Big model-based bicycle pile cooperative information feature identification and fusion method and system

Abstract

The invention relates to the technical field of large model application, and discloses a method and a system for identifying and fusing vehicle pile collaborative information features based on a large model. The semantic vector generation of the text information is realized through the pre-training large model, the text information matching accuracy is improved, meanwhile, the similarity vector corresponding to each pair of stations is generated by integrating multi-modal information such as text semantics, geographic positions and numerical attributes, and the same probability that each pair of stations are the same station is calculated, so that the matching accuracy of the text information is improved. The limitation of a traditional method in processing multi-modal information is solved, and more comprehensive and more accurate station comprehensive similarity evaluation is provided. And secondly, in the face of conflict information of different sources of the same station, the multi-modal information of the non-main station in each pair of stations can be fused to the main station, so that the problem of information loss after the stations are merged is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of large model application, in particular to a large model-based vehicle-pile cooperative information feature recognition and fusion method and system. BACKGROUND

[0002] With the rapid development and popularization of new energy vehicles, as a key support for the popularization of new energy vehicles, charging infrastructure, especially charging stations, is experiencing explosive growth. However, the current information management and utilization of charging stations face many challenges. Charging station information is scattered in different operators, third-party aggregation platforms, map service providers, and vehicle navigation systems, making it difficult for operation and maintenance personnel and users to manage and query uniformly. At the same time, charging station information not only includes structured data (such as latitude and longitude, number of charging piles, charging power, etc.), but also includes semi-structured data (such as user reviews, pictures, and videos) and unstructured data (such as station name, operator name, detailed address description, and service introduction).

[0003] In related technologies, to solve the above problems, methods based on rule matching, keyword search, simple geographic location clustering, or manual review are usually used. However, when dealing with massive, multi-source, heterogeneous, and dynamically changing charging station information, related technologies have difficulty in deeply understanding the implied semantic information of natural language texts such as station names, locations, and operator names, leading to problems such as synonym, near-synonym, and polysemy that cannot be effectively solved, affecting information matching and deduplication accuracy, and also unable to comprehensively and accurately assess the comprehensive similarity between stations. At the same time, in the face of conflicting information from different sources about the same station, related technologies are difficult to intelligently identify, judge, and resolve. SUMMARY

[0004] Therefore, the present application provides a large model-based vehicle-pile cooperative information feature recognition and fusion method and system to solve the problems of related technologies in deeply understanding the implied semantic information of natural language texts such as station names, locations, and operator names, leading to problems such as synonym, near-synonym, and polysemy that cannot be effectively solved, affecting information matching and deduplication accuracy, and also unable to comprehensively and accurately assess the comprehensive similarity between stations, and related technologies being difficult to intelligently identify, judge, and resolve conflicting information from different sources about the same station.

[0005] In a first aspect, the present application provides a large model-based vehicle-pile cooperative information feature recognition and fusion method, which comprises: Obtaining multi-modal information of a plurality of stations for vehicle-pile cooperation on one or more network platforms. The multi-modal information includes text information, geographic information, and numerical information.

[0006] According to a preset large model, a semantic vector corresponding to the text information of each station is obtained.

[0007] For each pair of stations in the plurality of stations, a same probability that each pair of stations is the same station is calculated based on the semantic vector corresponding to each pair of stations, the geographic information, and the numerical information.

[0008] If the same probability is greater than a preset threshold, it is determined that each pair of stations is the same station, and the multi-modal information of each pair of stations is fused.

[0009] The method for identifying and fusing information features of vehicle-pile coordination based on a large model provided in the embodiment realizes the generation of semantic vectors of text information by introducing a pre-trained large model such as SimCSE, improves the accuracy of text information matching, and solves the limitations of traditional methods in processing multi-modal information by integrating multi-modal information such as text semantics, geographic location, and numerical attributes and calculating the same probability that each pair of stations is the same station, thereby providing more comprehensive and accurate comprehensive similarity evaluation of stations. Secondly, when facing conflicting information from different sources of the same station, the multi-modal information from different sources can be fused.

[0010] In some optional embodiments, according to a preset large model, a semantic vector corresponding to the text information of each station is obtained, including: The text information of each station is standardized.

[0011] The standardized text information is input into the large model to construct a semantic vector corresponding to the text information of each station.

[0012] In some optional embodiments, for each pair of stations in the plurality of stations, a same probability that each pair of stations is the same station is calculated based on the semantic vector corresponding to each pair of stations, the geographic information, and the numerical information, including: The multi-modal information includes at least one type of text information and at least one category of numerical information.

[0013] The semantic similarity of each type of text information is calculated based on the semantic vector corresponding to the same type of text information of each pair of stations.

[0014] The geographic distance between each pair of stations is calculated based on the geographic information corresponding to each pair of stations.

[0015] The difference value of each category of numerical information is calculated based on the same category of numerical information of each pair of stations.

[0016] The same probability of each pair of stations is calculated based on the geographic distance, the semantic similarity of each type of text information, and the difference value of each category of numerical information.

[0017] In some optional embodiments, the difference value corresponding to the numerical information of each category is calculated according to the numerical information of each pair of stations in the same category, including: Obtaining the numerical information of each pair of stations, and processing the missing values and abnormal values in the obtained numerical information.

