Automatic driving and automatic charging fusion application scene evaluation method and system

Abstract

The invention relates to an automatic driving and automatic charging fusion application scene evaluation method and system, and the method comprises the steps: searching an association path between an automatic driving entity and an automatic charging entity in a fusion application knowledge graph, and summarizing an entity combination which accords with the logic and is higher in association degree as a candidate fusion application scene; extracting a numerical feature vector of each candidate fusion application scene from the fusion application knowledge graph; obtaining potential scores of each candidate fusion application scene under different evaluation indexes; obtaining an overall potential score according to the potential score of each candidate fusion application scene under different evaluation indexes; and obtaining a posterior mean value of each candidate fusion application scene according to the overall potential score and the potential scores under different evaluation indexes, sorting all the candidate fusion application scenes according to the posterior mean values, and identifying the most potential automatic driving and automatic charging fusion application scene. Compared with the prior art, the fusion application scene recognition accuracy is improved.

Description

Technical Field

[0001] This invention relates to the fields of intelligent transportation and new energy, and in particular to an evaluation method and system for integrated application scenarios of autonomous driving and automatic charging. Background Technology

[0002] With the rapid growth of the global electric vehicle market, the construction of charging infrastructure is accelerating. However, the current charging system still faces a series of pressing problems. For private charging, users typically need a fixed, dedicated parking space, but in urban residential areas and other scenarios, many users cannot meet this requirement and must rely on public charging facilities. Existing public charging facilities are mainly divided into two categories: public slow charging and public fast charging. Slow charging can support multiple vehicles charging simultaneously with a relatively small power capacity, but it suffers from problems such as non-charging vehicles occupying spaces for extended periods and vehicles not being moved promptly after charging, leading to inefficient use of parking resources and a poor user charging experience. Fast charging significantly shortens charging time, but it easily leads to vehicle queues during peak hours. Furthermore, fast charging requires high grid capacity, resulting in uneven facility utilization and misalignment between user fast charging times and off-peak electricity hours, hindering the efficient absorption of green electricity.

[0003] The aforementioned issues have, to some extent, hindered the widespread adoption of electric vehicles and exposed the shortcomings of current charging methods in terms of parking resource utilization efficiency and power system friendliness. With the rapid development of autonomous driving technology, vehicles are gradually acquiring advanced levels of autonomous perception, planning, and execution capabilities. Combining autonomous driving with automatic charging technology can enable vehicles to autonomously find available charging spots, automatically queue for charging during off-peak hours, and dynamically schedule charging. While improving user experience, this is expected to significantly alleviate problems such as uneven utilization of charging facilities, parking resource shortages, and limited green energy consumption, becoming a key breakthrough for the future integration of smart mobility and energy management.

[0004] Currently, autonomous driving and automatic charging systems are typically developed independently, without fully considering their potential for integration in practical applications. For example, patent application CN117668413A discloses a comprehensive decision-making evaluation method and device for autonomous driving that considers multiple driving factors. This method collects driving trajectory parameters and surrounding vehicle status, establishes a single-item evaluation model based on four indicators: safety, timeliness, comfort, and energy consumption, and fits weights with manual scoring to obtain the comprehensive decision-making evaluation result for autonomous driving. This method can only perform a posteriori evaluation of the autonomous vehicle's own driving behavior, without involving the automatic charging system, and cannot identify integration scenarios between autonomous driving and automatic charging. Considering the integration potential of autonomous driving and automatic charging requires not only the collaborative design of the two technology systems but also the effective identification, value assessment, and technical feasibility analysis of various potential integration scenarios. Although knowledge graph technology has been widely applied in fields such as intelligent question answering, recommendation systems, and medical assistance, in the field of autonomous driving and energy management, the integration scenarios of autonomous driving and automatic charging are diverse, but the technical feasibility, economic benefits, and user experience of different scenarios vary significantly, and a mature mechanism for constructing and reasoning about the integration graph is still lacking. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the existing technology by providing a method and system for evaluating the integrated application scenarios of autonomous driving and automatic charging. This method and system enables a systematic evaluation of the integrated application scenarios of autonomous driving and automatic charging, and improves the scientific accuracy of identifying the most promising integrated application scenarios.

