is completed, a charging service recommendation model is obtained, and then the accuracy of the charging service recommendation model is verified through offline sample data, and the verified charging service recommendation model is subjected to real-time simulation testing. The charging service recommendation model is once again optimized through targeted rearrangement combined with expert experience and the strategy of phased recall + fine sorting + rearrangement, and then used to execute the charging service (for example, the charging service recommendation model is deployed in the "New Electric Road" APP (Application) to provide users with charging needs in different scenarios).

[0080] The charging service recommendation method provided by the present invention is described below. The charging service recommendation method described below and the training method of the charging service recommendation model described above can be referred to each other.

[0081] Figure 3 is a flow chart of the charging service recommendation method provided by the present invention, such as Figure 3 As shown, the charging service recommendation method includes the following steps: Step 310: Obtain a user charging request, where the user input request includes charging station query information and order query information.

[0082] In this step, the user charging request can be a real-time charging request, including real-time charging station query information and real-time order query information, or it can be charging request data already existing in the charging request database, including historical charging station query information and historical order query information.

[0083] For example, the user's charging request may be "help me generate a list of nearby charging stations and charging prices."

[0084] Step 320: Process the user's charging request based on the charging service recommendation model to obtain a charging service recommendation result; wherein the charging service recommendation model is trained by a charging service recommendation model training method.

[0085] In this step, the charging service recommendation model is trained through the following steps: (1) Obtain charging station information and order information according to the reserved points in the charging business handling procedures; the charging station information includes the location information and supporting facilities information of the charging station, and the order information includes the charging order density and order success rate; (2) Based on the gravity model, an attractiveness feature is constructed according to the charging station information and order information. The attractiveness feature is used to count the frequency of users visiting the charging service application and the user conversion rate. (3) The order data that has been placed in the order information is used as positive samples, and the order data that has been exposed but not placed in the order information is used as negative samples. The deep learning model is iteratively trained with the attractiveness feature as the input feature. When the maximum number of iterations is met or the deep learning model converges, a charging service recommendation model is obtained.

[0086] It should be noted that the implementation method of each training step corresponds one-to-one to the above-mentioned steps 110 to 130, and will not be repeated in this embodiment.

[0087] In this embodiment, in response to the user input question "Help me generate a list of nearby charging piles and charging prices", the result output by the charging service recommendation model can be the following list data: a distribution map of charging stations within 1 km, within 1-2 km, within 2-3 km, and beyond 3 km from the user. Each charging station is marked with: charging location, charging unit price, fast charging or slow charging type, cumulative number of charging times, valid order share, surrounding facility recommendations and recommended comprehensive score and other information. Users can choose the charging pile with the highest score and navigate to charge according to the charging location.

[0088] The charging service recommendation method provided by the embodiment of the present invention processes the user's charging request through the charging service recommendation model trained by the charging service recommendation model training method to obtain charging service recommendations, thereby improving the efficiency and accuracy of charging service recommendations.

[0089] The training device for the charging service recommendation model provided by the present invention is described below. The training device for the charging service recommendation model described below and the training method for the charging service recommendation model described above can refer to each other.

[0090] Figure 4 is a structural diagram of a training device for a charging service recommendation model provided by the present invention, such as Figure 4 As shown, the training device of the charging service recommendation model includes: an information acquisition module 410, a feature construction module 420 and a training module 430.

[0091] The information acquisition module 410 is used to acquire the charging station information and order information according to the reserved buried points in the charging service handling procedure; wherein the charging station information includes the location information and supporting facilities information of the charging station, and the order information includes the charging order density and order success rate; A feature construction module 420 is used to construct an attractiveness feature based on the charging station information and the order information based on the gravity model; wherein the attractiveness feature is used to count the frequency of users accessing the charging service processing program and the user conversion rate; The training module 430 is used to iteratively train the deep learning model using the order data that has been placed in the order information as positive samples, the order data that has been exposed but not placed in the order information as negative samples, and the attractiveness feature as input feature, to obtain a charging service recommendation model when the maximum number of iterations is met or the deep learning model converges.

[0092] The training device for the charging service recommendation model provided by the present invention constructs an attractiveness feature for counting the frequency of users visiting the charging service processing program and the user conversion rate according to the charging station information and the order information through a gravity model, and then uses the order data that has been placed in the order information as a positive sample, and the order data that has been exposed but not placed in the order information as a negative sample, and uses the attractiveness feature as an input feature to iteratively train the deep learning model to obtain the charging service recommendation model. The model can recommend the most suitable charging station to the user based on information such as the user's preferences and historical behaviors, thereby improving the efficiency of charging service recommendations and the scalability of the charging service platform.

