Delivery processing method and system for vehicle rental business

Abstract

The present invention provides a delivery processing method and system for vehicle rental business, which relates to the field of vehicle rental technology, including receiving a vehicle rental request, and combining with a pre-established vehicle database, outputting a set of candidate vehicles whose matching degree with the vehicle rental request is higher than a preset threshold, obtaining the historical rental frequency, historical rental score and current vehicle position of each vehicle in the candidate vehicle set, calculating the comprehensive matching degree between each candidate vehicle and the vehicle rental request, determining the candidate vehicle with the highest comprehensive matching degree as a rental vehicle, and generating an intelligent vehicle delivery task; sending the intelligent vehicle delivery task to an intelligent vehicle management device corresponding to a vehicle identification, the intelligent vehicle management device returns vehicle delivery completion information after completing the vehicle delivery, automatically comparing and analyzing the changes in vehicle condition before and after the use of the vehicle, and automatically generating a rental bill based on the comparison result using an intelligent contract based on blockchain technology.

Description

Technical Field

[0001] The present invention relates to vehicle leasing technology, and in particular to a delivery processing method and system for vehicle leasing business. Background Art

[0002] With the rapid development of the sharing economy, the vehicle rental business is growing. The traditional vehicle rental process usually requires manual processing of vehicle rental requests, vehicle scheduling, vehicle condition inspection, bill generation and other links, which has problems such as low efficiency, many errors, and lack of transparency. In addition, under the traditional vehicle rental model, it is difficult to match the supply and demand of vehicles and optimize the scheduling, the vehicle utilization rate is not high, and the user rental experience is poor. Therefore, there is an urgent need for an intelligent and automated vehicle rental delivery processing method to improve the efficiency and quality of vehicle rental, reduce operating costs, and enhance user experience.

[0003] However, the existing vehicle leasing optimization solutions still have the following shortcomings: First, the intelligent level of vehicle supply and demand matching is not high, and it is difficult to accurately recommend vehicles that meet user needs; second, the automation level of vehicle delivery and acceptance is low, and manual operation and confirmation are still required; third, the efficiency and credibility of vehicle condition detection and bill generation are not high, and there are risks of disputes and cheating. Therefore, it is necessary to further innovate the vehicle leasing delivery processing model and comprehensively improve the intelligence, automation and credibility of the vehicle leasing business. Summary of the invention

[0004] The embodiment of the present invention provides a delivery processing method and system for vehicle rental business, which can solve the problems in the prior art.

[0005] According to a first aspect of the embodiments of the present invention,

[0006] A delivery processing method for a vehicle leasing business is provided, comprising:

[0007] Receive a vehicle rental request, input the vehicle rental request into a pre-built vehicle recommendation model, automatically extract key information from the vehicle rental request, and combine with a pre-built vehicle database to output a set of candidate vehicles whose matching degree with the vehicle rental request is higher than a preset threshold, wherein the vehicle database stores static attribute information and dynamic status information of multiple rentable vehicles, wherein the static attribute information includes vehicle information and in-vehicle configuration parameters, and the dynamic status information includes the current location of the vehicle and the available rental period;

[0008] Obtain the historical rental frequency, historical rental score and current vehicle location of each vehicle in the candidate vehicle set, calculate the comprehensive matching degree between each candidate vehicle and the vehicle rental request based on the fuzzy comprehensive evaluation method and the pick-up time and pick-up location in the vehicle rental request, determine the candidate vehicle with the highest comprehensive matching degree as the rental vehicle, and generate an intelligent vehicle delivery task;

[0009] The smart vehicle delivery task is sent to the smart vehicle management device corresponding to the vehicle identification. The smart vehicle management device returns the vehicle delivery completion information after completing the vehicle delivery. If the vehicle return information sent by the smart vehicle management device is received within the preset time after the return time specified in the vehicle rental request, the vehicle rental is determined to be completed. According to the vehicle condition information at the time of delivery in the vehicle delivery completion information and the vehicle condition information at the time of return in the vehicle return information, the changes in the vehicle condition before and after the use of the vehicle are automatically compared and analyzed, and the rental bill is automatically generated according to the comparison results using the smart contract based on blockchain technology.

[0010] In an optional embodiment,

[0011] Receive a vehicle rental request, input the vehicle rental request into a pre-built vehicle recommendation model, automatically extract key information from the vehicle rental request, and combine with a pre-built vehicle database to output a set of candidate vehicles that match the vehicle rental request with a degree higher than a preset threshold, including:

[0012] The ontology modeling method is used to define the core concepts and relationships in the vehicle field, and the vehicle entities and corresponding attributes are extracted from structured and unstructured data sources. The entity linking and knowledge alignment technology are used to fuse vehicle knowledge to form a fused vehicle knowledge graph.

[0013] Obtaining and preprocessing vehicle review data and vehicle image data, inputting the preprocessed vehicle review data and vehicle image data into a pre-trained convolutional neural network to extract vehicle review text features and image visual features, respectively;

[0014] Using a pre-built knowledge representation learning model, the vehicle entities and relationships in the fused vehicle knowledge graph are mapped to a low-dimensional semantic space to obtain a distributed representation of vehicle entities and relationships.

[0015] Design a knowledge-guided attention mechanism, use the distributed representation of vehicle entities and relations, calculate the importance weights of different words and image regions in vehicle review text features and image visual features, and build a multimodal graph neural network model. Use the graph convolutional network to learn the embedded representation of vehicle entities and relations, and fuse the vehicle review text features and image visual features weighted by the attention mechanism with the vehicle entity embedded representation to obtain a fused multimodal vehicle representation.

[0016] The fused multimodal vehicle representation is used as the input feature of the vehicle recommendation model, and the vehicle recommendation model is trained in combination with the user's historical interaction behavior data to obtain a trained vehicle recommendation model;

[0017] For the received vehicle rental request, key information in the vehicle rental request is extracted, and the key information includes the user characteristics of the requester and the vehicle demand characteristics. The user characteristics, vehicle demand characteristics and the fused multimodal vehicle representation are input into the trained vehicle recommendation model. At the same time, combined with the pre-established vehicle database, a set of candidate vehicles whose matching degree with the vehicle rental request is higher than a preset threshold is output.

[0018] In an optional embodiment,

[0019] The fused multimodal vehicle representation is used as the input feature of the vehicle recommendation model. The vehicle recommendation model is trained in combination with the user's historical interaction behavior data. The trained vehicle recommendation model includes:

[0020] Collect historical interaction behavior data of users browsing vehicle details pages, clicking on vehicle attributes, and inquiring about vehicle prices on the vehicle recommendation platform, and build a user-vehicle interaction behavior sequence based on the historical interaction behavior data;

[0021] Inputting the user-vehicle interaction behavior sequence into a bidirectional recurrent neural network integrated with a vehicle attribute attention mechanism in chronological order, dynamically modeling the user's vehicle preference through the hidden state of each time step in the bidirectional recurrent neural network, and obtaining a cumulative representation of the user's vehicle preference at the current moment;

[0022] The vehicle attribute attention mechanism is used to calculate the attention weights of different vehicle attributes in each time step, and the vehicle attribute attention weights are weighted and summed with the vehicle attribute embedding vector to obtain the vehicle attribute representation vector of the current time step; the vehicle attribute representation vector of the current time step is concatenated with the hidden state of the bidirectional recurrent neural network and input into the feedforward neural network to obtain the comprehensive interest representation of the vehicle at the current time step, and the comprehensive interest representation of the vehicle at each time step of the user-vehicle interaction behavior sequence is aggregated through the attention mechanism to obtain the user's final vehicle interest representation;

[0023] A paired learning paradigm is used to negatively sample the user-vehicle interaction behavior sequence and construct vehicle triplets. The interaction time difference between the positive and negative sample vehicles in the vehicle triplets is combined with the similarity difference of the multimodal vehicle representations of the positive and negative sample vehicles, and a time-series-aware multimodal contrast loss function is designed.

[0024] The vehicle click-through rate prediction task is introduced. Based on the multimodal vehicle representation of positive and negative sample vehicles and the user's final vehicle interest representation, the cross entropy function is used as the click-through rate prediction loss function. The multimodal contrast loss function and the click-through rate prediction loss function are weighted summed as the multi-task joint training objective function of the vehicle recommendation model. The parameters of the vehicle recommendation model are optimized and solved to obtain a trained vehicle recommendation model.

