An access resource point recommendation method and system based on multi-factor fusion and a medium

Abstract

The application discloses a kind of based on multi-factor fusion's access resource point recommendation method, system and medium, belong to communication technical field, based on service installation address, obtain the access resource point of candidate, and generate resource list;The address similarity between the address vector of access resource point address and service installation address is calculated;Access distance recommendation degree between access resource point address and service installation address is calculated;Based on residual available capacity index, the stability index of resource and maintenance cost index, the resource recommendation index of access resource point is calculated;Based on the address similarity, access distance recommendation degree and resource recommendation index of each access resource point in resource list, the comprehensive recommendation score of each access resource point is calculated, generates resource recommendation index ranking, obtains access resource point priority list and routing.The application has strong dynamic adaptability, can adapt to the differentiated needs of city / countryside, high-density / low-density user area, has good practicability.

Description

Technical Field

[0001] This invention belongs to the field of communication technology, specifically relating to a method, system, and medium for recommending access resource points based on multi-factor fusion. Background Technology

[0002] In the current telecommunications industry, competition among operators is becoming increasingly fierce. Against this backdrop, users' demands for rapid installation of communication services are becoming more prominent. This makes the selection of access resource points and service routing planning key challenges and crucial aspects of service solution planning, design, and delivery efficiency improvement. Choosing appropriate access resource points and improving service routing planning can effectively shorten installation time and enhance user experience. The selection of access resource points requires comprehensive consideration of factors such as network architecture, user distribution, and geographical environment to find the optimal solution under complex conditions.

[0003] Currently, in the scenario of recommending service access resources in the telecommunications industry, existing technologies have the following shortcomings:

[0004] 1. Low address matching accuracy: Traditional methods rely on string edit distance or exact match, which cannot handle address abbreviations (such as "Guangzhou City" → "Guangzhou"), word order reversal (such as "XX Road XX No." → "XX No. XX Road"), and differences in auxiliary information (such as "Room 605" being omitted), resulting in a matching accuracy of less than 60%, low search ranking, and high false positive rate.

[0005] 2. Distance calculation is out of touch with actual construction: The straight-line distance is calculated only by latitude and longitude without taking into account geographical obstacles such as roads, rivers, and mountains (e.g., the construction cost of crossing the main river is 80% higher than that without crossing it), which makes the construction of the recommended access resource points more difficult and extends the cycle by more than 30%.

[0006] 3. Oversimplified resource assessment: Using only "remaining capacity" as an indicator ignores stability (e.g., high-occupancy resources are prone to performance bottlenecks) and maintenance costs (e.g., the maintenance of resources around forests is more difficult than that of urban areas), leading to unreasonable resource allocation and an increase in maintenance costs of 15%-20% in the later stages. Summary of the Invention

[0007] The purpose of this invention is to provide a method, system, and medium for recommending access resource points based on multi-factor fusion, in order to solve the above-mentioned problems.

[0008] This invention is mainly achieved through the following technical solutions:

[0009] A method for recommending access resource points based on multi-factor fusion includes the following steps:

[0010] Step S1: Based on the service installation address, search for nearby GJ / GF optical facility resources with service access capabilities in the optical cable network, obtain candidate access resource points, and generate a resource list;

[0011] Step S2: Based on the BERT-CRF model, parse the address and quantize it into an address vector, and calculate the address similarity As between the address vectors of the candidate access resource point address and the service installation address;

[0012] Step S3: Calculate the access distance recommendation score (sim) between the candidate access resource point address and the service installation address. dist And the access distance recommendation score sim dist The spherical distance d between the latitude and longitude of the access resource point address and the service installation address is inversely proportional to the distance d between them.

[0013] Step S4: Based on the remaining available capacity index C score The resource stability index S and maintenance cost index M are used to calculate the resource recommendation index RI of candidate access resource points.

[0014] RI=w1·C score +w2·S+w3·M;

[0015] Where: w1, w2, w3 are the weights to be calculated, and w1+w2+w3=1;

[0016] Step S5: Based on the address similarity, access distance recommendation degree and resource recommendation index of each access resource point in the resource list, calculate the comprehensive recommendation score P of each access resource point, generate the resource recommendation index ranking, and obtain the access resource point priority list and route;

[0017] P=w4·sim dist +w5·RI+w6·As;

[0018] Where w4, w5, and w6 are the fusion weights, and w4+w5+w6=1.

