Vehicle charging recommendation method, device, equipment and product based on optical storage and charging system

Abstract

The invention belongs to the technical field of new energy, and provides a vehicle charging recommendation method and device based on an optical storage and charging system, electronic equipment and a program product, and the method comprises the steps: obtaining multi-modal data of a user, the multi-modal data comprising historical charging data, driving track data, social preference data and context environment data; processing the multi-modal data based on a large language model, and constructing a user portrait of the user, the user portrait being used for predicting a future charging behavior of the user; in response to the personalized charging recommendation request, vehicle data of a user and real-time state data of charging piles are obtained, the vehicle data are used for representing the current travel state and charging demand of the user, and the real-time state data of the charging piles are used for representing operation service capabilities and use conditions of a plurality of candidate charging piles in the optical storage and charging system; and generating personalized charging recommendation information for the user according to the user portrait, the vehicle data and the real-time state data.

Description

Technical Field

[0001] This application belongs to the field of new energy technology, and more specifically, it relates to a vehicle charging recommendation method, device, electronic equipment and program product based on a photovoltaic energy storage and charging system. Background Technology

[0002] With the rapid increase in the popularity of new energy vehicles, users' demands for personalized and convenient charging services are becoming increasingly prominent. Currently, vehicle charging recommendation schemes for photovoltaic-storage-charging systems mostly rely on single-dimensional data, resulting in insufficient recommendation accuracy. This can easily lead to situations where recommended charging stations do not match user habits or guide users to congested charging stations during peak hours.

[0003] Meanwhile, traditional vehicle charging recommendation methods lack the ability to deeply integrate multi-source data, making it impossible to accurately predict users' charging needs. In addition, the fluctuations in photovoltaic output and the constraints of energy storage capacity in photovoltaic-storage-charging systems are not effectively incorporated into the recommendation decision, resulting in low energy utilization. Furthermore, users need to manually select charging stations and compare prices and availability, leading to low decision-making efficiency and a poor user experience. Summary of the Invention

[0004] The purpose of this application is to provide a vehicle charging recommendation method, device, electronic device, and program product based on a photovoltaic energy storage and charging system, aiming to solve the technical problems in related technologies where vehicle charging recommendation schemes lack personalization and recommendation accuracy, resulting in low energy utilization efficiency and poor user experience.

[0005] To achieve the above objectives, according to the first aspect of this application, a vehicle charging recommendation method based on a photovoltaic-energy storage-charging system is provided, the method comprising: Acquire user's multimodal data, which includes historical charging data, driving trajectory data, social preference data, and contextual environment data; The multimodal data is processed based on a large language model to construct a user profile of the user, which is used to predict the user's future charging behavior. In response to a personalized charging recommendation request, the system obtains the user's vehicle data and the real-time status data of the charging piles. The vehicle data is used to characterize the user's current travel status and charging needs, and the real-time status data of the charging piles is used to characterize the operational service capabilities and usage conditions of multiple candidate charging piles within the photovoltaic-storage-charging system. Based on the user profile, the vehicle data, and the real-time status data, personalized charging recommendations are generated for the user. The beneficial effects of this application embodiment compared with the prior art are as follows: The vehicle charging recommendation method based on photovoltaic energy storage and charging system provided in this application embodiment achieves personalized, accurate and intelligent charging recommendation through multimodal data fusion, user profile construction driven by large language model and multi-dimensional data collaborative decision-making.

[0006] By integrating multimodal information such as historical charging data, driving trajectory data, social preference data, and contextual environment data, and combining the multimodal feature encoding and temporal modeling capabilities of large language models, the constructed user profile can accurately depict the user's core behavioral characteristics such as the distribution of average daily charging time periods, preferred charging locations, and sensitivity to discounts. The recommendation results can be deeply adapted to the user's travel habits (such as priority recommendations along commuting routes), charging preferences (such as targeted matching of fast charging / slow charging), and scenario needs (such as charging recommendations before long-distance travel), reducing user waiting time and adapting discount information to user sensitivity, further reducing user charging costs and significantly improving user satisfaction and loyalty.

[0007] Based on collaborative decision-making using real-time charging pile status data (such as load status, photovoltaic and energy storage output, and idle status) and user vehicle data (such as remaining battery power, travel route, and driving range), the recommendation process can dynamically avoid high-load charging piles and prioritize charging piles with sufficient photovoltaic and energy storage output, effectively achieving system load balancing and reducing the coexistence of idle and congested charging piles. At the same time, combined with charging demand predictions from user profiles, users can be guided to charge during off-peak hours in advance, reducing the peak-valley load difference of the power grid, improving the utilization rate of photovoltaic and energy storage energy, and reducing system operating costs.

[0008] According to a second aspect of this application, a vehicle charging recommendation device based on a photovoltaic energy storage and charging system is provided, the device comprising: The acquisition unit is used to acquire the user's multimodal data, which includes historical charging data, driving trajectory data, social preference data, and contextual environment data. The construction unit is used to process the multimodal data based on a large language model to construct a user profile of the user, which is used to predict the user's future charging behavior. The response unit is used to respond to a personalized charging recommendation request by obtaining the user's vehicle data and the real-time status data of the charging pile. The vehicle data is used to characterize the user's current travel status and charging needs, and the real-time status data of the charging pile is used to characterize the operational service capabilities and usage conditions of multiple candidate charging piles in the photovoltaic-storage-charging system. The generation unit is used to generate personalized charging recommendation information for the user based on the user profile, the vehicle data, and the real-time status data.

[0009] According to a third aspect of this application, an electronic device is provided, 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 electronic device causes the electronic device to perform the method as described in any one of the claims.

[0010] According to a fourth aspect of this application, a computer-readable storage medium is provided that stores a computer program, which, when executed by a processor, implements the method as described in any one of the claims.

[0011] According to a fifth aspect of this application, a computer program product is provided that, when run on an electronic device, causes the electronic device to perform the method described in any one of the first aspects above.

[0012] It is understandable that the beneficial effects of the second to fifth aspects mentioned above can be found in the relevant descriptions in the first aspect mentioned above, and will not be repeated here. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 This is a schematic flowchart of a vehicle charging recommendation method based on a photovoltaic energy storage and charging system provided in an embodiment of this application; Figure 2 This is a schematic flowchart of an optional vehicle charging recommendation method based on a photovoltaic energy storage and charging system provided in an embodiment of this application; Figure 3 This is a flowchart illustrating another optional vehicle charging recommendation method based on a photovoltaic energy storage and charging system provided in this application embodiment; Figure 4 This is a flowchart illustrating another optional vehicle charging recommendation method based on a photovoltaic energy storage and charging system provided in this application embodiment; Figure 5 This is a flowchart illustrating another optional vehicle charging recommendation method based on a photovoltaic energy storage and charging system provided in this application embodiment; Figure 6 This is a schematic diagram of the structure of a vehicle charging recommendation device based on a photovoltaic energy storage and charging system provided in an embodiment of this application; Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0015] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0016] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0017] It should also be understood that, in the description of this application, unless otherwise stated, the " / " used in the specification and appended claims indicates that the related objects are in an "or" relationship. For example, A / B can mean A or B. The "and / or" in this application is merely a description of the relationship between the related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone, where A and B can be singular or plural. Furthermore, in the description of this application, unless otherwise stated, "multiple" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0018] Furthermore, to facilitate a clear description of the technical solutions in the embodiments of this application, the terms "first" and "second" are used in the embodiments of this application to distinguish identical or similar items with essentially the same function and effect. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, but are only used for distinguishing descriptions, and the terms "first" and "second" do not necessarily imply that they are different, nor should they be construed as indicating or implying relative importance.

[0019] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0020] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0021] This application provides an example of a vehicle charging recommendation method based on a photovoltaic-energy storage-charging system. Please refer to [link / reference]. Figure 1 As shown, Figure 1 A schematic flowchart illustrating a vehicle charging recommendation method based on a photovoltaic energy storage and charging system provided in this application is shown. This is an example and not a limitation; the method can be applied to or operated in electronic devices. The method includes: S101, acquire the user's multimodal data.