[0006] The objective of this invention can be achieved through the following technical solutions: A method for evaluating integrated autonomous driving and automatic charging application scenarios includes the following steps: In the knowledge graph of integrated applications, the graph database query language is used to find the association path between autonomous driving entities and automatic charging entities, and entities that are logically consistent and highly related are grouped into candidate integrated application scenarios. Numerical feature vectors for each candidate fusion application scenario are extracted from the fusion application knowledge graph using a graph embedding algorithm. The potential score of each candidate fusion application scenario under different evaluation indicators is obtained based on the numerical feature vector of each candidate fusion application scenario. The overall potential score of each candidate fusion application scenario is obtained based on the weight of the evaluation indicators and the potential score of each candidate fusion application scenario under different evaluation indicators. The posterior mean of each candidate fusion application scenario is obtained based on its overall potential score and its potential score under different evaluation indicators. All candidate fusion application scenarios are then ranked according to the posterior mean to identify the most promising autonomous driving and automatic charging fusion application scenario.

[0007] Furthermore, the construction steps of the fusion application knowledge graph specifically include: Collect literature, patents, industry technical reports, industry standards and practical data related to the fields of autonomous driving and automatic charging, form multi-source basic data and perform data preprocessing; An encoder based on a bidirectional long short-term memory network is used to obtain the contextual information of the multi-source basic data to obtain the text sentence feature representation vector; The text sentence feature representation vector is input into a deep neural network to identify and classify different entities in the fields of autonomous driving and automatic charging. A convolutional neural network is used to capture the local context features between different entities in the autonomous driving domain and the automatic charging domain. Features are extracted through convolution and pooling operations to extract entity relationships between different entities in the autonomous driving domain and the automatic charging domain, and the entity relationship extraction results are obtained. Cosine similarity calculation based on word frequency statistics is used to calculate the similarity between different entities in the fields of autonomous driving and automatic charging. Similar entities are filtered by threshold and synonymous entities are merged to complete entity fusion. Different entities in the fields of autonomous driving and automatic charging are used as nodes in a fusion application knowledge graph. The entity relationships between these different entities are connected to the nodes in the form of edges, thus forming the fusion application knowledge graph.

[0008] Furthermore, the specific steps for capturing local contextual features between the different entities using a convolutional neural network, extracting features through convolution and pooling operations, and extracting entity relationships between different entities to obtain the entity relationship extraction results include: The feature map is obtained through the convolution operation of the convolutional layer. The feature map is as follows: In the formula, For feature maps, For activation function, Label prediction for text sentence feature representation vectors. The weights of the text sentence feature representation vector. This is a vector representing the features of a text sentence. For entity category labels; Max pooling is performed in the pooling layer to obtain local feature information in the feature map; Based on the local feature information in the feature map, entity relationships between different entities are extracted to obtain entity relationship extraction results. The entity relationship extraction results are the probability distribution of relationships between different entities: In the formula, The probability distribution of relationships between different entities. For the first The component contains relationships The weight, For the first The component contains relationships The weight, The number of components.

[0009] Furthermore, the similarity between the different entities is: In the formula, For entities and entity Similarity between them For entities The word frequency vector obtained after word segmentation is the first one. dimensional elements, For entities The word frequency vector obtained after word segmentation is the first one. dimensional elements, denoted as the dimension of the elements in the word frequency vector.

[0010] Furthermore, the evaluation metrics include multiple metrics such as autonomous driving task success rate, charging equipment utilization rate, user experience quantification metrics, and additional empty driving mileage of autonomous vehicles.

[0011] Furthermore, the numerical feature vector of each candidate fusion application scenario includes multiple features such as entity frequency within the scenario, semantic path length, and relation strength. The numerical feature vector of the fusion application scenario is as follows: In the formula, For the first Numerical feature vectors of candidate fusion application scenarios This serves as an identifier for candidate integrated application scenarios. For the first One characteristic, For feature dimensions.

[0012] Furthermore, the potential score for each candidate fusion application scenario under different evaluation metrics is as follows: In the formula, For the first The candidate fusion application scenario in the first Potential scores under each evaluation indicator It follows a normal distribution. For the first The intercept term of each evaluation indicator, For the first The regression coefficient vector corresponding to each evaluation index For the first Numerical feature vectors of candidate fusion application scenarios For the first The variance of potential scores under each evaluation indicator.

