Charging station dynamic recommendation method and system based on multi-dimensional real-time data
By constructing a dynamic recommendation method for charging stations based on multi-dimensional real-time data, and utilizing multi-source heterogeneous relationship graphs and hierarchical graph attention mechanisms, the problem of insufficient equipment availability and user intent matching in existing charging station recommendations is solved, achieving efficient and interpretable charging station recommendations.
Patent Information
- Application Number
- CN202511824033.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-05
- Publication Date
- 2026-05-08
AI Technical Summary
Existing charging station recommendation technologies fail to effectively integrate multi-source heterogeneous data, particularly failing to explicitly model the physical hierarchy of charging facilities and the complex relationship between user charging preferences and station service capabilities, resulting in insufficient recommendation results in terms of equipment availability and user intent matching.
A dynamic recommendation method for charging stations based on multi-dimensional real-time data is proposed. This method constructs a multi-source heterogeneous relationship graph of related orders, charging guns, charging piles, and charging stations, and uses a hierarchical graph attention mechanism for feature aggregation. By integrating real-time status and historical statistical features, dynamic modeling and recommendation of equipment status can be achieved.
It improves the availability of recommended charging stations and the matching accuracy of user intent, reduces the failure rate of users arriving at stations where charging is not possible, and enhances the interpretability and operational synergy of the recommendation system.
Smart Images
Figure CN121996849A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of big data analytics, and in particular to a method and system for dynamic recommendation of charging stations based on multidimensional real-time data. Background Technology
[0002] With the popularization of new energy vehicles, efficient and intelligent charging station recommendation systems are crucial for improving user experience and station operation efficiency. Existing charging station recommendation technologies mainly fall into two categories: one is location-based services and static rules, such as recommending the nearest, lowest-priced, or most available charging stations; the other is personalized recommendation models based on collaborative filtering or matrix factorization, which use historical user order data to predict preferences.
[0003] However, existing technologies have significant shortcomings. First, they generally treat charging equipment as static resources, using only historical average data or instantaneous idle states, lacking fine-grained modeling of the real-time operating status (such as faults and power fluctuations) and dynamic reliability of charging guns and charging piles. This can lead to users being recommended to devices that appear idle but are actually unavailable, resulting in a poor "failure upon arrival" experience. Second, existing methods fail to effectively integrate multi-source heterogeneous data, particularly failing to explicitly model the physical hierarchy of charging facilities ("gun-pile-station") and the complex relationship between user charging preferences and station service capabilities. This results in deficiencies in recommendation results regarding device availability, matching of user's true intent, and station operational synergy. Summary of the Invention
[0004] To overcome the shortcomings of the prior art, this invention proposes a dynamic recommendation method and system for charging stations based on multi-dimensional real-time data.
[0005] To achieve the above objectives, the present invention adopts the following technical solution, including: A dynamic recommendation method for charging stations based on multidimensional real-time data includes: S1: Based on the current user's charging request, obtain the charging data of all users, including historical order data, real-time charging gun data, charging pile data, and charging station data; S2: Construct a multi-source heterogeneous relationship graph G=(V,E) based on the charging data of all users, linking orders, charging guns, charging piles, and charging stations; where V is the set of nodes and E is the set of edges. The nodes include: order nodes, charging gun nodes, charging pile nodes, and charging station nodes; the edges include: order-order edges, order-charging gun edges, charging gun-charging pile edges, and charging pile-charging station edges. S3: Construct the corresponding initial feature vector for each node in the multi-source heterogeneous relationship graph; S4: Update the characteristics of each order node for all users based on adjacent order nodes; S5: Update the characteristics of all charging gun nodes based on the order node characteristics of each user; S6: Update the features of all charging pile nodes based on the features of the charging gun nodes; S7: Update the features of all charging station nodes based on the features of all charging pile nodes; S8: Iterate through steps S4-S7 L times to obtain the final updated features of all charging station nodes and calculate the recommendation index of all charging stations. S9: Recommend charging stations to the current user in descending order of their recommendation index.
[0006] Preferably, the order data includes order ID, vehicle ID, charging gun ID, start charging time, end charging time, charging duration, total electricity cost, total service fee, total charging amount, and reason for ending charging. The charging gun data includes the charging gun ID, the charging pile ID to which it belongs, the charging gun type, the rated power of the charging gun, the real-time power of the charging gun, the current status of the charging gun, and the duration of the current status. Charging pile data includes charging pile ID, charging station ID, charging pile type, rated power of charging pile, real-time power of charging pile, charging pile status information, and duration of charging pile status. The charging station data includes the charging station ID, charging station longitude, charging station latitude and longitude, equipment online rate, equipment failure rate, and user rating. A charging station consists of several charging piles, and a charging pile consists of several charging guns.
