Evaluation Method for Local Transmission Networks Based on AI Large Model Empowerment and Hierarchical Aggregation Algorithm

Abstract

This invention discloses a local transmission network evaluation method based on AI large-scale model empowerment and hierarchical aggregation algorithm, belonging to the field of communication network evaluation technology. The invention includes: constructing a three-level evaluation index system; collecting raw data from multi-source systems and performing intelligent cleaning using a weighted KNN algorithm and dynamic box plots; mapping data according to index type using corresponding functions; integrating subjective weights from AHP and objective weights from entropy weight method, and determining combined weights through a deviation maximization model; calculating the comprehensive capability index of the transmission network by a bottom-up, layer-by-layer aggregation based on the combined weights; and presenting the results visually through radar charts and combining random forest and SHAP analysis for intelligent diagnosis. This invention solves the problems of existing evaluation systems being single-dimensional, static, lacking intelligence, and having poor interpretability, achieving scientific quantification, dynamic evaluation, and intelligent diagnosis of multi-dimensional capabilities of transmission networks, and is applicable to network health assessment, precise investment, and optimization decision-making.

Description

Technical Field

[0001] This invention relates to the field of communication network evaluation technology, specifically to a local transmission network evaluation method based on AI large model empowerment and hierarchical aggregation algorithm. Background Technology

[0002] As the core of information infrastructure, transmission networks are evolving towards intelligence, service orientation, and green development. Traditional transmission network evaluation methods often focus on single performance indicators (such as latency and packet loss rate) or static equipment assessments, lacking multi-dimensional, dynamic, and intelligent comprehensive evaluation capabilities.

[0003] Existing evaluation systems often suffer from the following problems: First, they have a single evaluation dimension, focusing on technical performance and lacking integrated evaluation of multiple dimensions such as economic benefits, service agility, and green energy conservation; second, they are static analyses, mostly post-event assessments, and cannot predict or dynamically optimize the future network state; third, they have a low level of intelligence, relying on human experience and lacking AI-based data cleaning, weight optimization, and intelligent diagnostic capabilities; and fourth, they have poor interpretability, with evaluation results lacking visualization and root cause analysis, making it difficult to support accurate decision-making.

[0004] An existing patent (patent number: CN200410033129.9) discloses a method for evaluating the performance of a transmission network. It obtains the transmission quality of the network by acquiring relevant network security stability performance parameters, network management and maintenance parameters, and network resource utilization parameters. However, this application only evaluates network security performance and cannot provide an effective assessment of network infrastructure capabilities, efficiency, and cost.

[0005] To this end, this invention proposes a local transmission network evaluation method based on AI large model empowerment and hierarchical aggregation algorithm. Summary of the Invention

[0006] The purpose of this invention is to provide a local transmission network evaluation method based on AI large model empowerment and hierarchical aggregation algorithm, so as to realize the scientific quantification, dynamic evaluation, intelligent diagnosis and visualization of the multi-dimensional capabilities of the transmission network, and support the precise planning, intelligent operation and maintenance and optimization decision-making of the network.

[0007] According to a first aspect of the present invention, in order to achieve the above-mentioned objective, the present invention provides the following technical solution: a local transmission network evaluation method based on AI large model empowerment and hierarchical aggregation algorithm, comprising the following steps: A multi-level evaluation index system is constructed, specifically from eight dimensions: basic capabilities, safety capabilities, efficiency and cost, service capabilities, intelligent capabilities, technological advancement, green energy conservation, and engineering adaptability, which includes three levels of indicators. Based on the eight dimensions of the evaluation index system, the corresponding original index data in the live network operation system is obtained. The weighted KNN algorithm is used to fill in missing values ​​in the original index data, and the dynamic box plot method is used to detect and correct outliers, resulting in a cleaned index dataset. The indicators in the cleaned indicator dataset are classified into different types. Based on their characteristics of being benefit-oriented, cost-oriented, or moderate, the upper limit effect measurement function, lower limit effect measurement function, or piecewise normalization function are called to map them, and all indicator values ​​are uniformly quantized to the [0,1] interval to obtain normalized indicator data. The objective weights of each indicator in the normalized index data are calculated using the entropy weight method. At the same time, expert scoring data is introduced, and the subjective weights of each indicator are calculated using the analytic hierarchy process. Then, the subjective weights and the objective weights are fused and optimized using the deviation maximization model to output the combined weights of each indicator. Based on the combined weights, the normalized index data is aggregated layer by layer from bottom to top, and the scores of the third-level index, the second-level index, and the first-level index are calculated in sequence. Then, the sum of all the first-level index scores is calculated by the linear weighted summation method to obtain the comprehensive capability index of the local transmission network, and the capability level is divided according to the comprehensive capability index.

[0008] Furthermore, in the aforementioned three-level indicator evaluation system, each of the eight dimensions includes the following secondary indicators, as detailed below: The basic capability dimension includes four secondary indicators: spatial coverage capability, building coverage capability, bandwidth supply capability, and network latency capability. The security capability dimension includes five secondary indicators: building security, transmission system security, equipment security, network management security, and average network lifespan. The efficiency and cost dimension includes three secondary indicators: network utilization, network construction cost, and network construction competitiveness. The service-oriented capability dimension includes two secondary indicators: service activation timeliness and service adjustment capability. The intelligent capability dimension includes four secondary indicators: AI for planning and construction, AI for maintenance, AI for operation, and AI for optimization. The technological advancement dimension includes three secondary indicators: optical layer advancement, optical layer protection advancement, and network evolution advancement. The green energy-saving dimension includes a secondary indicator: single-bit power consumption. The engineering adaptability dimension includes two secondary indicators: air supply method and power supply method.

