Fast matching search method based on addressing matrix

Abstract

The invention belongs to the field of identification networks of the Internet of Things, and discloses a quick matching and searching method based on an addressing matrix, which comprises the following steps of: mapping each category attribute into a coordinate value of a corresponding dimension in a multi-dimensional space according to a rule, and determining the position of a point in the multi-dimensional space; filling an addressing request in a preset addressing matrix, and sending the addressing request to an edge layer multi-dimensional identification server; the server extracts the attribute content filled in the matrix, and converts the extracted attribute value into a coordinate value of a corresponding dimension in a multi-dimensional space according to a mapping rule; and taking a point obtained by mapping according to the addressing matrix as the position of the addressing request in the multi-dimensional space, carrying out distance calculation with spatial points corresponding to all objects which are mapped and stored in a database, and selecting a communication object corresponding to a nearest point with a distance within a preset threshold range as an optimal communication target. The method solves the problems that an existing addressing mechanism is low in searching efficiency and limits the instant response capability of a network under the condition of large-scale identification data.

Description

Technical Field

[0001] The present invention belongs to the field of Internet of Things identification networks, and particularly relates to a fast matching and searching method based on an addressing matrix. Background Art

[0002] The traditional TCP / IP network is designed as a point-to-point communication protocol, focusing on the mutual communication between two terminal nodes. With the emergence of new network scenarios such as mobile Internet, industrial Internet, Internet of Things, mobile edge computing network, vehicle networking, and space-ground integrated network, the essence of network users' Internet access needs has also changed, gradually shifting from simple network access to network communication needs centered on obtaining content and services. However, the existing IP network system has been difficult to meet the new requirements of current and future users for network communication.

[0003] In the existing Internet, the mismatch problem between the architecture function and its main use is also becoming increasingly prominent. Many studies are thinking about the development direction of future network architectures, and the network architecture urgently needs to evolve. The common problem of traditional IP networks and newly emerging network architectures in recent years, including IPv6, content-centric networks, etc., is that they all advocate using a single network protocol or communication entity as the primary communication method of the network and reconstructing a difficult-to-expand network layer to achieve the goal.

[0004] Based on the existing network architecture, IP as the only addressing method has been difficult to adapt to diverse, pluralistic, and specialized application requirements. The homogeneous pattern of "thousands of troops crossing the single-plank bridge of IP" weakens the inherent diversity of the network system, restricts technological innovation and ecological vitality, and leads to a slow network evolution process.

[0005] To solve the above problems, multi-dimensional identification networks have emerged. Among them, multi-dimensional identification is a network identification based on spatio-temporal benchmarks, divided into three types: object, connection, and application, and uses diverse attribute combinations to describe all nodes in the network. The multi-dimensional identification network is based on multi-dimensional identification, realizes diverse addressing and routing methods, and supports the efficient interconnection of each node in the network. However, in practical applications, there are still many challenges in fast matching under high-dimensional attribute combinations. Facing large-scale identification data, frequently changing attributes dynamically, and high real-time requirements, existing addressing mechanisms mostly rely on linear search or fixed rule matching, making it difficult to achieve an efficient and scalable multi-dimensional matching process, restricting the instant response ability and resource scheduling efficiency of the network. Summary of the Invention

[0006] To solve the above technical problems, the present invention provides a fast matching and searching method based on an addressing matrix, which realizes the efficient positioning and approximate matching of multi-dimensional identification through attribute mapping and spatial distance calculation, and significantly improves the addressing processing performance in scenarios of large data volume, high concurrency, and multi-attribute combination.