[0019] To better implement the present invention, step S1 further includes the following steps:

[0020] Step S11: Match the AOI data of the corresponding building based on the service installation address, and confirm the terminal resources GB through the building;

[0021] Step S12: Based on the hierarchical rules of user layer, access layer, aggregation layer and transport layer set from low to high, and taking the terminal resource GB as the center, search for accessible GJ / GF optical facilities within the radius of the three local fiber segments using the breadth-first search algorithm.

[0022] Step S13: Based on the search of GJ / GF optical facilities, obtain GJ / GF optical facility resources with service access capabilities, and construct a resource list.

[0023] To better realize the present invention, step S2 further includes the following steps:

[0024] Step S21: Based on the BERT-CRF model, perform address naming recognition on the service installation address and the access resource point address respectively, and perform structured parsing of the address;

[0025] First, for the input address text, the address text is encoded based on the BERT model to generate a hidden layer representation for each location;

[0026] Then, the CRF layer calculates the probability that each character or word belongs to a different entity label based on the hidden layer representation, and finally outputs a structured representation of the address;

[0027] Finally, the structured representation of the output address;

[0028] Step S22: After address structure parsing, the entity features of the address are quantized and represented to generate an address vector containing administrative division code and street code;

[0029] Step S23: Calculate the similarity between the address vectors of the service installation address and the access resource point address based on cosine similarity.

[0030] To better implement this invention, further, in step S3, based on the Haversine formula, the spherical distance d between the latitude and longitude of the candidate access resource point address and the service installation address is calculated. Then, based on the obstacle between the access resource point address and the service installation address, a scaling factor α is determined, and the access distance recommendation degree sim is calculated. dist ;

[0031] .

[0032] To better realize the present invention, step S3 further includes the following steps:

[0033] Step S31: Retrieve the latitude and longitude of the access resource point based on the resource location database;

[0034] Step S32: Obtain the latitude and longitude of the service installation address based on the POI data of the building where the service installation address is located;

[0035] Step S33: Based on the latitude and longitude of the access resource point and the service installation address, use the Haversine formula to calculate the spherical distance d between the access resource point address and the service installation address;

[0036] Step S34: Determine the scaling factor α based on the actual geographical situation between the access resource point address and the service installation address;

[0037] Step S35: Calculate the access distance recommendation score sim based on the spherical distance d and the scaling factor α. dist .

[0038] To better realize this invention, a PostGIS database is used to store vector data of urban road networks, water areas, and mountains; the ST_Crosses function of PostGIS is used to determine whether the user coordinates and the coordinate lines of the access resource points intersect with the GIS data of the road network and lakes, so as to determine the situation of crossing obstacles, and then determine the scaling factor α.

[0039] To better implement the present invention, further, in step S4, the remaining available capacity index C of the resources... score The calculation formula is:

[0040] ;

[0041] Where: C represents the actual capacity of the access resource point;

[0042] U represents the occupancy rate;

[0043] C max This represents the maximum capacity of the corresponding access resource point.

[0044] This invention is mainly achieved through the following technical solutions:

[0045] An access resource point recommendation system based on multi-factor fusion, based on the aforementioned access resource point recommendation method based on multi-factor fusion, includes:

[0046] Breadth-first search module: Used to extract nearby GJ / GF optical infrastructure resources with service access capabilities from the optical fiber network based on the service installation address, and generate a resource list;

[0047] Address similarity module: used to calculate the address similarity between candidate access resource point addresses and service installation addresses;

[0048] Access distance module: used to calculate the recommended access distance between candidate access resource point addresses and service installation addresses;

[0049] Resource Recommendation Index Module: Used to calculate the resource recommendation index (RI) for candidate access resource points;

[0050] The comprehensive recommendation ranking module is used to calculate the comprehensive recommendation score for each access resource point and obtain the access resource point priority list and routes.

[0051] A computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the aforementioned method for recommending access resource points based on multi-factor fusion.