[0025] In an optional embodiment,

[0026] The multi-task joint training objective function of the vehicle recommendation model is the weighted sum of the multi-modal contrast loss function and the click-through rate prediction loss function, which includes:

[0027] The multimodal contrast loss function calculation formula is as follows:

[0028] ;

[0029] Among them, L T represents the multimodal contrast loss function, a represents the user, b represents the positive sample vehicle, and b ' represents the negative sample vehicle, D represents the set of all triples in the training set, σ represents the Sigmoid function, γ represents the temporal weight factor, represents the time when user a interacts with the positive sample vehicle b, represents the virtual interaction time between user a and negative sample vehicle b, a a represents the final interest representation vector of user a, b x represents the multimodal representation vector of the positive sample vehicle, b ' x represents the multimodal representation vector of the negative sample vehicle, λ represents the L2 regularization coefficient, and θ represents the model parameter;

[0030] The calculation formula of the click rate prediction loss function is as follows:

[0031] ;

[0032] Among them, LC represents the click rate prediction loss function, c represents the click state, and C represents the set of click sample pairs. represents the probability predicted by the model that user a clicks on vehicle b;

[0033] The calculation formula of the multi-task joint training objective function is as follows:

[0034] L=L T +ηL C ;

[0035] Among them, L represents the multi-task joint training objective function, and η represents the weight coefficient of the click rate prediction task.

[0036] In an optional embodiment,

[0037] Based on the fuzzy comprehensive evaluation method, combined with the pick-up time and pick-up location in the vehicle rental request, the comprehensive matching degree between each candidate vehicle and the vehicle rental request is calculated, including:

[0038] Construct a multi-objective optimization problem for vehicle matching, with comprehensive matching degree, user satisfaction and vehicle utilization as optimization targets;

[0039] Designing a coding scheme for fuzzy rules, using a binary tree to represent the grammatical structure of fuzzy rules, wherein non-leaf nodes of the binary tree represent logical connectives between fuzzy propositions, and leaf nodes represent fuzzy propositions composed of input variables and their membership function parameters, and obtaining a fixed-length real-valued vector representing the fuzzy rules by traversing the binary tree;

[0040] The real-valued vector is optimized by using an evolutionary algorithm to obtain an optimal solution set of fuzzy rule combinations, and an optimal fuzzy rule combination is selected from the optimal solution set as an evaluation rule for vehicle matching. Based on the obtained evaluation rule and in combination with the pick-up time and pick-up location in the vehicle rental request, a comprehensive matching degree between each candidate vehicle and the vehicle rental request is calculated.

[0041] In an optional embodiment,

[0042] The calculation formula for calculating the comprehensive matching degree between each candidate vehicle and the vehicle rental request is as follows:

[0043] ;

[0044] Among them, M(V, R) represents the comprehensive matching degree between the candidate vehicle V and the vehicle rental request R, m represents the number of features in the fuzzy rule, Represents the feature f in the fuzzy rule i The weight coefficient of (V, R) represents the feature f i The membership degree between the candidate vehicle V and the vehicle rental request R, n represents the number of constraints, Represents the constraint d in the fuzzy rule j The weight coefficient of (V, R) represents the constraint d j The degree of membership between the candidate vehicle V and the vehicle rental request R.

[0045] In an optional embodiment,

[0046] Based on the vehicle condition information at the time of delivery in the vehicle delivery completion information and the vehicle condition information at the time of return in the vehicle return information, the changes in the vehicle condition before and after use are automatically compared and analyzed, and the rental bill is automatically generated based on the comparison results using the smart contract based on blockchain technology, including:

[0047] Build a vehicle condition data collection and chain-up mechanism. When the vehicle is delivered or returned, the intelligent vehicle management device automatically collects vehicle condition information, including vehicle appearance images, vehicle interior environment images, and vehicle sensor data. After preprocessing the collected vehicle condition information, the preprocessed vehicle condition information is uploaded to the blockchain in real time through the Internet of Vehicles communication module.

[0048] The blockchain node verifies and packages the received pre-processed vehicle condition information according to the pre-defined data format and identity authentication rules, writes the hash value and timestamp of the packaged vehicle condition information into the block, generates a new block and broadcasts it to the entire network, completing the on-chain storage of the vehicle condition data;

[0049] Obtain the vehicle delivery status information and return status information stored on the chain and input them into the pre-trained vehicle condition comparison analysis model, output the vehicle condition detection results under the two states and automatically generate a vehicle condition detection report;

[0050] According to the generated vehicle condition inspection report, the rental agreement and billing rules are encoded into a smart contract, and the smart contract is deployed on the blockchain. When the vehicle is returned, the smart contract automatically obtains the vehicle condition inspection report, determines whether the change in vehicle condition exceeds the normal use range, and calculates the compensation amount for the deterioration of the vehicle condition according to predefined rules. At the same time, the basic rental fee is calculated based on the actual duration and mileage of the vehicle rental. The compensation amount for the deterioration of the vehicle condition and the basic rental fee are summarized to automatically generate a rental bill, and the generated rental bill is written to the blockchain to form an unalterable bill record. Users and vehicle managers can view detailed electronic rental bill information through the blockchain browser.

[0051] According to a second aspect of the embodiments of the present invention,

[0052] A delivery processing system for vehicle leasing business is provided, comprising:

[0053] The first unit is used to receive a vehicle rental request, input the vehicle rental request into a pre-built vehicle recommendation model, automatically extract key information in the vehicle rental request, and combine with a pre-built vehicle database to output a set of candidate vehicles whose matching degree with the vehicle rental request is higher than a preset threshold, wherein the vehicle database stores static attribute information and dynamic status information of multiple rentable vehicles, wherein the static attribute information includes vehicle information and in-vehicle configuration parameters, and the dynamic status information includes a current location of the vehicle and a rental period;

[0054] The second unit is used to obtain the historical rental frequency, historical rental score and current vehicle location of each vehicle in the candidate vehicle set, and based on the fuzzy comprehensive evaluation method, combined with the pick-up time and pick-up location in the vehicle rental request, calculate the comprehensive matching degree between each candidate vehicle and the vehicle rental request, determine the candidate vehicle with the highest comprehensive matching degree as the rental vehicle, and generate an intelligent vehicle delivery task;

[0055] The third unit is used to send the smart vehicle delivery task to the smart vehicle management device corresponding to the vehicle identification. The smart vehicle management device returns the vehicle delivery completion information after completing the vehicle delivery. If the vehicle return information sent by the smart vehicle management device is received within a preset time after the return time specified in the vehicle rental request, the vehicle rental is determined to be completed. According to the vehicle condition information at the time of delivery in the vehicle delivery completion information and the vehicle condition information at the time of return in the vehicle return information, the changes in the vehicle condition before and after the use of the vehicle are automatically compared and analyzed, and the rental bill is automatically generated according to the comparison result using the smart contract based on blockchain technology.

[0056] According to a third aspect of the embodiments of the present invention,

[0057] An electronic device is provided, comprising:

[0058] processor;

[0059] a memory for storing processor-executable instructions;

[0060] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0061] A fourth aspect of the embodiments of the present invention is:

[0062] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0063] In this embodiment, by adopting a collaborative filtering algorithm, the static attributes and dynamic status of the vehicle and the historical behavior data of the user are integrated to achieve personalized and real-time vehicle recommendations. The intelligent vehicle management device uses on-board sensors and communication modules to realize the automatic execution of delivery tasks and the real-time upload of vehicle condition information. Vehicle condition detection is based on a deep learning algorithm, which analyzes vehicle images and sensor data to accurately identify changes in vehicle condition. The rental bill generation uses a blockchain smart contract to automatically trigger bill calculation and chain storage based on the vehicle condition detection results and the rental agreement to ensure that the bill is authentic and reliable. It breaks through the limitations of the traditional vehicle rental model and realizes the digitization, intelligence and automation of the entire process of vehicle rental delivery, greatly optimizes the vehicle rental business process, improves the efficiency of vehicle resource utilization and user rental experience, reduces operating management costs and risks, and is of great significance to the innovative development of the vehicle rental industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 A schematic diagram of a flow chart of a delivery processing method for a vehicle rental business according to an embodiment of the present invention;

[0065] Figure 2 It is a schematic diagram of the structure of a delivery processing system for vehicle rental business according to an embodiment of the present invention. DETAILED DESCRIPTION

[0066] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0067] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0068] Figure 1 FIG. 1 is a flow chart of a delivery processing method for vehicle rental business according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0069] S101. Receive a vehicle rental request, input the vehicle rental request into a pre-built vehicle recommendation model, automatically extract key information in the vehicle rental request, and combine with a pre-built vehicle database to output a set of candidate vehicles whose matching degree with the vehicle rental request is higher than a preset threshold, wherein the vehicle database stores static attribute information and dynamic status information of multiple rentable vehicles, the static attribute information includes vehicle information and in-vehicle configuration parameters, and the dynamic status information includes the current location of the vehicle and the available rental period.