[0007] A fast matching search method based on an addressing matrix according to the present invention includes the following steps: S1. Store the current multi-dimensional attributes for addressing in the edge layer database of the multi-dimensional identification network; S2. Map various category attributes into coordinate values of corresponding dimensions in the multi-dimensional space according to different rules to determine the positions of each point in the multi-dimensional space; wherein, each point in the multi-dimensional space corresponds to a registered or existing communication object, and the communication object is jointly located by multiple attribute dimensions, and its position in the multi-dimensional space is jointly determined by the attribute values of each dimension; S3. Fill in the addressing request in the preset addressing matrix, and send the addressing matrix filled with the addressing request to the edge layer multi-dimensional identification server; S4. The edge layer multi-dimensional identification server extracts the attribute content filled in the matrix, and converts the extracted attribute values into coordinate values of corresponding dimensions in the multi-dimensional space according to the mapping rules in S2; S5. Take the point mapped according to the addressing matrix as the position of the addressing request in the multi-dimensional space, calculate the distance from the space points corresponding to all the objects mapped and stored in the database in S2, and select the communication object corresponding to the nearest point within the preset threshold range as the best communication target.

[0008] Further, in S1, the current multi-dimensional attributes for addressing include device type attributes, geographical location attributes, and service identification attributes; wherein, the device type attributes include smart home, smart door lock, security camera, and smart electric meter; the geographical location attribute refers to the specific geographical area where the task occurs; the service identification attribute is used to mark a specific type of service, and different services are assigned independent identifiers.

[0009] Further, S2 is specifically as follows: S21. Based on the idea of Word2Vector, map the device type attributes into the multi-dimensional space through vectorization to generate the multi-dimensional coordinates of the device type, that is, the coordinate values of the first dimension; S22. Convert the longitude and latitude information of the geographical location attribute into a unique numerical representation as the coordinate values of the second dimension in the multi-dimensional space; S23. Judge the traffic volume of the service identification attribute, take the modulus in different ways according to the amount of traffic volume, and map the value to the predefined coordinate interval to obtain the multi-dimensional coordinates of the service identification, that is, the coordinate values of the third dimension.

[0010] Further, in S21, in the multi-dimensional identification network, the device type attributes need to be mapped into the multi-dimensional space through vectorization as strings. Based on the idea of Word2Vector, the generation of the multi-dimensional coordinates of the device type is completed, which specifically includes the following steps: S211. To ensure the standardization and consistency of the device type string, the original data is cleaned and standardized, including removing redundant characters, punctuation marks and irrelevant information; Defines the device type string , i represents the index of the device type string, and its value range is , where n is the total number of device type strings; Each device type string Represented as a sequence of characters, all of which come from the preset legal character set , ;in It is the complete set of all characters, including the valid character set and the redundant character set; A string indicating the device type The lth character in , m is the device type string The actual number of characters that satisfies , N is the maximum length defined by the system; define the valid character set , including English letters, numbers or other specific characters; Traverse the device type string , retain the characters that belong to the valid character set, and remove other characters. The screening formula is: ; Define the set of punctuation marks to be removed and irrelevant character sets , traverse and match from valid characters Remove the Characters: ; Convert the cleaned string to lowercase to complete the standardization process: ; S212, encoding based on semantic similarity to capture semantic relationships between device types; Defining the semantic similarity matrix ,in, A string indicating the device type Device classes related to their semantics The semantic similarity of the type is calculated by cosine similarity: , in, and A string indicating the device type And its related equipment The word vector of and for and The vector norm; d is the dimension of the word vector, is the set of real numbers; S213. Map the semantic feature vector to the coordinate values in the multi-dimensional space using the word vector model; Construct a corpus and train word vectors based on the device type set, with the goal of maximizing the conditional probability, where, represents the set of device types included in the context window centered on the device type string : , where the conditional probability is calculated by the Softmax function: , where, represents the index of all candidate device types in the device type set, is the word vector representation of the k-th device type; after training, organize the word vectors of all device types into a matrix V, where : ; To eliminate the influence of different dimensional value ranges, normalize the matrix V column-wise, where and are the mean and standard deviation of the j-th dimension respectively, and the element in the matrix represents the value of the device type string on the j-th dimension of the word vector. The normalized matrix is : , , ; Calculate the covariance matrix of the normalized matrix , represents the transpose matrix of the normalized matrix : ; Perform eigenvalue decomposition on the covariance matrix C: , , ; where, is the eigenvector matrix, is the transpose matrix of the eigenvector matrix , is the eigenvalue matrix, and the first principal component direction Corresponding to the maximum eigenvalue; select the first principal component direction , project the normalized word vector of each device type string onto this direction to obtain the coordinate value : ; Finally, the high-dimensional word vector of each device type is reduced to a single numerical value , and this value is used as the coordinate value of the first dimension of the point in the multi-dimensional space to participate in subsequent point distance calculation and matching.