[0009] Furthermore, the weighted KNN algorithm is used to impute missing values ​​in the original indicator data, and outlier detection and correction are performed based on the dynamic boxplot method, resulting in the cleaned indicator dataset, as follows: Based on the target city, cities with matching network scale, business structure, and operating model are selected as candidate cities; Calculate the Euclidean distance between each candidate city and the target city, eliminate cities with excessively large differences, and select a preset number of cities with the smallest distance as the similar nearest neighbor set; The correlation between each feature and the target variable was calculated using the Pearson correlation coefficient, and feature weights were assigned based on the correlation coefficient normalization method, as follows: The weights are allocated using the correlation coefficient normalization method, and the formula is as follows: in, =1, w i Let r be the weight of the i-th key feature. i The correlation coefficient between this feature and the target variable is n, where n is the number of key features; Calculate the weighted Euclidean distance between the target city and each of its nearest neighboring cities, and convert the distance into similarity weights, as follows: The weighted Euclidean distance is used to calculate the comprehensive difference between the target city and its neighboring cities. The calculation formula is as follows. The weighted distance is then converted into similarity weights (the smaller the distance, the higher the similarity), using the following formula: Among them, S k Let be the similarity weight of the k-th nearest neighbor city, and ; Based on the aforementioned similarity weights, a weighted average is applied to the corresponding data of neighboring cities to obtain the completion results for missing values, as detailed below: The missing values ​​are obtained by weighting the similarity scores, and the formula is as follows: Among them, Y A,t Y represents the missing indicator value to be completed for target city A in the t-th statistical period; k,t For the k-th neighboring city in the similar nearest neighbor set, the measured index value of the same indicator corresponding to the target city A in the same statistical period t; t is the statistical period in the time dimension.

[0010] Furthermore, outlier detection and correction are performed based on the dynamic boxplot method, as detailed below: Based on the time series characteristics of the indicator data, a dynamic sliding window is set; Calculate the upper quartile, lower quartile, and interquartile range of the indicator data within each sliding window; Data points that exceed the range of the upper quartile plus 1.5 times the interquartile range, or are below the range of the lower quartile minus 1.5 times the interquartile range, are identified as outliers. For identified outliers, the mean or median of the nearest normal data is used for correction, or they are marked as missing values ​​and then filled in again using the weighted KNN algorithm.

[0011] Furthermore, the benefit-type indicators are normalized using an upper limit effect measurement function, as follows: In the formula, x' is the normalized standard quantified value of the benefit-type indicator, with a value range of [0,1]; x is the original measured indicator data corresponding to the benefit-type indicator; min(x) is the original minimum value of all samples in the current indicator dataset; max(x) is the original maximum value of all samples in the current indicator dataset, and satisfies max(x) ≠ min(x).

[0012] Furthermore, the cost-related indicators are normalized using a lower bound effect measurement function, as follows: In the formula, x' is the normalized standard quantified value of the cost indicator, with a value range of [0,1]; x is the original measured indicator data corresponding to the cost indicator; min(x) is the original minimum value of all samples in the current indicator dataset; max(x) is the original maximum value of all samples in the current indicator dataset, and satisfies max(x) ≠ min(x).

[0013] Furthermore, the appropriateness index is normalized using a piecewise function, as follows: In the formula, z0 is the optimal utilization threshold.

[0014] Furthermore, the objective weights of each indicator in the normalized index data are calculated using the entropy weight method, while expert scoring data is introduced. The subjective weights of each indicator are calculated using the analytic hierarchy process (AHP). Then, the subjective weights and objective weights are fused and optimized using a deviation maximization model to output the combined weights of each indicator, as detailed below: (81) The objective weights are calculated using the entropy weight method, as follows: (81.1) Construct a normalized index data matrix and calculate the information entropy value of each index; Let the total number of evaluation objects in the local transmission network be m, and the total number of evaluation indicators be n. Construct a normalized indicator matrix X = (x' ij )mxn , where x' ij Let x' represent the normalized value of the i-th evaluation object and the j-th indicator, where 0 ≤ x' ij ≤ 1; (81.2) Calculate the index weight matrix The normalized index is standardized by weighting to eliminate zero-value interference. The calculation formula is as follows: In the formula, p ij Let the weight of the indicator for the i-th evaluation object under the j-th indicator satisfy the following condition: If x' ij =0, then let p ij lnp ij =0; (81.3) Calculate the information entropy value of each indicator. The entropy value e of the j-th index is calculated using the information entropy formula. j The calculation formula is: In the formula, e j Let be the information entropy value of the j-th indicator, with a value range of [0,1]. The smaller the entropy value, the higher the dispersion and the stronger the discrimination of the indicator. (81.4) Calculate the coefficient of difference of the indicators The difference coefficient, derived from entropy, reflects the distinguishing ability of the indicator. The calculation formula is as follows: In the formula, g j denoted as the difference coefficient of the j-th indicator. The larger the difference coefficient, the higher the indicator's evaluation and identification. (81.5) Calculate the objective weight of the indicators After normalizing the difference coefficients, the objective weights of the entropy weight method are obtained. The calculation formula is: In the formula, Let j be the objective weight of the j-th indicator, satisfying And the weights are non-negative; (82) Subjective weights are calculated using the analytic hierarchy process, as follows: An expert group was organized to conduct pairwise comparisons of indicators at each level and to construct a judgment matrix using the 1-9 scale method. Calculate the largest eigenvalue and the corresponding eigenvector of the judgment matrix, and obtain the subjective weights of each indicator after normalization. A consistency check is performed on the judgment matrix. If the consistency ratio is lower than a preset threshold, the expert scoring logic is considered consistent; otherwise, the judgment matrix needs to be adjusted. (83) The subjective and objective weights are optimized by using the deviation maximization model, as follows: Let the preference coefficient α represent the degree of preference for subjective weights, and 1−α represent the degree of preference for objective weights; Construct a deviation maximization objective function to maximize the comprehensive deviation of each indicator under subjective and objective weights; Solve the optimization model and output a combined weight w that balances expert experience and objective data patterns. in, The subjective weights calculated by the analytic hierarchy process (AHP). The objective weights are calculated using the entropy weight method.