[0052] The beneficial effects of this invention are as follows:

[0053] (1) The present invention has strong dynamic adaptability, supports dynamic adjustment of search range, indicator weight, and exploration strategy, and can adapt to the differentiated needs of urban / rural and high-density / low-density user areas; the present invention provides efficient and accurate resource allocation basis for service activation scheme design, reduces delivery construction and maintenance costs, and has good practicality. Specifically, the present invention improves delivery efficiency, can better meet user needs, and secondly, the reduction in delivery time also means a reduction in delivery costs, and reasonable routing can also reduce later maintenance costs.

[0054] (2) This invention combines the dynamic adjustment of scaling factor α based on geographical obstacles and dynamically calculates the access distance recommendation degree to ensure the feasibility of construction of recommended access resource points, thereby shortening the construction cycle of recommended access resource points by 25%-30%. If it is necessary to expand the optical cable capacity, access resource points that cross geographical obstacles such as rivers will be avoided, and the construction difficulty and time will be reduced.

[0055] (3) This invention performs structured parsing of addresses and calculates address similarity As based on the BERT-CRF model, which solves the problems of abbreviation and word order. The address similarity accuracy reaches 92%-95%, which is more than 50% higher than the traditional method (<60%), and avoids "invalid exploration" caused by address mismatch.

[0056] (4) This invention integrates a multi-factor calculation of resource recommendation index based on "capacity-stability-maintenance cost" to achieve optimal resource allocation. The multi-factor calculated resource recommendation index increases resource utilization by 12%-15% and reduces subsequent maintenance costs by 10%-12% (e.g., prioritizing "high capacity + low maintenance cost" resources to reduce the number of fault repairs). Finally, the TOP3 access resource points are dynamically output, providing an efficient and accurate basis for resource allocation in the design of service activation schemes, and reducing delivery construction and maintenance costs. Attached Figure Description

[0057] Figure 1 This is a block diagram illustrating the principle of the address resolution-based access resource point recommendation method of the present invention.

[0058] Figure 2 This is a diagram illustrating the scope and hierarchy of a breadth-first search.

[0059] Figure 3 A schematic diagram of the optical infrastructure accessible from GB as the center;

[0060] Figure 4A flowchart for generating a resource list;

[0061] Figure 5 This is a block diagram illustrating the principle of calculating address similarity.

[0062] Figure 6 A block diagram illustrating the principle of calculating the access distance recommendation index;

[0063] Figure 7 A flowchart illustrating the principle of calculating the resource recommendation index;

[0064] Figure 8 A flowchart illustrating the principle of calculating the comprehensive recommendation score. Detailed Implementation

[0065] Example 1:

[0066] A method for recommending access resource points based on multi-factor fusion, such as... Figure 1 As shown, it includes the following steps:

[0067] Step S1: Based on the service installation address, search for nearby GJ / GF optical infrastructure resources with service access capabilities in the optical fiber network to obtain candidate access resource points and generate a resource list. Specifically, in the service access resource point survey stage, this invention obtains the building and end resources (GB) of the user's service installation address, and searches for nearby access resource points (GJ / GF) through the optical fiber network.

[0068] Step S2: Based on the BERT-CRF model, parse the address and quantize it into an address vector, and calculate the address similarity As between the address vectors of the candidate access resource point address and the service installation address;

[0069] Step S3: Calculate the access distance recommendation score (sim) between the candidate access resource point address and the service installation address. dist And the access distance recommendation score sim dist The spherical distance d between the latitude and longitude of the access resource point address and the service installation address is inversely proportional to the distance d between them.

[0070] Step S4: Based on the remaining available capacity index C score The resource stability index S and maintenance cost index M are used to calculate the resource recommendation index RI of candidate access resource points.

[0071] RI=w1·C score +w2·S+w3·M;

[0072] Where: w1, w2, w3 are the weights to be calculated, and w1+w2+w3=1;

[0073] Step S5: Based on the address similarity, access distance recommendation score, and resource recommendation index of each access resource point in the resource list, calculate the comprehensive recommendation score P for each access resource point.

[0074] P=w4·sim dist +w5·RI+w6·As;

[0075] Where: w4, w5, and w6 are the fusion weights, and w4+w5+w6=1;

[0076] Based on the comprehensive recommendation score of resources, a resource recommendation index ranking is generated, and a priority list and route of access resource points are obtained.