[0022] The multimodal data includes historical charging data, driving trajectory data, social preference data, and contextual environment data.

[0023] S102 processes multimodal data based on a large language model to construct user profiles, which are used to predict users' future charging behavior.

[0024] S103 responds to personalized charging recommendation requests by obtaining the user's vehicle data and the real-time status data of the charging station.

[0025] Among them, vehicle data is used to characterize the user's current travel status and charging needs, while real-time status data of charging piles is used to characterize the operational service capabilities and usage conditions of multiple candidate charging piles within the photovoltaic-storage-charging system.

[0026] S104 generates personalized charging recommendations for users based on user profiles, vehicle data, and real-time status data.

[0027] In some embodiments, the multimodal data includes historical charging data, driving trajectory data, social preference data, and contextual environment data. Historical charging data is collected in real-time through charging pile platforms and user charging apps, and includes structured data such as timestamps, charging volume, voltage, and cost. Driving trajectory data, with user authorization, is obtained through the vehicle's GPS or a third-party map API (map service application interface), presented as a time-series latitude and longitude path. Social preference data comes from user-authorized social media analysis results, extracting textual data such as interest tags and keywords related to charging and travel. Contextual environment data covers structured or semi-structured data such as real-time weather, holiday markers, and regional traffic flow, obtained synchronously through public data interfaces.

[0028] Next, feature embedding encoding is performed on the multimodal data. A multimodal encoder converts different types of data into vector representations of a unified dimension. Structured data uses linear embedding, textual data such as social preferences extracts feature vectors through a large language model, and driving trajectory data uses trajectory convolution or position embedding for spatial encoding, ensuring a consistent data format for subsequent models. Then, the embedded vector sequence is input into the Transformer model for temporal modeling. Using recent user behavior data (e.g., 72 hours, 48 ​​hours, 96 hours) as time windows, the model's self-attention mechanism captures temporal dependencies in behavior, highlighting high-frequency key patterns such as charging after get off work and recharging before long weekend trips. Finally, the context vectors output by the Transformer model are weighted and summarized to construct a user behavior feature vector, i.e., a user profile, containing dimensions such as daily charging time distribution, preferred charging scenarios, power demand characteristics, discount sensitivity, and high-frequency travel routes. This profile is used to accurately predict users' future charging behavior.

[0029] Then, in response to personalized charging recommendation requests (such as when a user initiates a charging station search via a charging app), the system obtains the user's vehicle data and the real-time status data of the charging stations. Specifically, vehicle data can be collected in real-time via the vehicle-to-everything (V2X) interface, including current latitude and longitude, remaining battery power, battery capacity, current driving route, and remaining driving range, accurately representing the user's current travel status and charging needs. Real-time status data of the charging stations can be synchronized from the photovoltaic-storage-charging system management platform, covering multiple dimensions such as the location coordinates of candidate charging stations, their idle status, charging power, real-time electricity price, and the output of the photovoltaic-storage system, comprehensively reflecting the charging station's operational service capabilities and usage conditions.

[0030] Finally, based on the user profile, vehicle data, and real-time status data, personalized charging recommendations are generated for the user. For example, a gated recurrent unit (GRU) or long short-term memory (LSTM) recurrent neural network (RNN) model can be used to model the user's historical behavior sequence (location, battery level, time, etc.), capture the behavioral evolution trend, and output the temporal context vector of the user's current state. Then, the feature vectors of candidate charging piles are matched with this context vector through attention matching, and charging piles that match the user's habits are selected by calculating preference weights. Finally, a multi-task output structure is adopted, combined with optimization objectives such as charging efficiency, cost economy, and system load balancing, to generate a Top-N recommended charging pile list, the optimal charging time period (matching pile load and user trip), and customized preferential schemes (such as peak and off-peak electricity price discounts, preferential offers for preferred scenarios), and other personalized recommendation information.

[0031] The beneficial effects of this application embodiment compared with the prior art are as follows: The vehicle charging recommendation method based on photovoltaic energy storage and charging system provided in this application embodiment achieves personalized, accurate and intelligent charging recommendation through multimodal data fusion, user profile construction driven by large language model and multi-dimensional data collaborative decision-making.

[0032] By integrating multimodal information such as historical charging data, driving trajectory data, social preference data, and contextual environment data, and combining the multimodal feature encoding and temporal modeling capabilities of large language models, the constructed user profile can accurately depict the user's core behavioral characteristics such as the distribution of average daily charging time periods, preferred charging locations, and sensitivity to discounts. The recommendation results can be deeply adapted to the user's travel habits (such as priority recommendations along commuting routes), charging preferences (such as targeted matching of fast charging / slow charging), and scenario needs (such as charging recommendations before long-distance travel), reducing user waiting time and adapting discount information to user sensitivity, further reducing user charging costs and significantly improving user satisfaction and loyalty.

[0033] Based on collaborative decision-making using real-time charging pile status data (such as load status, photovoltaic and energy storage output, and idle status) and user vehicle data (such as remaining battery power, travel route, and driving range), the recommendation process can dynamically avoid high-load charging piles and prioritize charging piles with sufficient photovoltaic and energy storage output, effectively achieving system load balancing and reducing the coexistence of idle and congested charging piles. At the same time, combined with charging demand predictions from user profiles, users can be guided to charge during off-peak hours in advance, reducing the peak-valley load difference of the power grid, improving the utilization rate of photovoltaic and energy storage energy, and reducing system operating costs.

[0034] One possible implementation is, such as Figure 2 As shown, the user's multimodal data is obtained, including: S201, obtain the user's historical charging data from the charging pile platform and / or associated application of the photovoltaic-storage-charging system. The historical charging data includes charging timestamp, charging amount, charging voltage and charging cost.

[0035] S202 obtains the user's driving trajectory data through the vehicle's global positioning system and / or map service application interface. The driving trajectory data includes time-series latitude and longitude path information.

[0036] S203, with the user's authorization, extracts the user's social preference data through the application programming interface of the user's social media platform.

[0037] S204 obtains contextual data from meteorological service platforms, traffic management systems, and calendar tools. The contextual data includes weather information, holiday schedule information, and real-time traffic flow information.

[0038] S205 standardizes the collected historical charging data, driving trajectory data, social preference data, and contextual environment data to obtain multimodal data.

[0039] In some embodiments, the social preference data is presented in text form such as interest tags and keywords related to charging and travel.

[0040] In some embodiments, the collected historical charging data, driving trajectory data, social preference data, and contextual environment data are standardized. Specifically, numerical data such as charging power and voltage are normalized to unify the data volume; textual data such as social preferences are cleaned and deduplicated to retain effective keywords; and time series data such as driving trajectories are interpolated to ensure uniform time granularity, ultimately resulting in standardized multimodal data.

[0041] In some embodiments, acquiring multimodal user data can be achieved through multi-source data collection and standardized integration. Firstly, historical charging data of the target user can be synchronously obtained from the charging pile operation platform of the photovoltaic-storage-charging system and the user charging application (APP) associated with the system. It should be understood that historical charging data is a complete record of the user's past use of the photovoltaic-storage-charging system's charging service, including charging timestamps (accurate to the second, used to locate the specific start and end times of each charge), charging capacity (unit: kWh, recording the total charging energy of a single charge), charging voltage (unit: V, reflecting voltage stability and charging mode during the charging process, such as fast charging / slow charging), and charging cost (unit: yuan, including basic electricity fees, service fees, and the final amount after discounts), ensuring that the data completely covers the core quantitative indicators of the user's charging behavior.