[0015] Furthermore, based on the combined weights, the normalized index data is aggregated layer by layer from bottom to top, calculating the scores of the third-level index, the second-level index, and the first-level index in sequence. Then, the sum of all first-level index scores is calculated using a linear weighted summation method to obtain the comprehensive capability index of the local transmission network. The capability levels are then classified according to the comprehensive capability index, as follows: (91) The normalized index data is aggregated layer by layer, and the scores of the third-level index, the second-level index, and the first-level index are calculated in sequence, as follows: Calculation of Level 3 Indicator Scores: Let S ijk Let x' be the score of the kth tertiary indicator within the jth secondary indicator under the i-th primary indicator. ijk w is the normalized value of this third-level indicator. ijk Let S be the combined weight of its respective secondary indicator. ijk =x' ijk ×w ijk This score directly reflects the performance of a single basic evaluation unit; Secondary indicator score calculation: The score of a secondary indicator is obtained by linearly weighting and summing the scores of all its subordinate tertiary indicators. Let S be the score of a secondary indicator. ij K represents the score of the j-th secondary indicator under the i-th primary indicator. ij S represents the number of tertiary indicators contained in this secondary indicator. ij =∑S ijk k ranges from 1 to K ij This enables the aggregation of basic evaluation units into intermediate-level indicators; Primary indicator score calculation: The score of a primary indicator is obtained by weighted aggregation of the scores of all its subordinate secondary indicators, let S be... i J is the score of the i-th primary indicator. i S represents the number of secondary indicators contained in this primary indicator. i =∑S ij j from 1 to J i; Based on the hierarchical aggregation results, the scores of the four primary indicators are combined to obtain the Transmission Network Comprehensive Capability Index (TN-CCI), calculated as follows: TN-CCI=S A1 +S A2 +S A3 +...+S A8 In the formula, S AX These are the eight dimensions of the indicator evaluation system, where x = 1, 2, 3, ..., 8; (92) The range of TN-CCI is [0,1]. The closer the value is to 1, the stronger the integrated capability of the transmission network; the closer the value is to 0, the weaker the integrated capability of the network. The capability level is divided according to the integrated capability index TN-CCI as follows: Excellent TN-CCI ≥ 0.8: The network has balanced development in all aspects and is at a leading level; Good (0.6 ≤ TN-CCI < 0.8): Performance in the main capability dimensions is good, but there is room for improvement in some areas; Qualified: 0.4 ≤ TN-CCI < 0.6: Basically meets operational requirements, but requires systematic improvement; TN-CCI < 0.4: There is a significant shortcoming that requires focused investment in improvement.

[0016] Furthermore, after obtaining the Comprehensive Competency Index (TN-CCI) and classifying competency levels, the following steps are also included: A radar chart is constructed based on the eight dimensions of the evaluation index system. The scores of the eight dimensions and the comprehensive ability index are input and presented visually. The random forest algorithm combined with the analytic hierarchy process is used to perform root cause analysis on the scores of each dimension to identify the weak indicators and their influencing factors that lead to low scores. The AI ​​big model is integrated, and the comprehensive capability index, scores of each dimension and root cause analysis results are input. The AI ​​big model is then invoked to execute intelligent applications. The intelligent applications include at least one of automatically generating evaluation reports, responding to natural language questions and answers, or making decision-making inferences on changes in indicators after the introduction of optimization measures. Among them, the scores of the eight dimensions in the evaluation index system are the scores of the first-level indicators.

[0017] This invention has at least the following beneficial effects: 1. This invention abandons the traditional static evaluation model and achieves intelligent data cleaning through the weighted KNN algorithm, effectively handling the problems of missing and anomalies in multi-source heterogeneous data; it adopts a deviation maximization weight optimization model that integrates AHP and entropy weight method, which respects expert experience and reflects the objective laws of data; it introduces a hierarchical aggregation algorithm to achieve bottom-up layer-by-layer quantification calculation, so that the evaluation results can dynamically respond to changes in network status and more realistically reflect the actual operating level of the transmission network.

[0018] 2. This invention visually displays capability scores across various dimensions using multi-dimensional visualization methods such as radar charts, enabling network administrators to quickly identify strengths and weaknesses. Simultaneously, it introduces the Random Forest algorithm and SHAP analysis method to deeply mine and analyze the evaluation results, accurately pinpointing the key indicators and influencing factors leading to low scores. This effectively solves the problem of traditional evaluation methods that "know what but not why," providing clear guidance for precise policy implementation.

[0019] 3. This invention breaks through the limitations of traditional evaluation methods that only focus on technical performance or a single security dimension. It constructs a three-level evaluation index system from eight dimensions: basic capabilities, security capabilities, efficiency and cost, service capabilities, intelligent capabilities, technological advancement, green energy saving, and engineering adaptability. This system not only covers basic technical indicators such as network coverage, bandwidth capacity, and latency performance, but also incorporates cutting-edge considerations such as economic benefits, service agility, AI empowerment level, and green environmental protection. It achieves a comprehensive and three-dimensional evaluation of the overall capabilities of the local transmission network, providing a more comprehensive decision-making basis for network planning, construction, and operation and maintenance.

[0020] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0021] Fig. 1 This is a flowchart illustrating the method described in this invention; Fig. 2 This is a radar chart of the comprehensive capability evaluation index of the local transmission network of this invention. Detailed Implementation