[0077] Preferably, in step S1, as follows: Figure 4 As shown, the building AOI data is first matched with the service installation address to determine the terminal resource GB; for example... Figure 2 and Figure 3 As shown, according to the hierarchical rules from the user layer to the transport layer, with the target terminal resource GB as the center, based on the breadth-first search algorithm, GJ / GF optical facilities within the 3 configurable local fiber optic radii are searched, GJ / GF optical facilities with access capabilities are screened, candidate access resource points (GJ / GF) are obtained, and a resource list is generated.

[0078] Specifically, the AOI data of the corresponding building is confirmed by the user's service installation address, and the terminal resources are confirmed by the building, which are usually optical terminal boxes or optical distribution frames in the user's equipment room. For example, "Room XXX, No. 30, Tianhe First Street, Tianhe District, Guangzhou" corresponds to the terminal resource "XX Building GB1".

[0079] Determine the search scope and level: such as Figure 2 As shown, the layers are defined as user layer, access layer, aggregation layer, and transport layer, with the hierarchy from low to high. Routing is only allowed between the same layer or upwards within the same layer.

[0080] like Figure 3 As shown, optical facilities (GJ / GF) can be reached within a radius of GB via three (configurable) local optical fibers.

[0081] Filtering the available list: As shown in Table 1, nearby GJ (optical distribution box) and GF (optical fiber splitter box) resources with service access capabilities are extracted from the optical cable network and used as the resource list for subsequent comprehensive recommendation calculation.

[0082] Table 1 Example of a resource list

[0083]

[0084] Preferably, in step S2, as follows: Figure 5As shown, the BERT-CRF model is used to perform structured parsing of the service installation address and the candidate access resource point address. The parsed address is a structured entity such as province, city, district, etc., which is quantized into a vector to generate an address vector containing administrative division code and street code. Then, the matching degree between the address vectors of the service installation address and the access resource point address is calculated by cosine similarity, that is, the address similarity is obtained, so as to solve the limitations of traditional string matching.

[0085] Specifically, in network resource access scenarios, accurately determining the similarity between a user's address and the access resource address is crucial for precisely recommending suitable access resource points.

[0086] Traditional address matching methods often employ simple string matching techniques, such as directly comparing whether two address strings are identical or calculating the edit distance between strings. However, this approach is limited by factors such as word segmentation quality. First, addresses are expressed in diverse ways, including abbreviations, alternative names, and different writing orders. For example, "Guangzhou City" might be written as "Guangzhou," and "Tianhe District" might sometimes be expressed as "Tianhe." Second, there is insufficient semantic understanding of addresses, making it impossible to identify the meaning and relationships between different parts of an address. For instance, "XX Road XX Number" and "XX Number XX Road" might be considered significantly different under simple string matching, but they actually point to the same location. Furthermore, traditional methods struggle to accurately handle complex address structures, such as addresses containing multiple levels and supplementary information.

[0087] This invention utilizes deep learning-based Named Entity Recognition (NER) technology, employing BERT-CRF for address name recognition. BERT leverages its contextual representation capabilities, while CRF enhances its label dependency modeling capabilities. BERT extracts the contextual features of each word, while CRF models the label dependencies between words, thereby improving the effectiveness of named entity recognition and achieving structured address parsing. The BERT model, through pre-training, learns a large amount of semantic information from the text, enabling bidirectional encoding of the input address text to obtain a contextual representation vector h for each character or word. i For the input address sequence x=[x1, x2, ..., x... n After processing by the BERT model, the corresponding hidden layer representation H = [h1, h2, ..., h] is obtained. n ].

[0088] Each hidden layer vector h in the BERT model i This involves integrating the semantic information from the context of the entire address text, and the semantic relationships between different parts of the address. For example, for "No. 123, Tiyu West Road, Tianhe District, Guangzhou", h iIt can represent the semantic relationship between "Tianhe District", "Guangzhou City", and "Tiyu West Road". The CRF layer is based on the hidden layer representation of BERT output. It predicts the entity label of each character or word in the address by calculating the conditional probability P(y|x), where is the input address text and is the corresponding entity label sequence, such as "province", "city", "district", "street", "house number", etc.

[0089] Step S2 specifically includes the following steps:

[0090] (1) First, input the address text to be parsed into the pre-trained BERT-CRF model.