[0042] Subsequently, the GPS module on the user's vehicle collects the vehicle's location information in real time during its journey. Simultaneously, with the user's authorization, third-party map service application programming interfaces (APIs, such as those for Gaode Maps and Baidu Maps Open Platform) are invoked to supplement the data with high-precision driving trajectory data. It should be understood that the driving trajectory data is organized primarily in a time-series format, with each data record containing the latitude and longitude coordinates (accurate to the meter level) of the corresponding time point, forming complete time-series latitude and longitude path information. This clearly reconstructs the user's travel routes, driving range, and stopping areas before and after charging, providing spatial dimensional support for subsequent association with charging scenarios. Furthermore, with the user's explicit authorization, social preference data related to the user's charging and travel can be extracted through the official application programming interfaces of the user's linked social media platforms.

[0043] It should be noted that the data collection process strictly follows data security and privacy protection regulations, extracting only content that users have publicly published or authorized to share, such as posts, comments, or interest tags containing keywords such as "fast charging recommendation," "long-distance charging guide," and "low-cost charging piles." Redundant information unrelated to charging and travel is filtered out to ensure that social preference data is highly relevant to the personalized service needs in this application embodiment.

[0044] Afterwards, real-time and forecast weather information, including indicators such as temperature, humidity, precipitation probability, and wind force level, can be obtained from professional meteorological service platforms (such as the China Weather Network interface); real-time regional traffic flow information can be obtained from the public data interface of the traffic management system, with the traffic flow index (0-10 levels) representing the degree of road congestion; and holiday arrangement information (including statutory holidays, adjusted workdays, weekends, etc.) can be obtained through the system's built-in calendar tool and public holiday database, forming contextual environmental data covering three dimensions: weather, holidays, and traffic, comprehensively capturing external environmental factors that affect users' charging decisions.

[0045] Finally, by standardizing the four types of raw data collected above, we ensured that the data format and magnitude were consistent, meeting the requirements of subsequent model input. For example, for numerical data such as charging power, voltage, and cost in historical charging data, the Min-Max normalization method was used to map the data to the [0,1] interval, eliminating the magnitude differences between different indicators. Another example is the latitude and longitude coordinates in driving trajectory data, which were converted to Cartesian coordinates using Gaussian projection and then normalized. Simultaneously, linear interpolation was performed on the time series, unifying the time granularity to 1 minute per data point. This approach addresses the issue of missing values ​​during data collection. For example, textual information in social preference data is converted into numerical word vectors using a bag-of-words model or TF-IDF algorithm, removing invalid information such as stop words and punctuation marks while retaining core keyword features. Categorical data in contextual data (such as weather types and holiday markers) is converted into numerical features using one-hot encoding. Continuous data, such as traffic flow indices, is also normalized. Ultimately, all data is integrated into structured multimodal data, providing high-quality data input for subsequent user profile construction and recommendation model inference.

[0046] In one possible implementation, the user profile includes a user behavior feature vector, such as... Figure 3 As shown, multimodal data is processed based on a large language model to construct user profiles, including: S301 uses a multimodal encoder in a large language model to perform feature embedding encoding on multimodal data, resulting in multiple encoded vectors. S302, input multiple encoded vectors into the multi-layer transformer in the large language model in a time series order to obtain multiple feature vectors output by the multi-layer transformer.

[0047] Among them, the multi-layer transformer is used to obtain the temporal dependencies and behavioral patterns of user behavior from the encoding vector based on the self-attention mechanism, so as to generate feature vectors. S303 calculates the mean and weights multiple feature vectors to construct a user behavior feature vector.

[0048] Among them, the user behavior feature vector includes the distribution of average daily charging time periods, preferred charging locations, average battery capacity preference, sensitivity to discounts, and high-frequency travel routes.

[0049] In some embodiments, user profiles include user behavior feature vectors, constructed through deep processing of multimodal data using a large language model. For example, the multimodal encoder built into the large language model can be used to perform targeted feature embedding encoding on the standardized multimodal data, uniformly converting data of different types and dimensions into encoded vectors of the same dimension, laying the foundation for subsequent time-series modeling. For instance, for structured numerical data such as historical charging data and contextual environment data, transformation can be performed using a linear embedding layer in the multimodal encoder, utilizing linear transformation formulas. (in, This is the weight matrix. For bias terms, For standardized numerical features, the data is mapped into fixed-dimensional vectors. For textual information (such as interest tags and keywords) in social preference data, the text encoding layer of a large language model is used to extract semantic feature vectors. For example, the output of the last hidden layer of the BERT model is used as the text encoding result to ensure accurate capture of user preference semantics related to charging and travel. For time-series latitude and longitude path information in driving trajectory data, a trajectory convolution combined with location embedding encoding method is adopted. First, the spatial features of the trajectory are extracted through a one-dimensional convolutional layer, and then a location encoding vector is added to represent the chronological order in the time series, finally generating an encoding vector that integrates spatial and temporal features. After the above encoding processing, all multimodal data are converted into 512-dimensional encoding vectors to achieve unified data format adaptation.

[0050] Next, the multiple encoded vectors are arranged in order of their corresponding timestamps to form time-series data, which is then input into the multi-layer transformer in the large language model for temporal modeling. This multi-layer transformer contains six encoder layers, each with eight attention heads. It uses a self-attention mechanism to capture the temporal dependencies and potential behavioral patterns of user behavior. The self-attention mechanism calculates the association weights between each encoded vector and all other vectors in the sequence, using the following formula: (Where Q is the query vector matrix, K is the key vector matrix, and V is the value vector matrix,) (As the key vector dimension), it can highlight the importance of users' recent high-frequency behaviors and behaviors at key time nodes. For example, it can identify typical behavioral patterns such as "frequently charging near the company on weekdays from 18:00 to 20:00" and "prioritizing charging stations in highway service areas before long-distance travel during holidays". After the encoded vector sequence is processed by components such as the fully connected layer and normalization layer of the multi-layer transformer, multiple 512-dimensional feature vectors with the same length as the input sequence are output. Each feature vector integrates the user behavior features at the corresponding time point and the temporal correlation information of the entire sequence.

[0051] Then, by averaging and weighting multiple feature vectors output by the multi-layer transformer, a user behavior feature vector is formed. Specifically, the mean of all feature vectors is first calculated to obtain the basic feature vector, which reflects the user's long-term stable behavioral characteristics. Based on the attention weights output by the self-attention mechanism, each feature vector is assigned a corresponding weight coefficient (the higher the attention weight, the greater the contribution of the corresponding feature vector to the representation of user behavior), and a weighted summation operation is performed. Combining the mean vector and the weighted summation result, feature fusion is performed through a fully connected layer, and finally a 512-dimensional user behavior feature vector is constructed, which can accurately depict the user's charging behavior habits and core preferences.

[0052] For example, user behavior feature vectors may include: daily average charging time distribution (such as the proportion of charging frequency during morning peak, evening peak, and nighttime off-peak hours), preferred charging locations (such as preference coefficients for scenarios such as residential areas, office areas, business districts, and highway service areas), average battery capacity preference (such as the target battery capacity for a single charge and the frequency of charging to full capacity), discount sensitivity (such as the probability of responding to peak and off-peak discounts and full-reduction offers), and high-frequency travel routes (such as charging demand characteristics corresponding to commuting routes and commonly used long-distance travel routes).

[0053] One possible implementation is, such as Figure 4 As shown, based on user profiles, vehicle data, and real-time status data, personalized charging recommendations are generated for each user, including: S401 obtains the location coordinates, idle status, current charging cost, and load status of multiple candidate charging piles based on real-time status data.

[0054] S402, based on the location coordinates, idle status, current charging cost and load status of multiple candidate charging piles, construct the candidate charging pile feature vectors for each of the multiple candidate charging piles.

[0055] S403, based on the user behavior feature vector in the user profile and the user's historical behavior sequence data, generate a temporal context vector related to the user's current behavior characteristics.