[0007] Preferably, the specific construction process of the multi-source heterogeneous relationship graph G of the associated order-charging gun-charging pile-charging station is as follows: Node definition: Each valid historical order or ongoing order creates an order node; each physical charging gun creates a charging gun node; each physical charging pile creates a charging pile node; each charging station creates a charging station node. Edge definition: Order-order edge: For all order nodes under the same vehicle, establish a bidirectional edge between them; Order-gun edge: Based on the charging gun ID in the order data, establish a directed edge from the order node to its corresponding charging gun node; Gun-pile edge: Based on the charging pile ID in the charging gun data, establish a directed edge from the charging gun node to its corresponding charging pile node. Charging pile-station edge: Based on the charging station ID in the charging pile data, a directed edge is established from the charging pile node to its corresponding charging station node.
[0008] Preferably, in step S3, the characteristic of the i-th order node is: ;in, Indicates the first The charging time for each order. Indicates the first Total electricity cost for the order Indicates the first Total service fee for each order Indicates the first Total charging amount for each order Indicates the first Reason for charging termination for each order; Features of the j-th charging gun node ;in, Indicates the first The type of charging gun for each charging gun. Indicates the first The rated power of each charging gun. Indicates the first Real-time power of each charging gun Indicates the first Charging gun status information for each charging gun. Indicates the first The duration of the charging gun status of each charging gun; Features of the r-th charging pile node ;in, Indicates the first The type of charging station. Indicates the first The rated power of each charging station. Indicates the first Real-time power of each charging station Indicates the first Charging station status information for each charging station Indicates the first The duration of the charging status of each charging pile; Features of the kth charging station node ; Indicates the first The longitude of each charging station Indicates the first The charging station dimension of an individual charging station. Indicates the first The online rate of equipment at each charging station Indicates the first The offline rate of equipment at each charging station Indicates the first The rating of each charging station.
[0009] Preferably, step S4 includes: S41: For each order node By using order-order edges, we find all neighboring order nodes of the current order node, and denote them as the order set. ; S42: Aggregate all order node information pointing to order node i, and capture the attention coefficient of each order to its adjacent order nodes in the multi-source heterogeneous relationship graph. :
[0010] in, Indicates the activation function; Represents an exponential function; Indicates order node characteristics transpose; , Indicates pointing to the first A set of order nodes; Indicates pointing to the first Features of the other p-th order node of the vehicle corresponding to each order; express The number of sets, that is, the total number of orders associated with the vehicle corresponding to the current order; Indicates the pointer to the first Attention coefficients of the p-th order node for the vehicle corresponding to the p-th order; S43: Merge the attention coefficients of all neighboring order nodes and update the features of each order node for all users:
[0011] in, This represents a feedforward neural network; Update the i-th order node features for each user; For the feature of the i-th order node obtained in the previous iteration, in the first iteration... Right now .
[0012] Preferably, step S5 includes: S51: For each charging gun node j, find all order nodes pointing to it through order-gun edge, denoted as set M(j); S52: Calculate the attention coefficient of each order node in the order combination M(j) associated with each charging gun node j. :
[0013] Representing the characteristics of the charging gun node Transpose of a matrix; , Indicates pointing to the first The set of order nodes associated with each charging gun node; Indicates pointing to the first The first charging gun node Individual order node characteristics; express The number of sets; Indicates the pointer to the first The first charging gun Attention coefficient of each order node; S53: Aggregate all historical order information associated with the current charging gun node and update the characteristics of all charging gun nodes:
[0014] in, The updated features of the j-th charging gun node; The feature of the j-th charging gun node obtained in the previous iteration; in the first iteration... Right now .
[0015] Preferably, step S6 includes: S61: For each charging pile node r, find all subordinate charging gun nodes pointing to it through the gun-pile edge, and denote it as the set S(r); S62: Calculate the value of each charging gun node in the set S(r) of charging gun nodes associated with each charging pile node r. Attention coefficient :
[0016] in, Characteristics of charging pile nodes transpose; Indicates pointing to the first A set of charging gun nodes for each charging pile node; Indicates pointing to the first The first charging pile node Node characteristics of a charging gun; express The number of sets; Indicates the pointer to the first The first charging pile Attention coefficient of each charging gun node; S63: Aggregate subordinate charging gun information and update charging pile node characteristics:
[0017] in, The updated characteristics of the r-th charging pile node; The features of the r-th charging pile node obtained in the previous cycle; in the first cycle Right now .