[0018] The numerical information of the same category is normalized and calculated for each pair of stations, and the difference value of each pair of the same category is determined.

[0019] In some optional embodiments, the geographical distance between each pair of stations is calculated according to the geographical information corresponding to each pair of stations, including: The geographical information of each pair of stations is converted into two sets of latitude and longitude.

[0020] According to the two sets of latitude and longitude, the original distance between each pair of stations is calculated, and standardized processing is performed to obtain the geographical distance between each pair of stations.

[0021] In some optional embodiments, the same probability of each pair of stations is calculated by the geographical distance of each pair of stations, the semantic similarity corresponding to each type of text information, and the difference value corresponding to each category of numerical information, including: The semantic similarity corresponding to each pair of the same type of text information of each pair of stations, the geographical distance between each pair of stations, and the difference value of each pair of the same category of numerical information of each pair of stations are constructed into a similarity feature vector of each pair of stations.

[0022] The same probability of each pair of stations is calculated according to the similarity feature vector corresponding to each pair of stations.

[0023] In some optional embodiments, the same probability is calculated in the following manner: wherein y is the same probability, is the algorithm for calculating the same probability for each pair of S1 and S2 stations, wherein the similarity feature vector , represents the semantic similarity corresponding to the text information of the same type, is the geographical distance, represents the difference value of the numerical information of the same category; a is the index of the semantic similarity, a = 1, 2, …, n, n is the number of text information types in the multi-modal information; b is the index of the difference value of the numerical information, b = 1, 2, …, m, m is the number of numerical information categories in the multi-modal information.

[0024] In some optional embodiments, the multi-modal information of each pair of stations is fused, including: Obtaining multimodal information of each pair of stations, determining the main station in each pair of stations.

[0025] Using multimodal information of another station other than the main station to complement the multimodal information of the main station.

[0026] Generating a unique identifier for the main station, and corresponding the original identifier of each pair of stations with the unique identifier.

[0027] In some optional embodiments, after the feature recognition and fusion of the plurality of stations are completed, the method further comprises: Calculating the deduplication rate and data integrity to evaluate the quality of the fused multimodal information of the stations.

[0028] According to the quality of the multimodal information, adjusting the preset large model and adjusting the algorithm for calculating the same probability.

[0029] The method for feature recognition and fusion of vehicle-pile collaborative information based on a large model provided in the embodiment realizes the generation of semantic vectors of text information by introducing a pre-trained large model such as SimCSE, improves the accuracy of text information matching, and at the same time integrates multimodal information such as text semantics, geographic location, and numerical attributes to generate a similarity vector corresponding to each pair of stations and calculate the same probability of each pair of stations for the same station, solving the limitations of traditional methods in processing multimodal information and providing more comprehensive and accurate comprehensive similarity evaluation of stations. Secondly, when facing conflicting information from different sources of the same station, the multimodal information of the non-main station in each pair of stations can be fused to the main station, solving the problem of information loss after station merging.

[0030] In a second aspect, the present application provides a system for feature recognition and fusion of vehicle-pile collaborative information based on a large model, which comprises: An information acquisition module is configured to acquire multimodal information of a plurality of stations for vehicle-pile collaboration from one or more network platforms. The multimodal information includes text information, geographic information, and numerical information.

[0031] A semantic vector module is configured to obtain semantic vectors corresponding to the text information of each station according to a preset large model.

[0032] A same probability module is configured to calculate the same probability of each pair of stations for the same station by the semantic vectors, geographic information, and numerical information corresponding to each pair of stations.

[0033] An information fusion module is configured to determine that each pair of stations is the same station if the same probability is greater than a preset threshold, and fuse the multimodal information of each pair of stations.

[0034] In a third aspect, the present application provides a computer device, comprising a memory and a processor, which are communicatively connected with each other, and the memory stores computer instructions, and the processor executes the computer instructions to perform the large model-based vehicle-pile coordination information feature recognition and fusion method in the first aspect or any of the corresponding embodiments thereof.

[0035] In a fourth aspect, the present application provides a computer readable storage medium, which stores computer instructions for causing a computer to perform the large model-based vehicle-pile coordination information feature recognition and fusion method in the first aspect or any of the corresponding embodiments thereof.

[0036] In a fifth aspect, the present application provides a computer program product, which comprises computer instructions for causing a computer to perform the large model-based vehicle-pile coordination information feature recognition and fusion method in the first aspect or any of the corresponding embodiments thereof. BRIEF DESCRIPTION OF DRAWINGS

[0037] In order to more clearly illustrate the specific embodiments or prior art technical solutions of the present application, the following will briefly introduce the drawings needed to be used in the specific embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creative labor.

[0038] Figure 1 is a flowchart of a large model-based vehicle-pile coordination information feature recognition and fusion method according to an embodiment of the present application; Figure 2 is a flowchart of another large model-based vehicle-pile coordination information feature recognition and fusion method according to an embodiment of the present application; Figure 3 is a structural block diagram of a large model-based vehicle-pile coordination information feature recognition and fusion system according to an embodiment of the present application; Figure 4 is a hardware structure schematic diagram of a computer device according to an embodiment of the present application. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical scheme and advantages of the embodiments of the present application more clear, the technical scheme in the embodiments of the present application will be clearly and completely described below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are some of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0040] The application provides a large model-based vehicle and pile cooperative information feature recognition and fusion method, which integrates and calculates the same probability of each pair of stations being the same station through a pre-trained large model, and provides comprehensive similarity evaluation. Secondly, when facing conflict information from different sources of the same station, the multi-modal information from different sources can be fused.

