Resource benefit evaluation method and device, electronic equipment and storage medium

Abstract

The invention discloses a resource benefit evaluation method and device, electronic equipment and a storage medium, and relates to the technical field of resource benefit evaluation, and the method comprises the steps: building a three-dimensional feature space containing a business type, coverage quality and user density based on multi-dimensional feature data, and outputting the prediction results of resource benefits in different business scenes through a dynamic mapping model. And benefit grading is carried out through a preset grading algorithm, so that the problems that a mapping relation between flow and physical resource block utilization rate is distorted and a 5G network complex environment is difficult to adapt due to the fact that a service classification mode based on 4G experience is adopted and a static threshold value is depended in an existing 5G cell resource benefit evaluation method can be solved; the technical effect of improving the accuracy, adaptability and flexibility of resource benefit evaluation is achieved.

Description

Technical Field

[0001] This disclosure relates to the field of resource benefit assessment technology, and in particular to a resource benefit assessment method and apparatus, electronic equipment and storage medium. Background Technology

[0002] With the rapid development of 5G networks, their position in the mobile communications industry is becoming increasingly prominent, and they are widely used in many key areas such as the Industrial Internet, smart cities, and the Internet of Vehicles. Compared with 4G, 5G networks have not only achieved significant improvements in bandwidth, latency, and connection density, but also introduced key technologies such as Massive MIMO, network slicing, and spatial multiplexing, promoting the deep integration of the radio access network and the core network. Currently, a cell-level resource benefit assessment system is typically constructed through the collaborative work of service classification, coverage quality assessment, and user density monitoring.

[0003] Currently, the resource efficiency assessment for 5G cells adopts a service classification method based on 4G experience, simply dividing 5G services into three categories: large packets, medium packets, and small packets. Static thresholds for user numbers, traffic, and PRB utilization are set based on expert experience. However, this method leads to a distortion in the mapping relationship between traffic and PRB utilization due to the differences in bandwidth, latency, and protocol types required by different 5G services. It is difficult to adapt to the complex environment of 5G networks, which features sudden changes in user behavior, diverse service types, and dynamic changes in coverage quality. Summary of the Invention

[0004] This disclosure provides a resource efficiency assessment method, apparatus, electronic device, and storage medium. Its main purpose is to address the problem that the varying demands of different 5G services in terms of bandwidth, latency, and protocol type lead to a distorted mapping relationship between traffic and PRB utilization, making it difficult to adapt to the complex environment of 5G networks characterized by sudden changes in user behavior, diverse service types, and dynamically changing coverage quality.

[0005] According to a first aspect of this disclosure, a resource benefit assessment method is provided, comprising:

[0006] Acquire multidimensional feature data of the target area, and construct a three-dimensional feature space based on the multidimensional feature data; wherein, the three-dimensional feature space includes service type features, coverage quality features, and user density features;

[0007] The three-dimensional feature space is input into the dynamic mapping model to obtain the prediction results of resource efficiency under different business scenarios output by the dynamic mapping model.

[0008] The prediction results are classified into benefit levels based on a preset level classification algorithm to obtain the evaluation results of the target area.

[0009] Optionally, before inputting the three-dimensional feature space into the dynamic mapping model to obtain the prediction results of resource efficiency under different business scenarios output by the dynamic mapping model, the method further includes:

[0010] A dynamic mapping model between physical resource block utilization and traffic under different scenarios is established based on the random forest regression algorithm.

[0011] Optionally, acquiring multidimensional feature data of the target region and constructing a three-dimensional feature space based on the multidimensional feature data further includes:

[0012] The business type features are clustered and dimensionality reduced using a Gaussian mixture model based on Mahalanobis distance.

[0013] The coverage quality characteristics are dynamically classified based on the measurement reports of the target area and the cumulative distribution function of the physical uplink shared channel;

[0014] The user density features are classified in a two-stage density classification based on K-means clustering and quantile regression models.

[0015] Optionally, acquiring multidimensional feature data of the target region and constructing a three-dimensional feature space based on the multidimensional feature data further includes:

[0016] The business type features are dimensionality reduced, reducing the uplink business feature dataset to three business types: real-time interaction, data transmission, and transaction processing; and reducing the downlink business feature dataset to five business types: real-time interaction, streaming media transmission, large file transmission, transaction processing, and lightweight information service. The business type features include both the uplink and downlink business feature datasets.