[0011] Furthermore, in S22, in the multi-dimensional identification network, the geographical location is described by longitude and latitude data. The longitude and latitude information is converted into a unique numerical representation as the coordinate value of the second dimension in the multi-dimensional space, which specifically includes the following steps: S221. Let the longitude of the geographical location be x and the latitude be y, and their value ranges are respectively: ; Perform normalization processing on the longitude and latitude and map them to the interval : ; The normalized longitude , the normalized latitude ; S222. According to the normalization result, convert into an integer grid coordinate. Let the coding precision be b bits, and the grid coordinate calculation formula is: , , where, represents the position of the longitude and latitude in the grid, is the floor operation; S223. Binary encode the grid coordinates and respectively, and the encoding results are respectively represented as and : , ; S224. Use the Z-shaped curve to alternately splice the binary bits to generate an overall binary string, representing the unique encoding of the geographical location: ; S225. The spliced binary string Convert it to a decimal number as the unique mapping value of the geographical location: , This value is used as the second-dimensional coordinate value of the point in the multi-dimensional space.

[0012] Furthermore, in S23, in the multi-dimensional identification network, the service identifier h is a 128-bit digital code. When the type of the service identifier is less than three digits, directly take the modulus of the 128-bit number of the service identifier by 10 to obtain a single digit as the mapping value, that is, the third-dimensional coordinate value; then ; When the type of the service identifier gradually increases to three digits or more, to avoid uneven distribution caused by direct modulo operation, it is necessary to first perform a hash process on the service identifier h, denoted as , and then perform a modulo operation on by 10, and assign the final result to as the coordinate value of the third dimension in the multi-dimensional space.

[0013] Furthermore, in S4, using the Euclidean distance as the standard for evaluating the best object, assuming the point obtained by attribute mapping in the addressing matrix is , and any point in the multi-dimensional space is , then the distance between the two points is defined as: ; where , , respectively represent the device type, geographical location, and service identifier coordinate value of the point obtained by attribute mapping in the addressing matrix; , , respectively represent the device type, geographical location, and service identifier coordinate value of any point in the multi-dimensional space; Select the point with the shortest distance from point A within the limited range as the best communication object.

[0014] The beneficial effects of the present invention are as follows: 1) The present invention performs node addressing through the combination of multi-dimensional attributes, maps the multi-dimensional attributes of the node into multi-dimensional coordinate points, which is different from the traditional single ID addressing method, ensures the accurate matching of task requirements, and thus improves the success rate of task execution and the overall performance of the system; 2) The present invention constructs a dynamic matching mechanism by calculating the Euclidean distance between the node mapping point and the target point, accurately quantifies the adaptability between the node and the task, and optimizes the rationality of task scheduling; Compared with the traditional multi-condition database polling mechanism, the present invention has certain advantages in reducing communication latency and resource waste caused by matching errors; 3) Through a unified numerical mapping and a simple distance calculation model, the present invention transforms the complex multi-attribute addressing problem into a point distance problem in geometric space, avoiding the computational overhead of complex rules and conditional judgments in traditional methods, significantly improving the operating efficiency of the system, and providing guarantee for real-time task processing at the same time; In summary, through the hierarchical organization and matrix expression of the multi-dimensional attributes of nodes, the present invention effectively improves the information matching efficiency, realizes fast search in scenarios of large scale, high concurrency, and multi-attribute combination, guarantees the efficient operation and wide applicability of the network system, and is applicable to fields with complex task requirements such as dynamic resource matching of Internet of Things devices and high requirements for real-time performance, and has broad application prospects. Brief Description of the Drawings