[0070] In an optional embodiment,

[0071] Receive a vehicle rental request, input the vehicle rental request into a pre-built vehicle recommendation model, automatically extract key information from the vehicle rental request, and combine with a pre-built vehicle database to output a set of candidate vehicles that match the vehicle rental request with a degree higher than a preset threshold, including:

[0072] The ontology modeling method is used to define the core concepts and relationships in the vehicle field, and the vehicle entities and corresponding attributes are extracted from structured and unstructured data sources. The vehicle knowledge is integrated using entity linking and knowledge alignment technology to form a fused vehicle knowledge graph.

[0073] Obtaining vehicle review data and vehicle image data and performing preprocessing, inputting the preprocessed vehicle review data and vehicle image data into a pre-trained convolutional neural network to extract vehicle review text features and image visual features, respectively;

[0074] Using a pre-built knowledge representation learning model, the vehicle entities and relationships in the fused vehicle knowledge graph are mapped to a low-dimensional semantic space to obtain a distributed representation of vehicle entities and relationships.

[0075] Design a knowledge-guided attention mechanism, use the distributed representation of vehicle entities and relations, calculate the importance weights of different words and image regions in vehicle review text features and image visual features, and build a multimodal graph neural network model. Use the graph convolutional network to learn the embedded representation of vehicle entities and relations, and fuse the vehicle review text features and image visual features weighted by the attention mechanism with the vehicle entity embedded representation to obtain a fused multimodal vehicle representation.

[0076] The fused multimodal vehicle representation is used as the input feature of the vehicle recommendation model, and the vehicle recommendation model is trained in combination with the user's historical interaction behavior data to obtain a trained vehicle recommendation model;

[0077] For the received vehicle rental request, key information in the vehicle rental request is extracted, and the key information includes the user characteristics of the requester and the vehicle demand characteristics. The user characteristics, vehicle demand characteristics and the fused multimodal vehicle representation are input into the trained vehicle recommendation model. At the same time, combined with the pre-established vehicle database, a set of candidate vehicles whose matching degree with the vehicle rental request is higher than a preset threshold is output.

[0078] For example, first, determine some top-level concepts in the vehicle field, such as car brands, car models, car configurations, usage scenarios, etc. Then, further refine the sub-concepts of each top-level concept. For example, car brands can be subdivided into joint venture brands, domestic brands, imported brands, etc.; car models can be subdivided into sedans, SUVs, MPVs, sports cars, etc. Next, define the key attributes of each concept. For example, car models have attributes such as manufacturers, levels, sizes, engines, and gearboxes. Finally, define the semantic relationships between concepts. For example, there is a belonging relationship between car brands and car models, and there is a containment relationship between car models and car configurations.

[0079] Through ontology modeling, a conceptual hierarchy system and relationship network of vehicle domain knowledge are formed. After defining the vehicle ontology, entities, attributes and relationship knowledge in the vehicle domain are extracted from structured and unstructured data sources. For structured data sources, such as the model parameter comparison page of the automobile vertical website, structured vehicle knowledge can be directly extracted based on the page structure and predefined extraction rules. For unstructured data sources, such as vehicle news reports, vehicle forum posts, etc., natural language processing technology is used for knowledge extraction. First, the named entity recognition model is used to identify entity mentions such as vehicle brands, models, configurations, etc. that appear in the text. Then, semantic role labeling, dependency syntax analysis and other technologies are used to extract entity attributes and relationships between entities.

[0080] After knowledge extraction, vehicle entities and associated attribute values ​​and relationship triples are obtained to preliminarily construct a vehicle knowledge base. Since the vehicle entities and relationships obtained from knowledge extraction may have redundancy and inconsistency, knowledge fusion is required to build a high-quality vehicle knowledge graph. First, the entity linking technology is used to map the vehicle entities mentioned in the unstructured text with the existing entities in the knowledge base to eliminate the ambiguity of the referents. Then, the knowledge alignment technology is used to identify equivalent entities, attributes, and relationships in the knowledge bases of different data sources, and a mapping link is established. Based on the mapping link, redundant attribute values ​​are merged, and contradictory attribute values ​​are adjudicated to form a consistent attribute description. Furthermore, using existing vehicle knowledge, new implicit knowledge is mined and the knowledge graph is expanded through methods such as ontology reasoning and rule reasoning.

[0081] Ultimately, the fused vehicle knowledge graph contains a large number of vehicle entities, rich attribute information, and diverse semantic associations between vehicle entities, which can provide structured and semantic background knowledge in the vehicle field.

[0082] Obtain vehicle review data and vehicle image data and perform preprocessing, including cleaning, word segmentation, and stop word removal. Use word vector models (such as Word2Vec and GloVe) to learn features of review texts and obtain distributed representations at the word level. Use convolutional neural networks (CNNs) or long short-term memory networks (LSTMs) to encode review texts and extract semantic feature representations of reviews.

[0083] Perform preprocessing operations such as size normalization and data enhancement on vehicle image data. Use pre-trained convolutional neural networks (such as ResNet and Inception) to extract features from vehicle images and obtain high-level semantic representations of the images. Fine-tune the pre-trained model on the vehicle image dataset through transfer learning to improve the domain adaptability of image features.

[0084] Using named entity recognition technology, vehicle entity mentions are identified from vehicle reviews and image descriptions. By building a vehicle entity linking model, such as a similarity-based matching model or a deep learning-based sequence labeling model, the entity mentions in vehicle reviews and images are linked to vehicle entities in the knowledge graph to establish an association between multimodal data and structured knowledge.

[0085] Attribute extraction technology is used to extract vehicle attribute information from vehicle reviews and image descriptions. By building an attribute alignment model, such as rule-based template matching or machine learning-based classification model, the extracted attributes are aligned with the vehicle attributes in the knowledge graph to further enhance the relevance of multimodal data and structured knowledge.

[0086] The TransE knowledge representation learning model is used to map entities and relations in the vehicle knowledge graph to a low-dimensional continuous space. The TransE model models relations as translation operations between entity embeddings and learns distributed representations of entities and relations by minimizing the translation error. Optionally, the TransR knowledge representation learning model is used. Considering that different relations may be in different semantic subspaces, the TransR model introduces an independent mapping matrix for each relation to map entity embeddings to relation-specific subspaces. Relation-specific entity representations are learned by minimizing the translation error after mapping.

[0087] By using the attribute and relationship information in the vehicle knowledge graph, a knowledge-guided attention mechanism is designed. When processing vehicle reviews and image features, the importance weights of different words and image regions are calculated based on the attributes and relationships of vehicle entities in the knowledge graph, highlighting key information related to vehicle features and suppressing the influence of irrelevant information.

[0088] A multimodal graph neural network model is designed to map information of different modalities such as vehicle knowledge graph, text, and image into a unified semantic space. With the vehicle knowledge graph as the backbone, entities and relationships are used as graph nodes and edges, and the graph convolutional network (GCN) is used to learn the embedded representation of entities and relationships. At the same time, vehicle reviews and image features are aligned with knowledge graph entities through the attention mechanism as additional features of graph nodes. Through the information propagation and fusion of multimodal graph neural networks, the effective integration of vehicle knowledge of different modalities is achieved.