[0017] The traffic of the uplink business feature dataset and the downlink business feature dataset in the business feature dataset is summed and normalized;

[0018] The normalized flow rate is subjected to feature transformation based on the central logarithmic ratio transform.

[0019] Optionally, the step of classifying the prediction results into benefit levels based on a preset level classification algorithm to obtain the evaluation results of the target area includes:

[0020] The prediction results are classified into benefit levels based on the variance-weighted quartile method to obtain the classification level of the target area.

[0021] The health index of the target area is calculated based on the classification level to obtain the evaluation result of the target area; wherein, the proportion of the health index is different for different classification levels.

[0022] Optionally, after classifying the prediction results into benefit levels based on a preset classification algorithm to obtain the evaluation results of the target area, the method further includes:

[0023] A three-level benefit ratio heatmap is generated based on the health index ratio and displayed on a preset screen.

[0024] Optionally, the step of classifying the prediction results into benefit levels based on the variance-weighted quartile method to obtain the classification level of the target area includes:

[0025] The measurement reports and physical uplink shared channel values ​​in the coverage quality dataset are reconstructed into a continuous probability density function through probability density estimation, and the cumulative distribution function is calculated.

[0026] Based on the cumulative distribution function curve, a preset quantile value is extracted as the coverage quality feature threshold;

[0027] Based on the distribution characteristics of the upstream and downstream business feature datasets, the four-quadrant method is used to classify the coverage quality levels.

[0028] Optionally, the two-stage density classification of the user density features based on the K-means clustering and quantile regression model further includes:

[0029] The elbow rule is used to determine the optimal number of clusters, and the average number of users is divided into a preset number of density levels.

[0030] Based on the average number of users at each density level, the quantile regression model is constructed to quantify the maximum user number fluctuation threshold.

[0031] According to a second aspect of this disclosure, a resource benefit assessment apparatus is provided, comprising:

[0032] The acquisition unit is used to acquire multi-dimensional feature data of the target area and construct a three-dimensional feature space based on the multi-dimensional feature data; wherein, the three-dimensional feature space includes service type features, coverage quality features, and user density features;

[0033] The input unit is used to input the three-dimensional feature space into the dynamic mapping model to obtain the prediction results of resource efficiency under different business scenarios output by the dynamic mapping model.

[0034] The evaluation unit is used to classify the prediction results into benefit levels based on a preset level classification algorithm to obtain the evaluation results of the target area.

[0035] Optionally, the device further includes:

[0036] The establishment unit is used to establish the dynamic mapping model between physical resource block utilization and traffic under different scenarios based on the random forest regression algorithm before the input unit inputs the three-dimensional feature space into the dynamic mapping model and obtains the prediction results of resource benefits under different business scenarios output by the dynamic mapping model.

[0037] Optionally, the acquisition unit is further configured to:

[0038] The business type features are clustered and dimensionality reduced using a Gaussian mixture model based on Mahalanobis distance.

[0039] The coverage quality characteristics are dynamically classified based on the measurement reports of the target area and the cumulative distribution function of the physical uplink shared channel;

[0040] The user density features are classified in a two-stage density classification based on K-means clustering and quantile regression models.

[0041] Optionally, the acquisition unit is further configured to:

[0042] The business type features are dimensionality reduced, reducing the uplink business feature dataset to three business types: real-time interaction, data transmission, and transaction processing; and reducing the downlink business feature dataset to five business types: real-time interaction, streaming media transmission, large file transmission, transaction processing, and lightweight information service. The business type features include both the uplink and downlink business feature datasets.

[0043] The traffic of the uplink business feature dataset and the downlink business feature dataset in the business feature dataset is summed and normalized;

[0044] The normalized flow rate is subjected to feature transformation based on the central logarithmic ratio transform.

[0045] Optionally, the evaluation unit is also used for:

[0046] The prediction results are classified into benefit levels based on the variance-weighted quartile method to obtain the classification level of the target area.

[0047] The health index of the target area is calculated based on the classification level to obtain the evaluation result of the target area; wherein, the proportion of the health index is different for different classification levels.