[0015] Figure 1 is a flowchart of the method described in the present invention; Figure 2 is a schematic diagram of the mapping rule of device types in the method described in the present invention; Figure 3 is a schematic diagram of the mapping rule of geographical locations in the method described in the present invention; Figure 4 is a schematic diagram of the performance comparison between the present invention and other search algorithms under different database scales. Detailed Embodiments

[0016] In order to make the content of the present invention easier to be clearly understood, the present invention will be further described in detail below according to specific embodiments in conjunction with the accompanying drawings.

[0017] As Figure 1 shown, a fast matching search method based on an addressing matrix described in the present invention includes the following steps: S1. In a multi-dimensional identification network, the multi-dimensional attributes currently available for addressing mainly include device type, geographical location, and service identification. Therefore, the edge layer database stores the above attributes and other relevant information to ensure the accuracy and efficiency of network addressing. The device types mainly include: various nodes such as smart home, smart door lock, security camera, smart electric meter, etc.; the geographical location refers to the specific geographical area where the task occurs; the service identification is used to mark a specific type of service, and different services are assigned independent identifications, and this identification, as one of the three types of multi-dimensional identifications, is assigned to each node by the multi-dimensional identification server.

[0018] S2. Before search and matching, different attributes in the edge layer database need to adopt corresponding mapping mechanisms according to their data types or their own characteristics, and finally be mapped to numerical values as the coordinate values of the corresponding dimensions in the multi-dimensional space. These coordinate values together determine the position of this point in the multi-dimensional space; S3. When the addressing request in the addressing matrix is ​​sent to the edge layer multi-dimensional identification server, the server extracts the attribute content filled in the matrix and maps the attributes to coordinate values ​​of different dimensions in the multi-dimensional space according to the mapping rules described in S2; S4. Taking Euclidean distance as the criterion for evaluating the best object, let the point obtained by attribute mapping in the addressing matrix be , any point in the multidimensional space is , then the distance between two points is defined as: ; The point with the shortest distance to point A within the limited range is selected as the best communication partner.

[0019] Among them, the specific implementation steps of S2 are as follows:

[0020] S21. In the multidimensional identification network, the device type attribute as a string needs to be mapped to the multidimensional space through vectorization; based on the idea of ​​Word2Vector, the multidimensional coordinate generation of the device type is completed, such as Figure 2 As shown, the specific steps include: S211. To ensure the standardization and consistency of the device type string, the original data needs to be cleaned and standardized, including removing redundant characters, punctuation marks and irrelevant information. Define the device type string , i represents the index of the device type string, and its value range is , where n is the total number of device type strings; Each device type string Represented as a sequence of characters, all of which come from the preset legal character set , ;in It is the complete set of all characters, including the valid character set and the redundant character set; A string indicating the device type The lth character in , m is the device type string The actual number of characters that satisfies , N is the maximum length defined by the system; define the valid character set , including English letters, numbers or other specific characters; Traverse the device type string , retain the characters that belong to the valid character set, and remove other characters. The screening formula is: ; Define the set of punctuation marks to be removed and irrelevant character sets , traverse and match from valid characters Remove the Characters: ; Convert the cleaned string to lowercase to complete the standardization process: ; S212. Define the semantic similarity matrix , where represents the semantic similarity between the device type string and its semantically related device class type, and the similarity is calculated by cosine similarity: , where and represent the word vectors of the device type string and its related device ; and are and the vector norms of S213. Use the word vector model to map the semantic feature vector to the coordinate values in the multi-dimensional space; Build a corpus based on the device type set and train the word vectors. The goal is to maximize the conditional probability, where represents the set of device types included in the context window centered on the device type string : , where the conditional probability is calculated by the Softmax function: , where represents the index of all candidate device types in the device type set, is the word vector representation of the k-th device type; after training, organize the word vectors of all device types into a matrix V, where : ; To eliminate the influence of different dimensional value ranges, standardize the matrix V column-wise, where and are the mean and standard deviation of the j-th dimension respectively, and the element in the matrix represents the value of the device type string on the j-th dimension of the word vector. The standardized matrix is : , , ; Calculate the covariance matrix of the standardized matrix , denote the transpose matrix of the standardized matrix : ; Perform eigenvalue decomposition on the covariance matrix C: , , ; where is the eigenvector matrix, is the transpose matrix of the eigenvector matrix , is the eigenvalue matrix, and the first principal component direction corresponds to the largest eigenvalue; select the first principal component direction , and project the standardized word vector of each device type string onto this direction to obtain the coordinate value : ; Finally, the high-dimensional word vector of each device type is reduced to a single value , and this value is used as the coordinate value of the first dimension of the point in the multi-dimensional space to participate in subsequent point distance calculation and matching.