[0022] The technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0023] Please see Figs. 1-2 This invention provides a technical solution: a local transmission network evaluation method based on AI large model empowerment and hierarchical aggregation algorithm, comprising the following steps: S1. Construct a multi-layered evaluation index system, specifically from eight dimensions: basic capabilities, safety capabilities, efficiency and cost, service capabilities, intelligent capabilities, technological advancement, green energy saving, and engineering adaptability. This system includes three levels of indicators, as shown in Table 1 below. Table 1 Evaluation System for Local Transmission Networks S2. Based on the eight dimensions of the evaluation index system, obtain the corresponding original index data from the live network operation system. Use the weighted KNN algorithm to fill in missing values ​​in the original index data, and use the dynamic box plot method to detect and correct outliers, resulting in a cleaned index dataset, as follows: S21. Data collection primarily uses raw data from the City-A live network operation system during the evaluation period as its core source, covering data from all scenarios including network operation, business support, maintenance management, and cost control. The collection method combines automatic system extraction with manual verification to avoid data bias. Examples of some core indicator data collection are shown in Table 2. Table 2 Examples of Systematic Data Collection S22. After data collection, the raw data needs to be preprocessed, focusing on solving problems such as missing data, anomalies, and redundancy, to ensure that the data quality meets the evaluation requirements; Based on the target city, cities with matching network scale, business structure, and operation mode are selected as candidate cities; based on City-A, candidate cities with matching network scale (number of hub buildings, length of optical cable lines), business structure (proportion of leased line business, user structure), and operation mode are selected. The Euclidean distance between each candidate city and City-A is calculated, cities with too large differences are eliminated, and finally the 5 cities with the smallest distance are selected as the similar nearest neighbor set. Calculate the Euclidean distance between each candidate city and the target city, eliminate cities with excessively large differences, and select a preset number of cities with the smallest distance as the similar nearest neighbor set; Using tributary port utilization as the target variable, candidate features were selected, and their correlation was quantified using the Pearson correlation coefficient. Features with high correlation coefficients (such as line port utilization r=0.82 and peak-to-average traffic ratio r=0.79) were retained, while weakly correlated features (such as tributary port traffic volume r=0.33) were removed. The correlation between each feature and the target variable was calculated using the Pearson correlation coefficient, and feature weights were assigned based on the correlation coefficient normalization method, as detailed below: The weights are allocated using the correlation coefficient normalization method, and the formula is as follows: in, =1, w i Let r be the weight of the i-th key feature. i The correlation coefficient between this feature and the target variable is n, where n is the number of key features; Calculate the weighted Euclidean distance between the target city and each of its nearest neighboring cities, and convert the distance into similarity weights, as follows: The weighted Euclidean distance is used to calculate the comprehensive difference between the target city and its neighboring cities. The calculation formula is as follows. The weighted distance is then converted into similarity weights (the smaller the distance, the higher the similarity), using the following formula: Among them, S k Let be the similarity weight of the k-th nearest neighbor city, and ; Based on the aforementioned similarity weights, a weighted average is applied to the corresponding data of neighboring cities to obtain the completion results for missing values, as detailed below: The missing values ​​are obtained by weighting the similarity scores, and the formula is as follows: Among them, Y A,t Y represents the missing indicator value to be completed for target city A in the t-th statistical period; k,t For the kth neighboring city in the similar neighbor set, the measured index value corresponding to the same indicator in the tth statistical period, which is the same as that of the target city A; t is the statistical period in the time dimension (such as month, quarter, year), and 5 is the total number of cities in the similar neighbor set. In this embodiment, the number of neighboring cities selected is set to 5. S23. Outlier detection and correction based on dynamic boxplot method, as detailed below: Based on the time series characteristics of the indicator data, a dynamic sliding window is set; Calculate the upper quartile, lower quartile, and interquartile range of the indicator data within each sliding window; Data points that exceed the range of the upper quartile plus 1.5 times the interquartile range, or are below the range of the lower quartile minus 1.5 times the interquartile range, are identified as outliers. For identified outliers, the mean or median of the nearest normal data is used for correction, or they are marked as missing values ​​and then filled in again using the weighted KNN algorithm; S3. The indicators in the cleaned dataset are categorized by type. Based on their characteristics (benefit-oriented, cost-oriented, or moderate), the upper limit effect measurement function, lower limit effect measurement function, or piecewise normalization function is called for mapping. All indicator values ​​are uniformly quantized to the [0,1] interval to eliminate the influence of units and ensure the rationality of the evaluation results. The final normalized indicator data is as follows: Benefit-oriented indicators: The higher the value of these indicators, the better the evaluation effect. An upper limit effect measurement function is used. Typical indicators include optical network accessibility, direct access rate of critical facilities, OXC / ROADM ratio, and electrical layer recovery success rate. The calculation formula is as follows: In the formula, x' is the normalized standard quantified value of the benefit-type indicator, with a value range of [0,1]; x is the original measured indicator data corresponding to the benefit-type indicator; min(x) is the original minimum value of all samples in the current indicator dataset; max(x) is the original maximum value of all samples in the current indicator dataset, and satisfies max(x) ≠ min(x); Nodes with wider network coverage, more advanced technology, and stronger protection capabilities receive higher scores on this metric; Cost-based metrics: The smaller the value of this type of metric, the better the evaluation effect. The lower limit effect measurement function is applied to cost-based metrics. Representative metrics include unit bit construction cost, unit service bandwidth maintenance cost, inter-node transmission latency, and single bit power consumption. Nodes with lower costs, higher energy efficiency, and lower latency receive higher evaluations. The calculation formula for this metric is: In the formula, x' is the normalized standard quantified value of the cost indicator, with a value range of [0,1]; x is the original measured indicator data corresponding to the cost indicator; min(x) is the original minimum value of all samples in the current indicator dataset; max(x) is the original maximum value of all samples in the current indicator dataset, and satisfies max(x) ≠ min(x); Moderate indicators: These indicators have an optimal value range; the closer the value is to the optimal range, the better the evaluation effect. Examples include line port utilization, branch port utilization, and spectrum space utilization. Their values ​​should not be too high or too low; they should be as close to the ideal value as possible. Therefore, piecewise functions are needed for normalization. Where z0 is the optimal utilization threshold, which can be set according to the operation and maintenance experience of the transmission network. This algorithm can avoid resource waste caused by too low utilization and network congestion risk caused by too high utilization. Following the above method, complete the normalization calculation of all tertiary indicators to form a normalized score matrix, ensuring that the scores of each indicator are comparable. S4. The objective weights of each indicator in the normalized index data are calculated using the entropy weight method. Simultaneously, expert scoring data is introduced, and the subjective weights of each indicator are calculated using the analytic hierarchy process (AHP). Then, the subjective and objective weights are fused and optimized using a deviation maximization model to output the combined weights of each indicator. This approach reflects both the strategic guidance of expert experience and the objective laws inherent in the data itself, avoiding the limitations of a single weighting method. The details are as follows: S41. The objective weights are calculated using the entropy weight method, as follows: S41.1. Construct a normalized index data matrix and calculate the information entropy value of each index; Let the total number of evaluation objects in the local transmission network be m, and the total number of evaluation indicators be n. Construct a normalized indicator matrix X = (x' ij ) mxn , where x' ij Let x' represent the normalized value of the i-th evaluation object and the j-th indicator, where 0 ≤ x' ij ≤ 1.