[0018] Preferably, step 7 includes: S71: For each charging station node k, find all its subordinate charging pile nodes pointing to it through the pile-station edge, and denote it as the set D(k); S72: Calculate the attention coefficient of each charging station node k for each charging pile node in the set D(k) of charging pile nodes associated with it. :
[0019] in, Representing the characteristics of charging station nodes transpose; , Indicates pointing to the first A set of charging pile nodes for a charging station; Indicates pointing to the first The characteristics of the f-th charging pile node among charging station nodes; F represents... The number of sets; Indicates the pointer to the first The attention coefficient of the f-th charging pile node of a charging station; S73: Integrates the characteristics of charging pile nodes with all attention coefficients, and updates the first... Characteristic representation of a charging station node:
[0020] in, For the updated number Characteristics of each charging station node; The features of the k-th charging pile node obtained in the previous cycle; in the first cycle Right now .
[0021] Preferably, in step S8, the formula for calculating the recommendation index of the charging station is:
[0022] in, This represents the transpose of the features of the k-th charging station node after the final update; Represents the learnable parameter matrix; Indicates a learnable bias; Indicates the first Recommendation index for each charging station.
[0023] A dynamic recommendation system for charging stations based on multidimensional real-time data is disclosed, which implements a dynamic recommendation method for charging stations based on multidimensional real-time data. The system includes a feature selection and acquisition module, a graph construction module, a charging station node feature acquisition module, and a charging station recommendation module. The feature selection and acquisition module is used to select and acquire charging parameters. The graph construction module is used to construct a multi-source heterogeneous relationship graph of the charging parameters. The charging station node feature acquisition module is used to acquire charging station node features based on order node features, charging gun node features, and charging pile node features. The charging station recommendation module is used to recommend charging stations based on the current user's charging request using a charging station recommendation index.
[0024] The advantages of this invention are: (1) This invention constructs a multi-source heterogeneous relational graph structure for charging scenarios, realizing multi-level semantic fusion of orders, charging guns, charging piles, and charging stations. Existing recommendation methods are mostly based on isolated site scoring or simple collaborative filtering, ignoring the physical hierarchy of charging facilities (charging gun → charging pile → charging station) and user behavior paths (orders associated with vehicles and equipment). This application explicitly constructs a multi-source heterogeneous relational graph containing four types of nodes: orders, charging guns, charging piles, and charging stations, and establishes directed edge relationships based on actual business logic (such as order → order, order → charging gun, charging gun → charging pile, charging pile → charging station), effectively modeling the full-link dependency relationship from micro-equipment status to macro-site operation, providing structured semantic support for recommendations, making the recommendation results consistent with equipment availability, better meeting the true intentions of customers and the synergy of site operation.
[0025] (2) This invention introduces a hierarchical graph attention mechanism to achieve dynamic and interpretable feature aggregation. Unlike traditional GNNs that use mean or summation aggregation, this application designs attention coefficients at each level (order → order, order → charging gun, charging gun → charging pile, charging pile → charging station), and adaptively weights neighbor information based on node feature similarity. For example, the weight of faulty orders on the evaluation of charging gun status is automatically reduced, and the impact of high service fee orders on user preferences is reasonably suppressed. This mechanism not only improves the model's expressive power but also enhances the interpretability of the recommendation results.