[0048] Optionally, the device further includes:

[0049] The display unit is used to generate a three-level benefit ratio heatmap based on the health index ratio after the evaluation unit classifies the prediction results into benefit levels according to the preset level classification algorithm and obtains the evaluation results of the target area, and then displays it on a preset screen.

[0050] Optionally, the evaluation unit is also used for:

[0051] The measurement reports and physical uplink shared channel values ​​in the coverage quality dataset are reconstructed into a continuous probability density function through probability density estimation, and the cumulative distribution function is calculated.

[0052] Based on the cumulative distribution function curve, a preset quantile value is extracted as the coverage quality feature threshold;

[0053] Based on the distribution characteristics of the upstream and downstream business feature datasets, the four-quadrant method is used to classify the coverage quality levels.

[0054] Optionally, the acquisition unit is further configured to:

[0055] The elbow rule is used to determine the optimal number of clusters, and the average number of users is divided into a preset number of density levels.

[0056] Based on the average number of users at each density level, the quantile regression model is constructed to quantify the maximum user number fluctuation threshold.

[0057] According to a third aspect of this disclosure, an electronic device is provided, comprising:

[0058] At least one processor; and

[0059] A memory communicatively connected to the at least one processor; wherein,

[0060] The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method described in the first aspect above.

[0061] According to a fourth aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions, wherein the computer instructions are configured to cause the computer to perform the method described in the first aspect above.

[0062] According to a fifth aspect of this disclosure, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the method described in the first aspect above.

[0063] The resource benefit assessment method, apparatus, electronic device, and storage medium disclosed herein mainly include the following technical solutions: acquiring multi-dimensional feature data of a target area, and constructing a three-dimensional feature space based on the multi-dimensional feature data; wherein the three-dimensional feature space includes service type features, coverage quality features, and user density features; inputting the three-dimensional feature space into a dynamic mapping model to obtain the predicted resource benefits under different service scenarios output by the dynamic mapping model; and classifying the predicted results into benefit levels based on a preset level classification algorithm to obtain the assessment results of the target area. Compared with related technologies, the embodiments of this application construct a three-dimensional feature space containing service type, coverage quality, and user density based on multi-dimensional feature data, and use a dynamic mapping model to output the predicted resource benefits under different service scenarios, and then classify the benefit levels using a preset level classification algorithm. Therefore, this can solve the problem in existing 5G cell resource benefit assessment methods that use service classification methods based on 4G experience and rely on static thresholds, resulting in distorted mapping relationships between traffic and physical resource block utilization, making it difficult to adapt to the complex environment of 5G networks. This achieves the technical effect of improving the accuracy, adaptability, and flexibility of resource benefit assessment.

[0064] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent from the following description. Attached Figure Description

[0065] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:

[0066] Figure 1 A schematic flowchart illustrating a resource benefit assessment method provided in an embodiment of this disclosure;

[0067] Figure 2 A schematic flowchart illustrating a resource benefit assessment method provided in an embodiment of this disclosure;

[0068] Figure 3 A schematic flowchart illustrating a resource benefit assessment method provided in an embodiment of this disclosure;

[0069] Figure 4 This is a schematic diagram of the structure of a resource benefit assessment device provided in an embodiment of the present disclosure;

[0070] Figure 5 A schematic diagram of another resource benefit assessment device provided in this disclosure embodiment;

[0071] Figure 6 A schematic block diagram of an example electronic device provided for embodiments of this disclosure. Detailed Implementation

[0072] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0073] The resource benefit assessment method, apparatus, electronic device, and storage medium of this disclosure are described below with reference to the accompanying drawings.

[0074] Figure 1 This is a schematic flowchart of a resource benefit assessment method provided in an embodiment of the present disclosure.

[0075] like Figure 1 As shown, the method includes the following steps:

[0076] Step 101: Obtain multi-dimensional feature data of the target area, and construct a three-dimensional feature space based on the multi-dimensional feature data; wherein, the three-dimensional feature space includes service type features, coverage quality features, and user density features;

[0077] To obtain multidimensional feature data of the target area and construct a three-dimensional feature space containing service type features, coverage quality features, and user density features, the specific process is as follows: First, the multidimensional feature data is collected in 15-minute increments and labeled using a spatiotemporal coding method (i.e., cell ID + time series) to associate various datasets. The labeling format is CXTySz (where CX represents the cell identifier, Ty represents the time identifier, Sz represents each service feature dataset, and the subscript Z represents the data dimension or feature of different datasets), thereby ensuring the accurate correspondence of various types of data in time and space dimensions.