[0013] Furthermore, the overall potential score for each of the aforementioned candidate fusion application scenarios is: In the formula, For the first The overall potential score of each candidate integrated application scenario. For the first The weight of each evaluation indicator, The number of evaluation indicators, For the first The candidate fusion application scenario in the first Potential scores under each evaluation indicator The attenuation coefficient is... For the first The evaluation metric is the shortest path length from the integrated application knowledge graph to the core entities in the field of autonomous driving. As a core entity in the field of autonomous driving, For the first The evaluation metric is the shortest path length from the integrated application knowledge graph to the core entity in the automatic charging field. As a core entity in the field of automatic charging, For the first The evaluation metric is the shortest path length from the integrated application knowledge graph to the core entities in the field of autonomous driving. For the first The evaluation metric is the shortest path length from the integrated application knowledge graph to the core entity in the field of automatic charging.

[0014] Furthermore, the specific steps for obtaining the posterior mean of each candidate fusion application scenario based on its potential score under different evaluation metrics and its overall potential score include: Using the Markov Monte Carlo sampling method, a posterior sample set of potential scores and overall potential scores for each candidate fusion application scenario under different evaluation metrics was obtained, and the posterior mean was calculated based on the posterior sample set. The posterior sample set is: In the formula, For the first The candidate fusion application scenario in the first Under the evaluation index, the first Potential score at the next sampling time For the first The first candidate fusion application scenario Overall potential score at the time of sampling. The number of samples in the Markov Monte Carlo sampling process. For the first Secondary sampling; The posterior mean is: In the formula, For the first The posterior mean of each candidate fusion application scenario.

[0015] According to another aspect of the present invention, an evaluation system for the integrated application scenario of autonomous driving and automatic charging is provided, comprising: The candidate fusion application scenario acquisition module is used to find the association path between autonomous driving entities and automatic charging entities in the fusion application knowledge graph using graph database query language, and to group entities that are logically consistent and highly related into candidate fusion application scenarios. The numerical feature vector extraction module is used to extract the numerical feature vector of each candidate fusion application scenario from the fusion application knowledge graph through a graph embedding algorithm. The potential score acquisition module for different evaluation indicators is used to acquire the potential score of each candidate fusion application scenario under different evaluation indicators based on the numerical feature vector of each candidate fusion application scenario. The overall potential score acquisition module is used to obtain the overall potential score of each candidate fusion application scenario based on the weight of the evaluation index and the potential score of each candidate fusion application scenario under different evaluation indexes. The module for identifying the most promising integrated application scenarios is used to obtain the posterior mean of each candidate integrated application scenario based on the overall potential score of each candidate integrated application scenario and its potential score under different evaluation indicators, and to rank all candidate integrated application scenarios based on the posterior mean to identify the most promising integrated application scenarios of autonomous driving and automatic charging.

[0016] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention constructs an integrated application knowledge graph and uses graph database query language to automatically find the relationship paths between entities. It also uses graph embedding algorithm to extract numerical features of the scene. By combining knowledge graph technology with multi-objective decision-making algorithm, it constructs a complete analysis framework from automatic scene discovery to potential quantitative assessment. This enables a systematic evaluation of the integrated application scenario of autonomous driving and automatic charging, overcoming the limitations of traditional experience-driven methods.

[0017] 2. This invention introduces an evaluation index system covering multiple dimensions such as the success rate of autonomous driving tasks and the utilization rate of charging equipment. It calculates the potential scores of candidate fusion application scenarios and ranks them according to the posterior mean. This establishes a mechanism for multi-dimensional quantitative evaluation of candidate fusion application scenarios, effectively balancing the weights and uncertainties among different evaluation indicators, and improving the scientificity and accuracy of identifying the most promising fusion application scenarios. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating an evaluation method for the integrated application scenario of autonomous driving and automatic charging proposed in this invention. Figure 2 This is a schematic diagram of the process for constructing the integrated application knowledge graph in this invention. Detailed Implementation

[0019] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0020] Example 1 This embodiment provides a method for evaluating integrated application scenarios of autonomous driving and automatic charging, such as Figure 1 As shown, it includes the following steps: S1. In the knowledge graph of the integrated application, the graph database query language is used to find the association path between autonomous driving entities and automatic charging entities, and the logically consistent and highly related entity combinations are summarized as candidate integrated application scenarios.