[0089] The vehicle knowledge graph representation, comment text features, and image visual features are integrated to construct a multimodal vehicle representation. Fusion strategies such as splicing and attention weighting are used to integrate information from different modalities to obtain semantically rich vehicle feature representations. A vehicle recommendation model is constructed, such as classic recommendation algorithms such as collaborative filtering and matrix decomposition, or a deep learning neural network recommendation model. The multimodal vehicle representation is used as an input feature and combined with user historical interaction data to train the vehicle recommendation model. By optimizing the recommendation objective function (such as BPR loss and cross entropy loss), the potential representation of users and vehicles is learned to capture user preferences for vehicles.

[0090] For a given user's vehicle rental request, extract key information in the request (such as user characteristics, vehicle requirements), and combine the vehicle knowledge graph and multimodal vehicle representation to use the trained recommendation model to predict the user's interest in candidate vehicles. According to the predicted interest score, select the Top-N vehicles as the recommendation results to meet the accuracy and diversity requirements of the recommendation. Furthermore, use the semantic relationship in the vehicle knowledge graph to reason and explain the recommendation results. By mining the correlation between the user's historical rental vehicles and the current recommended vehicle, such as the same model, similar configuration, etc., generate a reasonable recommendation explanation, such as "Recommended model A because it is the same high-end SUV model as the model B you rented before."

[0091] Combine vehicle reviews and image information to generate multimodal recommendation explanations. Use keywords and sentiment in reviews to extract users' evaluations and feelings about vehicles, such as "Vehicle A is spacious and suitable for family outings." At the same time, select key areas in vehicle images, such as body appearance and interior details, as a supplement to visual explanations to enhance the intuitiveness and persuasiveness of explanations.

[0092] In this embodiment, the structured knowledge provided by the vehicle knowledge graph is combined with multimodal data such as vehicle review text and images to achieve knowledge-driven multimodal feature learning and vehicle recommendation. In the recommendation process, the introduction of the knowledge graph enhances the recommendation performance and improves the accuracy and diversity of the recommendation. At the same time, the knowledge graph and multimodal information are used to explain the recommendation results, which enhances the transparency and credibility of the recommendation. It provides new ideas for knowledge-driven vehicle recommendation and improves user satisfaction and acceptance of vehicle recommendations.

[0093] In an optional embodiment,

[0094] The fused multimodal vehicle representation is used as the input feature of the vehicle recommendation model. The vehicle recommendation model is trained in combination with the user's historical interaction behavior data. The trained vehicle recommendation model includes:

[0095] Collect historical interaction behavior data of users browsing vehicle details pages, clicking on vehicle attributes, and inquiring about vehicle prices on the vehicle recommendation platform, and build a user-vehicle interaction behavior sequence based on the historical interaction behavior data;

[0096] Inputting the user-vehicle interaction behavior sequence into a bidirectional recurrent neural network integrated with a vehicle attribute attention mechanism in chronological order, dynamically modeling the user's vehicle preference through the hidden state of each time step in the bidirectional recurrent neural network, and obtaining a cumulative representation of the user's vehicle preference at the current moment;

[0097] The vehicle attribute attention mechanism is used to calculate the attention weights of different vehicle attributes in each time step, and the vehicle attribute attention weights are weighted and summed with the vehicle attribute embedding vector to obtain the vehicle attribute representation vector of the current time step; the vehicle attribute representation vector of the current time step is concatenated with the hidden state of the bidirectional recurrent neural network and input into the feedforward neural network to obtain the comprehensive interest representation of the vehicle at the current time step, and the comprehensive interest representation of the vehicle at each time step of the user-vehicle interaction behavior sequence is aggregated through the attention mechanism to obtain the user's final vehicle interest representation;

[0098] A paired learning paradigm is used to negatively sample the user-vehicle interaction behavior sequence and construct vehicle triplets. The interaction time difference between the positive and negative sample vehicles in the vehicle triplets is combined with the similarity difference of the multimodal vehicle representations of the positive and negative sample vehicles, and a time-series-aware multimodal contrast loss function is designed.

[0099] The vehicle click-through rate prediction task is introduced. Based on the multimodal vehicle representation of positive and negative sample vehicles and the user's final vehicle interest representation, the cross entropy function is used as the click-through rate prediction loss function. The multimodal contrast loss function and the click-through rate prediction loss function are weighted summed as the multi-task joint training objective function of the vehicle recommendation model. The parameters of the vehicle recommendation model are optimized and solved to obtain a trained vehicle recommendation model.

[0100] For example, we first collect historical interaction behavior data of users on the vehicle recommendation platform, including browsing vehicle details pages, clicking on vehicle attributes, and inquiring about vehicle prices, and then build a user-vehicle interaction behavior sequence based on these historical interaction behavior data. The user-vehicle interaction behavior sequence reflects the user's interaction behavior with different vehicles at different time points, and contains the temporal characteristics of the evolution of the user's vehicle preference.

[0101] In order to capture the temporal dynamic characteristics of user vehicle preferences, a bidirectional recurrent neural network with vehicle attribute attention mechanism is used to model the user-vehicle interaction behavior sequence. Specifically, the user-vehicle interaction behavior sequence is input into the bidirectional recurrent neural network in chronological order, and the user vehicle preference is dynamically modeled through the hidden state of each time step in the bidirectional recurrent neural network to obtain the cumulative representation of the user's vehicle preference at the current moment. The bidirectional recurrent neural network can simultaneously consider the forward and backward information of the interaction behavior sequence, and more comprehensively characterize the temporal dependency of the user vehicle preference.

[0102] In addition, in order to highlight the impact of different vehicle attributes on user preferences, a vehicle attribute attention mechanism is introduced to dynamically adjust the importance of different vehicle attributes. At each time step of the bidirectional recurrent neural network, the attention weights of different vehicle attributes are calculated using the attention mechanism, and the attention weights are weighted and summed with the vehicle attribute embedding vector to obtain the vehicle attribute representation vector of the current time step. The vehicle attribute representation vector integrates the semantic information of different vehicle attributes and can more accurately characterize the comprehensive characteristics of the vehicle. Then, the vehicle attribute representation vector of the current time step is concatenated with the hidden state of the bidirectional recurrent neural network and input into the feedforward neural network to obtain the comprehensive interest representation of the vehicle at the current time step. The comprehensive interest representation of the vehicle combines the temporal characteristics of user preferences and the semantic characteristics of vehicle attributes, and can more comprehensively characterize the user's interest in the vehicle. Finally, the comprehensive interest representation of the vehicle at each time step of the user-vehicle interaction behavior sequence is aggregated through the attention mechanism to obtain the user's final vehicle interest representation.

[0103] In order to further enhance the representation learning ability of the vehicle recommendation model, a paired learning paradigm is used to negatively sample the user-vehicle interaction behavior sequence to construct vehicle triples. The vehicle triples consist of positive sample vehicles, negative sample vehicles and users, where positive sample vehicles are vehicles with which users have interacted, and negative sample vehicles are randomly sampled vehicles with which users have not interacted. When constructing vehicle triples, not only the similarity difference between the multimodal vehicle representations of positive and negative sample vehicles is considered, but also the interaction time difference between positive and negative sample vehicles is introduced, and a time-aware multimodal contrast loss function is designed. This loss function encourages the multimodal representation of positive sample vehicles to be closer to the user's vehicle interest representation in the semantic space, while considering the impact of the interaction time difference on vehicle similarity, so that the model can learn time-aware vehicle representation.

[0104] Furthermore, the vehicle click rate prediction task is introduced as an auxiliary task to form a multi-task learning framework with the vehicle representation learning task. Specifically, based on the multimodal vehicle representation of positive and negative sample vehicles and the user's final vehicle interest representation, the cross entropy function is used as the click rate prediction loss function, and the multimodal contrast loss function and the click rate prediction loss function are weighted summed as the multi-task joint training objective function of the vehicle recommendation model. By optimizing this objective function, the model can accurately predict the user's click probability on the vehicle while learning the time-series perception vehicle representation, thereby improving the accuracy of vehicle recommendation.