[0026] (3) This invention deeply integrates real-time status and historical statistical features, significantly improving the usability of recommendations. This solution jointly encodes dynamic indicators such as the real-time power, current status, and duration of status of the charging gun / pile with behavioral features such as electricity cost, service fee, charging amount, and termination reason in historical orders, enabling the model to accurately identify devices that are "ostensibly idle but actually unusable". Compared with existing systems that rely solely on static idle status or average online rate, this application significantly reduces the failure rate of users "reaching a point where charging is impossible", improving the actual user experience. Attached Figure Description
[0027] Figure 1 This is a flowchart of the method steps of the present invention; Figure 2 This is a schematic diagram of the method structure of the present invention; Figure 3 This is a schematic diagram illustrating the connection between the charging station, charging pile, charging gun, and order in this invention. Detailed Implementation
[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0029] Against the backdrop of the rapid popularization of new energy vehicles, the intelligent operation of charging infrastructure has become a key link in improving user experience and station efficiency. Existing charging station recommendation methods are mainly based on two technical approaches: one is a recommendation system based on location services and static rules, such as prioritizing the nearest charging station, the one with the lowest service fee, or the one with the most available charging guns based on the user's current location; the other is a personalized recommendation model based on collaborative filtering or matrix factorization, which uses the user's historical charging orders (such as charging duration, consumption amount, and station preference) to construct a user-site interaction matrix and predict the user's preference score for unvisited stations. Some improvement schemes attempt to introduce simple contextual features, such as time period, electricity price, or station rating, but these features are usually treated as independent inputs, without establishing a structured relationship between the physical level of equipment (gun-pile-station). More importantly, existing technologies generally treat charging equipment as static available resources, using only historical average online rate or idle status as filtering conditions, lacking fine-grained modeling of the real-time status of charging guns and piles (such as whether they are currently faulty, real-time power output capability) and dynamic reliability (such as recent fault frequency, order anomaly rate). Furthermore, these methods typically employ isomorphic graph or non-graph structure modeling, which fails to effectively capture high-order heterogeneous relationships and information propagation paths among users, orders, and devices. This results in significant deficiencies in recommendation results regarding device availability, matching degree of users' true intent, and site operation synergy.
[0030] Existing charging station recommendation systems generally suffer from the following technical pain points: First, most systems rely solely on static rules (such as nearest or lowest price) or simple collaborative filtering for recommendations, neglecting the real-time operational status of charging equipment (such as charging gun / pile online / offline status, malfunction, and power fluctuations). This often leads users to unavailable equipment, resulting in a negative "arrival equals failure" experience. Second, existing methods fail to effectively integrate multi-source heterogeneous data, including the physical hierarchy of equipment (charging gun-pile-station), users' historical behavioral characteristics, and station operational indicators (such as star rating, failure rate, and online rate), making it difficult to model the complex relationship between user charging preferences and station service capabilities. Therefore, there is an urgent need for an intelligent recommendation method that can deeply integrate real-time status, equipment topology, and operational goals to achieve synergistic optimization of high availability, high satisfaction, and high operational efficiency.
[0031] like Figures 1-3 As shown, the first aspect of this invention proposes a dynamic recommendation method for charging stations based on multi-dimensional real-time data, comprising: S1: Based on the current user's charging request, retrieve all users' historical order data, real-time charging gun data, charging pile data, and charging station data from the database or data stream. Perform preprocessing such as data cleaning and normalization on the data.
[0032] Order data includes fields such as order ID, vehicle ID, charging gun ID, start charging time, end charging time, charging duration, total electricity cost, total service fee, total charging volume, and reason for charging termination (e.g., normal termination, user manual stop, equipment failure interruption). The charging gun data includes the charging gun ID, the charging pile ID to which it belongs, the charging gun type (e.g., AC gun, DC gun), the rated power of the charging gun, the real-time power of the charging gun, the current status of the charging gun (e.g., idle, charging, offline, faulty, scheduled), and the duration of the current status. The charging pile data includes the charging pile ID, the charging station ID to which it belongs, the charging pile type, the rated power of the charging pile, the real-time power of the charging pile, the charging pile status information, and the duration of the charging pile status. The charging station data includes the charging station ID, charging station longitude, charging station latitude and longitude, equipment online rate, equipment failure rate, and user rating.
[0033] A charging station consists of several charging piles, and a charging pile consists of several charging guns.
[0034] S2: Construct a multi-source heterogeneous relationship diagram of related orders, charging guns, charging piles, and charging stations.
[0035] This step is fundamental to the invention, and its goal is to construct a graph structure G=(V,E), where V is the set of nodes and E is the set of edges. For example... Figure 3 As shown, the specific construction process is as follows: (1) Node definition: Order node: Create an order node for each valid historical order or an ongoing order; Charging gun node: Create a charging gun node for each physical charging gun; Charging pile node: Create a charging pile node for each physical charging pile; Charging station node: Create a charging station node for each charging station.
[0036] (2) Edge definition: Order-order edge: For all order nodes under the same vehicle, establish a bidirectional edge between them. This edge is used to capture the charging behavior patterns and preferences of the same user.
[0037] Order-Charging Gun Edge: Based on the charging gun ID in the order data, a directed edge is established from the order node to its corresponding charging gun node. This edge associates user behavior with a specific device.
[0038] Charger-Charging Pile Edge: Based on the charging pile ID in the charging gun data, a directed edge is established from the charging gun node to its corresponding charging pile node. This edge establishes the physical hierarchy of the devices.