[0021] S22. In the multi-dimensional identification network, the geographical location is described by longitude and latitude data, where the longitude range is , and the latitude range is . Convert the longitude and latitude information into a unique numerical representation as the coordinate value of the second dimension in the multi-dimensional space. As shown in Figure 3 , the specific steps are as follows: S221. Let the longitude of the geographical location be x and the latitude be y, and their value ranges are respectively: ; Perform normalization on the longitude and latitude and map them to the interval : ; The normalized longitude , and the normalized latitude ; S222. According to the normalization result, convert into integer grid coordinates. Let the coding precision be b bits, and the grid coordinate calculation formula is: , , Among them, represents the position in the grid where the longitude and latitude are in the grid, is the floor operation; S223. Binary-encode the grid coordinates and respectively, and the encoding results are represented as and : , ; S224. Use the Z-shaped curve to alternately splice the binary bits to generate an overall binary string, representing the unique encoding of the geographical location: ; S225. Convert the spliced binary string into a decimal number as the unique mapping value of the geographical location: , This value is used as the second-dimensional coordinate value of the point in the multi-dimensional space.

[0022] S23. In the multi-dimensional identification network, the service identification is a 128-bit digital code. When the current number of services is small, directly take the modulus of these 128 bits by 10 to obtain a single-digit number as the mapping value. Let the service identification be h, then . If the service identification increases to three digits or more, to avoid uneven distribution caused by direct modulo operation, first perform a hash processing on the service identification h, denoted as , and then take the modulus of by 10, and finally assign the result to as the third-dimensional coordinate value in the multi-dimensional space.

[0023] To make the present invention easier to understand, the following takes the addressing scenario of a certain device in a smart industrial park as an example for illustration.

[0024] The edge node receives the following addressing request: Device type: "Smart_Conveyor_V1" Geographical location: longitude x = 113.9356, latitude y = 22.5460 Service identification: hexadecimal encoding is 0xDAE345AF03BCD99012 The system performs mapping processing according to the preset rules: The string is standardized to "smartconveyorv1", and its high-dimensional word vector is obtained using the trained word vector model. After principal component dimensionality reduction, the projection value on the first principal component is: , which serves as the first-dimensional coordinate of the device type in the multi-dimensional space.

[0025] In terms of geographical location, after normalizing the original longitude and latitude, the system maps them to integer coordinate values using the Z-curve encoding method. Assuming the encoding precision is 10 bits, the geographical space is correspondingly divided into grids. After binary interleaved splicing and decimal conversion of the normalized coordinates, the finally generated geographical mapping value is 1039232, which serves as the second-dimensional coordinate of the multi-dimensional space.