[0091] The BERT model encodes the address text and generates a hidden layer representation for each location;

[0092] Then, the CRF layer calculates the probability that each character or word belongs to a different entity label based on these hidden layer representations;

[0093] Finally, the structured representation of the output address.

[0094] For example, if you input "No. 123, Tiyu West Road, Tianhe District, Guangzhou", the model will output the province (Guangdong Province), city (Guangzhou City), district (Tianhe District), street (Tiyu West Road), and house number (No. 123).

[0095] (2) After the address is structured and parsed, the individual entity features of the address are quantized and represented to form an address vector.

[0096] For example, for "No. 123, Tiyu West Road, Tianhe District, Guangzhou" and "No. 456, Tiyu East Road, Tianhe District, Guangzhou", "Guangzhou" and "Tianhe District" can be quantized into a fixed value, while "Tiyu West Road" and "Tiyu East Road" can be quantized into different values ​​to construct a vector.

[0097] (3) For two address vectors A=(a1, a2, ..., a2), n ) and B=(b1, b2, ..., b n The cosine similarity was used to calculate the address similarity between two address vectors. If the calculated address similarity result is close to 1, it indicates that the two addresses are relatively similar in structure and semantics; if the calculated address similarity result is close to 0, it indicates that the differences are large. Address similarity matching calculations were performed on all access resource point addresses and installation addresses, as shown in Table 2.

[0098] The cosine similarity formula is as follows:

[0099] ;

[0100] Where: a i b iThese are the i-th quantization features of the address vectors for the installation address and the access resource point address, respectively.

[0101] Table 2 Address Similarity Information

[0102]

[0103] Preferably, in step S3, as follows: Figure 6 As shown, the spherical distance d between the user's service installation address and the candidate access resource point address is calculated based on the Haversine formula; combined with PostGIS, it is determined whether the route crosses obstacles such as roads and rivers, and the scaling factor α is determined (for example, α=1 for no crossing, α=1.4 for crossing main roads, α=1.8 for crossing main rivers, etc.); the distance recommendation degree is calculated based on the spherical distance d and the scaling factor α.

[0104] Specifically, the distance between the user's location and the location of the access resource is calculated and converted into an access distance recommendation that can be used for comparison.

[0105] Based on geographic information technology, the distance between two points is calculated by spatial distance calculation.

[0106] Latitude and longitude of access resource points (GJ / GF): retrieved from the operator's "resource location database";

[0107] Latitude and longitude of user address: POI data of the building where the user address is located (enter "Room 1601, No. 30, Tianhe First Street, Tianhe District, Guangzhou" to get the building data "No. 30, Tianhe First Street").

[0108] If the latitude and longitude of the user's location (lat1, lon1) and the latitude and longitude of the access resource location (lat2, lon2) are known, these coordinate values ​​are substituted into the Haversine formula, and the values ​​of each part are calculated sequentially according to the order of operations in the formula, finally obtaining the spherical distance d kilometers between the two:

[0109] .

[0110] Meanwhile, the construction of optical cables needs to consider whether they cross roads, rivers, lakes, etc. When planning the route, priority should be given to resources that do not cross roads, rivers, lakes, or mountains. After calculating the distance d, an appropriate scaling factor α is determined based on the actual geographical conditions, ultimately yielding a recommended distance value.

[0111] GIS data: The PostGIS database is used to store vector data (format: Shapefile) of urban road network (highway / main road / ordinary road), water area (main river / tributary / lake), and mountains.

[0112] Judgment logic: Use the ST_Crosses function of PostGIS to determine whether the user coordinates and the coordinate lines of the access resource points intersect with GIS data such as road networks and lakes. That is, the ST_Crosses(ST_MakeLine(user coordinates, access resource point coordinates), obstacle vector) function returns "True / False" to determine whether it crosses.

[0113] α value: There is no standard α value between two points. Based on the cost difference statistics of 1000+ actual construction cases (e.g., the construction cost of crossing the main river is 80% higher than that without crossing, so α=1.8; construction across mountains is the most difficult, so α=2), the definition is shown in Table 3.

[0114] Substituting d and α into the formula yields the distance recommendation score sim. dist :

[0115] ;

[0116] As shown in Table 4, if sim dist A value close to 1 indicates that the user is relatively close to the access resource and will be given priority.