[0021] The steps for constructing a knowledge graph for integrated applications are as follows: Figure 2 As shown, it specifically includes: S101. Collect literature, patents, industry technical reports, industry standards and practical data related to the fields of autonomous driving and automatic charging to form multi-source basic data and perform data preprocessing.

[0022] The specific steps for preprocessing multi-source basic data include format unification, data cleaning, time synchronization and alignment, data standardization and normalization, data classification and labeling, and data fusion and noise reduction.

[0023] Data preprocessing ensures that multi-source foundational data from literature, patents, industry technical reports, industry standards, and practical materials are consistent, usable, and interpretable, providing support for subsequent knowledge graph construction.

[0024] S102. An encoder based on a bidirectional long short-term memory network is used to obtain contextual information from multiple sources of basic data, and the text sentence feature representation vector is obtained.

[0025] Based on multi-source basic data that has undergone data preprocessing, the original text sentences are first encoded and transformed. Through the encoding process, the original text sentences are uniformly converted into numerical feature representation vectors. Using a deep learning model based on a bidirectional long short-term memory network, information is extracted from both the beginning and end of the text, and then merged to form the text sentence feature representation vector.

[0026] The encoder forward propagation output value can be expressed by the formula: In the formula, This is the encoder forward propagation output value. For the forward encoding function of a long short-term memory neural network, For the first Bidirectional feature vectors of each text sentence. For the first The embedding vector of each text sentence.

[0027] The encoder's backpropagation output value can be expressed by the formula: In the formula, This is the encoder's backpropagation output value. For the back-encoding function of a long short-term memory neural network, For the first The bidirectional feature vectors of each text sentence.

[0028] By concatenating the forward propagation output and the back propagation output, the complete text sentence feature representation vector is obtained as follows: In the formula, This is a vector representing the features of a text sentence.

[0029] S103. Input the text sentence feature representation vector into a deep neural network to identify and classify different entities in the fields of autonomous driving and automatic charging.

[0030] Using the text sentence feature representation vector as input to a deep neural network, the feature representation of the input text sentence feature representation vector is obtained through neural network layers as follows: In the formula, The feature representation of a text sentence feature representation vector. For feature weights, This is a bias term.

[0031] The label prediction obtained from the text sentence feature representation vector in the output layer of the deep neural network is expressed by the formula: In the formula, Label prediction for text sentence feature representation vectors. For the input entity category label, The text feature vectors of the output layer. For activation function, For the characteristic transformation function, These are the training parameters for the hidden layer. These are the training parameters for the output layer.

[0032] The predicted label values ​​of the text sentence feature representation vectors correspond to the entity category labels of different entities in the fields of autonomous driving and automatic charging.

[0033] Entities in the field of autonomous driving include perception layer entities, execution layer entities, and application scenario entities. Perception layer entities include LiDAR, millimeter-wave radar, cameras, ultrasonic sensors, and GPS / IMU integrated navigation, among others. Execution layer entities include brake-by-wire systems, steering-by-wire systems, electronic throttles, and transmission controllers, among others. Application scenario entities include Robotaxi, long-haul logistics, unmanned delivery, last-mile delivery, autonomous driving in ports / mines, and automated parking, among others.

[0034] Entities in the field of automatic charging include charging equipment entities, energy entities, and charging strategy entities. Charging equipment entities include AC charging piles, DC fast charging piles, high-power supercharging piles, battery swapping stations, automatic charging robots, and wireless charging devices. Energy entities include power grids, photovoltaic systems, and energy storage systems. Charging strategy entities include scheduled charging, vehicle-to-grid (V2G) systems, scheduled charging, off-peak charging, and dynamic power regulation.

[0035] S104. A convolutional neural network is used to capture the local contextual features between different entities in the autonomous driving and automatic charging domains. Features are extracted through convolution and pooling operations to extract entity relationships between different entities in the autonomous driving and automatic charging domains, and the entity relationship extraction results are obtained.