[0026] For the business identification field, to improve discreteness and stability, the system first performs hashing using the hash algorithm and takes the modulus of the obtained result with 10. The finally generated coordinate value is 7, corresponding to the third-dimensional coordinate. Thus, the addressing request is mapped to a point (0.384, 1039232, 7) in the multi-dimensional space.

[0027] The system retrieves all registered communication objects from the edge layer database and calculates the Euclidean distance between each of them and the current request point one by one. Suppose the spatial coordinates of a candidate object are (0.390, 1039240, 6), and the calculated distance between the two is approximately 8.06. According to the maximum matching distance threshold set by the system (which can be set to 10 for this search), it is determined that the candidate object meets the matching condition and is finally determined as the best communication target node.

[0028] To verify the actual performance advantages of the method of the present invention in multi-dimensional matching scenarios, the method of the present invention is compared and evaluated with three typical retrieval algorithms (Euclidean distance algorithm, Manhattan distance algorithm, and K-means clustering algorithm) in terms of operation efficiency under different database scales. The experiment is implemented in the Python 3.8 environment, using NumPy for data generation and distance calculation, scikit-learn to implement K-means clustering, and Matplotlib for result visualization. To fairly evaluate the response efficiency of different algorithms to addressing requests in the multi-dimensional space, the experiment constructs data sets of multiple scales, simulates the multi-dimensional coordinate information of pre-registered communication objects in the edge database, and statistically analyzes the matching time of each algorithm.

[0029] A set of three-dimensional random coordinate points in the range of [0,1000] is generated by np.random.rand(db_size,3)*1000 to simulate the three-dimensional vector representation of the combination of device type mapping, geographic location encoding and business identification. The test point is fixed to [500,500,500] to simulate the spatial location of a typical addressing request. The experiment selects six groups of database sizes, namely 1000, 2000, 3000, 5000, 8000 and 10000 data points. Each group of experiments re-randomly generates data and uses np.random.seed(42) to ensure that the results are reproducible.

[0030] The experimental results are as follows Figure 4 As shown in the figure, the curve of the method of the present invention is always close to the horizontal axis. As the number of data points increases from 1000 to 10000, the execution time is always maintained at about 0.002 seconds, showing only a very slow linear growth trend. This shows that the method has good scalability and efficiency, and can maintain fast response capabilities even when the amount of data increases sharply.

[0031] In contrast, the performance of the traditional Euclidean distance search method shows a clear linear growth trend. The results corresponding to the orange curve show that when the data size reaches 5,000, the search time exceeds 0.03 seconds, and when it reaches 10,000 entries, it is almost close to 0.05 seconds. The processing logic of this method is to traverse all data points to calculate the Euclidean distance and select the smallest one. The complexity is , when the data scale increases, the search time increases sharply, which is not suitable for business scenarios with high concurrency or high response time requirements. Similar to Euclidean search, the operating efficiency of the Manhattan distance search method also increases rapidly with the increase in the number of data points. Although the Manhattan distance calculation itself is simpler than the Euclidean distance in theory, the actual experimental results show that the execution time difference between the two is not obvious under the scale of large data, and the overall time consumption of the Manhattan method is even slightly higher. The K-means clustering search method showed certain disadvantages in the early stage of the experiment. Its initial running time was higher than other methods, reaching about 0.01 seconds at the scale of 1000 data points. This is mainly due to the overhead caused by its cluster initialization and center point recalculation. However, as the amount of data increases, the curve of this method changes relatively smoothly, indicating that it has a certain ability to resist scale. However, since this method only makes a rough judgment based on the distance between the cluster center and the query point, it lacks a fine capture of the original data distribution, so it has certain limitations in matching accuracy and applicable scenarios. It is mainly suitable for application scenarios with low accuracy requirements, heavy rough positioning or candidate screening.