[0117] Table 3 Scaling Factor Values

[0118]

[0119] Table 4

[0120]

[0121] Preferably, in step S4, as follows: Figure 7 As shown, the resource recommendation index mainly involves the analysis and evaluation of network resources. In a network resource access system, multiple factors need to be considered to evaluate the quality of access resources, thereby recommending the most suitable resources to users. This module calculates and analyzes indicators such as resource capacity and occupancy rate to derive the resource recommendation index.

[0122] Traditional network resource allocation often considers only a single factor to assess resource availability, such as focusing solely on remaining capacity. However, this approach neglects other crucial factors, such as resource utilization, stability, and maintenance costs. Evaluating a single factor can lead to inefficient resource allocation, failing to fully meet user needs and optimize overall system performance. For instance, a resource with ample remaining capacity but high utilization may experience performance bottlenecks in actual use.

[0123] The resource recommendation index of this invention considers the following factors:

[0124] Factor 1: Remaining Available Resource Capacity Index C scoreThe calculation formula is:

[0125] ;

[0126] Where: C is the actual capacity of the access resource point, and U is the occupancy rate. max This represents the maximum capacity for this type of resource.

[0127] Example: The capacity of resource GJ001 is C=100 ports, the utilization rate is U=20%, and the maximum capacity of the same type is C. max =100, then C score =(100×0.8) / 100=0.8.

[0128] Factor 2: Resource stability index S, with a score of [0, 1]. This index is calculated based on historical fault analysis to quantify the stability and reliability of network resources during historical operation. The closer the value is to 1, the more stable the resource; the closer it is to 0, the higher the risk of failure. As shown in Table 5, fault data from six months were used, and the index scores were categorized according to fault level.

[0129] As shown in Table 6, the fault data of access resources can be obtained from historical fault work orders. After data processing, the data is integrated into fault-related network resource data, which is dynamically updated daily.

[0130] Factor 3: Maintenance Cost M, with a score of [0, 1]. This indicator is calculated by uniformly assessing the maintenance difficulty based on the geographical characteristics of the resource (city, county, plain, hills (including urban mountains), mountainous areas, and islands). As shown in Table 7, this indicator is directly benchmarked against the maintenance fee settlement coefficient. The higher the maintenance fee settlement coefficient, the greater the maintenance cost and the lower the score.

[0131] Table 5

[0132]

[0133] Table 6

[0134]

[0135] Table 7

[0136]

[0137] Finally, the resource recommendation index is incorporated into the calculation using a weighted method:

[0138] RI=w1·C score +w2·S+w3·M;

[0139] Where: w1, w2, w3 are the weights to be calculated, and w1+w2+w3=1;

[0140] Preferably, the weights w1, w2, and w3 are determined as follows: using the Analytic Hierarchy Process (AHP) and combined with the operator's business needs survey, the default weights are:

[0141] w1 (capacity metric weight) = 0.5 (core metric, directly affecting access capability);

[0142] w2 (stability weight) = 0.3 (affects long-term service quality);

[0143] w3 (maintenance cost weight) = 0.2 (impact on long-term operating costs);

[0144] Supports dynamic adjustments (e.g., the weight of maintenance costs in rural areas can be increased to 0.3).

[0145] Preferably, in step S5, as follows: Figure 8 As shown, the recommendation score is based on the address similarity As and access distance sim of each access resource point. dist The comprehensive recommendation score P for each access resource point is calculated using different types of indicators such as the resource recommendation index (RI) to obtain a list of access resource recommendation priorities and routes.

[0146] The overall recommendation score P for the resource is:

[0147] P=w4·sim dist +w5·RI+w6·As;

[0148] Where: w4, w5, and w6 are the fusion weights, and w4+w5+w6=1;

[0149] Based on the comprehensive recommendation score of resources, a resource recommendation index ranking is generated, and a priority list and route of access resource points are obtained.

[0150] Example 2:

[0151] A multi-factor fusion-based access resource point recommendation method, taking "Broadband installation service at Room 1601, No. 30, Tianhe First Street, Tianhe District, Guangzhou" as an example, includes the following specific steps:

[0152] Step 1: Data Acquisition and Breadth-Based Search;

[0153] The building where the installation is located is "No. 30, Tianhe First Street", with coordinates: 23.1302°N, 113.3298°E, and associated terminal resource GB1 for room 1605 (coordinates 23.1302°N, 113.3298°E).