[0024] S41.2. Calculate the index weight matrix The normalized index is standardized by weighting to eliminate zero-value interference. The calculation formula is as follows: In the formula, p ij Let the weight of the indicator for the i-th evaluation object under the j-th indicator satisfy the following condition: If x' ij =0, then let p ij lnp ij =0.

[0025] S41.3. Calculate the information entropy value of each indicator. The entropy value e of the j-th index is calculated using the information entropy formula. j The calculation formula is: In the formula, e j Let be the information entropy value of the j-th indicator, with a value range of [0,1]. The smaller the entropy value, the higher the dispersion and the stronger the discrimination of the indicator.

[0026] S41.4. Calculate the coefficient of difference of indicators The difference coefficient, derived from entropy, reflects the distinguishing ability of the indicator. The calculation formula is as follows: In the formula, g j Let be the difference coefficient of the j-th indicator. The larger the difference coefficient, the higher the indicator's evaluation and identification.

[0027] S41.5. Calculate the objective weight of the indicator After normalizing the difference coefficients, the objective weights of the entropy weight method are obtained. The calculation formula is: In the formula, Let j be the objective weight of the j-th indicator, satisfying And the weights are non-negative.

[0028] S42. Subjective weights are calculated using the Analytic Hierarchy Process (AHP), as follows: An expert group was organized to conduct pairwise comparisons of indicators at each level and to construct a judgment matrix using the 1-9 scale method. Calculate the largest eigenvalue and the corresponding eigenvector of the judgment matrix, and obtain the subjective weights of each indicator after normalization. A consistency check is performed on the judgment matrix. If the consistency ratio is lower than a preset threshold, the expert scoring logic is considered consistent; otherwise, the judgment matrix needs to be adjusted. S43. The subjective and objective weights are fused and optimized using a deviation maximization model, as detailed below: Let the preference coefficient α represent the degree of preference for subjective weights, and 1−α represent the degree of preference for objective weights; Construct a deviation maximization objective function to maximize the comprehensive deviation of each indicator under subjective and objective weights; Solve the optimization model and output a combined weight w that balances expert experience and objective data patterns. in, The subjective weights calculated by the analytic hierarchy process (AHP). The objective weights calculated using the entropy weight method; In this embodiment, the preference coefficient α is set to 0.6, and the subjective and objective weights are fused based on maximizing the deviation. Ultimately, the combined weight of "optical layer protection coverage" is 0.11, and "tributary port utilization" is 0.09, reflecting both the strategic orientation of security and reliability and the key differentiating role of this indicator in actual operation. S5. Based on the combined weights, the normalized index data is aggregated layer by layer from bottom to top, calculating the scores of the third-level index, the second-level index, and the first-level index in sequence. Then, the sum of all first-level index scores is calculated using the linear weighted summation method to obtain the comprehensive capability index of the local transmission network. The capability levels are then classified according to the comprehensive capability index, as follows: Calculation of Level 3 Indicator Scores: Let S ijk Let x' be the score of the kth tertiary indicator within the jth secondary indicator under the i-th primary indicator. ijkw is the normalized value of this third-level indicator. ijk Let S be the combined weight of its respective secondary indicator. ijk =x' ijk ×w ijk This score directly reflects the performance of a single basic evaluation unit; Secondary indicator score calculation: The score of a secondary indicator is obtained by linearly weighting and summing the scores of all its subordinate tertiary indicators. Let S be the score of a secondary indicator. ij K represents the score of the j-th secondary indicator under the i-th primary indicator. ij S represents the number of tertiary indicators contained in this secondary indicator. ij =∑S ijk (k from 1 to K) ij This enables the aggregation of basic evaluation units into intermediate-level indicators; Primary indicator score calculation: The score of a primary indicator is obtained by weighted aggregation of the scores of all its subordinate secondary indicators, let S be... i J is the score of the i-th primary indicator. i S represents the number of secondary indicators contained in this primary indicator. i =∑S ij (j from 1 to J) i ); Based on the above hierarchical aggregation results, the scores of the four primary indicators are combined to obtain the Transmission Network Comprehensive Capability Index (TN-CCI), calculated as follows: TN-CCI = S A1 +S A2 +S A3 +...+S A8 S AX These are the eight dimensions of the evaluation system mentioned above; The TN-CCI value ranges from [0,1]. A value closer to 1 indicates a stronger overall transmission network capability; a value closer to 0 indicates a weaker overall network capability. To achieve a tiered assessment of the overall capabilities of local networks, the following four-level capability classification standard is established based on the distribution characteristics of the index score: Excellent (TN-CCI≥0.8): The network has balanced development in all aspects and is at a leading level; Good (0.6≤TN-CCI<0.8): Performance in the main capability dimensions is good, but there is room for improvement in some areas; Pass (0.4≤TN-CCI<0.6): Basically meets operational requirements, but requires systematic improvement; Areas for improvement (TN-CCI<0.4): There are obvious shortcomings that require focused investment in development.