[0039] Charging Pile-Station Edge: Based on the charging station ID in the charging pile data, a directed edge is established from the charging pile node to its corresponding charging station node. This edge completes the aggregation path from the micro-device to the macro-station.
[0040] The order data records which charging gun was used for the order; the charging gun data includes which charging pile the charging gun belongs to; the charging pile data includes which charging station the charging pile belongs to; therefore, an edge is obtained: order -> charging gun -> charging pile -> charging station.
[0041] S3: Construct an initial feature vector for each node in the multi-source heterogeneous relationship graph. These features are the foundation for the model's learning and inference.
[0042] Order node characteristics ;in, Indicates the first The charging time for each order. Indicates the first Total electricity cost for the order Indicates the first Total service fee for each order Indicates the first Total charging amount for each order Indicates the first Reason for charging termination for each order; Charging gun node characteristics ;in, Indicates the first The type of charging gun for each charging gun. Indicates the first The rated power of each charging gun. Indicates the first Real-time power of each charging gun Indicates the first Charging gun status information for each charging gun. Indicates the first The duration of the charging gun status of each charging gun; Charging pile node characteristics ;in, Indicates the first The type of charging station. Indicates the first The rated power of each charging station. Indicates the first Real-time power of each charging station Indicates the first Charging station status information for each charging station Indicates the first The duration of the charging status of each charging pile; Charging station node characteristics ; Indicates the first The longitude of each charging station Indicates the first The charging station dimension of an individual charging station. Indicates the first The online rate of equipment at each charging station Indicates the first The offline rate of equipment at each charging station Indicates the first The rating of each charging station.
[0043] S4: Update the order nodes for all users based on adjacent order nodes. This step aims to enrich the current user's order characteristics by utilizing the historical charging records of the same user. This includes: S41: For each order node i, find all neighboring order nodes of the current order node through order-order edges, and denote them as the order set. ; S42: Aggregate all order node information pointing to order node i, and capture the attention coefficient of each order to its adjacent order nodes in the multi-source heterogeneous relationship graph. .
[0044] ; in, This represents the activation function. Represents an exponential function. Indicates order node characteristics transpose, , Indicates pointing to the first A set of order nodes for each order node. Indicates pointing to the first Features of the other p-th order node of the vehicle corresponding to each order. express The number of items in the set, that is, the total number of orders associated with the vehicle corresponding to the current order. Indicates the pointer to the first The attention coefficients of the p-th order node for the vehicle corresponding to the current order. The attention mechanism allows the model to focus on historical orders that are more similar to and have more reference value than the current order.
[0045] Find its "neighbors": that is, all other order nodes connected through the "same order edge", denoted as set N(i).
[0046] The data of the i-th order node will contain which vehicle the order belongs to. This vehicle contains multiple orders, and N(i) actually represents the set of other order nodes of the vehicle corresponding to the i-th order.
[0047] S43: Merge the attention coefficients of all neighboring order nodes and update the feature representation of each order node for all users.
[0048] ; in, This represents a feedforward neural network; Update the i-th order node features for each user; For the feature of the i-th order node obtained in the previous iteration, in the first iteration... Right now .
[0049] S5: Update the characteristics of all charging gun nodes based on each user's order node information. This step injects user behavior information into the device representation. This includes: S51: For each charging gun node j, find all the order nodes pointing to it through "order-gun edge", denoted as set M(j); these orders represent the service history of the charging gun.
[0050] S52: Calculate the order node in M(j) for each charging gun node j and its associated order combination M(j). Attention coefficient For example, the attention weight of an order that has just experienced a failure may be automatically reduced.
[0051] ; This represents the activation function. Represents an exponential function. Representing the characteristics of the charging gun node Matrix transpose , Indicates pointing to the first The set of order nodes associated with each charging gun node. Indicates pointing to the first The first charging gun node Individual order node characteristics express The number of sets Indicates the pointer to the first The first charging gun Attention coefficient of each order node; S53: Aggregate all historical order information associated with the current charging gun node and update the characteristics of all charging gun nodes. ;
[0052] in, This represents a feedforward neural network; The updated features of the j-th charging gun node; The feature of the j-th charging gun node obtained in the previous iteration; in the first iteration... Right now .
[0053] S6: Update the characteristics of all charging pile nodes based on the characteristics of the charging gun node. This step realizes the upward aggregation of the device status.