[0041] A large model-based vehicle and pile cooperative information feature recognition and fusion method is provided in the embodiment, which can be used in electronic devices such as computer devices, tablet computers, etc. Figure 1 The flowchart of the large model-based vehicle and pile cooperative information feature recognition and fusion method according to the embodiment of the application is shown in FIG. 1, which includes the following steps: Figure 1 In step S101, multi-modal information of a plurality of stations for vehicle and pile cooperation is obtained from one or more network platforms. The multi-modal information includes text information, geographic information, and numerical information.

[0042] In the application process of the embodiment, first, the multi-modal information of each station needs to be obtained. The information source can be obtained through one or more network platforms, such as an operator, a third-party aggregation platform, a map service provider, or a vehicle navigation system platform. The multi-modal information includes text information, geographic information, and numerical information. The text information can include station name, operator name, user evaluation, etc. The geographic information can include latitude and longitude, detailed address description, etc. The numerical information can include the number of charging piles, charging pile power, price, operating time, etc.

[0043] In step S102, a semantic vector corresponding to the text information of each station is obtained according to a pre-set large model.

[0044] After obtaining the text information of each station, the workload of directly comparing the original text in the comparison of the text information is very large, so a semantic vector corresponding to the text information of each station can be obtained through a pre-set large model. The semantic vector is a fixed-length numerical vector converted from the text, and the numerical vector can capture the deep semantic information of the text. The pre-set large model can be a semantic model such as Sentence-T5, BERT, or SimCSE. In the embodiment, SimCSE can be used as the pre-set large model.

[0045] In step S103, for each pair of stations in the plurality of stations, the same probability of each pair of stations being the same station is calculated through the semantic vector, geographic information, and numerical information corresponding to each pair of stations.

[0046] ​After obtaining the semantic vectors corresponding to the text information of each station, the probability that each pair of stations (i.e., two stations) in the plurality of stations is the same station can be obtained by calculating the semantic vectors, geographical information, and numerical information corresponding to the pair of stations. After calculating all possible combinations, the probability that each pair of stations is the same station, i.e., the same probability that each pair of stations is the same station, can be obtained.

[0047] In step S104, if the same probability is greater than the preset threshold, it is determined that each pair of stations is the same station, and the multi-modal information of each pair of stations is fused.

[0048] The preset threshold can be a default value or a value input by a user. The preset threshold has a value range of 0-1, for example, the preset threshold can be 80%. For each pair of stations in the plurality of stations, if the same probability of a pair of stations is greater than the preset threshold, it can be determined that the pair of stations is the same station, i.e., the pair of stations can be fused, i.e., the multi-modal information is fused. For example, the multi-modal information (i.e., text information, geographical information, and numerical information) of the other station in the pair of stations can be updated to the multi-modal information of one station in the pair of stations while keeping one station in the pair of stations unchanged.

[0049] The method for identifying and fusing vehicle-pile cooperative information features based on a large model provided in this embodiment realizes the generation of semantic vectors of text information by introducing a pre-trained large model such as SimCSE, improves the accuracy of text information matching, and solves the limitations of traditional methods in processing multi-modal information by integrating multi-modal information such as text semantics, geographical location, and numerical attributes and calculating the same probability that each pair of stations is the same station, thereby providing more comprehensive and accurate comprehensive similarity evaluation of stations. In addition, when facing conflicting information from different sources of the same station, the multi-modal information from different sources can be fused.

[0050] In this embodiment, a method for identifying and fusing vehicle-pile cooperative information features based on a large model is provided, which can be used in the electronic device described above, such as a computer device, a tablet computer, etc. Figure 2 The flowchart of the method for identifying and fusing vehicle-pile cooperative information features based on a large model according to an embodiment of the present application is shown in FIG. Figure 2 The flowchart includes the following steps: In step S201, multi-modal information of a plurality of stations for vehicle-pile cooperation is obtained from one or more network platforms. The multi-modal information includes text information, geographical information, and numerical information.

[0051] For details, please refer to the embodiment shown in Figure 1 The step S101 of the embodiment is not repeated here.

[0052] Step S202, according to a preset large model, obtaining the semantic vector corresponding to the text information of each station.

[0053] For details, please refer to Figure 1 Step S102 of the embodiment shown will not be described here.

[0054] In some optional embodiments, step S202 "according to a preset large model, obtaining the semantic vector corresponding to the text information of each station" includes step a1 and step a2.

[0055] Step a1, standardizing the text information of each station.

[0056] Step a2, inputting the standardized text information into the large model to construct the semantic vector corresponding to the text information of each station.

[0057] When obtaining the text information in the multi-modal information through the network platform, the text information may be incomplete or not uniform, for example, some text information may have HTML (Hyper Text Markup Language, Hyper Text Markup Language) tags, special symbols, extra spaces or line breaks, etc. Therefore, the text information needs to be standardized to unify the encoding format of the text; at the same time, if the text information is Chinese text, the text information can be segmented, which is the process of standardization. After the text information is standardized, the standardized text information can be input into the large model to construct the semantic vector corresponding to the text information of each station through the preset large model.