[0105] In the model training phase, the parameters of the vehicle recommendation model are optimized and solved using optimization algorithms such as stochastic gradient descent, and the model parameters are updated through the back propagation algorithm to finally obtain a trained vehicle recommendation model. In the recommendation phase, given a target user, the trained vehicle recommendation model is used to score and sort the candidate vehicles, and the top-N vehicles with the highest scores are selected as the recommendation results.

[0106] In an optional embodiment,

[0107] The multi-task joint training objective function of the vehicle recommendation model is the weighted sum of the multi-modal contrast loss function and the click-through rate prediction loss function, which includes:

[0108] The multimodal contrast loss function calculation formula is as follows:

[0109] ;

[0110] Among them, L T represents the multimodal contrast loss function, a represents the user, b represents the positive sample vehicle, and b ' represents the negative sample vehicle, D represents the set of all triples in the training set, σ represents the Sigmoid function, γ represents the temporal weight factor, represents the time when user a interacts with the positive sample vehicle b, represents the virtual interaction time between user a and negative sample vehicle b, a a represents the final interest representation vector of user a, b x represents the multimodal representation vector of the positive sample vehicle, b ' x represents the multimodal representation vector of the negative sample vehicle, λ represents the L2 regularization coefficient, and θ represents the model parameter;

[0111] The calculation formula of the click rate prediction loss function is as follows:

[0112] ;

[0113] Among them, LC represents the click rate prediction loss function, c represents the click state, and C represents the set of click sample pairs. represents the probability predicted by the model that user a clicks on vehicle b;

[0114] The calculation formula of the multi-task joint training objective function is as follows:

[0115] L=L T +ηL C ;

[0116] Among them, L represents the multi-task joint training objective function, and η represents the weight coefficient of the click rate prediction task.

[0117] In this embodiment, the temporal features and vehicle attribute information contained in the user-vehicle interaction behavior sequence are fully utilized to learn the time-aware vehicle representation and improve the accuracy of vehicle recommendation. The introduction of a bidirectional recurrent neural network and an attention mechanism can capture the temporal dynamic characteristics of user vehicle preferences and the importance differences of vehicle attributes. By adopting a time-aware multimodal contrastive learning paradigm, the time-aware vehicle representation is learned to improve the distinguishing ability of vehicle representation. By introducing the vehicle click-through rate prediction task and building a multi-task learning framework, the accuracy of vehicle recommendation is further improved, providing new ideas for the design of vehicle recommendation systems.

[0118] S102. Obtain the historical rental frequency, historical rental score and current vehicle location of each vehicle in the candidate vehicle set, calculate the comprehensive matching degree between each candidate vehicle and the vehicle rental request based on the fuzzy comprehensive evaluation method and the pick-up time and pick-up location in the vehicle rental request, determine the candidate vehicle with the highest comprehensive matching degree as the rental vehicle, and generate an intelligent vehicle delivery task.

[0119] Among them, the fuzzy comprehensive evaluation first determines the evaluation index system based on expert knowledge and experience, including the historical rental frequency of the vehicle, the historical rental score, the distance between the vehicle and the user, etc. The weight is set for each indicator, reflecting the importance of the indicator in the entire evaluation system. Then, each evaluation indicator is fuzzified. For quantitative indicators (such as rental frequency, geographical distance, etc.), fuzzy membership functions (such as triangle, trapezoid, etc.) are used to map them to the [0, 1] interval. For qualitative indicators (such as user scores), language variables (such as "poor", "average", "good", "excellent", etc.) are used to describe them, and corresponding membership values ​​are given based on expert experience.

[0120] Finally, the fuzzy evaluation matrix R is constructed, where the matrix element r_ij represents the membership of the i-th candidate vehicle on the j-th evaluation index. By combining the membership and weight of each index through weighted average or geometric average methods, the comprehensive evaluation value vector of the candidate vehicle is obtained.

[0121] In an optional embodiment,

[0122] Based on the fuzzy comprehensive evaluation method, combined with the pick-up time and pick-up location in the vehicle rental request, the comprehensive matching degree between each candidate vehicle and the vehicle rental request is calculated, including:

[0123] Construct a multi-objective optimization problem for vehicle matching, with comprehensive matching degree, user satisfaction and vehicle utilization as optimization targets;

[0124] Designing a coding scheme for fuzzy rules, using a binary tree to represent the grammatical structure of fuzzy rules, wherein non-leaf nodes of the binary tree represent logical connectives between fuzzy propositions, and leaf nodes represent fuzzy propositions composed of input variables and their membership function parameters, and obtaining a fixed-length real-valued vector representing the fuzzy rules by traversing the binary tree;

[0125] The real-valued vector is optimized by using an evolutionary algorithm to obtain an optimal solution set of fuzzy rule combinations, and an optimal fuzzy rule combination is selected from the optimal solution set as an evaluation rule for vehicle matching. Based on the obtained evaluation rule and in combination with the pick-up time and pick-up location in the vehicle rental request, a comprehensive matching degree between each candidate vehicle and the vehicle rental request is calculated.

[0126] For example, in a vehicle recommendation system, in order to provide a more comprehensive and balanced recommendation result, it is necessary to consider multiple optimization objectives at the same time: (1) Comprehensive matching degree: measures the matching degree between candidate vehicles and user rental requests in terms of time, space, vehicle characteristics, etc. The higher the matching degree, the more the vehicle meets the user's needs. (2) User satisfaction: reflects the user's acceptance of the recommended vehicle and the user experience. The higher the satisfaction, the more the recommendation result can meet the user's preferences. (3) Vehicle utilization rate: evaluates the efficiency of vehicle resource utilization. The higher the utilization rate, the more fully and reasonably the vehicle is dispatched and allocated.

[0127] In order to use evolutionary algorithms to optimize fuzzy rules, it is necessary to map the fuzzy rules into the encoding of evolutionary individuals. Fuzzy rules usually consist of several premises and a conclusion, such as "IF (condition1) AND (condition2)AND ... THEN (conclusion)". Among them, the premises and conclusions are fuzzy propositions about input variables, such as "the remaining power of the vehicle is high", "the user rating is excellent", etc. Optionally, a binary tree is used to represent the grammatical structure of the fuzzy rules. The non-leaf nodes of the tree represent the logical connectives between fuzzy propositions (such as AND, OR, etc.), and the leaf nodes represent specific fuzzy propositions. Each fuzzy proposition consists of the parameters of the corresponding input variables and their membership functions (such as triangles, trapezoids, etc.).

[0128] By traversing the binary tree in pre-order, the fuzzy rules can be encoded into a fixed-length real-valued vector as the individual representation of the evolutionary algorithm. The vector contains information such as the input variable selection, membership function type and parameter value of the fuzzy proposition.

[0129] According to the coding scheme of fuzzy rules, the implementation process of evolutionary optimization specifically includes randomly generating a certain number of evolutionary individuals, each of which represents a set of fuzzy rules. The coding vector of the individual is randomly generated according to the rule structure and membership function. For each individual in the population, the corresponding fuzzy rule combination is decoded according to the coding vector. According to the decoded fuzzy rules, multiple optimization target values ​​such as comprehensive matching degree, user satisfaction and vehicle utilization rate on the training data set are calculated as the fitness evaluation of the individual. The Fast Non-dominated Sorting algorithm is used to sort the individuals in the population, and the individuals are divided into multiple non-dominated sets (Pareto frontiers) according to the dominance relationship of the individuals in the target space. There is no mutual domination between individuals in the same frontier, but there is a dominance relationship between different frontiers. Individuals with higher fitness are selected as high-quality parents to prepare for the generation of the next generation of population.

[0130] For the selected parent individuals, crossover and mutation operators are randomly applied to generate new offspring individuals. The crossover operator combines and reorganizes the encoding vectors of the two parents, and the mutation operator randomly perturbs the encoding vectors of the parents to introduce new search information. Through the action of genetic operators, new fuzzy rule combinations are continuously generated to expand the search range. The parent individuals and offspring individuals are merged, and the non-dominated sorting is re-performed. Individuals with higher fitness are selected to form a new generation of populations, keeping the population size unchanged.

[0131] Repeat the above steps to iterate and evolve continuously until the preset maximum number of evolution generations is reached or the population converges (such as the Pareto front no longer changes significantly). From the population finally evolved, select the non-dominated individuals on the Pareto front, that is, the fuzzy rule combination that achieves the optimal balance on multiple objectives. These rules can give vehicle recommendation results with better comprehensive performance while taking into account matching degree, satisfaction and utilization rate.