[0029] This classification standard can clearly define the comprehensive capability level of local network transmission networks, providing a clear basis for subsequent targeted optimization and improvement and precise resource allocation. In order to effectively convey the complex multi-dimensional evaluation results to decision-makers, a visualization scheme with capability radar chart as the core has been designed. This scheme aims to transform quantitative evaluation results into intuitive graphical language, helping managers to quickly grasp the overall network status and identify strengths and weaknesses. S6. Construct a radar chart based on the eight dimensions of the evaluation index system. Input the scores of the eight dimensions and the comprehensive capability index for visualization. Then, use the random forest algorithm combined with the analytic hierarchy process to perform root cause analysis on the scores of each dimension, identify the weak indicators that lead to low scores and their influencing factors, as detailed below: Based on the scores obtained from hierarchical aggregation across various dimensions, the system automatically generates visual charts to intuitively present the comprehensive capabilities and performance of the City-A local transmission network across different dimensions. Simultaneously, it activates the intelligent diagnostic module, combining AI algorithms to identify weaknesses, analyze root causes, and recommend strategies, providing precise guidance for network optimization. The specific implementation process is as follows: Visual Presentation: The system automatically generates capability radar charts, bar charts of scores for each dimension, and trend charts. The radar charts, using eight primary indicators as their radius, clearly display the differences in scores across each dimension, intuitively reflecting the strengths and weaknesses of network capabilities. The radar charts show that the target city (e.g., City-A) performs well in "Technological Advancement" (score 0.75) and "Basic Capabilities" (score 0.70), approaching a good level; however, it exhibits significant weaknesses in "Service Capabilities" (score 0.45) and "Efficiency and Cost" (score 0.55), lagging far behind other dimensions, which are the core factors restricting the improvement of overall capabilities. Weakness Identification: The Weakness Index (SI) was used, calculated as SI = 1 - Score of that dimension / Average score of all dimensions. A higher SI value indicates a more significant weakness in that dimension. Calculations showed that the SI value for "Service Capability" was 0.42, and the SI value for "Efficiency and Cost" was 0.21, both significantly higher than other dimensions, identifying these two dimensions as core weaknesses. Further detailed analysis revealed that "Service Activation Duration" (a third-level indicator, score 0.38) and "Branch Port Utilization Rate" (a third-level indicator, score 0.41) were key indicators that lowered the scores of their respective first-level dimensions and hindered overall capability improvement. Root cause analysis: The random forest algorithm combined with SHAP value analysis is used to uncover the core influencing factors of the weakness index, ensuring the accuracy and pertinence of the root cause analysis and avoiding subjective judgment bias. S7. Integrate the AI ​​big model, input the comprehensive capability index, scores of each dimension and root cause analysis results, and call the AI ​​big model to execute intelligent applications. The intelligent applications include at least one of automatically generating evaluation reports, responding to natural language questions and answers, or making decision-making inferences on the changes in indicators after the introduction of optimization measures. Among them, the scores of the eight dimensions in the evaluation index system are the scores of the first-level indicators; In the era of artificial intelligence, this invention proposes to introduce large-scale AI models to achieve intelligent applications of evaluation results, encompassing three major functions: intelligent report generation, natural language question answering, and decision inference. This will drive the upgrade of network evaluation from "data quantification" to "intelligent decision-making," with specific applications as follows: Intelligent report generation: The Large Language Model (LLM) can automatically analyze the above data and scores, intelligently generate diagnostic results and strategy recommendations, and integrate them into a complete "City-A Local Transmission Network 2023 Annual Capability Evaluation and Optimization Recommendation Report".

[0030] Natural Language Question Answering: Users can freely ask questions in natural language without needing professional technical questioning skills. The AI ​​system can accurately understand the core intent of the question, automatically extract relevant and accurate information from the evaluation report, and quickly provide concise, easy-to-understand, and relevant answers, reducing the information access threshold and improving ease of use.

[0031] Decision simulation: Based on the trained and optimized artificial intelligence model, the system can simulate and extrapolate the evolution of indicators after the introduction of automated tools, quantitatively predict the improvement of each key indicator in the next year, and obtain the future improvement value of TN-CCI, providing a quantitative decision-making basis for operators' investment layout and project priority assessment.

[0032] Through the implementation of this embodiment, the present invention successfully transforms the massive and heterogeneous operational data of the City-A transmission network into an intuitive comprehensive capability index and a clear optimization path, realizing a leap from "human experience judgment" to "data and AI-driven decision-making", and significantly improving the scientificity, efficiency and accuracy of network evaluation.

[0033] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0034] For those skilled in the art, the specific meaning of the above terms in this invention can be understood according to the specific circumstances. When an element is referred to as being "assembled on," "mounted on," "fixed to," or "set on" another element, it may be directly on the other element or there may be an intermediate element present. When an element is considered to be "connected to" another element, it may be directly connected to the other element or there may be an intermediate element present. The terms "vertical," "horizontal," "upper," "lower," "left," "right," and similar expressions used herein are for illustrative purposes only and do not represent the only possible embodiments.

[0035] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

[0036] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this disclosure. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

Claims

1. A local transmission network evaluation method based on AI large model empowerment and hierarchical aggregation algorithm, characterized in that, Includes the following steps: A multi-level evaluation index system is constructed, specifically from eight dimensions: basic capabilities, safety capabilities, efficiency and cost, service capabilities, intelligent capabilities, technological advancement, green energy conservation, and engineering adaptability, which includes three levels of indicators. Based on the eight dimensions of the evaluation index system, the corresponding original index data in the live network operation system is obtained. The weighted KNN algorithm is used to fill in missing values ​​in the original index data, and the dynamic box plot method is used to detect and correct outliers, resulting in a cleaned index dataset. The indicators in the cleaned indicator dataset are classified into different types. Based on their characteristics of being benefit-oriented, cost-oriented, or moderate, the upper limit effect measurement function, lower limit effect measurement function, or piecewise normalization function are called to map them, and all indicator values ​​are uniformly quantized to the [0,1] interval to obtain normalized indicator data. The objective weights of each indicator in the normalized index data are calculated using the entropy weight method. At the same time, expert scoring data is introduced, and the subjective weights of each indicator are calculated using the analytic hierarchy process. Then, the subjective weights and the objective weights are fused and optimized using the deviation maximization model to output the combined weights of each indicator. Based on the combined weights, the normalized index data is aggregated layer by layer from bottom to top, and the scores of the third-level index, the second-level index, and the first-level index are calculated in sequence. Then, the sum of all the first-level index scores is calculated by the linear weighted summation method to obtain the comprehensive capability index of the local transmission network, and the capability level is divided according to the comprehensive capability index.