[0093] The charging service recommendation device provided by the present invention is described below. The charging service recommendation device described below and the charging service recommendation method described above can be referred to each other.

[0094] Figure 5 is a schematic diagram of the structure of the charging service recommendation device provided by the present invention, such as Figure 5 As shown, the charging service recommendation device includes: a charging request acquisition module 510 and a recommendation module 520.

[0095] A charging request acquisition module 510 is used to acquire a user charging request, where the user input request includes charging station query information and order query information; The recommendation module 520 is used to process the user's charging request based on the charging service recommendation model to obtain a charging service recommendation result; wherein the charging service recommendation model is trained by a charging service recommendation model training method.

[0096] The charging service recommendation device provided in the embodiment of the present invention processes user charging requests through a charging service recommendation model trained by a charging service recommendation model training method to obtain charging service recommendations, thereby improving the efficiency and accuracy of charging service recommendations.

[0097] Figure 6 is a schematic diagram of the structure of the electronic device provided by the present invention, such as Figure 6 As shown, the electronic device may include: a processor 610, a communication interface 620, a memory 630 and a communication bus 640, wherein the processor 610, the communication interface 620 and the memory 630 complete mutual communication through the communication bus 640. The processor 610 may call the logic instructions in the memory 630 to execute the training method of the charging service recommendation model, the method comprising: obtaining charging station information and order information according to the reserved buried points in the charging service handling program; wherein the charging station information includes the location information and supporting facilities information of the charging station, and the order information includes the charging order density and the order success rate; constructing an attractiveness feature according to the charging station information and the order information based on the gravity model; wherein the attractiveness feature is used to count the frequency of users visiting the charging service handling program and the user conversion rate; taking the order data that has been placed in the order information as the positive sample, taking the order data that has been exposed but not placed in the order information as the negative sample, taking the attractiveness feature as the input feature to iteratively train the deep learning model, and obtaining the charging service recommendation model when the maximum number of iterations is met or the deep learning model converges.

[0098] Or execute a charging service recommendation method, the method comprising: obtaining a user charging request, the user input request including charging station query information and order query information; processing the user charging request based on a charging service recommendation model to obtain a charging service recommendation result; wherein the charging service recommendation model is trained by a charging service recommendation model training method.

[0099] In addition, the logic instructions in the above-mentioned memory 630 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0100] On the other hand, the present invention also provides a computer program product, which includes a computer program, and the computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the training method of the charging service recommendation model provided by the above methods, the method including: obtaining charging station information and order information according to the reserved points in the charging service processing program; wherein the charging station information includes the location information and supporting facilities information of the charging station, and the order information includes the charging order density and the order success rate; constructing an attractiveness feature based on the charging station information and the order information based on the gravity model; wherein the attractiveness feature is used to count the frequency of users visiting the charging service processing program and the user conversion rate; taking the order data that has been placed in the order information as a positive sample, and taking the order data that has been exposed but not placed in the order information as a negative sample, and taking the attractiveness feature as the input feature to iteratively train the deep learning model, and obtaining the charging service recommendation model when the maximum number of iterations is met or the deep learning model converges.

[0101] Or execute a charging service recommendation method, the method comprising: obtaining a user charging request, the user input request including charging station query information and order query information; processing the user charging request based on a charging service recommendation model to obtain a charging service recommendation result; wherein the charging service recommendation model is trained by a charging service recommendation model training method.

[0102] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a training method for a charging service recommendation model provided by the above-mentioned methods, the method comprising: obtaining charging station information and order information according to the reserved points in the charging service processing program; wherein the charging station information includes the location information and supporting facilities information of the charging station, and the order information includes the charging order density and the order success rate; constructing an attractiveness feature according to the charging station information and the order information based on a gravity model; wherein the attractiveness feature is used to count the frequency of users visiting the charging service processing program and the user conversion rate; taking the order data that has been placed in the order information as a positive sample, taking the order data that has been exposed but not placed in the order information as a negative sample, and taking the attractiveness feature as an input feature to iteratively train the deep learning model, and obtaining the charging service recommendation model when the maximum number of iterations is met or the deep learning model converges.