[0132] In an optional embodiment,

[0133] The calculation formula for calculating the comprehensive matching degree between each candidate vehicle and the vehicle rental request is as follows:

[0134] ;

[0135] Among them, M(V, R) represents the comprehensive matching degree between the candidate vehicle V and the vehicle rental request R, m represents the number of features in the fuzzy rule, Represents the feature f in the fuzzy rule i The weight coefficient of (V, R) represents the feature f i The membership degree between the candidate vehicle V and the vehicle rental request R, n represents the number of constraints, Represents the constraint d in the fuzzy rule j The weight coefficient of (V, R) represents the constraint d j The degree of membership between the candidate vehicle V and the vehicle rental request R.

[0136] In this embodiment, by constructing a multi-objective optimization problem, designing a fuzzy rule coding scheme, and using an evolutionary algorithm to search for the Pareto optimal rule combination, a fuzzy evaluation rule that balances various objectives is automatically generated while taking into account multiple objectives such as matching degree, user satisfaction, and vehicle utilization. Compared with traditional fuzzy comprehensive evaluation, this method can adaptively optimize fuzzy rules, does not require prior knowledge of expert experience, and simultaneously considers multiple important objectives of the recommendation results, thereby improving the comprehensive performance of vehicle recommendation.

[0137] S103. Send the smart vehicle delivery task to the smart vehicle management device corresponding to the vehicle identification. The smart vehicle management device returns the vehicle delivery completion information after completing the vehicle delivery. If the vehicle return information sent by the smart vehicle management device is received within the preset time after the return time specified in the vehicle rental request, the vehicle rental is determined to be completed. According to the vehicle condition information at the time of delivery in the vehicle delivery completion information and the vehicle condition information at the time of return in the vehicle return information, the changes in the vehicle condition before and after the use of the vehicle are automatically compared and analyzed, and the rental bill is automatically generated according to the comparison results using the smart contract based on blockchain technology.

[0138] Exemplarily, first, the system sends the generated smart vehicle delivery task to the smart vehicle management device corresponding to the vehicle ID. The smart vehicle management device is an embedded terminal installed on the vehicle, which has functions such as positioning, communication, and vehicle condition monitoring, and can collect information such as the vehicle's location, mileage, fuel level, and fault code in real time.

[0139] After receiving the delivery task, the intelligent vehicle management device controls the vehicle to automatically navigate to the designated delivery location and hands the vehicle over to the user for use. After the delivery is completed, the intelligent vehicle management device returns the vehicle delivery completion information to the system, including the delivery time, delivery location, and vehicle condition information at the time of delivery. The vehicle condition information at the time of delivery is collected through on-board sensors and cameras, recording detailed information such as the vehicle's appearance, interior, and component status at the time of delivery, which serves as a reference for subsequent vehicle condition comparison.

[0140] After the user has used the vehicle, he / she needs to drive the vehicle back to the return location within the specified return time. The system continuously monitors the status of the intelligent vehicle management device. If the system receives the vehicle return information sent by the intelligent vehicle management device within the preset time (such as 30 minutes) after the return time, it is determined that the user has completed the vehicle return. The vehicle return information includes the return time, return location, vehicle condition information at the time of return, etc. The vehicle condition information at the time of return is collected in the same way as the vehicle condition information at the time of delivery.

[0141] After confirming that the vehicle rental is complete, the system automatically triggers the vehicle condition comparison and rental bill generation process. First, the system extracts the vehicle condition information at the time of delivery and the vehicle condition information at the time of return from the delivery completion information and vehicle return information, and uses computer vision technology for intelligent comparison and analysis. Computer vision technology uses algorithms such as image processing, feature extraction, and pattern recognition to automatically detect and quantify changes in the vehicle condition before and after use, such as body scratches, tire wear, interior damage, etc., and generates a vehicle condition change report. If the vehicle condition changes beyond the normal use range, the corresponding repair or compensation fees will be added to the rental bill.

[0142] The generation of rental bills is automatically completed through blockchain-based smart contracts. Smart contracts are self-executable digital contracts that pre-define the terms and billing rules of the rental agreement in the form of program code on the blockchain. Once the contract execution is triggered, the rental fee can be automatically calculated and an electronic invoice that cannot be tampered with can be generated. The system uses information such as vehicle usage time, mileage, and vehicle condition changes as input, calls smart contracts to calculate fees, and writes the calculation results to the blockchain to form a permanent certificate. Users can view their rental bill details through the blockchain browser to ensure transparent and fair billing.

[0143] In an optional embodiment,

[0144] Based on the vehicle condition information at the time of delivery in the vehicle delivery completion information and the vehicle condition information at the time of return in the vehicle return information, the changes in the vehicle condition before and after use are automatically compared and analyzed, and the rental bill is automatically generated based on the comparison results using the smart contract based on blockchain technology, including:

[0145] Build a vehicle condition data collection and chain-up mechanism. When the vehicle is delivered or returned, the intelligent vehicle management device automatically collects vehicle condition information, including vehicle appearance images, vehicle interior environment images, and vehicle sensor data. After preprocessing the collected vehicle condition information, the preprocessed vehicle condition information is uploaded to the blockchain in real time through the Internet of Vehicles communication module.

[0146] The blockchain node verifies and packages the received pre-processed vehicle condition information according to the pre-defined data format and identity authentication rules, writes the hash value and timestamp of the packaged vehicle condition information into the block, generates a new block and broadcasts it to the entire network, completing the on-chain storage of the vehicle condition data;

[0147] Obtain the vehicle delivery status information and return status information stored on the chain and input them into the pre-trained vehicle condition comparison analysis model, output the vehicle condition detection results under the two states and automatically generate a vehicle condition detection report;

[0148] According to the generated vehicle condition inspection report, the rental agreement and billing rules are encoded into a smart contract, and the smart contract is deployed on the blockchain. When the vehicle is returned, the smart contract automatically obtains the vehicle condition inspection report, determines whether the change in vehicle condition exceeds the normal use range, and calculates the compensation amount for the deterioration of the vehicle condition according to predefined rules. At the same time, the basic rental fee is calculated based on the actual duration and mileage of the vehicle rental. The compensation amount for the deterioration of the vehicle condition and the basic rental fee are summarized to automatically generate a rental bill, and the generated rental bill is written to the blockchain to form an unalterable bill record. Users and vehicle managers can view detailed electronic rental bill information through the blockchain browser.

[0149] For example, in terms of vehicle condition data collection and chain-up mechanism, a set of intelligent vehicle management equipment is designed, equipped with vehicle-mounted sensors, image collectors and vehicle networking communication modules. When the vehicle is delivered and returned, the intelligent vehicle management equipment automatically collects vehicle condition information, including vehicle appearance images, interior environment images and vehicle-mounted sensor data. The vehicle appearance image is collected by high-definition cameras on various parts of the vehicle body, reflecting the overall condition of the vehicle appearance; the interior environment image is collected by the in-vehicle camera, reflecting the cleanliness and integrity of the interior environment; the in-vehicle sensor data includes engine operating parameters, battery power, tire pressure, etc., reflecting the mechanical condition and performance level of the vehicle. After preprocessing, the collected vehicle condition information is uploaded to the blockchain in real time through the vehicle networking communication module. The preprocessing process includes image compression, data format conversion, encryption, etc., to reduce the overhead of data transmission and storage and protect the security and privacy of data.

[0150] After the blockchain node receives the pre-processed vehicle condition information, it verifies and packages it according to the predefined data format and identity authentication rules. The data format rules define the data structure and field type of the vehicle condition information to ensure the standardization and consistency of the data; the identity authentication rules verify the identity of the data provider to ensure the credibility and non-repudiation of the data source. After the verification, the blockchain node writes the hash value and timestamp of the packaged vehicle condition information into the block, generates a new block and broadcasts it to the entire network. The hash value is the unique identifier of the vehicle condition information, and the timestamp records the time when the data is uploaded to the chain, ensuring the integrity and non-tamperability of the data. The nodes of the entire blockchain network verify and confirm the new block through the consensus algorithm. Once written into the blockchain, the vehicle condition information is permanently stored, forming a traceable and verifiable data certificate.