2. The local transmission network evaluation method based on AI large model empowerment and hierarchical aggregation algorithm according to claim 1, characterized in that: The three-level indicator evaluation system includes the following secondary indicators for each of the eight dimensions: The basic capability dimension includes four secondary indicators: spatial coverage capability, building coverage capability, bandwidth supply capability, and network latency capability. The security capability dimension includes five secondary indicators: building security, transmission system security, equipment security, network management security, and average network lifespan. The efficiency and cost dimension includes three secondary indicators: network utilization, network construction cost, and network construction competitiveness. The service-oriented capability dimension includes two secondary indicators: service activation timeliness and service adjustment capability. The intelligent capability dimension includes four secondary indicators: AI for planning and construction, AI for maintenance, AI for operation, and AI for optimization. The technological advancement dimension includes three secondary indicators: optical layer advancement, optical layer protection advancement, and network evolution advancement. The green energy-saving dimension includes a secondary indicator: single-bit power consumption. The engineering adaptability dimension includes two secondary indicators: air supply method and power supply method.

3. The local transmission network evaluation method based on AI large model empowerment and hierarchical aggregation algorithm according to claim 2, characterized in that: The weighted KNN algorithm was used to impute missing values ​​in the original indicator data, and outlier detection and correction were performed based on the dynamic boxplot method, resulting in the cleaned indicator dataset, as follows: Based on the target city, cities with matching network scale, business structure, and operating model are selected as candidate cities; Calculate the Euclidean distance between each candidate city and the target city, eliminate cities with excessively large differences, and select a preset number of cities with the smallest distance as the similar nearest neighbor set; The correlation between each feature and the target variable was calculated using the Pearson correlation coefficient, and feature weights were assigned based on the correlation coefficient normalization method, as follows: The weights are allocated using the correlation coefficient normalization method, and the formula is as follows: in, =1, w i Let r be the weight of the i-th key feature. i The correlation coefficient between this feature and the target variable is n, where n is the number of key features; Calculate the weighted Euclidean distance between the target city and each of its nearest neighboring cities, and convert the distance into similarity weights, as follows: The weighted Euclidean distance is used to calculate the comprehensive difference between the target city and its neighboring cities. The calculation formula is as follows. The weighted distance is then converted into similarity weights (the smaller the distance, the higher the similarity), using the following formula: Among them, S k Let be the similarity weight of the k-th nearest neighbor city, and ; Based on the aforementioned similarity weights, a weighted average is applied to the corresponding data of neighboring cities to obtain the completion results for missing values, as detailed below: The missing values ​​are obtained by weighting the similarity scores, and the formula is as follows: Among them, Y A,t Y represents the missing indicator value to be completed for target city A in the t-th statistical period; k,t For the k-th neighboring city in the similar nearest neighbor set, the measured index value of the same indicator corresponding to the target city A in the same statistical period t; t is the statistical period in the time dimension.

4. The local transmission network evaluation method based on AI large model empowerment and hierarchical aggregation algorithm according to claim 3, characterized in that: Outlier detection and correction are performed based on a dynamic boxplot method, as detailed below: Based on the time series characteristics of the indicator data, a dynamic sliding window is set; Calculate the upper quartile, lower quartile, and interquartile range of the indicator data within each sliding window; Data points that exceed the range of the upper quartile plus 1.5 times the interquartile range, or are below the range of the lower quartile minus 1.5 times the interquartile range, are identified as outliers. For identified outliers, the mean or median of the nearest normal data is used for correction, or they are marked as missing values ​​and then filled in again using the weighted KNN algorithm.

5. The local transmission network evaluation method based on AI large model empowerment and hierarchical aggregation algorithm according to claim 4, characterized in that: The benefit-type indicators are normalized using an upper limit effect measurement function, as follows: In the formula, x' is the normalized standard quantified value of the benefit-type indicator, with a value range of [0,1]; x is the original measured indicator data corresponding to the benefit-type indicator; min(x) is the original minimum value of all samples in the current indicator dataset; max(x) is the original maximum value of all samples in the current indicator dataset, and satisfies max(x) ≠ min(x).

6. The local transmission network evaluation method based on AI large model empowerment and hierarchical aggregation algorithm according to claim 4, characterized in that: The cost-related indicators are normalized using a lower bound effect measurement function, as follows: In the formula, x' is the normalized standard quantified value of the cost indicator, with a value range of [0,1]; x is the original measured indicator data corresponding to the cost indicator; min(x) is the original minimum value of all samples in the current indicator dataset; max(x) is the original maximum value of all samples in the current indicator dataset, and satisfies max(x) ≠ min(x).

7. The local transmission network evaluation method based on AI large model empowerment and hierarchical aggregation algorithm according to claim 4, characterized in that: The appropriateness index is normalized using a piecewise function, as follows: In the formula, z0 is the optimal utilization threshold.