[0154] Search scope: such as Figure 2As shown in Table 8, with GB1 as the center, within a 3-hop optical fiber network, 3 candidate access resource points (GJ001 / GJ002 / GJ003) were selected, and their basic information is shown in Table 8.

[0155] Table 8. Filtered Resource Information

[0156]

[0157] Step 2: Calculate address similarity;

[0158] BERT-CRF structured parsing: The installation address outputs "Province (Guangdong), City (Guangzhou), District (Tianhe), Street (Tianhe First Street), House Number (No. 30), Room Number (Room 1601)", and the candidate address is processed similarly;

[0159] Cosine similarity calculation: GJ001=0.95, GJ002=0.91, GJ003=0.88.

[0160] Step 3: Calculate the access distance recommendation score;

[0161] Spherical distance (Haversine formula): GJ001=0.3km, GJ002=0.8km, GJ003=0.5km;

[0162] Geographical obstacle assessment: GJ001 has no crossing (α=1), GJ002 crosses a regular road (α=1.2), and GJ003 has no crossing (α=1).

[0163] Distance Recommendation Rating:

[0164] GJ001: 1 / (1+1×0.3)=0.769;

[0165] GJ002: 1 / (1+1.2×0.8)=0.510;

[0166] GJ003: 1 / (1+1×0.5)=0.667.

[0167] Step 4: Calculate the resource recommendation index;

[0168] Weights: w1=0.5, w2=0.3, w3=0.2;

[0169] Index calculation:

[0170] GJ001: 0.5×0.8+0.3×0.9+0.2×1=0.87;

[0171] GJ002: 0.5×0.6+0.3×0.7+0.2×1=0.71;

[0172] GJ003: 0.5×0.7+0.3×0.8+0.2×1=0.79.

[0173] Step 5: Comprehensive Recommendation (Integration Weights: Address Similarity 0.3, Distance 0.3, Resource Index 0.4);

[0174] Overall Recommendation Rating:

[0175] GJ001: 0.3×0.95+0.3×0.769+0.4×0.87≈0.864;

[0176] GJ002: 0.3×0.91+0.3×0.510+0.4×0.71≈0.71;

[0177] GJ003: 0.3×0.88+0.3×0.667+0.4×0.79≈0.78;

[0178] Output TOP3: GJ001 (first choice) → GJ003 (second choice) → GJ002 (alternative).

[0179] Example 3:

[0180] An access resource point recommendation system based on multi-factor fusion, based on the method in Example 1, aims to accurately recommend access resources based on the user's service installation address, and includes the following five core modules:

[0181] Breadth-first search module: Used to extract nearby GJ / GF optical infrastructure resources with service access capabilities from the optical fiber network based on the service installation address, and generate a resource list;

[0182] Address similarity module: used to calculate the address similarity between candidate access resource point addresses and service installation addresses;

[0183] Access distance module: used to calculate the recommended access distance between candidate access resource point addresses and service installation addresses;

[0184] Resource Recommendation Index Module: Used to calculate the resource recommendation index (RI) for candidate access resource points;

[0185] The comprehensive recommendation ranking module is used to calculate the comprehensive recommendation score for each access resource point and obtain the access resource point priority list and routes.

[0186] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Any simple modifications or equivalent changes made to the above embodiments based on the technical essence of the present invention shall fall within the protection scope of the present invention.