[0078] In terms of acquiring and constructing business type feature data, we abandoned the traditional crude classification method of simply dividing businesses into large packets, medium packets, and small packets. Instead, we comprehensively considered multiple factors such as bandwidth requirements and latency sensitivity of businesses, collected all B-domain data from the cell, and performed data association. From the XDR data, we extracted 23 types of businesses (such as live streaming, video, payment, and games) to form a business feature dataset. Subsequently, the dataset underwent dimensionality reduction. The uplink business feature dataset was reduced to three types: real-time interaction (including live streaming, games, instant messaging, VoIP, etc., characterized by high real-time performance, low latency, and primarily using UDP protocol), data transmission (including cloud storage, characterized by high bandwidth, primarily using TCP protocol, and emphasizing reliable transmission), and transaction processing (including payment, navigation, characterized by low latency but not real-time, hybrid protocols, and emphasizing security). The downlink business feature dataset was reduced to five types: real-time interaction, streaming media transmission, large file transmission, transaction processing, and lightweight information services. Next, the dimensionality-reduced dataset was standardized, mapping the values ​​to the [0,1] interval. Then, a business traffic ratio vector was constructed using the central logarithmic ratio transformation (CLR), and the geometric mean was calculated to obtain the transformed vector. Finally, a Gaussian mixture model (GMM) based on Mahalanobis distance was used for cluster analysis, and the optimal number of clusters was automatically determined using the Bayesian information criterion (BIC), thereby achieving accurate classification of business types and forming business type features.

[0079] For coverage quality characteristics, MR data and PUSCH RSRP data from all cells in the network are collected to form a coverage quality dataset (Z=148). This dataset contains the occurrence frequency of different RSRP values ​​(ranging from -60dBm to -140dBm). The discrete MR and PUSCH RSRP values ​​are reconstructed into continuous probability density functions through probability density estimation (CDF), and the cumulative distribution function (CDF) is calculated. Then, the 80th quantile value in the CDF curve is extracted as the coverage feature threshold (e.g., CDF80th ≥ 90dBm in MR-RSRP, CDF80th ≥ -125dBm in PUSCH-RSRP). The four-quadrant method is used to classify the coverage quality levels, thereby constructing the coverage quality characteristics.

[0080] To construct user density features, data on the maximum number of users and the average number of active users in the target area were collected at a granularity of 15 minutes, forming a user density dataset (Z=2). The K-means clustering algorithm was used to classify the average number of users by density, and the elbow rule was used to determine the optimal number of clusters. Then, based on quantile regression, the maximum user fluctuation threshold within each density level was quantified, establishing a two-stage density classification model for the maximum and average number of users, thereby obtaining user density levels (e.g., high density, medium density, low density), which constitute the user density features.

[0081] Finally, the business type features, coverage quality features, and user density features constructed above are correlated and fused through spatiotemporal coding (CXTy) to construct a three-dimensional feature space of "business type-coverage quality-user density", thereby realizing an accurate profile of the target area's community scenarios and laying the foundation for subsequent resource benefit assessment.

[0082] Step 102: Input the three-dimensional feature space into the dynamic mapping model to obtain the prediction results of resource efficiency under different business scenarios output by the dynamic mapping model;