[0058] Step S203, for each pair of stations in the plurality of stations, calculating the same probability of each pair of stations for the same station through the semantic vector, geographical information and numerical information corresponding to each pair of stations.

[0059] For details, please refer to Figure 1 Step S103 of the embodiment shown will not be described here.

[0060] In some optional embodiments, in the case where the multi-modal information includes at least one type of text information and at least one category of numerical information, step S203 "for each pair of stations in the plurality of stations, calculating the same probability of each pair of stations for the same station through the semantic vector, geographical information and numerical information corresponding to each pair of stations" includes steps S2031 to S2034.

[0061] Step S2031, according to the semantic vector corresponding to the same type of text information of each pair of stations, calculating the semantic similarity corresponding to each type of text information respectively.

[0062] In the case where the multi-modal information includes at least one type of text information, it is necessary to match the types corresponding to each pair of station text information first, for example, the types can be station names, operators, etc. After the matching is completed, the semantic similarity corresponding to each pair of station text information of the same type can be calculated according to the semantic vectors corresponding to the text information of the same type of each pair of stations. The value of the semantic similarity is used to measure the semantic similarity degree corresponding to the text information of the same type of two stations. In the case of having multiple types of text information, it is necessary to calculate the semantic similarity corresponding to multiple types of text information respectively. In the process of calculating the semantic similarity, the cosine similarity calculation method can be used, or the Euclidean distance or dot product calculation method can be used. In this embodiment, the cosine similarity calculation method is taken as an example, that is, the cosine similarity is calculated through the semantic vectors, and the formula is as follows:

[0063] wherein, is the cosine similarity between the semantic vectors A and B, for example, it can be the station name A and the station name B in a pair of stations, is the semantic vector A, is the semantic vector B, the numerator is the dot product of the two vectors, and the denominator is the product of the lengths of the two vectors. The result range of the cosine similarity is [ 1,1], the closer to 1, the more similar the semantics, the closer to 0, the less relevant, and the closer to -1, the more opposite the semantics (which rarely occurs in actual application).

[0064] Step S2032, according to the geographical information corresponding to each pair of stations, the geographical distance between each pair of stations is calculated.

[0065] According to the geographical information of each pair of stations in the multi-modal information, the geographical distance between each pair of stations can be calculated, for example, the geographical distance between each pair of stations can be calculated through the latitude and longitude, or the geographical distance between each pair of stations can be calculated through the detailed address description by the large model.

[0066] In some optional embodiments, the step S2032 of "calculating the geographical distance between each pair of stations according to the geographical information corresponding to each pair of stations" includes steps b1 and b2.

[0067] Step b1, the geographical information of each pair of stations is converted into two sets of latitude and longitude.

[0068] Step b2, according to the two sets of latitude and longitude, the original distance between each pair of stations is calculated, and standardized processing is performed to obtain the geographical distance between each pair of stations.

[0069] In the process of converting geographic information into two sets of latitude and longitude, if the geographic information is latitude and longitude, it can be used directly, if the geographic information is a detailed address description, the two sets of latitude and longitude of the detailed address description can be obtained through a large model.

[0070] In the calculation of the original distance between each pair of stations, the Haversine formula can be used to calculate the great circle distance of geographic coordinate points on the surface of the earth, which takes into account the curvature of the earth and is more accurate than simple Euclidean distance in long-distance calculation. The Haversine formula is as follows:

[0071]

[0072]

[0073] wherein, and are the latitude in radian (corresponding to the two sets of latitude), and are the longitude in radian (corresponding to the two sets of longitude), and R is the radius of the earth, which can take an average value of 6371 kilometers. The d obtained is the original distance between each pair of stations.

[0074] After obtaining the original distance d between each pair of stations, standardization can be performed, such as Min-Max standardization (i.e., scaling the original distance d to the interval [0, 1]) or logarithmic transformation (log(1+d)) to reduce the influence of extreme distance values. The original distance d after standardization is the geographic distance between each pair of stations.

[0075] Step S2033, according to the numerical information of the same category of each pair of stations, the difference value corresponding to the numerical information of each category is calculated.

[0076] According to the numerical information of each pair of stations in the multi-modal information, the difference value corresponding to the numerical information can be calculated. As mentioned in the foregoing, in the case where the multi-modal information includes at least one type of text information, it is necessary to match the type corresponding to the text information of each pair of stations. Similarly, when the multi-modal information includes at least one category of numerical information, the difference value of the numerical information of the corresponding category needs to be calculated according to the corresponding category, wherein the category of the numerical information can be total pile number, direct current pile number, alternating current pile number, charging power, etc.

[0077] In some optional embodiments, step S2033 "according to the numerical information of the same category of each pair of stations, the difference value corresponding to the numerical information of each category is calculated", includes step c1 and step c2.

[0078] Step c1, obtain the numerical information of each pair of stations, and process the missing values and abnormal values in the obtained numerical information.

[0079] After obtaining the numerical information of a pair of stations, it is possible that one station in the pair has numerical information of a certain category, but the other station lacks numerical information of the category, or that a certain numerical information is not reasonable, for example, the total number of piles of a station is 4.5 (i.e. the total number of piles should not have a decimal), at which time the missing values and abnormal values in the numerical information need to be processed, for example, the missing values can be filled with average values, median values or specific marker values, and the abnormal values can be truncated or replaced.