[0054] S61: For each charging pile node r, find all subordinate charging gun nodes pointing to it through "gun-pile edge", denoted as set S(r); S62: Calculate the value of each charging gun node in the set S(r) of charging gun nodes associated with each charging pile node r. Attention coefficient For example, a high-power, stable charging gun will receive more attention than a faulty charging gun.
[0055] ; This represents the activation function. Represents an exponential function. Characteristics of charging pile nodes transpose, , Indicates pointing to the first A set of charging gun nodes for each charging pile node. Indicates pointing to the first The first charging pile node Features of each charging gun node express The number of sets Indicates the pointer to the first The first charging pile Attention coefficient of each charging gun node.
[0056] S63: Aggregate information on subordinate charging guns and update charging pile node characteristics. ; This represents a feedforward neural network; The updated characteristics of the r-th charging pile node; The features of the r-th charging pile node obtained in the previous cycle; in the first cycle Right now .
[0057] S7: Update the features of all charging station nodes based on the features of all charging pile nodes. This step is the key to the final decision. The overall recommended value of a station is determined by the status of all its subordinate charging piles.
[0058] S71: For each charging station node k, find all its subordinate charging pile nodes pointing to it through "pile-station edge", denoted as set D(k); S72: Calculate the Z of each charging pile node in the set D(k) of charging pile nodes associated with each charging station node k. j Attention coefficient .
[0059] ; This represents the activation function. Represents an exponential function. Representing the characteristics of charging station nodes transpose, , Indicates pointing to the first A set of charging pile nodes for a charging station. Indicates pointing to the first The f-th charging pile node of a charging station node, where F represents The number of sets Indicates the pointer to the first The first charging station Attention coefficient of each charging station node; S73: Integrates the characteristics of charging pile nodes with all attention coefficients, and updates the first... Feature representation of a charging station node ; in, This represents a feedforward neural network; For the updated number Characteristics of each charging station node; The features of the k-th charging pile node obtained in the previous cycle; in the first cycle Right now .
[0060] S8: Iterate through steps S4-S7 L times to obtain the final features of all charging station nodes. Calculate the recommendation index for all charging stations.
[0061] ; in, This represents the transpose of the features of the k-th charging station node after the final update. This represents the learnable parameter matrix. This represents a learnable bias. Indicates the first Recommendation index for each charging station.
[0062] Step 9: Recommend charging stations to users in descending order of their recommendation index.
[0063] This invention employs a method for constructing a multi-source heterogeneous relationship graph for charging pile recommendations. For the first time, it explicitly models four types of entities in the charging business—orders, charging guns, charging piles, and charging stations—as heterogeneous graph nodes. Based on real business logic (such as "orders are associated with vehicles and charging guns," "charging guns belong to charging piles," and "charging piles belong to charging stations"), directed edge relationships are constructed, forming a graph structure with both physical and semantic meaning. This overcomes the limitation of traditional recommendation systems that treat charging stations as isolated entities.
[0064] Simultaneously, a hierarchical graph attention-driven dynamic feature aggregation mechanism is adopted. Learnable attention coefficients are designed at each level of the heterogeneous graph (order → order, order → charging gun, charging gun → charging pile, charging pile → charging station), enabling the model to adaptively weight and aggregate neighbor information based on the semantic similarity and business rationality of node features, thereby achieving automatic suppression of noisy data and enhancement of key signals.
[0065] A multi-dimensional node feature system that integrates real-time status and historical behavior. It systematically integrates dynamic real-time features (such as the current status of the gun / pile, real-time power, and duration of status) with historical statistical features (such as order electricity cost, service fee, charging amount, and termination reason) to construct a fine-grained and highly discriminative node representation, significantly improving the modeling accuracy of equipment availability and user preferences.
[0066] A dynamic recommendation system for charging stations based on multidimensional real-time data, to implement the aforementioned dynamic recommendation method for charging stations based on multidimensional real-time data, includes a feature selection and acquisition module, a graph construction module, a charging station node feature acquisition module, and a charging station recommendation module; the feature selection and acquisition module is used to select and acquire charging parameters; the graph construction module is used to construct a multi-source heterogeneous relationship graph of charging parameters; the charging station node feature acquisition module is used to acquire charging station node features based on order node features, charging gun node features, and charging pile node features; the charging station recommendation module is used to recommend charging stations based on the current user's charging request based on the charging station recommendation index.
[0067] Of course, those skilled in the art will recognize that the present invention is not limited to the details of the exemplary embodiments described above, but also includes the same or similar structures that can be implemented in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.