[0083] The dynamic mapping model is a scenario-based dynamic prediction model for the "PRB utilization rate-traffic" efficiency ratio, built on the random forest regression algorithm. It is constructed separately for each sub-segment of the three-dimensional feature space (i.e., a specific scenario formed by the combination of cluster labels of service type features, coverage quality level, and user density level). First, during the model training phase, the cell-level PRB utilization rate and traffic data corresponding to each scenario in the three-dimensional feature space (obtained from the original dataset for cell efficiency ratio evaluation, including uplink and downlink PRB utilization rates and traffic, and synchronously collected at a 15-minute granularity) are divided into training and test sets. Next, the random forest regression model is trained using the training set. Through ensemble learning of multiple decision trees, the nonlinear mapping relationship between PRB utilization rate and traffic under different scenarios is captured, especially the complex dynamic correlation brought about by spatial division multiplexing technology in 5G networks. During training, the mean squared error of the model is calculated using the test set to evaluate model performance and ensure the model's prediction accuracy. When the three-dimensional feature space is input into this dynamic mapping model, the model matches the corresponding sub-scenario model based on the specific service type, coverage quality, and user density characteristics of the input, and then outputs the predicted resource efficiency results for that scenario, namely, the prediction of the dynamic correlation between PRB utilization and traffic, providing a data foundation for subsequent classification of resource efficiency levels. Among them, the random forest regression algorithm, with its ability to process high-dimensional data and capture nonlinear relationships, can effectively adapt to the complex scenarios brought about by the diversity of services, dynamic coverage, and changes in user density in 5G networks, ensuring the accuracy of the prediction results.

[0084] Step 103: Based on a preset level classification algorithm, the prediction results are classified into benefit levels to obtain the evaluation results of the target area.

[0085] The predicted results output by the dynamic mapping model are classified into benefit levels based on a preset level classification algorithm to obtain the evaluation results of the target area. The specific process is as follows: The preset level classification algorithm adopts the variance-weighted quartile method, which can adaptively determine the classification threshold by combining the distribution characteristics of the data. After obtaining the predicted results of PRB utilization and traffic under different business scenarios, the predicted results of each scenario are divided into three benefit levels: Good, Normal, and Bad, using the variance-weighted quartile method. Good represents excellent resource benefit, Normal represents average resource benefit, and Bad represents poor resource benefit.

[0086] After completing the classification, the number of Good, Normal, and Bad grades for each community within the target area under different three-dimensional feature scenarios of "service type-coverage quality-user density" throughout the entire time period is counted. Subsequently, the corresponding proportions are calculated based on the number of each grade: the proportion of Good is the ratio of the number of that grade to the total number, and the proportions of Normal and Bad are calculated similarly.

[0087] The resource efficiency of each community is comprehensively evaluated by combining the Multidimensional Health Index (MHI). The formula for calculating MHI is: MHI = 0.68 × Good percentage + 0.24 × Normal percentage - 0.08 × Bad percentage. This formula quantifies the overall resource efficiency level of the community by assigning different weights to different levels (Good has the highest weight and Bad has a negative weight).

[0088] Finally, based on the above calculation results, a three-level benefit ratio heatmap of "cell-bad scenario-dimensional" is output. This heatmap intuitively shows that each cell in the target area has poor resource efficiency (Bad level) under which specific scenarios (i.e. specific service type, coverage quality, user density combination) and the corresponding impact dimensions, thus forming a complete evaluation result of the target area and providing clear data support for the precise optimization of 5G network.

[0089] In some embodiments, before inputting the three-dimensional feature space into a dynamic mapping model to obtain the prediction results of resource efficiency under different business scenarios output by the dynamic mapping model, the method further includes:

[0090] A dynamic mapping model between physical resource block utilization and traffic under different scenarios is established based on the random forest regression algorithm.

[0091] A dynamic mapping model between Physical Resource Block (PRB) utilization and traffic under different scenarios is established based on the random forest regression algorithm. The specific process is as follows: First, based on the three-dimensional feature scenario profile of "service type-coverage quality-user density", corresponding PRB utilization and traffic data are collected for each sub-scenario (i.e., a specific scenario formed by the combination of service type cluster labels, coverage quality level, and user density level) as the raw data for model training. These data come from the original dataset of cell benefit ratio evaluation. This dataset is collected synchronously with a granularity of 15 minutes and includes indicators such as uplink PRB utilization, uplink traffic, downlink PRB utilization, and downlink traffic. The accuracy and relevance of the data are ensured by using an uplink / downlink decoupling collection strategy.