[0080] Step c2, for the numerical information of each pair of stations, the numerical information of the same category is normalized and calculated to determine the difference value of each pair of the same category.

[0081] When calculating the difference value of the numerical information of the corresponding category according to the corresponding category, normalization calculation can be performed, and the result obtained is the difference after normalization. In this embodiment, the difference value of each pair of the same category can be determined by a plurality of normalization calculation methods. For example, relative difference calculation can be performed, which is suitable for measuring the difference in relative proportion, and the algorithm is as follows:

[0082] wherein, and are the numerical information of the same category in a pair of stations, for example, the total number of piles in a pair of stations, is the difference value of the numerical information calculated by the relative difference between and When =0, the difference value can be set to 0.

[0083] Range normalization difference calculation can also be performed. The maximum and minimum values of the numerical attribute in the entire data set can map the difference to the range of [0, 1], which is convenient for comparison between different numerical characteristics, and the algorithm is as follows:

[0084] wherein, and are the numerical information of the same category in a pair of stations, for example, the total number of piles in a pair of stations, is all values in the entire data set, for example, the total number of piles of all stations, is the difference value of the numerical information calculated by the relative difference between and a difference value of the numerical information calculated by the range normalized difference between and, when the denominator is 0, the difference value can be set to 0.

[0085] Z-score standardization and absolute difference calculation can also be performed, which can eliminate the influence of dimension, make the difference of different numerical characteristics more comparable, and have certain robustness to abnormal values. First, the Z-score standardization of each numerical value is performed, the standardized value obtained by Z-score standardization is :

[0086] wherein, is the mean of all numerical values in the numerical category, for example, it can be the mean of the total pile number of all stations, is the standard deviation. Then the standardized absolute difference is calculated:

[0087] wherein, is the difference value of the numerical information calculated by the range normalized difference between and, and is the same category of numerical information in a pair of stations.

[0088] Step S2034, the same probability of each pair of stations is calculated by the geographical distance of each pair of stations, the semantic similarity corresponding to each type of text information, and the difference value corresponding to each category of numerical information.

[0089] After obtaining the geographical distance of each pair of stations, the semantic similarity corresponding to each type of text information, and the difference value corresponding to each category of numerical information, the same probability of each pair of stations can be calculated according to the algorithm, for example, the same probability of each pair of stations can be calculated by direct comparison.

[0090] In some optional embodiments, step S2034 “calculating the same probability of each pair of stations by the geographical distance of each pair of stations, the semantic similarity corresponding to each type of text information, and the difference value corresponding to each category of numerical information” includes step d1 and step d2.

[0091] Step d1, the similarity feature vector of each pair of stations is constructed by the semantic similarity corresponding to each pair of the same type of text information of each pair of stations, the geographical distance between each pair of stations, and the difference value of each pair of the same category of numerical information of each pair of stations.

[0092] Step d2, the same probability of each pair of stations is calculated according to the similarity feature vector corresponding to each pair of stations.​​

[0093] After obtaining the semantic similarity corresponding to each pair of the same type of text information of each pair of stations, the geographical distance between each pair of stations, and the difference value of each pair of the same category of numerical information of each pair of stations, a similarity feature vector of each pair of stations can be constructed, that is, the semantic similarity, the geographical distance, and the difference value are integrated to generate a similarity feature vector. For example, the similarity feature vector between a pair of stations S1 and S2 in each pair of stations can be generated as follows:

[0094] wherein represents the semantic similarity corresponding to the same type of text information, n is the number of text information types in the multi-modal information, for example may be the station name text semantic similarity, is the geographical distance, represents the difference value of the numerical information of the same category, m is the number of numerical information categories in the multi-modal information, for example may be the total pile number difference value between the stations S1 and S2. After obtaining the similarity feature vector , whether the two stations belong to the same entity can be determined by a large model, that is, the probability y that the two stations belong to the same entity (i.e., the same probability of a pair of stations) is predicted:

[0095] wherein, is an algorithm for determining whether the two stations belong to the same entity according to the large model for a pair of stations S1 and S2 in each pair of stations.

[0096] Step S204: If the same probability is greater than a preset threshold, it is determined that each pair of stations is the same station, and the multi-modal information of each pair of stations is fused.

[0097] For details, please refer to the step S104 of the embodiment shown in Figure 1 , which will not be described here again.

[0098] In some optional embodiments, the step S204 of “fusing the multi-modal information of each pair of stations” includes steps e1 to e3.

[0099] Step e1: Obtain the multi-modal information of each pair of stations, and determine the main station in each pair of stations.

[0100] Step e2: Use the multi-modal information of another station other than the main station to complete the multi-modal information of the main station.

[0101] Step e3: Generate a unique identifier for the main station, and correspond the original identifier of each pair of stations to the unique identifier.​

[0102] In the case of information fusion, the priority of the multi-modal information of the pair of stations to be fused can be determined to determine the main station in the pair of stations to be fused, and the priority can be the station with the highest completeness of data categories (the most non-empty categories), the station with the latest information update time, and the station with information from an authoritative data source.