[0068] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.
[0069] The technologies, shapes, and structures not described in detail in this invention are all known technologies.
Claims
1. A dynamic recommendation method for charging stations based on multi-dimensional real-time data, characterized in that, include: S1: Based on the user's charging request, obtain the charging data of all users, including historical order data, real-time charging gun data, charging pile data, and charging station data; S2: Construct a multi-source heterogeneous relationship graph G=(V,E) based on the charging data of all users, linking orders, charging guns, charging piles, and charging stations; Where V is the set of nodes and E is the set of edges; The nodes include: order nodes, charging gun nodes, charging pile nodes, and charging station nodes; the edges include: order-order edges, order-charging gun edges, charging gun-charging pile edges, and charging pile-charging station edges. S3: Construct the corresponding initial feature vector for each node in the multi-source heterogeneous relationship graph; S4: Update the characteristics of each order node for all users based on adjacent order nodes; S5: Update the characteristics of all charging gun nodes based on the order node characteristics of each user; S6: Update the features of all charging pile nodes based on the features of the charging gun nodes; S7: Update the features of all charging station nodes based on the features of all charging pile nodes; S8: Iterate through steps S4-S7 L times to obtain the final updated features of all charging station nodes and calculate the recommendation index of all charging stations. S9: Recommend charging stations to users in descending order of their recommendation index.
2. The charging station dynamic recommendation method based on multi-dimensional real-time data as described in claim 1, characterized in that, Order data includes order ID, vehicle ID, charging gun ID, start charging time, end charging time, charging duration, total electricity cost, total service fee, total charging volume, and reason for ending charging. The charging gun data includes the charging gun ID, the charging pile ID to which it belongs, the charging gun type, the rated power of the charging gun, the real-time power of the charging gun, the current status of the charging gun, and the duration of the current status. Charging pile data includes charging pile ID, charging station ID, charging pile type, rated power of charging pile, real-time power of charging pile, charging pile status information, and duration of charging pile status. The charging station data includes the charging station ID, charging station longitude, charging station latitude and longitude, equipment online rate, equipment failure rate, and user rating. A charging station consists of several charging piles, and a charging pile consists of several charging guns.
3. The charging station dynamic recommendation method based on multi-dimensional real-time data as described in claim 1, characterized in that, The specific construction process of the multi-source heterogeneous relationship graph G of the associated order-charging gun-charging pile-charging station is as follows: Node definition: Create an order node for each valid historical order or an ongoing order; create a charging gun node for each physical charging gun; create a charging pile node for each physical charging pile. For each charging station, create one charging station node; Edge definition: Order-order edge: For all order nodes under the same vehicle, establish a bidirectional edge between them; Order-gun edge: Based on the charging gun ID in the order data, establish a directed edge from the order node to its corresponding charging gun node; Gun-pile edge: Based on the charging pile ID in the charging gun data, establish a directed edge from the charging gun node to its corresponding charging pile node. Charging pile-station edge: Based on the charging station ID in the charging pile data, a directed edge is established from the charging pile node to its corresponding charging station node.
4. The charging station dynamic recommendation method based on multi-dimensional real-time data as described in claim 1, characterized in that, In step S3, the characteristic of the i-th order node is ;in, Indicates the first The charging time for each order. Indicates the first Total electricity cost for the order Indicates the first Total service fee for each order Indicates the first Total charging amount for each order Indicates the first Reason for charging termination for each order; Features of the j-th charging gun node ;in, Indicates the first The type of charging gun for each charging gun. Indicates the first The rated power of each charging gun. Indicates the first Real-time power of each charging gun Indicates the first Charging gun status information for each charging gun. Indicates the first The duration of the charging gun status of each charging gun; Features of the r-th charging pile node ;in, Indicates the first The type of charging station. Indicates the first The rated power of each charging station. Indicates the first Real-time power of each charging station Indicates the first Charging station status information for each charging station Indicates the first The duration of the charging status of each charging pile; Features of the kth charging station node ; Indicates the first The longitude of each charging station Indicates the first The charging station dimension of an individual charging station. Indicates the first The online rate of equipment at each charging station Indicates the first The offline rate of equipment at each charging station Indicates the first The rating of each charging station.