[0056] S404 uses a recurrent neural network and attention model to calculate the matching degree between the temporal context vector and the feature vector of each candidate charging pile, and assigns preference weights to multiple candidate charging piles. The preference weights are used to reflect the user's priority in choosing different candidate charging piles. S405 generates personalized charging recommendations based on the preference weights of multiple candidate charging piles, real-time status data, vehicle data, and the average daily charging distribution time period in the user behavior feature vector.

[0057] The personalized charging recommendation information, based on vehicle data including the vehicle's current latitude and longitude coordinates, remaining battery percentage, total battery capacity, and the user's current planned travel route, includes Top-N recommended charging stations, suggested charging time periods, and applicable discount information.

[0058] In this embodiment, personalized charging recommendation information is generated collaboratively based on user profiles, vehicle data, and real-time status data of charging piles. The core is achieved through feature construction, time-series modeling, matching calculation, and multi-dimensional integration.

[0059] For example, based on real-time status data synchronized from the photovoltaic-storage-charging system management platform, multiple candidate charging piles in normal operation (non-fault, non-maintenance) can be selected. Key operational information for each candidate charging pile can be extracted: location coordinates can be precise latitude and longitude data down to the meter level, used to calculate the distance to the user's current location and the degree of matching with their travel route. Real-time updated operational status includes three categories: "idle," "occupied," and "reservation pending," prioritizing "idle" and "available within 1 hour" charging piles. Current charging costs include basic electricity fees, service fees, and the final unit price after dynamic pricing adjustments (unit: yuan / kWh), clearly indicating whether peak / off-peak discounts, full-reduction offers, etc., are included. Load status is calculated by the ratio of currently occupied interfaces to the total number of interfaces (0-100%). Combined with the load change trend over the past hour, the load fluctuation situation for the next 1-3 hours is predicted, providing a basis for selecting recommended time periods.

[0060] For each candidate charging pile, its location coordinates, idle status, current charging cost, and load status are converted into a standardized feature vector, i.e., the candidate charging pile feature vector, ensuring consistency with the vector dimensions related to subsequent users and computability. For spatial feature processing, the location coordinates can be converted into Cartesian coordinates through Gaussian projection and then mapped to the [0,1] interval using Min-Max normalization, serving as the spatial feature dimension. For status feature encoding, the idle status can be one-hot encoded, with "idle" corresponding to [1,0,0], "occupied" to [0,1,0], and "reserved" to [0,0,1], converting it into numerical features. For numerical feature standardization, the current charging cost and load rate can be standardized using Z-score to eliminate magnitude differences. For vector integration, the spatial features, status features, and standardized numerical features can be concatenated sequentially to construct a 512-dimensional candidate charging pile feature vector, comprehensively covering the core attributes of the charging pile operation.

[0061] Subsequently, combining user behavior feature vectors from the user profile with the user's historical behavior sequence data over the past 30 days (including historical charging time, location, battery level changes, travel routes, pricing responses, etc.), a recurrent neural network (RNN) is used for temporal modeling to generate a temporal context vector highly correlated with the user's current behavioral characteristics. In some embodiments, a GRU (Gated Recurrent Unit) model can be preferentially used (an LSTM model can be used if the historical sequence is long). The user's historical behavior sequence is sorted by timestamp and converted into a 128-dimensional temporal feature sequence as the model input. The GRU model controls the retention and updating of historical information through update gates and adjusts the influence of the current input through reset gates, iteratively updating the hidden layer state to capture the temporal evolution pattern of user behavior, such as "prioritizing fast charging stations near the company when the battery level is below 30% on weekday evenings" and "charging near highway entrances 2 hours before long-distance travel on weekends." Finally, the hidden state of the last time step of the GRU model is taken as the temporal context vector, with a dimension of 512. This vector integrates the user's long-term behavioral habits and recent behavioral trends, which can accurately represent the user's current charging needs.

[0062] By working in conjunction with an attention model through a recurrent neural network (i.e., the GRU model mentioned above), the matching degree between the temporal context vector and the feature vectors of each candidate charging station is calculated, and a preference weight is assigned to each candidate charging station. The temporal context vector can be used as the query vector (Q), and the feature vectors of each candidate charging station can be used as the key vector (K) and value vector (V), respectively, ensuring that the vector dimensions are consistent (both are 512-dimensional).

[0063] Through the formula of attention mechanism Q represents the user's current state vector, and Ki and Vi are the embedding vectors of the candidate charging stations. Weights are assigned to each candidate station (calculated using the formula above, i.e., the dot product of the user's behavior sequence and the candidate charging station's feature vector, then converted to corresponding weights using softmax), representing the recommendation priority. Parameters : Represents the user's attention / preference weights for candidate stakes, parameter Q: Represents the user's behavior sequence, i.e., the output of the RNN above, parameter : Represents the transpose of the vector matrix of features of the i-th candidate charging station. The softmax function ensures that the sum of the preference weights of all candidate charging stations is 1; the larger the preference weight, the higher the degree of fit between the candidate charging station and the user's current behavior characteristics, historical habits, and needs, and the greater the probability that the user will choose it first. For example, if the user's time-series context vector shows "preferring fast charging stations in office areas and being sensitive to off-peak discounts", then candidate charging stations near office areas that support off-peak discounts and are fast charging types will receive higher preference weights.

[0064] Based on multi-dimensional constraints, and considering the daily average charging time distribution in the user behavior feature vector, a complete personalized charging recommendation is generated. Vehicle data includes current latitude and longitude coordinates, remaining battery percentage, total battery capacity, and the user's planned travel route. Key verification criteria include whether the distance from the user's current location to the candidate charging station is within the remaining driving range and whether the charge level can meet the user's subsequent travel needs. Candidate charging stations that cannot support the user's travel are eliminated. The Top-N recommended charging stations are first sorted in descending order of preference weight, selecting the top 10. Then, real-time status data (prioritizing charging stations with a load rate below 60% and clear idle status) and vehicle travel route matching (prioritizing charging stations along the route or within 2 kilometers of the route) are used to finally select the Top-3 to Top-5 recommended charging stations, ensuring the practicality and accessibility of the recommendations. Suggested charging time periods can be combined with the daily average charging time distribution in the user behavior feature vector (e.g., user habit of charging between 18:00-20:00) and the load prediction of candidate charging stations (avoiding high load rates in the next 1-3 hours). Based on 80% of the time period and the user's current travel plan (e.g., if the user plans to depart at 19:00, charging from 17:30 to 18:30 is recommended), the optimal charging time period is determined, and the expected idle rate and charging time (calculated based on the remaining battery power and charging pile power) of this time period are also marked. The matching of appropriate discount information can be based on the discount sensitivity in the user's behavioral feature vector (e.g., multiple discounts are pushed to highly sensitive users first) and the current pricing policy of the candidate charging piles (e.g., discounts for purchases over a certain amount, time-based discounts, new customer subsidies, etc.) to match appropriate discount information, and clearly mark the discount strength, usage conditions and expected cost after the discount. Finally, the Top-N recommended charging piles (including name, location, distance, charging power, and idle status), suggested charging time periods (including time range, expected time, and predicted idle rate) and appropriate discount information (including discount type, strength, and expected savings) are integrated into structured recommendation content, which is pushed to users through associated applications, while providing quick access to navigation jumps, scheduled charging and other operations.

[0065] One possible implementation is, such as Figure 5 As shown, the method also includes: S501 acquires historical load data of multiple candidate charging piles, user response behavior data to charging services, and electricity demand data for different time periods.

[0066] In some embodiments, the present application embodiments use deep learning models to mine charging pile operation data and user behavior patterns to achieve dynamic optimization of pricing strategies, ensuring that charging pricing not only meets user needs but also balances charging pile load and operating revenue.

[0067] In some embodiments, firstly, hourly load records of multiple candidate charging piles over the past 30 days can be collected, including the load rate (current number of occupied interfaces / total number of interfaces × 100%), average charging time, and peak and valley values ​​of charging volume for each time period of the day (divided into peak, flat, and valley periods), forming a historical load time series dataset for analyzing load change patterns (e.g., the load rate is generally higher than 80% during the morning peak of 7:00-9:00 and the evening peak of 18:00-20:00 on weekdays).