[0103] After determining the main station, the multi-modal information of the main station is supplemented using the multi-modal information of the other station. For example, if the main station does not have charging pile power information, but the other station has this information, the charging pile power information of the other station can be used to supplement the corresponding information of the main station. If there is a conflict (i.e., for the same type or category of text or numerical information, the main station and the other station both have information but are inconsistent), conflict processing can be performed. For example, numerical information (such as the total number of piles) can be subjected to maximum value, weighted average, and summation operations; the station name in the text information can retain the naming of the main station, and the naming of the other station as an alias, etc. After the main station information is supplemented, a unique identifier is generated for the main station, such as by generating a UUID or a hash value to generate a unique identifier, and the original identifier of each pair of stations is corresponded to the unique identifier.

[0104] In some optional embodiments, the method for identifying and fusing vehicle-pile coordination information features based on a large model further includes steps f1 and f2.

[0105] Step f1, calculate the deduplication rate and data integrity to evaluate the quality of the fused multi-modal information of the station.

[0106] Step f2, adjust the preset large model according to the quality of the multi-modal information, and adjust the algorithm for calculating the same probability.

[0107] After completing the construction of the method, training and verification can also be performed in advance. For example, the quality of the fused multi-modal information of the station can be evaluated by the deduplication rate and the data integrity, where the deduplication rate is the ratio of the total number of original stations to the total number of unique stations after fusion, reflecting the effectiveness of the fusion work, i.e.,

[0108] The data integrity is the filling rate of each key field (i.e., the type of text information and the category of numerical information) in the multi-modal information of the fused station, and whether the missing field is effectively supplemented compared to the original data source, i.e.,

[0109] According to the deduplication rate and the data integrity, the multi-modal information quality is obtained, according to the multi-modal information quality, the preset large model is adjusted, and the algorithm for calculating the same probability, that is, the model used for calculating the same probability, can also be evaluated by the accuracy, precision and recall.

[0110] The method provided by the embodiment realizes semantic vector generation of text information by introducing a pre-trained large model such as SimCSE, improves the accuracy of text information matching, and at the same time, generates a similarity vector corresponding to each pair of stations by integrating multi-modal information such as text semantics, geographic positions and numerical attributes, and calculates the same probability of each pair of stations for the same station, solves the limitations of traditional methods in processing multi-modal information, and provides more comprehensive and accurate comprehensive similarity evaluation of stations. Secondly, when facing conflict information of the same station from different sources, the multi-modal information of the non-main station in each pair of stations can be fused to the main station, solving the problem of information loss after station merging.

[0111] In the embodiment, a large model-based vehicle-pile cooperative information feature recognition and fusion system is also provided, which is used to implement the above-mentioned embodiments and preferred embodiments, and will not be described again. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the system described in the following embodiments is preferably implemented in software, hardware or a combination of software and hardware is also possible and is contemplated.

[0112] The embodiment provides a large model-based vehicle-pile cooperative information feature recognition and fusion system, as shown in Figure 3 , comprising: An information acquisition module 301 is configured to acquire multi-modal information of a plurality of stations for vehicle-pile cooperation on one or more network platforms. The multi-modal information includes text information, geographic information and numerical information.

[0113] A semantic vector module 302 is configured to acquire semantic vectors corresponding to text information of each station according to a preset large model.

[0114] A same probability module 303 is configured to calculate, for each pair of stations in the plurality of stations, a same probability of each pair of stations for the same station through semantic vectors, geographic information and numerical information corresponding to each pair of stations.

[0115] An information fusion module 304 is configured to determine that each pair of stations is the same station if the same probability is greater than a preset threshold, and fuse multi-modal information of each pair of stations.

[0116] In some optional embodiments, the semantic vector module 302 comprises: a standardization submodule configured to standardize the text information of each station.

[0117] a semantic vector submodule configured to input the standardized text information into a large model to construct semantic vectors corresponding to the text information of each station.

[0118] In the case where the multi-modal information includes at least one type of text information and at least one category of numerical information, in some optional embodiments, the same probability module 303 includes: a semantic similarity submodule configured to calculate semantic similarities corresponding to various types of text information according to semantic vectors corresponding to the same type of text information of each pair of stations.

[0119] a geographic distance submodule configured to calculate a geographic distance between each pair of stations according to geographic information corresponding to each pair of stations.

[0120] a numerical difference submodule configured to calculate difference values corresponding to various categories of numerical information according to the same category of numerical information of each pair of stations.

[0121] a same probability submodule configured to calculate a same probability of each pair of stations by the geographic distance, the semantic similarities corresponding to various types of text information, and the difference values corresponding to various categories of numerical information.

[0122] In some optional embodiments, the numerical difference submodule includes: a numerical repair unit configured to obtain numerical information of each pair of stations and process missing values and abnormal values in the obtained numerical information.

[0123] a difference calculation unit configured to normalize the same category of numerical information in the numerical information of each pair of stations to determine a difference value of each pair of the same category of numerical information.

[0124] In some optional embodiments, the geographic distance submodule includes: a longitude and latitude conversion unit configured to convert the geographic information of each pair of stations into two sets of longitude and latitude.

[0125] a geographic distance calculation unit configured to calculate an original distance between each pair of stations according to the two sets of longitude and latitude, and perform standardization processing to obtain a geographic distance between each pair of stations.