[0092] The dataset for each scenario is divided into a training set and a test set. The training set is used for model training, and the test set is used for model performance validation. During model construction, the characteristics of the random forest regression algorithm are utilized to improve the model's generalization ability and prediction accuracy by integrating the prediction results of multiple decision trees. This effectively handles the nonlinear mapping relationship between PRB utilization and traffic caused by factors such as 5G network spatial multiplexing technology and service diversity, overcoming the shortcomings of traditional linear regression models in capturing complex dynamic relationships.

[0093] After model training, the model is validated using a test set. The predictive performance is evaluated by calculating metrics such as mean squared error. If the performance falls short of expectations, the parameters of the random forest regression model (such as the number of decision trees and tree depth) are adjusted, and the model is retrained until it meets the preset accuracy requirements. Finally, for each three-dimensional feature scenario of "business type-coverage quality-user density," a corresponding PRB utilization-traffic random forest regression model, i.e., a dynamic mapping model, is established. This model can output the dynamic correlation prediction results between PRB utilization and traffic based on the input scenario features and related parameters, providing accurate model support for subsequent benefit level classification.

[0094] Please see Figure 2 , Figure 2 This is a flowchart illustrating a resource benefit assessment method provided in an embodiment of the present disclosure, as shown below. Figure 2 As shown, it includes:

[0095] Step 201: Cluster the business type features based on the Gaussian mixture model of Mahalanobis distance for dimensionality reduction;

[0096] The process of clustering and dimensionality reduction of business type features based on Gaussian Mixture Model (GMM) with Mahalanobis distance is as follows: First, the business feature data that has been preprocessed is prepared. This data is obtained by parsing 23 types of business from XDR data after abandoning the traditional coarse classification and taking into account factors such as business bandwidth requirements and latency sensitivity. The uplink business feature dataset is reduced to 3 types: real-time interaction, data transmission, and transaction processing (CXTyBzu'), and the downlink business feature dataset is reduced to 5 types, including real-time interaction and streaming media transmission (CXTyBzd'). Then, these dimensionality-reduced datasets are standardized to map the values ​​to the [0,1] interval. Then, the business traffic ratio vector is constructed by the central logarithmic ratio transformation (CLR) and the geometric mean is calculated to obtain the transformed vector (such as the vector φ=(φ1,φ2,φ3) corresponding to CXTyBzu'), which provides a suitable data foundation for clustering.

[0097] Next, a clustering analysis based on Mahalanobis distance is performed. Mahalanobis distance can take into account the correlation between features and more accurately measure the similarity between data points, making it suitable for data with complex relationships, such as business features. Simultaneously, to automatically determine the optimal number of clusters, the Bayesian Information Criterion (BIC) is introduced. By calculating the BIC value of the model under different numbers of clusters (between 2 and 10), the cluster number corresponding to the minimum BIC value is selected as the optimal number of clusters (for example, implemented using the GaussianMixture module of the sklearn library in Python, iteratively calculating the BIC value of different n_components and taking the cluster number corresponding to the minimum value).

[0098] Finally, the optimal GMM model is used to cluster the service feature data after CLR transformation, and the clustering results are output and labeled as CXTyBui (i is the cluster label). Each cluster corresponds to a combination of service types with similar network demand characteristics (such as interaction-driven, transaction-intensive, etc.), which realizes further clustering and dimensionality reduction of service type features, making service classification more accurate and able to more accurately reflect the network resource requirements of different services, laying the foundation for the subsequent construction of a three-dimensional feature space of "service type-coverage quality-user density".

[0099] Step 202: Dynamically classify the coverage quality characteristics based on the measurement report of the target area and the cumulative distribution function of the physical uplink shared channel;

[0100] The process of dynamically classifying coverage quality characteristics based on the Measurement Report (MR) and the Cumulative Distribution Function (CDF) of the Physical Uplink Shared Channel (PUSCH) of the target area is as follows: First, coverage quality data of cells in the target area is collected. This data is granular in 15-minute intervals and is labeled as CXTyMz (Z=148) using spatiotemporal coding (Cell ID + time series). It includes the number of occurrences of MR RSRP and PUSCH RSRP corresponding to different RSRP values ​​(from -60dBm to -140dBm). For example, the number of occurrences of each value of MR RSRP from -60dBm to -128dBm and the number of occurrences of each value of PUSCH RSRP from -60dBm to -140dBm.