[0068] Subsequently, user behavior feedback records for different pricing strategies were collected over the past 30 days, including behaviors such as accepting pricing and completing charging, accepting pricing and canceling reservations, refusing pricing and not charging, and changing charging stations due to price adjustments. At the same time, corresponding pricing schemes (such as base price, discounted price, and full reduction price), charging scenarios (commuting, long distance, emergency) and user profile tags were associated to build a user response behavior database and mine user price sensitivity characteristics (such as discount-sensitive users with a 75% response rate to full reduction activities, and users with essential needs having a higher tolerance for price fluctuations). Next, the electricity demand is divided into three periods according to the natural day: peak (08:00-11:00, 18:00-21:00), average (06:00-08:00, 11:00-18:00, 21:00-23:00), and valley (23:00-06:00 the next day). Combined with the charging volume statistics of the past 90 days, the average electricity demand, demand volatility, and peak demand of each period are calculated. At the same time, factors such as holidays, weather, and regional activities are considered (for example, during holidays, the demand for long-distance travel surges, and the electricity demand in the valley is only 30% of that in the peak period), forming a periodic electricity demand feature library containing electricity demand data for different time periods.

[0069] The S502 uses a deep learning model to dynamically adjust the pricing strategy of multiple candidate charging piles based on historical load data, user response behavior data to charging services, and electricity demand data at different times.

[0070] In some embodiments, the DeepQ-Learning (DQN) model from deep reinforcement learning can be used as the deep learning model. The model is trained and the policy is learned with the joint optimization objective of "maximizing user satisfaction, balancing charging pile load, and improving operational revenue." For example, the DQN model includes a target network and an evaluation network, both consisting of three fully connected layers. The input layer has a 128-dimensional dimension (integrating historical load features, user response features, and time-period demand features), the hidden layer has a 64-dimensional dimension (using the ReLU activation function), and the output layer has a 4-dimensional dimension (corresponding to four types of candidate pricing strategies). Historical interaction data is stored in an experience replay pool to avoid interference from sample correlation during training.

[0071] Subsequently, the load rate, charging duration, and other numerical features in the historical load data are standardized to the [0,1] interval. User response behavior data is converted into numerical features through behavior type encoding (e.g., 1 for completing charging, 0 for refusing pricing) and sensitivity coefficient (calculated based on historical response frequency). Time-based electricity demand data is integrated into a feature vector through demand level encoding (3, 2, 1 for high, medium, and low) and time period identifiers (3, 2, 1 for peak, flat, and valley). Finally, these are concatenated into a 128-dimensional model input vector. During training, the learning rate can be set to 5e-4, the discount factor γ to 0.9, and the experience replay pool capacity to 10000. Mini-batch gradient descent (Batch Size=32) is used for training, with 1000 iterations. Every 200 iterations, the target network and evaluation network parameters are synchronized. During training, the constraints of "user acceptance rate ≥60%, load balance (standard deviation of load rate in each time period) ≤20%, and operating revenue 10% higher than fixed pricing" are used to ensure the practicality of the model's output strategy.

[0072] Based on the trained DQN model, the pricing strategies of each candidate charging station are dynamically generated and adjusted by combining real-time data. The DQN model can read the current time period (peak / flat / valley), the current load rate of the candidate charging station, the recent user response behavior trend (such as 30% of users refusing to charge due to high prices in the past hour), and real-time electricity demand data in real time, and update the model input feature vector. The DQN model calculates the candidate pricing strategies (such as basic pricing, discount pricing, time-limited discount pricing, tiered pricing) for each candidate charging station through forward propagation, as well as the expected revenue Q value corresponding to each of the multiple candidate pricing strategies.

[0073] It should be understood that the Q value can comprehensively reflect the long-term cumulative benefits (including user acceptance, load balancing effect, and direct benefits) under the corresponding candidate pricing strategy. The candidate pricing strategy with the highest Q value is selected as the current optimal pricing strategy, i.e., the target pricing strategy.

[0074] S503 outputs corresponding charging pricing data based on the pricing strategies of multiple candidate charging stations. If the current charging pile load rate is higher than 85% (peak-hour congestion), the DQN model automatically prioritizes "time-of-use discount pricing" (e.g., a 10% increase during peak hours to encourage users to avoid peak times); if the load rate is lower than 40% (off-peak idle period), it selects "spend-and-refund pricing" (e.g., a 15 RMB discount for every 40 RMB spent on charging to attract users); if user response data shows high price sensitivity (rejection rate ≥ 40% in the past 3 days), the pricing adjustment is reduced (e.g., a 5% increase instead of the originally planned 15%) to ensure the target pricing strategy adapts to the real-time operational scenario. Finally, the dynamically adjusted pricing strategy is converted into standardized charging pricing data and synchronized to the photovoltaic-storage-charging system management platform and user-related applications.

[0075] In some embodiments, the charging pricing data includes a base unit price (yuan / kWh), discount rules (such as minimum spending threshold, discount percentage, applicable time period), effective time (accurate to the minute), and applicable scenarios (such as only for commuter users, only supporting fast charging piles). For example, "Base unit price is 1.2 yuan / kWh, no discount during peak hours (08:00-11:00), 10% discount during off-peak hours (11:00-18:00), 20% discount during off-peak hours (23:00-06:00 the next day) + 12 yuan off for charging over 50 yuan, effective time is 00:00-24:00 today."

[0076] In some embodiments, pricing data can also be synchronized in real time to the control terminals of each candidate charging pile and the user's charging APP through an encrypted interface. The APP clearly displays the current pricing, discount details, and estimated charging cost, while also marking "dynamic pricing" and the reason for the price adjustment (such as "currently during peak electricity consumption, the price is 8% higher than during off-peak hours"), ensuring the user's right to know. Furthermore, the pricing data is updated every 30 minutes based on real-time load and user response. In case of emergencies, such as charging pile failure causing sudden load changes or extreme weather causing a surge in demand, an instant update mechanism can be triggered to ensure the timeliness and accuracy of the pricing data.

[0077] In one possible implementation, step S502 above employs a deep learning model to dynamically adjust the pricing strategy for multiple candidate charging piles based on historical load data, user response behavior data to charging services, and electricity demand data for different time periods. This includes: Obtain multiple candidate pricing strategies for each candidate charging pile, and the expected revenue Q value corresponding to each candidate pricing strategy.

[0078] An ε-greedy strategy is adopted to select the candidate pricing strategy with the best Q value from the expected revenue Q values ​​corresponding to multiple candidate pricing strategies for each candidate charging pile as the target pricing strategy for each candidate charging pile.

[0079] The system acquires user behavior feedback data under the target pricing strategy for each candidate charging station, and combines this data with historical response behavior data from multiple candidate charging stations to update user response behavior data for charging services.

[0080] Based on the updated user response behavior data and the actual operation results of multiple candidate charging piles, the reward value of each candidate charging pile is calculated.

[0081] Among them, the reward value is positively correlated with the charging success rate, negatively correlated with the user's abandonment of charging behavior, and positively correlated with the adaptability of electricity demand data in different time periods.

[0082] The network parameters of the Q network are updated based on the reward value, and the target pricing strategy of each candidate charging pile is iteratively optimized to form a dynamic pricing scheme that adapts to the actual load of each candidate charging pile, the electricity demand in the corresponding time period, and user behavior preferences.

[0083] First, a multi-tiered candidate strategy library is constructed for each candidate charging pile, including basic pricing, discounts for purchases over a certain amount, time-based discounts, and tiered pricing. The historical load data, historical user response behavior data, and time-based electricity demand data of the charging pile are integrated and standardized to form an environmental state vector, which is then input into the Q-network of the DQN model to calculate the long-term expected return Q value corresponding to each candidate strategy.