[0036] The specific steps for using a convolutional neural network to capture local contextual features between different entities, extracting features through convolution and pooling operations, and extracting entity relationships between different entities to obtain entity relationship extraction results include: The feature map is obtained through the convolution operation of the convolutional layer. The feature map is as follows: In the formula, For feature maps, For activation function, Label prediction for text sentence feature representation vectors. The weights of the text sentence feature representation vector. This is a vector representing the features of a text sentence. For entity category labels; Max pooling is performed in the pooling layer to obtain local feature information in the feature map; Based on the local feature information in the feature map, entity relationships between different entities are extracted, resulting in entity relationship extraction results. The entity relationship extraction results are the probability distribution of relationships between different entities: In the formula, The probability distribution of relationships between different entities. For the first The component contains relationships The weight, For the first The component contains relationships The weight, The number of components.

[0037] S105. Using cosine similarity calculation based on word frequency statistics, calculate the similarity between different entities in the autonomous driving field and the automatic charging field, filter similar entities through threshold, and merge synonymous entities to complete entity fusion.

[0038] The similarity between different entities is: In the formula, For entities and entity Similarity between them For entities The word frequency vector obtained after word segmentation is the first one. dimensional elements, For entities The word frequency vector obtained after word segmentation is the first one. dimensional elements, denoted as the dimension of the elements in the word frequency vector.

[0039] The cosine similarity method based on word frequency statistics maps word vectors from the semantic space to the vector space, and evaluates entity similarity based on the cosine value between vectors; that is, the larger the cosine value, the more similar the two entities are. This step effectively maps semantic information to the vector space, ensuring that similar entities can be accurately identified and fused, thus supporting the construction of knowledge graphs.

[0040] S106. Take different entities in the autonomous driving field and the automatic charging field as nodes of the integrated application knowledge graph, and connect the nodes with the entity relationships between different entities in the autonomous driving field and the automatic charging field in the form of edges to form the integrated application knowledge graph.

[0041] A graph database (Neo4j) is used as the storage and query platform for the knowledge graph. In the graph database, different entities in the fields of autonomous driving and automatic charging are represented as nodes, and the entity relationships between different entities in the fields of autonomous driving and automatic charging are connected as edges to form a complete knowledge graph, realizing the structured expression of key concepts such as autonomous driving and automatic charging and their related content.

[0042] In the knowledge graph for integrated applications, graph database query language is used to find the association paths between autonomous driving entities and automatic charging entities, and entities that are logically consistent and highly related are grouped into candidate integrated application scenarios. The set of candidate integrated application scenarios is shown in Table 1: Table 1 Candidate Set of Application Scenarios for the Integration of Autonomous Driving and Automatic Charging S2. Extract the numerical feature vector of each candidate fusion application scenario from the fusion application knowledge graph using a graph embedding algorithm.

[0043] The numerical feature vector of each candidate fusion application scenario includes multiple features such as entity frequency within the scenario, semantic path length, and relation strength. The numerical feature vector of the fusion application scenario is as follows: In the formula, For the first Numerical feature vectors of candidate fusion application scenarios This serves as an identifier for candidate integrated application scenarios. For the first One characteristic, For feature dimensions.

[0044] S3. Obtain the potential score of each candidate fusion application scenario under different evaluation indicators based on the numerical feature vector of each candidate fusion application scenario.

[0045] The evaluation metrics include several of the following: success rate of autonomous driving tasks, utilization rate of charging equipment, quantitative indicators of user experience, and additional empty driving mileage of autonomous vehicles.

[0046] The potential scores of candidate fusion application scenarios under different evaluation metrics are not fixed values, but fluctuate randomly around a certain mean. Therefore, they are assumed to follow a normal distribution, and there is a linear relationship between scenario characteristics and metric potential. Considering the existence of random error, the potential scores of each candidate fusion application scenario under different evaluation metrics are as follows: In the formula, For the first The candidate fusion application scenario in the first Potential scores under each evaluation indicator It follows a normal distribution. For the first The intercept term of each evaluation indicator, For the first The regression coefficient vector corresponding to each evaluation index For the first Numerical feature vectors of candidate fusion application scenarios For the first The variance of potential scores under each evaluation indicator.

[0047] S4. Based on the weight of the evaluation indicators and the potential scores of each candidate fusion application scenario under different evaluation indicators, obtain the overall potential score of each candidate fusion application scenario.