[0126] In some optional embodiments, the same probability submodule includes: The feature vector construction unit is configured to construct a similarity feature vector of each pair of stations according to the semantic similarity of each pair of same type of text information of the pair of stations, the geographical distance between the pair of stations, and the difference value of each pair of same category of numerical information of the pair of stations.

[0127] The same probability calculation unit is configured to calculate the same probability of each pair of stations according to the similarity feature vector corresponding to each pair of stations.

[0128] In some optional embodiments, the same probability calculation unit uses the following calculation method of the same probability: wherein y is the same probability, is an algorithm for calculating the same probability for each pair of stations S1 and S2, wherein the similarity feature vector , represents the semantic similarity of the same type of text information, is the geographical distance, represents the difference value of the same category of numerical information. a is the index of the semantic similarity, a = 1, 2, …, n, n is the number of text information types in the multi-modal information. b is the index of the difference value of the numerical information, b = 1, 2, …, m, m is the number of numerical information categories in the multi-modal information.

[0129] In some optional embodiments, the information fusion module 304 comprises: The main station determination sub-module is configured to obtain the multi-modal information of each pair of stations, and determine the main station in each pair of stations.

[0130] The information completion sub-module is configured to use the multi-modal information of another station other than the main station to complete the multi-modal information of the main station.

[0131] The identification generation sub-module is configured to generate a unique identification for the main station, and correspond the original identification of each pair of stations to the unique identification.

[0132] In some optional embodiments, the big model-based vehicle-pile cooperative information feature recognition and fusion system further comprises: The evaluation module is configured to calculate the deduplication rate and the data integrity, and evaluate the quality of the fused multi-modal information of the station.

[0133] The model adjustment module is configured to adjust the preset big model and the algorithm for calculating the same probability according to the quality of the multi-modal information.

[0134] The further function description of each module and unit is the same as the above-mentioned corresponding embodiments, and will not be repeated here.

[0135] In this embodiment, the vehicle-pile collaborative information feature identification and fusion system based on a large model is presented in the form of functional units. Here, a unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0136] This invention also provides a computer device having the above-described features. Figure 3 The system shown is a vehicle-pile collaborative information feature identification and fusion system based on a large model.

[0137] Please see Figure 4 , Figure 4 This is a schematic diagram of the structure of a computer device provided in an optional embodiment of the present invention, such as... Figure 4 As shown, the computer device includes one or more processors 10, memory 20, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components communicate with each other via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on external input / output devices (such as display devices coupled to the interfaces). In some alternative implementations, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system). Figure 4 Take a processor 10 as an example.

[0138] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GDA), or any combination thereof.

[0139] The memory 20 stores instructions executable by at least one processor 10 to cause the at least one processor 10 to perform the method shown in the above embodiments.

[0140] The memory 20 can include a program storage area and a data storage area, where the program storage area can store an operating system, application programs required for at least one function, and the data storage area can store data created according to the use of the computer device, etc. In addition, the memory 20 can include a high-speed random access memory, and can also include a non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state memory device. In some alternative embodiments, the memory 20 can optionally include a memory disposed remotely from the processor 10, which can be connected to the computer device through a network. Examples of the network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0141] The memory 20 can include a volatile memory, such as a random access memory, and can also include a non-volatile memory, such as a flash memory, a hard disk, or a solid state disk, and can also include a combination of the above-mentioned kinds of memories.

[0142] The computer device also includes an input device 30 and an output device 40. The processor 10, the memory 20, the input device 30, and the output device 40 can be connected by a bus or other means, Figure 4 The connection by the bus is taken as an example.

[0143] The input device 30 can receive inputted digital or character information, and generate key signal inputs related to the user settings and function controls of the computer device, such as a touch screen, a keypad, a mouse, a trackpad, a touchpad, a pointing stick, one or more mouse buttons, a trackball, a joystick, etc. The output device 40 can include a display device, an auxiliary lighting device (e.g., an LED), a tactile feedback device (e.g., a vibration motor), etc. The display device includes, but is not limited to, a liquid crystal display, a light-emitting diode, a display, and a plasma display. In some alternative embodiments, the display device can be a touch screen.

[0144] The embodiments of the present application further provide a computer readable storage medium, and the method according to the embodiments of the present application can be implemented in hardware, firmware, or recorded in a storage medium, or stored in a remote storage medium or a non-transitory machine readable storage medium and downloaded to a local storage medium through network, so that the method described herein can be processed by such software on a storage medium using a general purpose computer, a special purpose processor, or programmable or special hardware. The storage medium can be a magnetic disk, an optical disk, a read-only memory, a random access memory, a flash memory, a hard disk, or a solid state disk, etc. Further, the storage medium can also include a combination of the above-mentioned memories. It can be understood that the computer, the processor, the microprocessor controller, or the programmable hardware includes a storage component that can store or receive software or computer code, when the software or computer code is accessed and executed by the computer, the processor, or the hardware, the method shown in the above embodiments is implemented.

[0145] Part of the present application can be applied as a computer program product, for example, computer program instructions, when executed by a computer, through the operation of the computer, the method and / or technical solutions according to the present application can be called or provided. Those skilled in the art should understand that the form of computer program instructions in computer readable medium includes but is not limited to source file, executable file, installation package file, etc. Correspondingly, the way of computer program instructions executed by computer includes but is not limited to: the computer directly executes the instructions, or the computer compiles the instructions and then executes the corresponding compiled program, or the computer reads and executes the instructions, or the computer reads and installs the instructions and then executes the corresponding installed program. Here, the computer readable medium can be any available computer readable storage medium or communication medium accessible to the computer.