[0084] Next, an ε-greedy strategy is used to select the target pricing strategy. For example, by dynamically adjusting the exploration probability ε, the strategy of "utilizing the known optimal strategy" and "exploring potential better strategies" are balanced. In the early stage, the ε value is relatively high (e.g., 0.5), focusing on randomly selecting strategies to accumulate feedback data. As the iteration progresses, the ε decays exponentially to the minimum value (e.g., 0.01), and the strategy with the highest Q value is selected first to ensure the stability of returns.

[0085] Then, user behavior feedback under the target pricing strategy is collected in real time (such as completing charging, refusing pricing, etc.), and the user response behavior data is updated using a 7-day sliding window method to ensure that the data reflects the latest preferences. Based on the updated data and the actual operation results of the charging piles, reward values ​​are calculated through a multi-dimensional reward function (positively correlated with charging success rate and electricity demand matching, and negatively correlated with user abandonment rate) to quantify the effectiveness of the strategy implementation.

[0086] Finally, based on the reward value, the Q-network parameters are updated through temporal differential error. Every 100 iterations, the target network and evaluation network parameters are synchronized to continuously optimize the prediction accuracy of the strategy's Q-value. Through multiple iterations, the pricing strategy gradually adapts to the actual load conditions of each charging station (high load guides traffic diversion, low load attracts users), the electricity demand during corresponding time periods (moderate increase during peak hours, larger discounts during off-peak hours), and user behavior preferences (increase discounts for discount-sensitive users, maintain stable pricing for users with essential needs), ultimately forming a dynamically adjusted personalized pricing scheme.

[0087] One possible implementation involves obtaining multiple candidate pricing strategies for each candidate charging station, and the expected revenue Q value corresponding to each of the multiple candidate pricing strategies, including: Multiple candidate charging piles are treated as independent intelligent agents, and a corresponding environmental state vector is constructed for each intelligent agent.

[0088] Multiple candidate pricing strategies are set as optional actions for each smart agent. The candidate pricing strategies include basic pricing, discount pricing for purchases over a certain amount, time-limited discount pricing, and tiered pricing. The environmental state vector corresponding to each intelligent agent is input into the Q-network of the deep learning model, and the expected return Q value corresponding to multiple candidate pricing strategies is calculated through the Q-network.

[0089] The environmental state vector includes current time information, electricity demand data for the corresponding time period, historical load data of the corresponding candidate charging pile, the number of remaining vacant spaces of the candidate charging pile, queuing status, remaining charging time, user behavior feature vectors in the user profile, and historical pricing data of the candidate charging pile.

[0090] In this embodiment, a deep reinforcement learning model (DeepQ-Learning, DQN) is used to dynamically optimize the pricing strategy of candidate charging piles. The strategy is continuously adjusted by combining user behavior and charging pile operation data, following a closed-loop process of "strategy generation-selection-feedback-iteration", to ensure that the pricing is adapted to the load conditions, time period demand and user preferences.

[0091] First, for each candidate charging pile, multiple candidate pricing strategies are preset to cover the needs of different operating scenarios. These include basic pricing (standard pricing based on the local grid benchmark electricity price), discount pricing (such as "10 yuan off for charging over 50 yuan" or "20 yuan off for charging over 80 yuan"), time-based discount pricing (such as 5%-10% increase during peak hours, maintaining the benchmark price during off-peak hours, and 10%-15% decrease during off-peak hours), and tiered pricing (such as 0-50kWh at the benchmark price, 50-100kWh at a 10% discount, and over 100kWh at a 15% discount). Each strategy corresponds to 3-5 sub-tiers, forming a candidate pricing strategy library for each candidate charging pile.

[0092] The historical load data (hourly load rate and charging volume for the past 30 days), user historical response behavior data (acceptance rate and completion rate of various pricing methods), and electricity demand data for the corresponding time period (demand level and demand volatility during peak / flat / valley periods) of each candidate charging pile are standardized and integrated into a 128-dimensional environmental state vector. This vector is then input into the Q-network of the DQN model (consisting of 3 fully connected layers, with 128 input dimensions, 64 hidden layer dimensions, and the same number of output dimensions as the candidate pricing strategies).

[0093] The expected revenue Q value for each candidate pricing strategy is obtained through forward propagation calculation using the Q network. This Q value comprehensively represents the long-term cumulative revenue after the strategy is implemented (including user acceptance, load balancing effect, and operational revenue), providing a quantitative basis for strategy selection.

[0094] An ε-greedy strategy is employed to balance the "exploratory" and "utilitarian" aspects of pricing strategies, ensuring that both the validated optimal strategy is fully utilized and potential better solutions are explored. The initial value of ε is set to 0.5, gradually decreasing exponentially to 0.01 with each iteration. For each candidate charging station, a random probability value between 0 and 1 is generated: if the random probability value is less than the current ε, a strategy is randomly selected from the candidate pricing strategy library to explore its actual effect; if the random probability value is greater than or equal to the current ε, the candidate pricing strategy with the highest Q-value output by the Q-network is selected as the target pricing strategy for that candidate charging station, maximizing the current expected revenue. For example, if the Q-value calculation results for a candidate charging station during peak hours show that the "8% peak-hour markup + full-reduction discount" strategy has the highest Q-value, then this strategy is preferentially selected as the target pricing strategy in the non-exploratory state.

[0095] After the target pricing strategy takes effect, user behavior feedback data under this strategy is collected in real time. This includes behaviors such as accepting the price and completing charging, accepting the price and canceling the reservation, rejecting the price and not charging, and switching to other charging stations due to price adjustments. The timestamps of these behaviors, charging scenarios (commuting, long-distance, emergency), and user profile tags (discount-sensitive, essential, etc.) are also recorded. The real-time collected behavior feedback data is then integrated with the historical response behavior data (records of the last 30 days) of the candidate charging station. A sliding window method (window size set to 7 days) is used for data updates, removing outdated data outside the window and retaining the latest 7 days of behavior feedback records. This ensures that user response behavior data reflects changes in user preferences for the pricing strategy in real time. For example, if a user has accepted the "15% discount during off-peak hours" strategy and completed charging three times recently, the updated data will increase the weight of that user's response to the off-peak discount strategy.

[0096] Based on the updated user response behavior data and the actual operational results of the candidate charging stations, a multi-dimensional reward function is constructed to calculate the reward value for each candidate charging station. The reward function formula is as follows: Where: S is the charging success rate, which is the ratio of the number of completed charging attempts to the number of pricing notifications, reflecting the user's acceptance of the pricing strategy and positively correlated with the reward value; C is the user abandonment charging rate, which is the ratio of the sum of the number of canceled reservations and the number of times pricing was rejected to the number of pricing notifications, reflecting the irrationality of the pricing strategy and negatively correlated with the reward value; M is the electricity demand matching degree, which is the ratio of the actual charging amount in the current period to the historical average electricity demand in the same period, reflecting the matching effect of the pricing strategy on the demand in the current period and positively correlated with the reward value; a, b, c is the weighting coefficient, all set to 1.0, which can be dynamically adjusted according to operational goals (e.g., the weight of M can be increased when focusing on load balancing); the reward value ranges from [-1, 1]. For example, if a candidate charging pile has a charging success rate S=85%, a charging abandonment rate C=10%, and an electricity demand matching degree M=120% during off-peak hours, then the reward value R=1.0×0.85-1.0×0.1+1.0×1.2=1.95 (which becomes 0.975 after being mapped proportionally to the [-1, 1] interval), indicating that the strategy is highly effective.