5. The charging station dynamic recommendation method based on multi-dimensional real-time data as described in claim 4, characterized in that, Step S4 includes: S41: For each order node By using order-order edges, we find all neighboring order nodes of the current order node, and denote them as the order set. ; S42: Aggregate all order node information pointing to order node i, and capture the attention coefficient of each order to its adjacent order nodes in the multi-source heterogeneous relationship graph. : in, Indicates the activation function; Represents an exponential function; Indicates order node characteristics transpose; , Indicates pointing to the first A set of order nodes; Indicates pointing to the first Features of the other p-th order node of the vehicle corresponding to each order; express The number of sets, that is, the total number of orders associated with the vehicle corresponding to the current order; Indicates the pointer to the first Attention coefficients of the p-th order node for the vehicle corresponding to the p-th order; S43: Merge the attention coefficients of all neighboring order nodes and update the features of each order node for all users: in, This represents a feedforward neural network; Update the i-th order node features for each user; For the feature of the i-th order node obtained in the previous iteration, in the first iteration... Right now .
6. The charging station dynamic recommendation method based on multi-dimensional real-time data as described in claim 5, characterized in that, Step S5 includes: S51: For each charging gun node j, find all order nodes pointing to it through order-gun edge, denoted as set M(j); S52: Calculate the attention coefficient of each order node in the order combination M(j) associated with each charging gun node j. : Representing the characteristics of the charging gun node Transpose of a matrix; , Indicates pointing to the first The set of order nodes associated with each charging gun node; Indicates pointing to the first The first charging gun node Individual order node characteristics; express The number of sets; Indicates the pointer to the first The first charging gun Attention coefficient of each order node; S53: Aggregate all historical order information associated with the current charging gun node and update the characteristics of all charging gun nodes: in, The updated features of the j-th charging gun node; The feature of the j-th charging gun node obtained in the previous iteration; in the first iteration... Right now .
7. The charging station dynamic recommendation method based on multi-dimensional real-time data as described in claim 6, characterized in that, Step S6 includes: S61: For each charging pile node r, find all subordinate charging gun nodes pointing to it through the gun-pile edge, and denote it as the set S(r); S62: Calculate the charging gun node in the set S(r) of charging gun nodes associated with each charging pile node r. Attention coefficient : in, Characteristics of charging pile nodes transpose; Indicates pointing to the first A set of charging gun nodes for each charging pile node; Indicates pointing to the first The first charging pile node Node characteristics of a charging gun; express The number of sets; Indicates the pointer to the first The first charging pile Attention coefficient of each charging gun node; S63: Aggregate subordinate charging gun information and update charging pile node characteristics: in, The updated characteristics of the r-th charging pile node; The features of the r-th charging pile node obtained in the previous cycle; in the first cycle Right now .
8. The charging station dynamic recommendation method based on multi-dimensional real-time data as described in claim 7, characterized in that, Step 7 includes: S71: For each charging station node k, find all its subordinate charging pile nodes pointing to it through the pile-station edge, and denote it as the set D(k); S72: Calculate the attention coefficient of each charging station node k for each charging pile node in the set D(k) of charging pile nodes associated with it. : in, Representing the characteristics of charging station nodes transpose; , Indicates pointing to the first A set of charging pile nodes for a charging station; Indicates pointing to the first The characteristics of the f-th charging pile node among charging station nodes; F represents... The number of sets; Indicates the pointer to the first The attention coefficient of the f-th charging pile node of a charging station; S73: Integrates the characteristics of charging pile nodes with all attention coefficients, and updates the first... Characteristic representation of a charging station node: in, For the updated number Characteristics of each charging station node; The features of the k-th charging pile node obtained in the previous cycle; in the first cycle Right now .
9. The charging station dynamic recommendation method based on multi-dimensional real-time data as described in claim 8, characterized in that, In step S8, the formula for calculating the recommendation index of the charging station is: in, This represents the transpose of the features of the k-th charging station node after the final update; Represents the learnable parameter matrix; Indicates a learnable bias; Indicates the first Recommendation index for each charging station.
10. A dynamic recommendation system for charging stations based on multi-dimensional real-time data, characterized in that, To achieve the dynamic recommendation method for charging stations based on multi-dimensional real-time data as described in any one of claims 1-9, It includes a feature selection and acquisition module, a graph construction module, a charging station node feature acquisition module, and a charging station recommendation module; The feature selection and acquisition module is used to select and acquire charging parameters; the graph construction module is used to construct a multi-source heterogeneous relationship graph of charging parameters; the charging station node feature acquisition module is used to acquire charging station node features based on order node features, charging gun node features, and charging pile node features. The charging station recommendation module is used to recommend charging stations based on the current user's charging request, according to the charging station recommendation index.