[0032] The above experimental results illustrate the feasibility and practicality of the present invention. The method described in the present invention maintains the lowest execution time-consuming under different data scales, far superior to the Euclidean and Manhattan methods, and significantly superior to the K-means clustering approximate search. The method described in the present invention has excellent scalability and real-time performance when dealing with high-dimensional, multi-attribute, large-scale addressing tasks, and is particularly suitable for being deployed in scenarios with high throughput and low latency requirements such as the Internet of Things, edge intelligence, and industrial device interconnection, and can effectively support the rapid response and precise matching of high-frequency dynamic addressing requests.

[0033] The above is only the preferred solution of the present invention, and is not intended to further limit the present invention. All equivalent changes made by using the content of the specification and drawings of the present invention are within the protection scope of the present invention.

Claims

1. A fast matching search method based on an addressing matrix, characterized in that It includes the following steps: S1. Store the multi-dimensional attributes currently used for addressing in the edge layer database of the multi-dimensional identification network; S2. Map various types of attributes into coordinate values of corresponding dimensions in the multi-dimensional space according to different rules to determine the positions of each point in the multi-dimensional space; among them, each point in the multi-dimensional space corresponds to a registered or existing communication object; S3. Fill in the addressing request in the preset addressing matrix and send the addressing matrix after filling in the addressing request to the edge layer multi-dimensional identification server; S4. The edge layer multi-dimensional identification server extracts the attribute content filled in the matrix and converts the extracted attribute values into coordinate values of corresponding dimensions in the multi-dimensional space according to the mapping rules in S2; S5. Take the point obtained by mapping according to the addressing matrix as the position of the addressing request in the multi-dimensional space, calculate the distance from the space points corresponding to all objects that have been mapped and stored in the database in S2, and select the communication object corresponding to the nearest point within the preset threshold range as the best communication target.

2. The fast matching search method based on an addressing matrix according to claim 1, wherein In S1, the multi-dimensional attributes currently used for addressing include device type attributes, geographical location attributes, and service identification attributes; among them, the device type attributes include smart home, smart door lock, security camera, and smart meter; the geographical location attribute refers to the specific geographical area where the task occurs; the service identification attribute refers to marking a specific type of service, and different services are assigned independent identifiers.

3. A fast matching search method based on an addressing matrix according to claim 2, characterized in that, S2 is specifically as follows: S21. Based on the idea of Word2Vector, map the device type attributes into the multi-dimensional space through vectorization to generate the multi-dimensional coordinates of the device type, that is, the coordinate values of the first dimension; S22. Convert the longitude and latitude information of the geographical location attribute into a unique numerical representation as the coordinate value of the second dimension in the multi-dimensional space; S23. Judge the traffic volume of the service identification attribute, take the modulus in different ways according to the amount of traffic volume, and map the value to the predefined coordinate interval to obtain the multi-dimensional coordinates of the service identification, that is, the coordinate values of the third dimension.