[0103] Or execute a charging service recommendation method, the method comprising: obtaining a user charging request, the user input request including charging station query information and order query information; processing the user charging request based on a charging service recommendation model to obtain a charging service recommendation result; wherein the charging service recommendation model is trained by a charging service recommendation model training method.

[0104] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0105] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0106] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A training method for a charging service recommendation model, characterized in that: include: Obtain charging station information and order information according to the reserved points in the charging service handling procedure; wherein the charging station information includes the location information and supporting facilities information of the charging station, and the order information includes the charging order density and order success rate; Constructing an attraction feature according to the charging station information and the order information based on a gravity model; wherein the attraction feature is used to count the frequency of users accessing the charging service processing program and the user conversion rate; The order data that has been placed in the order information is used as a positive sample, the order data that has been exposed but not ordered in the order information is used as a negative sample, and the deep learning model is iteratively trained with the attractiveness feature as an input feature. When the maximum number of iterations is met or the deep learning model converges, the charging service recommendation model is obtained.

2. The training method of the charging service recommendation model according to claim 1, characterized in that: The deep learning model is determined based on a relational database, a Torch architecture, and an ONNX architecture.

3. The training method of the charging service recommendation model according to claim 1, characterized in that: After acquiring the charging station information and order information according to the reserved buried points in the charging service handling procedure, the method further includes: The target features in the charging station information and the order information are cleaned to obtain cleaned charging station information and cleaned order information; wherein the target features include at least one of the distance between the user and the target charging station, the time it takes the user to arrive at the target charging station, the user's location information, and the user's usage behavior features; the target charging station is a charging station that has been exposed and has no user orders.

4. The training method of the charging service recommendation model according to claim 1, characterized in that: After obtaining the charging service recommendation model, the method further includes: Inputting the test sample into the charging service recommendation model to obtain multiple recommendation results; Scoring the multiple recommendation results based on expert experience, and sorting the multiple recommendation results according to the scores to obtain a target recommendation result; The charging service recommendation model is supervised and fine-tuned using the target recommendation result as a training sample to obtain an optimized charging service recommendation model.

5. The training method of the charging service recommendation model according to claim 4, characterized in that: After obtaining the plurality of recommendation results, the method further includes: A step recall strategy is adopted to filter the distance nodes corresponding to the plurality of recommendation results respectively with distance as a cutoff index to obtain a set of sites to be recalled; the distance nodes are distance representations between the user and the plurality of charging stations; Sorting each site to be recalled in the set of sites to be recalled according to the refined ranking model to obtain a sorted set of sites to be recalled; Re-shooting each site to be recalled in the sorted set of sites to be recalled according to the target indicator to obtain a re-ranked recommendation result; the target indicator includes at least one of AUC, AUC@3, RECALL@3 and MRR@3; The re-arranged recommendation results are used as training samples to perform supervised fine-tuning on the charging service recommendation model to obtain a new charging service recommendation model.

6. A charging service recommendation method, characterized in that: include: Obtaining a user charging request, where the user input request includes charging station query information and order query information; The user charging request is processed based on the charging service recommendation model to obtain a charging service recommendation result; wherein the charging service recommendation model is trained by the training method of the charging service recommendation model.

7. A training device for a charging service recommendation model, characterized in that: include: An information acquisition module, used to acquire charging station information and order information according to the reserved points in the charging service handling procedure; wherein the charging station information includes the location information and supporting facilities information of the charging station, and the order information includes the charging order density and order success rate; A feature construction module, configured to construct an attractiveness feature according to the charging station information and the order information based on a gravity model; wherein the attractiveness feature is used to count the frequency of users accessing the charging service processing program and the user conversion rate; A training module is used to iteratively train the deep learning model using the order data that has been placed in the order information as positive samples, the order data that has been exposed but not placed in the order information as negative samples, and the attractiveness features as input features, to obtain the charging service recommendation model when the maximum number of iterations is met or the deep learning model converges.

8. A charging service recommendation device, characterized in that: include: A charging request acquisition module, used to acquire a user's charging request, wherein the user input request includes charging station query information and order query information; The recommendation module is used to process the user charging request based on the charging service recommendation model to obtain a charging service recommendation result; wherein the charging service recommendation model is trained by the training method of the charging service recommendation model.

9. 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 according to any one of claims 1 to 6 is implemented.

10. 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 according to any one of claims 1 to 6 is implemented.

Images