Claims

1. A method for recommending access resource points based on multi-factor fusion, characterized in that, Includes the following steps: Step S1: Based on the service installation address, obtain candidate access resource points and generate a resource list; Step S2: Based on the BERT-CRF model, parse the address and quantize it into an address vector, and calculate the address similarity As between the address vectors of the candidate access resource point address and the service installation address; Calculate the address similarity As between two address vectors using cosine similarity. Step S3: Calculate the access distance recommendation score (sim) between the candidate access resource point address and the service installation address. dist And the access distance recommendation score sim dist The spherical distance d between the latitude and longitude of the access resource point address and the service installation address is inversely proportional to the distance d between them. In step S3, based on the Haversine formula, the spherical distance d between the latitude and longitude of the candidate access resource point address and the service installation address is calculated. Then, based on the obstacles between the access resource point address and the service installation address, a scaling factor α is determined, and the access distance recommendation degree sim is calculated. dist ; ； Step S4: Based on the remaining available capacity index C score The resource stability index S and maintenance cost index M are used to calculate the resource recommendation index RI of candidate access resource points. RI=w1·C score +w2·S+w3·M; Where: w1, w2, w3 are the weights to be calculated, and w1+w2+w3=1; The stability index S of the resource has a score of [0, 1]; The maintenance cost index M has a score of [0, 1]; In step S4, the remaining available capacity index C of the resources score The calculation formula is: ； Where: C represents the actual capacity of the access resource point; U represents the occupancy rate; C max This represents the maximum capacity of the corresponding access resource point; Step S5: Based on the address similarity, access distance recommendation degree and resource recommendation index of each access resource point in the resource list, calculate the comprehensive recommendation score P of each access resource point, generate the resource recommendation index ranking, and obtain the access resource point priority list and route; P=w4·sim dist +w5·RI+w6·As; Where w4, w5, and w6 are the fusion weights, and w4+w5+w6=1.

2. The access resource point recommendation method based on multi-factor fusion according to claim 1, characterized in that, Step S1 includes the following steps: Step S11: Match the AOI data of the corresponding building based on the service installation address, and confirm the terminal resources GB through the building; Step S12: Based on the hierarchical rules of user layer, access layer, aggregation layer and transport layer set from low to high, and taking the terminal resource GB as the center, search for accessible GJ / GF optical facilities within the radius of the three local fiber segments using the breadth-first search algorithm. Step S13: Based on the search of GJ / GF optical facilities, obtain GJ / GF optical facility resources with service access capabilities, and construct a resource list.

3. The access resource point recommendation method based on multi-factor fusion according to claim 1, characterized in that, Step S2 includes the following steps: Step S21: Based on the BERT-CRF model, perform address naming recognition on the service installation address and the access resource point address respectively, and perform structured parsing of the address; First, for the input address text, the address text is encoded based on the BERT model to generate a hidden layer representation for each location; Then, the CRF layer calculates the probability that each character or word belongs to a different entity label based on the hidden layer representation, and finally outputs a structured representation of the address; Step S22: After address structure parsing, the entity features of the address are quantized and represented to generate an address vector containing administrative division code and street code; Step S23: Calculate the similarity between the address vectors of the service installation address and the access resource point address based on cosine similarity.

4. The access resource point recommendation method based on multi-factor fusion according to claim 1, characterized in that, Step S3 includes the following steps: Step S31: Retrieve the latitude and longitude of the access resource point based on the resource location database; Step S32: Obtain the latitude and longitude of the service installation address based on the POI data of the building where the service installation address is located; Step S33: Based on the latitude and longitude of the access resource point and the service installation address, use the Haversine formula to calculate the spherical distance d between the access resource point address and the service installation address; Step S34: Determine the scaling factor α based on the actual geographical situation between the access resource point address and the service installation address; Step S35: Calculate the access distance recommendation score sim based on the spherical distance d and the scaling factor α. dist .

5. A method for recommending access resource points based on multi-factor fusion according to claim 1 or 4, characterized in that, The vector data of urban road network, water area and mountains are stored in PostGIS database; the ST_Crosses function of PostGIS is used to determine whether the user coordinates and the coordinate lines of the access resource points intersect with the GIS data of road network and lake, so as to determine the situation of crossing obstacles, and then determine the scaling factor α.

6. An access resource point recommendation system based on multi-factor fusion, comprising the access resource point recommendation method based on multi-factor fusion as described in any one of claims 1-5, characterized in that, include: Breadth-first search module: Used to extract nearby GJ / GF optical infrastructure resources with service access capabilities from the optical fiber network based on the service installation address, and generate a resource list; Address similarity module: used to calculate the address similarity between candidate access resource point addresses and service installation addresses; Access distance module: used to calculate the recommended access distance between candidate access resource point addresses and service installation addresses; Resource Recommendation Index Module: Used to calculate the resource recommendation index (RI) for candidate access resource points; The comprehensive recommendation ranking module is used to calculate the comprehensive recommendation score for each access resource point and obtain the access resource point priority list and routes.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the access resource point recommendation method based on multi-factor fusion as described in any one of claims 1-5.

Images