[0048] The potential scores of each indicator are weighted and summed using the determined indicator weights to obtain the overall potential score for each candidate fusion application scenario. The overall potential score for each candidate fusion application scenario is as follows: In the formula, For the first The overall potential score of each candidate integrated application scenario. For the first The weight of each evaluation indicator, The number of evaluation indicators, For the first The candidate fusion application scenario in the first Potential scores under each evaluation indicator The attenuation coefficient is... For the first The evaluation metric is the shortest path length from the integrated application knowledge graph to the core entities in the field of autonomous driving. As a core entity in the field of autonomous driving, For the first The evaluation metric is the shortest path length from the integrated application knowledge graph to the core entity in the automatic charging field. As a core entity in the field of automatic charging, For the first The evaluation metric is the shortest path length from the integrated application knowledge graph to the core entities in the field of autonomous driving. For the first The evaluation metric is the shortest path length from the integrated application knowledge graph to the core entity in the field of automatic charging.

[0049] The core entities in the autonomous driving field and the core entities in the automatic charging field can be set according to actual needs. The shortest path length of entities in the knowledge graph is used as the basis for weighting, allowing the weights to dynamically change with the structure of the knowledge graph. For example, if the evaluation metric of charging equipment utilization is directly connected to a core entity in the automatic charging field with a path length of 1, its weight will be higher than that of evaluation metrics with longer paths.

[0050] If an evaluation metric has no connection to the core entities of either domain, resulting in an infinitely long path, its weight is close to 0, and it is automatically eliminated, avoiding subjective bias. The path minimization operation considers both autonomous driving and charging domains simultaneously, ensuring unbiased weight allocation. For example, if an evaluation metric is close to a core entity in one domain but far from the other, the function minimizes the distance to select the closest metric, prioritizing those with strong relevance to either domain.

[0051] The correlation between path length and the degree of association is shown in Table 2: Table 2. Correlation between path length and degree of association To illustrate the close connection, consider this example: the success rate of autonomous driving tasks in the evaluation metrics may be directly connected to the core entities in the field of autonomous driving through a relationship edge that directly affects or is directly used for evaluation.

[0052] For an example of a strong correlation, the utilization rate of charging equipment in the evaluation index may need to be indirectly connected to the core entities in the field of autonomous driving through intermediate entities such as automatic charging stations and service autonomous vehicles.

[0053] Examples of weak or no direct correlation: Some general evaluation indicators, such as system cost, may be far removed from the core entities in the field of autonomous driving, or require a long path to be associated with them.

[0054] S5. Based on the overall potential score of each candidate fusion application scenario and its potential score under different evaluation indicators, obtain the posterior mean of each candidate fusion application scenario. Sort all candidate fusion application scenarios according to the posterior mean and identify the most promising autonomous driving and automatic charging fusion application scenario.

[0055] The specific steps for obtaining the posterior mean of each candidate fusion application scenario based on its potential score under different evaluation metrics and its overall potential score include: Using the Markov Monte Carlo sampling method, a posterior sample set of potential scores and overall potential scores for each candidate fusion application scenario under different evaluation metrics was obtained, and the posterior mean was calculated based on the posterior sample set. The posterior sample set is: In the formula, For the first The candidate fusion application scenario in the first Under the evaluation index, the first Potential score at the next sampling time For the first The first candidate fusion application scenario Overall potential score at the time of sampling. The number of samples in the Markov Monte Carlo sampling process. For the first Secondary sampling; The posterior mean is: In the formula, For the first The posterior mean of each candidate fusion application scenario.

[0056] All candidate fusion application scenarios are ranked from highest to lowest according to their calculated posterior mean. The higher the posterior mean, the greater the overall potential of the scenario. A threshold can be set according to actual needs, or the top K scenarios can be defined as a high-potential scenario group for decision-makers to prioritize.