[0101] The collected discrete MR RSRP and PUSCH RSRP data are subjected to probability density estimation, reconstructed into a continuous probability density function, and the corresponding cumulative distribution function (CDF) is calculated. The CDF can reflect the cumulative probability of different RSRP values. For example, the CDF of MR RSRP -88dBm at 18:00 in a certain cell is 68.80%, which means that the probability of RSRP value and above is 68.80%.

[0102] The 80th quantile of the CDF curve is extracted as a coverage feature threshold. This threshold effectively reflects the overall level of coverage quality. For example, a CDF 80th ≥ -90dBm for MR RSRP and a CDF 80th ≥ -125dBm for PUSCH RSRP are set as reference thresholds for good coverage. Finally, the four-quadrant method is used to dynamically classify coverage quality based on the above thresholds, dividing coverage quality into different levels (denoted as CXTyMi, i = 1, 2, 3, 4). This achieves accurate quantification of coverage quality features and provides accurate features for constructing a three-dimensional feature space of "service type - coverage quality - user density".

[0103] Step 203: Perform a two-stage density classification on the user density features based on K-means clustering and quantile regression models.

[0104] The process of performing a two-stage density classification of user density features based on K-means clustering and quantile regression models is as follows: First, a user density dataset CXTyUz (Z=2) is obtained. This dataset is collected at a 15-minute granularity and contains data on the maximum number of users and the average number of active users in each cell within the target area. The historical average user count sequence of this dataset is preprocessed; Z-score standardization can be used to eliminate the influence of data units, providing a more stable data foundation for subsequent clustering.

[0105] The first stage employs the K-means clustering algorithm: The sum of squared errors (ss) corresponding to different k values ​​(number of clusters) is calculated, and a curve showing ss changing with k is plotted. The optimal number of clusters k is determined using the elbow rule (usually divided into three levels: high density, medium density, and low density). Based on the optimal k value, the standardized average user count data is clustered to obtain k subsets, and the data is further divided according to cluster labels, thus classifying the average user count into different initial density levels.

[0106] The second stage introduces a quantile regression model: For each density level subset obtained from clustering, variables are defined (average number of users avg is the independent variable, and maximum number of users u_max is the dependent variable). An appropriate quantile τ is selected, and the regression problem is solved to obtain the regression equation u_max=β0+β1·avg for each level (where β0 and β1 are regression coefficients, and the coefficients are different for different levels, such as β0_h+β1_h·avg for high density level, β0_m+β1_m·avg for medium density level, and β0_l+β1_l·avg for low density level). This is used to quantify the maximum user fluctuation threshold within each density level.

[0107] Ultimately, a two-stage classification method is used to categorize user density features: first, a basic density level (e.g., high density, medium density, low density) is determined based on the cluster label of the average number of users; then, it is determined whether the maximum number of users exceeds the regression threshold of the corresponding level. If it does, an anomaly alarm is triggered; otherwise, the density level is confirmed. The categorization results include the average number of users range for each level (e.g., μ_high±δ, μ_mid±δ, μ_low±δ, where μ is the center value and δ is the cluster radius), the maximum number of users threshold, and the fluctuation range characteristics (high volatility, medium volatility, low volatility), thereby completing the accurate categorization of user density features and providing user-dimensional feature support for constructing a three-dimensional feature space of "business type-coverage quality-user density".

[0108] In some embodiments, acquiring multidimensional feature data of the target region and constructing a three-dimensional feature space based on the multidimensional feature data further includes:

[0109] The business type features are dimensionality reduced, reducing the uplink business feature dataset to three business types: real-time interaction, data transmission, and transaction processing; and reducing the downlink business feature dataset to five business types: real-time interaction, streaming media transmission, large file transmission, transaction processing, and lightweight information service. The business type features include both the uplink and downlink business feature datasets.

[0110] The traffic of the uplink business feature dataset and the downlink business feature dataset in the business feature dataset is summed and normalized;

[0111] The normalized flow rate is subjected to feature transformation based on the central logarithmic ratio transform.