4. A fast matching search method based on an addressing matrix according to claim 3, characterized in that S21. In the multi-dimensional identification network, the device type attribute, as a string, needs to be mapped into the multi-dimensional space through vectorization. Based on the idea of Word2Vector, the generation of the multi-dimensional coordinates of the device type is completed, specifically including the following steps: S211. To ensure the standardization and consistency of the device type string, clean and standardize the original data, including removing redundant characters, punctuation marks, and irrelevant information; Define the device type string , where i represents the index of the device type string, and the value range is , where n is the total number of device type strings; where each device type string is represented as a character sequence, and the characters are all from a preset legal character set , ; where is the universal set of all characters, including the valid character set and the redundant character set; represents the l-th character in the device type string , m is the actual number of characters in the device type string , satisfying , and N is the maximum length defined by the system; define the valid character set , including English letters, numbers, or other specific characters; ​ Traverse the device type string , retain the characters that belong to the valid character set, and remove other characters. The screening formula is: ； Define the set of punctuation marks to be removed and the set of irrelevant characters Traverse and match to remove characters belonging to from the valid characters : ; Convert the cleaned string to lowercase to complete the standardization process: ; S212. Encode based on semantic similarity to capture the semantic relationship between device types; Define the semantic similarity matrix , where represents the device type string and its semantically related device class type's semantic similarity. The similarity is calculated by cosine similarity: ， Among them, and represent the device type string and its related devices word vectors, and are and vector norms of; d is the dimension of the word vector, is the set of real numbers; S213. Use the word vector model to map the semantic feature vector to the coordinate value in the multi-dimensional space; Construct a corpus based on the set of device types and train word vectors with the goal of maximizing the conditional probability, where, represents the set of device types included in the context window centered on the device type string in the corpus: ， Among them, the conditional probability is calculated by the Softmax function: ， Among them, represents the index of all candidate device types in the set of device types, is the word vector representation of the k-th device type; after training, the word vectors of all device types are organized into a matrix V, where : ； To eliminate the influence of different dimensional value ranges, the matrix V is column-normalized, where and are the mean and standard deviation of the j-th dimension respectively, and the element in the matrix represents the numerical value of the device type string on the j-th dimension of the word vector. The normalized matrix is : ， ， ； Calculate the covariance matrix of the normalization matrix , denotes the transposed matrix of the normalization matrix : ； Perform eigenvalue decomposition on the covariance matrix C: ， ， ； Among them, is the eigenvector matrix, is the eigenvector matrix is the transpose matrix of is the eigenvalue matrix, and the first principal component direction corresponds to the largest eigenvalue; Select the first principal component direction , and project the normalized word vector of each device type string onto this direction to obtain the coordinate value : ； Finally, the high-dimensional word vectors of each device type are reduced to a single value , which is used as the coordinate value of the first dimension of the point in the multi-dimensional space and participates in subsequent point distance calculation and matching.

5. A fast matching search method based on an addressing matrix according to claim 4, characterized in that, S22. In the multi-dimensional identification network, the geographical location is described by longitude and latitude data. Convert the longitude and latitude information into a unique numerical representation as the coordinate value of the second dimension in the multi-dimensional space, specifically including the following steps: S221. Let the longitude of the geographical location be x and the latitude be y, and their value ranges are respectively: ； Normalize the longitude and latitude and map them to the interval : ； Normalized longitude , normalized latitude ; S222. According to the normalization result, convert into integer grid coordinates. Assuming the coding precision is b bits, the grid coordinate calculation formula is: ， ， Among them, represents the position in the grid of longitude and latitude, is the floor operation; S223. Binary-encode the grid coordinates and respectively, and the encoding results are represented as and : ， ； S224. Use a Z-shaped curve to alternately splice binary bits to generate an overall binary string, representing the unique encoding of the geographical location: ； S225. Convert the concatenated binary string into a decimal number as the unique mapping value of the geographical location: ， This value is used as the coordinate value of the second dimension of the point in the multi-dimensional space.

6. A fast matching search method based on an addressing matrix according to claim 5, characterized in that In S23, in the multi-dimensional identification network, the service identifier h is a 128-bit digital code. When the type of the service identifier is less than three digits, the 128 bits of the service identifier are directly modulo 10 to obtain a single digit as the mapping value, that is, the third-dimensional coordinate value; then ; When the types of business identifiers gradually increase to three digits or more, to avoid uneven distribution caused by direct modulo operation, the business identifier h needs to be hashed first, denoted as , and then is modulo 10, and the final result is assigned to , which is used as the coordinate value of the third dimension in the multi-dimensional space.

7. A fast matching search method based on an addressing matrix according to claim 6, characterized in that In S4, the Euclidean distance is used as the criterion for evaluating the best object. Let the point obtained by attribute mapping in the addressing matrix be , and any point in the multi-dimensional space be . Then the distance between the two points is defined as: ; Among them, , , respectively represent the device type, geographical location, and service identification coordinate values of the points obtained by attribute mapping within the addressing matrix; , , respectively represent the device type, geographical location, and service identification coordinate values of any point within the multi-dimensional space; Select the point with the shortest distance from point A within the limited range as the best communication object.

Images