[0057] Example 2 This embodiment provides an evaluation system for the integrated application scenario of autonomous driving and automatic charging, including: The candidate fusion application scenario acquisition module is used to find the association path between autonomous driving entities and automatic charging entities in the fusion application knowledge graph using graph database query language, and to group entities that are logically consistent and highly related into candidate fusion application scenarios. The numerical feature vector extraction module is used to extract the numerical feature vector of each candidate fusion application scenario from the fusion application knowledge graph through graph embedding algorithm; The module for obtaining potential scores under different evaluation indicators is used to obtain the potential scores of each candidate fusion application scenario under different evaluation indicators based on the numerical feature vector of each candidate fusion application scenario. The overall potential score acquisition module is used to obtain the overall potential score of each candidate fusion application scenario based on the weight of the evaluation indicators and the potential score of each candidate fusion application scenario under different evaluation indicators. The module for identifying the most promising integrated application scenarios is used to obtain the posterior mean of each candidate integrated application scenario based on its overall potential score and its potential score under different evaluation indicators. Based on the posterior mean, all candidate integrated application scenarios are ranked to identify the most promising integrated application scenarios of autonomous driving and automatic charging.

[0058] The rest is the same as in Example 1.

[0059] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. An automatic driving and automatic charging integrated application scenario evaluation method, characterized in that, The method comprises the following steps: In the fusion application knowledge graph, the association path between the automatic driving entity and the automatic charging entity is found by using a graph database query language, and entities that meet the logic and have a high correlation degree are combined and summarized as candidate fusion application scenarios; A numerical feature vector of each candidate fusion application scenario is extracted from the fusion application knowledge graph by using a graph embedding algorithm; A potential score of each candidate fusion application scenario under different evaluation indexes is obtained according to the numerical feature vector of each candidate fusion application scenario; An overall potential score of each candidate fusion application scenario is obtained according to the weight of the evaluation index and the potential score of each candidate fusion application scenario under different evaluation indexes; A posterior mean of each candidate fusion application scenario is obtained according to the overall potential score of each candidate fusion application scenario and the potential score of each candidate fusion application scenario under different evaluation indexes, all candidate fusion application scenarios are sorted according to the posterior mean, and the most potential automatic driving and automatic charging fusion application scenario is identified.

2. The automatic driving and automatic charging fusion application scenario evaluation method according to claim 1, characterized in that, The construction steps of the fusion application knowledge graph specifically include: Collecting literatures, patents, industry technical reports, industry standards and practical materials related to the automatic driving field and the automatic charging field to form multi-source basic data and performing data preprocessing; Using an encoder based on a bidirectional long short-term memory network to obtain the context information of the multi-source basic data, and obtaining a text sentence feature representation vector; Inputting the text sentence feature representation vector into a deep neural network to identify and classify different entities in the automatic driving field and the automatic charging field; Using a convolutional neural network to capture the local context features between different entities in the automatic driving field and the automatic charging field, extracting features through convolution and pooling operations, extracting entity relationships between different entities in the automatic driving field and the automatic charging field, and obtaining an entity relationship extraction result; Using a cosine similarity calculation based on word frequency statistics to calculate the similarity between different entities in the automatic driving field and the automatic charging field, screening similar entities through a threshold, and merging synonymous entities to complete entity fusion; Taking different entities in the automatic driving field and the automatic charging field as nodes of the fusion application knowledge graph, connecting the nodes in the form of edges according to the entity relationships between different entities in the automatic driving field and the automatic charging field, and forming a fusion application knowledge graph.

3. The automatic driving and automatic charging fusion application scenario evaluation method according to claim 2, characterized in that, The specific steps of capturing the local context features between different entities by using a convolutional neural network, extracting features through convolution and pooling operations, and extracting entity relationships between different entities to obtain an entity relationship extraction result include: Obtaining a feature map through a convolution operation of a convolutional layer, wherein the feature map is: In the formula, is a feature map, is an activation function, is a label prediction of a text sentence feature representation vector, is a weight of a text sentence feature representation vector, is a text sentence feature representation vector, is an entity category label; Performing maximum pooling sampling in a pooling layer to obtain local feature information in the feature map; Extracting entity relationships between different entities according to the local feature information in the feature map to obtain an entity relationship extraction result, wherein the entity relationship extraction result is a relationship probability distribution between different entities: wherein is a probability distribution of relationships between different entities, is a probability distribution of relationships between different entities, is a weight of a relationship included in the jth component, is a weight of a relationship included in the jth component, is a weight of a relationship included in the jth component, and N is the number of components.