[0097] Next, the calculated reward value and the environmental state vector for the next time step are input into the DQN model. The network parameters of the Q network are updated through temporal difference (TD) error, and the loss function is the mean squared error. Every 100 iterations, the parameters of the evaluation network are synchronized to the target network to ensure the stability of model training. Through multiple iterations, the accuracy of the Q-value calculation corresponding to each candidate pricing strategy is continuously optimized, so that the target pricing strategy gradually adapts to the actual load of the candidate charging pile (e.g., adjusting to a diversion pricing strategy during high load), the electricity demand during the corresponding time period (e.g., moderately increasing the price during peak demand and increasing the discount during off-peak demand), and user behavior preferences (e.g., increasing the discount for discount-sensitive users), ultimately forming a dynamically adjusted and precisely adapted pricing scheme. For example, after 500 iterations, the peak-hour pricing strategy of a certain candidate charging pile was optimized from "increase by 10%" to "increase by 5% + discount for commuter users", increasing the charging success rate by 20% and reducing the load rate from 90% to 75%, achieving a balance between user satisfaction and operational revenue.

[0098] In one possible implementation, the method further includes: Acquire user feedback data on personalized charging recommendations, including satisfaction scores, opinions, and behavioral selection data. Based on the evaluation data, the model parameters of at least one of the following are fine-tuned and optimized: large language model, recurrent neural network and attention model, and deep learning model.

[0099] In some embodiments, after personalized charging recommendation information is pushed, user evaluation data is obtained through methods such as pop-up windows in associated applications and questionnaires after charging is completed. The evaluation data includes satisfaction scores (quantitative scores of 1-5 points, with 5 points being the highest level of satisfaction), feedback (user suggestions or complaints in text form, such as "the recommended charging station is too far away" or "the discount is insufficient"), and behavioral selection data (whether to click on the recommendation list, whether to charge within the recommended time period, whether to use the appropriate discount, charging completion rate, etc.).

[0100] Subsequently, based on the aforementioned evaluation data, optimization objectives were constructed, and the model parameters of at least one of the following models were fine-tuned: large language model, recurrent neural network (GRU), attention model, and deep learning model (DQN). For the large language model, if the user satisfaction score is lower than 3 points and the feedback indicates that "the recommendation does not conform to the user's habits", the corresponding user's behavior selection data and multimodal data are used as supplementary training samples. The multimodal encoder and Transformer model parameters are fine-tuned using the mini-batch gradient descent method (Batch Size=8), with a learning rate of 1e-5 and 50 iterations to improve the accuracy of user profile characterization.

[0101] For the GRU and attention model, if the user does not click on the recommended charging station or abandons charging during the suggested time period, the weight calculation coefficient of the attention mechanism and the hidden layer parameters of the GRU model are adjusted to optimize the accuracy of the allocation of candidate charging station preference weights, with the loss target being "the deviation between the recommendation result and the user's choice".

[0102] For the DQN model, if users have low satisfaction ratings related to pricing or less than 30% of discounts are used, the pricing response records in the user behavior selection data are included in the reward function, the reward value weight coefficients a, b, and c are adjusted, and the parameters of the fully connected layer of the Q network are fine-tuned simultaneously to improve the user acceptance of the dynamic pricing strategy.

[0103] In some embodiments, after each round of fine-tuning, the model is validated using new user data from the past 7 days. If the recommended click-through rate and user satisfaction score increase by more than 5% compared to before the fine-tuning, the adjusted model parameters are retained; otherwise, the model is rolled back to the parameters before the fine-tuning to ensure the effectiveness of the model optimization.

[0104] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0105] Corresponding to the vehicle charging recommendation method based on the photovoltaic-storage-charging system described in the above embodiments, Figure 6This is a schematic diagram of a vehicle charging recommendation device based on a photovoltaic energy storage and charging system provided in an embodiment of this application. This device can be implemented as part or all of a computer device, which can be software, hardware, or a combination of both. Figure 7 The electronic device shown.

[0106] Reference Figure 6 The vehicle charging recommendation device based on the photovoltaic energy storage and charging system includes: The acquisition unit 601 is used to acquire the user's multimodal data, which includes historical charging data, driving trajectory data, social preference data, and contextual environment data.

[0107] Building unit 602 is used to process multimodal data based on a large language model to build user profiles, which are used to predict users' future charging behavior. The response unit 603 is used to respond to personalized charging recommendation requests by obtaining the user's vehicle data and the real-time status data of the charging pile. The vehicle data is used to characterize the user's current travel status and charging needs, and the real-time status data of the charging pile is used to characterize the operational service capabilities and usage conditions of multiple candidate charging piles in the photovoltaic-storage-charging system.

[0108] The generation unit 604 is used to generate personalized charging recommendation information for users based on user profiles, vehicle data, and real-time status data.

[0109] It is understood that the embodiments and any implementations of the vehicle charging recommendation device based on the photovoltaic energy storage and charging system correspond to the embodiments and any implementations of the vehicle charging recommendation method based on the photovoltaic energy storage and charging system. The technical effects corresponding to the embodiments and any implementations of the vehicle charging recommendation device based on the photovoltaic energy storage and charging system can be found in the aforementioned embodiments and any implementations of the vehicle charging recommendation method based on the photovoltaic energy storage and charging system, and will not be repeated here.

[0110] It should be noted that the vehicle charging recommendation device based on the photovoltaic energy storage and charging system provided in the above embodiments is only an example of the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0111] The functional units and modules in the above embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of the embodiments of this application.

[0112] It should be noted that the information interaction and execution process between the above-mentioned devices / units are based on the same concept as the method embodiments of this application. For details on their specific functions and technical effects, please refer to the method embodiments section, and they will not be repeated here.

[0113] This application also provides an electronic device, which includes one or more processors and a memory; The memory is coupled to one or more processors. The memory is used to store computer program code, which includes computer instructions. The one or more processors call the computer instructions to cause the electronic device to execute the vehicle charging recommendation method based on the photoelectric storage and charging system described above.

[0114] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device 700 can be a mobile phone, smart screen, tablet computer, wearable electronic device, in-vehicle electronic device, augmented reality (AR) device, virtual reality (VR) device, laptop computer, ultra-mobile personal computer (UMPC), netbook, personal digital assistant (PDA), projector, or a communication device such as a server, storage device, or base station, or a smart car, etc. This application embodiment does not impose any limitations on the specific type of electronic device.

[0115] The memory 701 can be used to store computer software programs 702 and modules. The processor 703 executes various functional applications and data processing of the electronic device by running the software programs and modules stored in the memory 701. The memory 701 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, application programs required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the electronic device (such as audio data, telephone directory, etc.). In addition, the memory 701 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0116] The processor 703 may include one or more processors such as a central processing unit (CPU), an application processor (AP), and a baseband processor. The processor can serve as the nerve center and command center of the wireless router. The processor 703 can generate operation control signals based on instruction opcodes and timing signals to control instruction fetching and execution. The memory 701 can be used to store executable program code, including instructions. The processor 703 executes various functional applications and data processing of the network device by running the instructions stored in the memory. The memory 701 may include a program storage area and a data storage area, such as storing data for audio signals to be played. For example, the memory may be Double Data Rate Synchronous Dynamic Random Access Memory (DDR) or Flash memory.

[0117] This application also provides a computer-readable storage medium storing computer instructions; when the computer-readable storage medium is used on an electronic device, it causes the electronic device to execute the aforementioned vehicle charging recommendation method based on a photoelectric energy storage and charging system.

[0118] The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or can include one or more data storage devices such as servers or data centers that can be integrated with media. The available medium can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media, or semiconductor media (e.g., solid-state disks (SSDs)).

[0119] This application also provides a computer program product containing computer instructions, which, when run on an electronic device, enables the electronic device to execute the aforementioned vehicle charging recommendation method based on a photoelectric energy storage and charging system.

[0120] The computer storage medium and computer program product provided in the above embodiments of this application are used to execute the methods provided above. Therefore, the beneficial effects they can achieve can be referred to the beneficial effects corresponding to the methods provided above, and will not be repeated here.

[0121] In the above embodiments, implementation can also be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, optical fiber, Digital Subscriber Line, DSL) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access, or a data storage device such as a server or data center that integrates one or more available media. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk drive (HDD), or solid-state drive (SSD), etc., and the storage medium can also include combinations of the above types of memory.