[0146] Although the embodiments of the present application are described in conjunction with the accompanying drawings, various modifications and changes can be made by those skilled in the art without departing from the spirit and scope of the present application, and such modifications and changes fall within the scope defined by the appended claims.

Claims

1. A method for identifying and fusing vehicle-pile cooperative information features based on a large model, characterized in that, The method includes: Multimodal information of multiple stations for vehicle-pile coordination is acquired through one or more network platforms; the multimodal information includes: text information, geographic information, and numerical information; Based on the pre-set large model, obtain the semantic vectors corresponding to the text information of each station; For each pair of stations among the plurality of stations, the probability that each pair of stations is the same station is calculated using the semantic vector, geographic information and numerical information corresponding to each pair of stations. If the same probability is greater than a preset threshold, then each pair of stations is determined to be the same station, and the multimodal information of each pair of stations is fused.

2. The method according to claim 1, characterized in that, The step of obtaining semantic vectors corresponding to the text information of each site based on a preset large model includes: Standardize the text information of each station; The standardized text information is input into the large model to construct semantic vectors corresponding to the text information of each station.

3. The method according to claim 1, characterized in that, For each pair of stations among the plurality of stations, the probability that each pair of stations belongs to the same station is calculated using the semantic vector, geographic information, and numerical information corresponding to each pair of stations, including: The multimodal information includes at least one type of text information and at least one category of numerical information; Based on the semantic vectors corresponding to the same type of text information for each pair of stations, calculate the semantic similarity of the text information for each type. Calculate the geographical distance between each pair of stations based on the geographical information corresponding to each pair of stations; Based on the numerical information of the same category for each pair of stations, calculate the difference value corresponding to the numerical information of each category; The probability of each pair of stations being identical is calculated by considering the geographical distance between each pair of stations, the semantic similarity of each type of text information, and the difference in numerical information of each category.

4. The method according to claim 3, characterized in that, The step of calculating the difference value corresponding to the numerical information of each category for each pair of stations, based on the numerical information of the same category, includes: Obtain the numerical information of each pair of stations, and process the missing and outlier values ​​in the obtained numerical information; For each pair of stations, the numerical information of the same category is normalized to determine the difference value of each pair of numerical information of the same category.

5. The method according to claim 3, characterized in that, The step of calculating the geographical distance between each pair of stations based on the geographical information corresponding to each pair of stations includes: The geographical information of each pair of stations is converted into two sets of latitude and longitude. Based on the two sets of latitude and longitude, the original distance between each pair of stations is calculated and then standardized to obtain the geographical distance between each pair of stations.

6. The method according to claim 3, characterized in that, The step of calculating the probability of each pair of stations being the same, based on the geographical distance between each pair of stations, the semantic similarity corresponding to each type of text information, and the difference value corresponding to each category of numerical information, includes: For each pair of stations, construct a similarity feature vector based on the semantic similarity of each pair of text information of the same type, the geographical distance between each pair of stations, and the difference value of each pair of numerical information of the same category. Based on the similarity feature vectors corresponding to each pair of stations, the probability of each pair of stations being the same is calculated.

7. The method according to claim 6, characterized in that, The method for calculating the same probability is as follows: Where y is the same probability. For each pair of S1 and S2 stations, an algorithm is used to calculate the same probability, where the similarity feature vectors... , The semantic similarity represents the semantic similarity of the text information of the same type. Geographical distance, The difference value represents the numerical information of the same category; 'a' is the semantic similarity index, a=1,2,...,n, where n is the number of text information types in the multimodal information; 'b' is the index of the difference value of the numerical information, b=1,2,...,m, where m is the number of numerical information categories in the multimodal information.

8. The method according to claim 1, characterized in that, The fusion of multimodal information for each pair of stations includes: Obtain the multimodal information of each pair of stations and determine the master station in each pair of stations; The multimodal information of the main station is supplemented using the multimodal information of another station besides the main station; A unique identifier is generated for the main station, and the original identifiers of each pair of stations are matched with the unique identifier.

9. The method according to claim 1 or 8, characterized in that, After feature identification and fusion are completed at multiple sites, the method further includes: Calculate the deduplication rate and data integrity, and evaluate the quality of the fused multimodal information of the stations; Based on the quality of the multimodal information, the preset large model is adjusted, and the algorithm for calculating the same probability is also adjusted.

10. A vehicle-pile cooperative information feature identification and fusion system based on a large model, characterized in that, The system includes: The information acquisition module is used to acquire multimodal information from multiple stations for vehicle-pile coordination on one or more network platforms; the multimodal information includes: text information, geographic information and numerical information; The semantic vector module is used to obtain the semantic vectors corresponding to the text information of each station based on the preset large model; The same probability module is used to calculate the same probability that each pair of stations is the same station for each pair of stations in the plurality of stations, based on the semantic vector, geographical information and numerical information corresponding to each pair of stations. The information fusion module is used to determine that each pair of stations is the same station if the same probability is greater than a preset threshold, and to fuse the multimodal information of each pair of stations.

Images