4. The automatic driving and automatic charging fusion application scenario evaluation method according to claim 2, characterized in that, The similarity between different entities is: In the formula, is the similarity between entities and entities , is the similarity between entities and entities , is the element in the word frequency vector obtained after word segmentation is the element in the word frequency vector obtained after word segmentation , is the element dimension in the word frequency vector.

5. The automatic driving and automatic charging fusion application scenario evaluation method according to claim 1, characterized in that, The evaluation indexes include multiple of automatic driving task success rate, charging device utilization rate, user experience quantitative index and additional driving mileage of the automatic driving vehicle.

6. The automatic driving and automatic charging fusion application scenario evaluation method according to claim 1, characterized in that, The features in the numerical feature vector of each candidate fusion application scenario include multiple of entity frequency, semantic path length and relationship strength in the scenario, and the numerical feature vector of the fusion application scenario is: In the formula, is the numerical feature vector of the first candidate fusion application scenario, is the identification of the candidate fusion application scenario, is the numerical feature vector of the first feature, is the feature dimension.

7. The automatic driving and automatic charging fusion application scenario evaluation method according to claim 1, characterized in that, The potential score of each candidate fusion application scenario under different evaluation indexes is: In the formula, For the first The candidate fusion application scenario in the first Potential scores under each evaluation indicator It follows a normal distribution. For the first The intercept term of each evaluation indicator, For the first The regression coefficient vector corresponding to each evaluation index For the first Numerical feature vectors of candidate fusion application scenarios For the first The variance of potential scores under each evaluation indicator. 8.The method of claim 1, wherein, The overall potential score of each candidate fusion application scenario is: In the formula, is the overall potential score of the first candidate fusion application scenario, is the weight of the first evaluation index, is the number of evaluation indexes, is the potential score of the first candidate fusion application scenario under the first evaluation index, is the decay coefficient, is the shortest path length of the first evaluation index to the core entity in the automatic driving field in the fusion application knowledge graph, is the core entity in the automatic driving field, is the shortest path length of the first evaluation index to the core entity in the automatic charging field in the fusion application knowledge graph, is the core entity in the automatic charging field, is the shortest path length of the first evaluation index to the core entity in the automatic driving field in the fusion application knowledge graph, is the shortest path length of the first evaluation index to the core entity in the automatic charging field in the fusion application knowledge graph.​​​​​​​​ 9.The automatic driving and automatic charging fusion application scenario evaluation method according to claim 1, characterized in that, The specific steps of obtaining the posterior mean of each candidate fusion application scenario according to the potential score of each candidate fusion application scenario under different evaluation indexes and the overall potential score include: The posterior sample set of the potential score of each candidate fusion application scenario under different evaluation indexes and the overall potential score is obtained through a Markov Monte Carlo sampling method, and the posterior mean is calculated according to the posterior sample set, The posterior sample set is: In the formula, For the first The candidate fusion application scenario in the first Under the evaluation index, the first Potential score at the next sampling time For the first The first candidate fusion application scenario Overall potential score at the time of sampling. The number of samples in the Markov Monte Carlo sampling process. For the first Secondary sampling; The posterior mean is: In the formula, is the posterior mean for the th candidate fusion application scenario.

10. An automatic driving and automatic charging integrated application scenario evaluation system, characterized in that, It includes: A candidate fusion application scenario acquisition module is configured to find an associated path between an automatic driving entity and an automatic charging entity in a fusion application knowledge graph by using a graph database query language, and combine entities with high logicality and high correlation into candidate fusion application scenarios; A numerical feature vector extraction module is configured to extract a numerical feature vector of each candidate fusion application scenario from the fusion application knowledge graph by using a graph embedding algorithm; A different evaluation index potential score acquisition module is configured to acquire a potential score of each candidate fusion application scenario under different evaluation indexes according to the numerical feature vector of each candidate fusion application scenario; An overall potential score acquisition module is configured to acquire an overall potential score of each candidate fusion application scenario according to the weight of an evaluation index and the potential score of each candidate fusion application scenario under different evaluation indexes; A most potential fusion application scenario identification module is configured to obtain a posterior mean of each candidate fusion application scenario according to the overall potential score of each candidate fusion application scenario and the potential score of each candidate fusion application scenario under different evaluation indexes, sort all candidate fusion application scenarios according to the posterior mean, and identify a most potential automatic driving and automatic charging fusion application scenario.

Images