[0122] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0123] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments claimed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0124] In the embodiments provided in this application, it should be understood that the disclosed apparatus / network devices and methods can be implemented in other ways. For example, the apparatus / network device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0125] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0126] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A vehicle charging recommendation method based on a photovoltaic-energy storage-charging system, characterized in that, include: Acquire user's multimodal data, which includes historical charging data, driving trajectory data, social preference data, and contextual environment data; The multimodal data is processed based on a large language model to construct a user profile of the user, which is used to predict the user's future charging behavior. In response to a personalized charging recommendation request, the system obtains the user's vehicle data and the real-time status data of the charging piles. The vehicle data is used to characterize the user's current travel status and charging needs, and the real-time status data of the charging piles is used to characterize the operational service capabilities and usage conditions of multiple candidate charging piles within the photovoltaic-storage-charging system. Based on the user profile, the vehicle data, and the real-time status data, personalized charging recommendations are generated for the user.

2. The method according to claim 1, characterized in that, The acquisition of user multimodal data includes: The user's historical charging data is obtained from the charging pile platform and / or associated application of the photovoltaic energy storage and charging system. The historical charging data includes charging timestamp, charging amount, charging voltage and charging cost. The user's driving trajectory data is obtained through the vehicle's global positioning system and / or map service application interface, and the driving trajectory data includes time-series latitude and longitude path information; With the user's authorization, the user's social preference data is extracted through the application programming interface of the user's social media platform. Contextual environment data is obtained from meteorological service platforms, traffic management systems, and calendar tools. This contextual environment data includes weather information, holiday arrangement information, and real-time traffic flow information. The collected historical charging data, driving trajectory data, social preference data, and contextual environment data are standardized to obtain the multimodal data.

3. The method according to claim 1, characterized in that, The user profile includes user behavior feature vectors. The process of processing the multimodal data based on a large language model to construct the user profile includes: The multimodal encoder in the large language model is used to perform feature embedding encoding on the multimodal data to obtain multiple encoding vectors; Multiple encoded vectors are input into the multi-layer transformer in the large language model in a time sequence to obtain multiple feature vectors output by the multi-layer transformer. The multi-layer transformer is used to obtain the temporal dependency and behavior pattern of user behavior from the encoded vectors based on the self-attention mechanism to generate the feature vectors. The user behavior feature vector is constructed by averaging and weighting multiple feature vectors. The user behavior feature vector includes the distribution of average daily charging time periods, preferred charging locations, average battery capacity preference, sensitivity to discounts, and high-frequency travel routes.

4. The method according to claim 1, characterized in that, The step of generating personalized charging recommendations for the user based on the user profile, the vehicle data, and the real-time status data includes: Based on the real-time status data, the location coordinates, idle status, current charging cost, and load status of multiple candidate charging piles are obtained. Based on the location coordinates, idle status, current charging cost and load status of multiple candidate charging piles, a candidate charging pile feature vector is constructed for each of the multiple candidate charging piles. Based on the user behavior feature vector in the user profile and the user's historical behavior sequence data, a temporal context vector related to the user's current behavior features is generated. By using a recurrent neural network and an attention model, the matching degree between the temporal context vector and the feature vectors of each candidate charging pile is calculated, and preference weights are assigned to each of the candidate charging piles. The preference weights are used to reflect the user's preference for different candidate charging piles. The personalized charging recommendation information is generated based on the preference weights of multiple candidate charging piles, the real-time status data, the vehicle data, and the average daily charging distribution time period in the user behavior feature vector. The vehicle data includes the vehicle's current latitude and longitude coordinates, the remaining battery power percentage, the total battery capacity, and the currently planned user travel route. The personalized charging recommendation information includes Top-N recommended charging piles, suggested charging time periods, and applicable discount information.

5. The method according to any one of claims 1 to 4, characterized in that, The method further includes: Acquire historical load data of multiple candidate charging piles, user response behavior data to charging services, and electricity demand data for different time periods; A deep learning model is used to dynamically adjust the pricing strategy of multiple candidate charging piles based on historical load data of multiple candidate charging piles, user response behavior data to charging services, and electricity demand data at different time periods. Based on the pricing strategies of the multiple candidate charging piles, output the corresponding charging pricing data.

6. The method according to claim 5, characterized in that, The method employs a deep learning model to dynamically adjust the pricing strategy for multiple candidate charging piles based on historical load data, user response behavior data to charging services, and electricity demand data over different time periods. This includes: Obtain multiple candidate pricing strategies for each candidate charging pile, and the expected revenue Q value corresponding to each of the multiple candidate pricing strategies; An ε-greedy strategy is adopted to select the candidate pricing strategy with the optimal Q value from the expected revenue Q values ​​corresponding to multiple candidate pricing strategies for each candidate charging pile as the target pricing strategy for each candidate charging pile. The user's behavioral feedback data under the target pricing strategy of each candidate charging pile is obtained respectively, and the user's response behavior data to the charging service is updated by combining the historical response behavior data of multiple candidate charging piles. Based on the updated user response behavior data and the actual operation results of the multiple candidate charging piles, the reward value of each of the multiple candidate charging piles is calculated. The reward value is positively correlated with the charging success rate, negatively correlated with the user's abandonment of charging behavior, and positively correlated with the adaptability of electricity demand data in different time periods. The network parameters of the Q network are updated based on the reward value, and the target pricing strategy of each candidate charging pile is iteratively optimized to form a dynamic pricing scheme that adapts to the actual load of each candidate charging pile, the electricity demand in the corresponding time period, and user behavior preferences.

7. The method according to claim 6, characterized in that, The step of obtaining multiple candidate pricing strategies for each candidate charging pile, and the expected revenue Q value corresponding to each of the multiple candidate pricing strategies, includes: Each of the candidate charging piles is treated as an independent smart agent, and a corresponding environmental state vector is constructed for each smart agent. The environmental state vector includes the current time information, the electricity demand data for the corresponding time period, the historical load data of the corresponding candidate charging pile, the number of remaining slots of the candidate charging pile, the queuing status, the remaining charging time, the user behavior feature vector in the user profile, and the historical pricing data of the candidate charging pile. For each of the intelligent agents, multiple candidate pricing strategies are set as optional actions, including basic pricing, discount pricing for purchases over a certain amount, time-limited discount pricing, and tiered pricing; The environmental state vector corresponding to each of the intelligent agents is input into the Q network of the deep learning model, and the expected revenue Q value corresponding to multiple candidate pricing strategies is calculated through the Q network.

8. The method according to any one of claims 1 to 4, characterized in that, The method further includes: Obtain the user's evaluation data on the personalized charging recommendation information, the evaluation data including satisfaction score, feedback and behavior selection data; Based on the evaluation data, the model parameters of at least one of the large language model, recurrent neural network and attention model, and deep learning model are fine-tuned and optimized.

9. A vehicle charging recommendation device based on a photovoltaic energy storage and charging system, characterized in that, The apparatus implements the method as described in any one of claims 1 to 8, comprising: The acquisition unit is used to acquire the user's multimodal data, which includes historical charging data, driving trajectory data, social preference data, and contextual environment data. The construction unit is used to process the multimodal data based on a large language model to construct a user profile of the user, which is used to predict the user's future charging behavior. The response unit is used to respond to a personalized charging recommendation request by obtaining the user's vehicle data and the real-time status data of the charging pile. The vehicle data is used to characterize the user's current travel status and charging needs, and the real-time status data of the charging pile is used to characterize the operational service capabilities and usage conditions of multiple candidate charging piles in the photovoltaic-storage-charging system. The generation unit is used to generate personalized charging recommendation information for the user based on the user profile, the vehicle data, and the real-time status data.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it causes the electronic device to implement the method as described in any one of claims 1 to 8.

11. A computer program product, characterized in that, Includes a computer program, which, when run, causes the method as described in any one of claims 1 to 8 to be performed.

12. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 8.

Images