[0151] The vehicle condition comparison analysis and test report generation mechanism uses the vehicle delivery status information and return status information stored on the chain, and with the help of the pre-trained vehicle condition comparison analysis model, automatically outputs the vehicle condition detection results in the two states and generates a vehicle condition detection report. The vehicle condition comparison analysis model uses a deep learning algorithm to extract features and compare and analyze the vehicle appearance image, the interior environment image and the on-board sensor data, identify the changes in the vehicle condition during the rental period, and locate and evaluate the location, degree and type of vehicle condition changes. The test results output by the model include quantitative indicators and details of vehicle condition changes, such as the area of ​​paint scratches, the depth of dents, the degree of seat wear, odor concentration, etc. The test report is generated in the form of a structured document, which records the information of vehicle condition changes in detail, and is accompanied by visual annotation images of the changed areas, which is convenient for users and managers to review and verify.

[0152] The smart contract automatic billing and bill management mechanism encodes the lease agreement and billing rules into smart contracts and deploys them on the blockchain to realize the automatic calculation and settlement of vehicle rental fees. The smart contract expresses the terms and billing rules in the lease agreement in the form of program code, such as the lease term, billing method, overtime penalty, mileage limit, vehicle condition requirements, etc. When the vehicle is returned, the smart contract automatically obtains the vehicle condition inspection report stored on the chain to determine whether the change in vehicle condition exceeds the normal use range. If the vehicle condition changes beyond the agreed range, such as serious scratches, component damage, mileage exceeding the standard, etc., the smart contract automatically calculates the compensation amount for the deterioration of the vehicle condition according to the predefined compensation terms and calculation formula. At the same time, the smart contract calculates the basic rental fee based on the actual duration and mileage of the vehicle rental. Finally, the compensation amount for the deterioration of the vehicle condition is summarized with the basic rental fee to automatically generate a rental bill. The generated bill information is written to the blockchain in the form of a transaction, forming an electronic bill record that cannot be tampered with and can be traced. Users and vehicle managers can view detailed bill information through the blockchain browser, including bill number, generation time, rental fee details, compensation amount, total amount, etc., to achieve openness, transparency and real-time settlement of rental fees.

[0153] In this embodiment, a decentralized, efficient and transparent vehicle rental management system is constructed through key technologies such as vehicle condition data collection and chain-up, vehicle condition comparative analysis and test report generation, and smart contract automatic billing and bill management. By using blockchain technology to achieve tamper-proof storage of vehicle condition data and rental information, the authenticity and credibility of the data are improved; by using smart contract technology to realize the automatic execution of vehicle rental business rules, the efficiency and accuracy of rental management are improved; by using the Internet of Vehicles and deep learning technology to realize automatic collection and intelligent analysis of vehicle condition information, the cost and error rate of manual operation are reduced; fourth, the decentralized blockchain network is used to realize the openness, transparency and real-time sharing of rental information, which improves the fairness and traceability of the rental process.

[0154] Figure 2 FIG. 1 is a schematic diagram of a delivery processing system for vehicle rental business according to an embodiment of the present invention. Figure 2 As shown, the system comprises:

[0155] The first unit is used to receive a vehicle rental request, input the vehicle rental request into a pre-built vehicle recommendation model, automatically extract key information in the vehicle rental request, and combine with a pre-built vehicle database to output a set of candidate vehicles whose matching degree with the vehicle rental request is higher than a preset threshold, wherein the vehicle database stores static attribute information and dynamic status information of multiple rentable vehicles, wherein the static attribute information includes vehicle information and in-vehicle configuration parameters, and the dynamic status information includes a current location of the vehicle and a rental period;

[0156] The second unit is used to obtain the historical rental frequency, historical rental score and current vehicle location of each vehicle in the candidate vehicle set, and based on the fuzzy comprehensive evaluation method, combined with the pick-up time and pick-up location in the vehicle rental request, calculate the comprehensive matching degree between each candidate vehicle and the vehicle rental request, determine the candidate vehicle with the highest comprehensive matching degree as the rental vehicle, and generate an intelligent vehicle delivery task;

[0157] The third unit is used to send the smart vehicle delivery task to the smart vehicle management device corresponding to the vehicle identification. The smart vehicle management device returns the vehicle delivery completion information after completing the vehicle delivery. If the vehicle return information sent by the smart vehicle management device is received within a preset time after the return time specified in the vehicle rental request, the vehicle rental is determined to be completed. According to the vehicle condition information at the time of delivery in the vehicle delivery completion information and the vehicle condition information at the time of return in the vehicle return information, the changes in the vehicle condition before and after the use of the vehicle are automatically compared and analyzed, and the rental bill is automatically generated according to the comparison result using the smart contract based on blockchain technology.

[0158] According to a third aspect of the embodiments of the present invention,

[0159] An electronic device is provided, comprising:

[0160] processor;

[0161] a memory for storing processor-executable instructions;

[0162] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0163] A fourth aspect of the embodiments of the present invention is:

[0164] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0165] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0166] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A delivery processing method for vehicle leasing business, characterized in that: include: Receive a vehicle rental request, input the vehicle rental request into a pre-built vehicle recommendation model, automatically extract key information from the vehicle rental request, and combine with a pre-built vehicle database to output a set of candidate vehicles that match the vehicle rental request with a degree higher than a preset threshold, including: The ontology modeling method is used to define the core concepts and relationships in the vehicle field, and the vehicle entities and corresponding attributes are extracted from structured and unstructured data sources. The vehicle knowledge is integrated using entity linking and knowledge alignment technology to form a fused vehicle knowledge graph. Obtaining vehicle review data and vehicle image data and performing preprocessing, inputting the preprocessed vehicle review data and vehicle image data into a pre-trained convolutional neural network to extract vehicle review text features and image visual features, respectively; Using a pre-built knowledge representation learning model, the vehicle entities and relationships in the fused vehicle knowledge graph are mapped to a low-dimensional semantic space to obtain a distributed representation of vehicle entities and relationships. Design a knowledge-guided attention mechanism, use the distributed representation of vehicle entities and relations, calculate the importance weights of different words and image regions in vehicle review text features and image visual features, and build a multimodal graph neural network model. Use the graph convolutional network to learn the embedded representation of vehicle entities and relations, and fuse the vehicle review text features and image visual features weighted by the attention mechanism with the vehicle entity embedded representation to obtain a fused multimodal vehicle representation. The fused multimodal vehicle representation is used as the input feature of the vehicle recommendation model. The vehicle recommendation model is trained in combination with the user's historical interaction behavior data. The trained vehicle recommendation model includes: Collect historical interaction behavior data of users browsing vehicle details pages, clicking on vehicle attributes, and inquiring about vehicle prices on the vehicle recommendation platform, and build a user-vehicle interaction behavior sequence based on the historical interaction behavior data; Inputting the user-vehicle interaction behavior sequence into a bidirectional recurrent neural network integrated with a vehicle attribute attention mechanism in chronological order, dynamically modeling the user's vehicle preference through the hidden state of each time step in the bidirectional recurrent neural network, and obtaining a cumulative representation of the user's vehicle preference at the current moment; The vehicle attribute attention mechanism is used to calculate the attention weights of different vehicle attributes in each time step, and the vehicle attribute attention weights are weighted and summed with the vehicle attribute embedding vector to obtain the vehicle attribute representation vector of the current time step; the vehicle attribute representation vector of the current time step is concatenated with the hidden state of the bidirectional recurrent neural network and input into the feedforward neural network to obtain the comprehensive interest representation of the vehicle at the current time step, and the comprehensive interest representation of the vehicle at each time step of the user-vehicle interaction behavior sequence is aggregated through the attention mechanism to obtain the user's final vehicle interest representation; A paired learning paradigm is used to negatively sample the user-vehicle interaction behavior sequence and construct vehicle triplets. The interaction time difference between the positive and negative sample vehicles in the vehicle triplets is combined with the similarity difference of the multimodal vehicle representations of the positive and negative sample vehicles, and a time-series-aware multimodal contrast loss function is designed. The vehicle click-through rate prediction task is introduced. Based on the multimodal vehicle representation of positive and negative sample vehicles and the user's final vehicle interest representation, the cross entropy function is used as the click-through rate prediction loss function. The multimodal contrast loss function and the click-through rate prediction loss function are weighted summed as the multi-task joint training objective function of the vehicle recommendation model. The parameters of the vehicle recommendation model are optimized and solved to obtain a trained vehicle recommendation model. The vehicle database stores static attribute information and dynamic status information of multiple rentable vehicles, wherein the static attribute information includes vehicle information and in-vehicle configuration parameters, and the dynamic status information includes the current location of the vehicle and the available rental period; Obtain the historical rental frequency, historical rental score and current vehicle location of each vehicle in the candidate vehicle set, calculate the comprehensive matching degree between each candidate vehicle and the vehicle rental request based on the fuzzy comprehensive evaluation method and the pick-up time and pick-up location in the vehicle rental request, determine the candidate vehicle with the highest comprehensive matching degree as the rental vehicle, and generate an intelligent vehicle delivery task; The smart vehicle delivery task is sent to the smart vehicle management device corresponding to the vehicle identification. The smart vehicle management device returns the vehicle delivery completion information after completing the vehicle delivery. If the vehicle return information sent by the smart vehicle management device is received within the preset time after the return time specified in the vehicle rental request, the vehicle rental is determined to be completed. According to the vehicle condition information at the time of delivery in the vehicle delivery completion information and the vehicle condition information at the time of return in the vehicle return information, the changes in the vehicle condition before and after the use of the vehicle are automatically compared and analyzed, and the rental bill is automatically generated according to the comparison results using the smart contract based on blockchain technology.