8. The local transmission network evaluation method based on AI large model empowerment and hierarchical aggregation algorithm according to claim 7, characterized in that: The objective weights of each indicator in the normalized index data are calculated using the entropy weight method. Simultaneously, expert scoring data is introduced, and the subjective weights of each indicator are calculated using the analytic hierarchy process (AHP). Then, the subjective and objective weights are fused and optimized using a deviation maximization model to output the combined weights of each indicator, as detailed below: (81) The objective weights are calculated using the entropy weight method, as follows: (81.1) Construct a normalized index data matrix and calculate the information entropy value of each index; Let m be the total number of evaluation objects in the local transmission network and n be the total number of evaluation indicators. Construct a normalized indicator matrix X = (x' ij ) mxn , where x' ij Let x' represent the normalized value of the i-th evaluation object and the j-th indicator, where 0 ≤ x' ij ≤ 1; (81.2) Calculate the index weight matrix The normalized index is standardized by weighting to eliminate zero-value interference. The calculation formula is as follows: In the formula, p ij Let the weight of the indicator for the i-th evaluation object under the j-th indicator satisfy the following condition: If x' ij =0, then let p ij lnp ij =0; (81.3) Calculate the information entropy value of each indicator. The entropy value e of the j-th index is calculated using the information entropy formula. j The calculation formula is: In the formula, e j Let be the information entropy value of the j-th indicator, with a value range of [0,1]. The smaller the entropy value, the higher the dispersion and the stronger the discrimination of the indicator. (81.4) Calculate the coefficient of difference of the indicators The difference coefficient, derived from entropy, reflects the distinguishing ability of the indicator. The calculation formula is as follows: In the formula, g j denoted as the difference coefficient of the j-th indicator. The larger the difference coefficient, the higher the indicator's evaluation and identification. (81.5) Calculate the objective weight of the indicators After normalizing the difference coefficients, the objective weights of the entropy weight method are obtained. The calculation formula is: In the formula, Let j be the objective weight of the j-th indicator, satisfying And the weights are non-negative; (82) Subjective weights are calculated using the analytic hierarchy process, as follows: An expert group was organized to conduct pairwise comparisons of indicators at each level and to construct a judgment matrix using the 1-9 scale method. Calculate the largest eigenvalue and the corresponding eigenvector of the judgment matrix, and obtain the subjective weights of each indicator after normalization. A consistency check is performed on the judgment matrix. If the consistency ratio is lower than a preset threshold, the expert scoring logic is considered consistent; otherwise, the judgment matrix needs to be adjusted. (83) The subjective and objective weights are optimized by using the deviation maximization model, as follows: Let the preference coefficient α represent the degree of preference for subjective weights, and 1−α represent the degree of preference for objective weights; Construct a deviation maximization objective function to maximize the comprehensive deviation of each indicator under subjective and objective weights; Solve the optimization model and output a combined weight w that balances expert experience and objective data patterns. in, The subjective weights calculated by the analytic hierarchy process (AHP). The objective weights are calculated using the entropy weight method.

9. The local transmission network evaluation method based on AI large model empowerment and hierarchical aggregation algorithm according to claim 8, characterized in that: Based on the aforementioned combined weights, the normalized index data is aggregated layer by layer from bottom to top, calculating the scores of the third-level, second-level, and first-level indicators in sequence. Then, the sum of all first-level indicator scores is calculated using a linear weighted summation method to obtain the comprehensive capability index of the local transmission network. The capability levels are then classified according to the comprehensive capability index, as follows: (91) The normalized index data is aggregated layer by layer, and the scores of the third-level index, the second-level index, and the first-level index are calculated in sequence, as follows: Calculation of Level 3 Indicator Scores: Let S ijk Let x' be the score of the kth tertiary indicator within the jth secondary indicator under the i-th primary indicator. ijk w is the normalized value of this third-level indicator. ijk Let S be the combined weight of its respective secondary indicator. ijk =x' ijk ×w ijk This score directly reflects the performance of a single basic evaluation unit; Secondary indicator score calculation: The score of a secondary indicator is obtained by linearly weighting and summing the scores of all its subordinate tertiary indicators. Let S be the score of a secondary indicator. ij K represents the score of the j-th secondary indicator under the i-th primary indicator. ij S represents the number of tertiary indicators contained in this secondary indicator. ij =∑S ijk k ranges from 1 to K ij This enables the aggregation of basic evaluation units into intermediate-level indicators; Primary indicator score calculation: The score of a primary indicator is obtained by weighted aggregation of the scores of all its subordinate secondary indicators, let S be... i J is the score of the i-th primary indicator. i S represents the number of secondary indicators contained in this primary indicator. i =∑S ij j ranges from 1 to J i ; Based on the hierarchical aggregation results, the scores of the four primary indicators are combined to obtain the Transmission Network Comprehensive Capability Index (TN-CCI), calculated as follows: TN-CCI=S A1 +S A2 +S A3 +...+S A8 In the formula, S AX These are the eight dimensions of the indicator evaluation system, where x = 1, 2, 3, ..., 8; (92) The range of TN-CCI is [0,1]. The closer the value is to 1, the stronger the integrated capability of the transmission network; the closer the value is to 0, the weaker the integrated capability of the network. The capability level is divided according to the integrated capability index TN-CCI as follows: Excellent TN-CCI ≥ 0.8: The network has balanced development in all aspects and is at a leading level; Good (0.6 ≤ TN-CCI < 0.8): Performance in the main capability dimensions is good, but there is room for improvement in some areas; Qualified: 0.4 ≤ TN-CCI < 0.6: Basically meets operational requirements, but requires systematic improvement; TN-CCI < 0.4: There is a significant shortcoming that requires focused investment in improvement.

10. The local transmission network evaluation method based on AI large model empowerment and hierarchical aggregation algorithm according to claim 9, characterized in that: After obtaining the Comprehensive Competency Index (TN-CCI) and classifying the competency levels, the following steps are also included: A radar chart is constructed based on the eight dimensions of the evaluation index system. The scores of the eight dimensions and the comprehensive ability index are input and presented visually. The random forest algorithm combined with the analytic hierarchy process is used to perform root cause analysis on the scores of each dimension to identify the weak indicators and their influencing factors that lead to low scores. The AI ​​big model is integrated, and the comprehensive capability index, scores of each dimension and root cause analysis results are input. The AI ​​big model is then invoked to execute intelligent applications. The intelligent applications include at least one of automatically generating evaluation reports, responding to natural language questions and answers, or making decision-making inferences on changes in indicators after the introduction of optimization measures. Among them, the scores of the eight dimensions in the evaluation index system are the scores of the first-level indicators.

Images