2. The method according to claim 1, characterized in that The multi-task joint training objective function of the vehicle recommendation model is the weighted sum of the multi-modal contrast loss function and the click-through rate prediction loss function, which includes: The multimodal contrast loss function calculation formula is as follows: ； Among them, L T represents the multimodal contrast loss function, a represents the user, b represents the positive sample vehicle, and b ' represents the negative sample vehicle, D represents the set of all triples in the training set, σ represents the Sigmoid function, γ represents the temporal weight factor, represents the time when user a interacts with the positive sample vehicle b, represents the virtual interaction time between user a and negative sample vehicle b, a a represents the final interest representation vector of user a, b x represents the multimodal representation vector of the positive sample vehicle, b ' x represents the multimodal representation vector of the negative sample vehicle, λ represents the L2 regularization coefficient, and θ represents the model parameter; The calculation formula of the click rate prediction loss function is as follows: ； Among them, LC represents the click rate prediction loss function, c represents the click state, and C represents the set of click sample pairs. represents the probability predicted by the model that user a clicks on vehicle b; The calculation formula of the multi-task joint training objective function is as follows: ； Among them, L represents the multi-task joint training objective function, and η represents the weight coefficient of the click rate prediction task.

3. The method according to claim 1, characterized in that Based on the fuzzy comprehensive evaluation method, combined with the pick-up time and pick-up location in the vehicle rental request, the comprehensive matching degree between each candidate vehicle and the vehicle rental request is calculated, including: Construct a multi-objective optimization problem for vehicle matching, with comprehensive matching degree, user satisfaction and vehicle utilization as optimization targets; Designing a coding scheme for fuzzy rules, using a binary tree to represent the grammatical structure of fuzzy rules, wherein non-leaf nodes of the binary tree represent logical connectives between fuzzy propositions, and leaf nodes represent fuzzy propositions composed of input variables and their membership function parameters, and obtaining a fixed-length real-valued vector representing the fuzzy rules by traversing the binary tree; The real-valued vector is optimized by using an evolutionary algorithm to obtain an optimal solution set of fuzzy rule combinations, and an optimal fuzzy rule combination is selected from the optimal solution set as an evaluation rule for vehicle matching. Based on the obtained evaluation rule and in combination with the pick-up time and pick-up location in the vehicle rental request, a comprehensive matching degree between each candidate vehicle and the vehicle rental request is calculated.

4. The method according to claim 3, characterized in that: The calculation formula for calculating the comprehensive matching degree between each candidate vehicle and the vehicle rental request is as follows: ； Among them, M(V, R) represents the comprehensive matching degree between the candidate vehicle V and the vehicle rental request R, m represents the number of features in the fuzzy rule, Represents the feature f in the fuzzy rule i The weight coefficient of (V, R) represents the feature f i The membership degree between the candidate vehicle V and the vehicle rental request R, n represents the number of constraints, Represents the constraint d in the fuzzy rule j The weight coefficient of (V, R) represents the constraint d j The degree of membership between the candidate vehicle V and the vehicle rental request R.

5. The method according to claim 1, characterized in that Based on the vehicle condition information at the time of delivery in the vehicle delivery completion information and the vehicle condition information at the time of return in the vehicle return information, the changes in the vehicle condition before and after use are automatically compared and analyzed, and the rental bill is automatically generated based on the comparison results using the smart contract based on blockchain technology, including: Build a vehicle condition data collection and chain-up mechanism. When the vehicle is delivered or returned, the intelligent vehicle management device automatically collects vehicle condition information, including vehicle appearance images, vehicle interior environment images, and vehicle sensor data. After preprocessing the collected vehicle condition information, the preprocessed vehicle condition information is uploaded to the blockchain in real time through the Internet of Vehicles communication module. The blockchain node verifies and packages the received pre-processed vehicle condition information according to the pre-defined data format and identity authentication rules, writes the hash value and timestamp of the packaged vehicle condition information into the block, generates a new block and broadcasts it to the entire network, completing the on-chain storage of the vehicle condition data; Obtain the vehicle delivery status information and return status information stored on the chain and input them into the pre-trained vehicle condition comparison analysis model, output the vehicle condition detection results under the two states and automatically generate a vehicle condition detection report; According to the generated vehicle condition inspection report, the rental agreement and billing rules are encoded into a smart contract, and the smart contract is deployed on the blockchain. When the vehicle is returned, the smart contract automatically obtains the vehicle condition inspection report, determines whether the change in vehicle condition exceeds the normal use range, and calculates the compensation amount for the deterioration of the vehicle condition according to predefined rules. At the same time, the basic rental fee is calculated based on the actual duration and mileage of the vehicle rental. The compensation amount for the deterioration of the vehicle condition and the basic rental fee are summarized to automatically generate a rental bill, and the generated rental bill is written to the blockchain to form an unalterable bill record. Users and vehicle managers can view detailed electronic rental bill information through the blockchain browser.

6. A delivery processing system for vehicle rental business, used to implement the method according to any one of claims 1 to 5, characterized in that: include: The first unit is used to receive a vehicle rental request, input the vehicle rental request into a pre-built vehicle recommendation model, automatically extract key information in the vehicle rental request, and combine with a pre-built vehicle database to output a set of candidate vehicles whose matching degree with the vehicle rental request is higher than a preset threshold, wherein the vehicle database stores static attribute information and dynamic status information of multiple rentable vehicles, wherein the static attribute information includes vehicle information and in-vehicle configuration parameters, and the dynamic status information includes a current location of the vehicle and a rental period; The second unit is used to obtain the historical rental frequency, historical rental score and current vehicle location of each vehicle in the candidate vehicle set, and based on the fuzzy comprehensive evaluation method, combined with the pick-up time and pick-up location in the vehicle rental request, calculate the comprehensive matching degree between each candidate vehicle and the vehicle rental request, determine the candidate vehicle with the highest comprehensive matching degree as the rental vehicle, and generate an intelligent vehicle delivery task; The third unit is used to send the smart vehicle delivery task to the smart vehicle management device corresponding to the vehicle identification. The smart vehicle management device returns the vehicle delivery completion information after completing the vehicle delivery. If the vehicle return information sent by the smart vehicle management device is received within a preset time after the return time specified in the vehicle rental request, the vehicle rental is determined to be completed. According to the vehicle condition information at the time of delivery in the vehicle delivery completion information and the vehicle condition information at the time of return in the vehicle return information, the changes in the vehicle condition before and after the use of the vehicle are automatically compared and analyzed, and the rental bill is automatically generated according to the comparison result using the smart contract based on blockchain technology.

7. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method described in any one of claims 1 to 5.

8. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images