[0112] When processing business type features to construct the business type dimension in the three-dimensional feature space, dimensionality reduction is first performed: For the uplink business feature dataset CXTyBzu, traditional coarse classification is abandoned. Considering factors such as business bandwidth requirements and latency sensitivity, the seven business types it contains—live streaming, games, instant messaging, VoIP, cloud storage, payment, and navigation—are dimensionalized into three types based on network characteristics: real-time interaction (the sum of live streaming, games, instant messaging, and VoIP, characterized by high real-time performance, low latency, and primarily using UDP protocol), data transmission (cloud storage, characterized by high bandwidth, primarily using TCP protocol, and emphasizing reliable transmission), and transaction processing (the sum of payment and navigation, characterized by low latency but not real-time, hybrid protocols, and emphasizing security), forming the dimensionality-reduced uplink business feature dataset CXTyBzu'; For the downlink business feature dataset CXTyBzd, its contents are... The dataset includes 18 service types such as instant messaging, microblogging communities, video, music, animation, app stores, cloud storage services, P2P services, browsing and downloading, shopping, payment, finance, security and antivirus, reading, microblogging, email, MMS, and travel. Based on network characteristics, these are dimensionalized into 5 types: Real-time interactive (the sum of instant messaging and microblogging communities, high real-time performance, low latency, primarily using UDP protocol), streaming media transmission (the sum of video, music, and animation, continuous high bandwidth), large file transmission (the sum of app stores, cloud storage services, P2P services, and browsing and downloading, bursty throughput), transaction processing (the sum of shopping, payment, finance, and security and antivirus, low latency but not real-time, mixed protocols, emphasizing security), and lightweight information services (the sum of reading, microblogging, email, MMS, and travel, intermittent access, low bandwidth). This forms the dimensionality-reduced downlink service characteristic dataset CXTyBzd'.

[0113] The values ​​of real-time interaction traffic, data transmission traffic, and transaction processing traffic in the uplink business feature dataset CXTyBzu', and the values ​​of the five types of traffic in the downlink business feature dataset CXTyBzd', are summed and normalized according to their respective dimensions. That is, each traffic value is divided by the sum of all types of traffic in the corresponding dataset, and the value is mapped to the interval [0,1]. The formulas are CXTyBzu'_z=CXTyBzu_z / Σ(CXTyBzu_z) (z=1,2,3) and CXTyBzd'_z=CXTyBzd_z / Σ(CXTyBzd_z) (z=1,2,3,4,5). This eliminates the influence of data of different magnitudes and facilitates subsequent feature analysis.

[0114] For the normalized uplink business feature dataset CXTyBzu' (3 dimensions) and downlink business feature dataset CXTyBzd' (5 dimensions), business traffic ratio vectors P (uplink P1, P2, P3, downlink P1, P2, P3, P4, P5) are constructed respectively. The geometric mean GM of the samples is calculated, with the formula GM = EXP(1 / nΣln(pi)) (n is the number of dimensions, n = 3 for uplink and n = 5 for downlink). The transformed vector φ is calculated based on the geometric mean, where φ_i = ln(pi / GM) (i corresponds to each dimension). This transformation converts the original traffic ratio data into feature vectors suitable for subsequent clustering analysis, effectively handling the correlation and ratio characteristics between data, and providing high-quality feature input for subsequent Gaussian mixture model clustering based on Mahalanobis distance.

[0115] Please see Figure 3 , Figure 3 This is a flowchart illustrating a resource benefit assessment method provided in an embodiment of the present disclosure, as shown below. Figure 3 As shown, it includes:

[0116] Step 301: Based on the variance-weighted quartile method, the prediction results are classified into benefit levels to obtain the classification level of the target area;

[0117] The process of classifying the prediction results output by the dynamic mapping model based on the variance-weighted quartile method to obtain the classification level of the target area is as follows: First, it is clarified that the prediction results are dynamic correlation data of PRB utilization and traffic under different three-dimensional feature scenarios of "business type-coverage quality-user density". These data reflect the benefit relationship between resource use and business traffic in each scenario.

[0118] The core of the variance-weighted quartile method lies in adjusting the traditional quartile thresholds based on the variance characteristics of the data to more accurately adapt to the distribution characteristics of data in different scenarios. Specifically, the quartiles of the predicted data are first calculated (including the first quartile Q1, the second quartile Q2, and the third quartile Q3), where Q1 is the value at the 25th percentile, Q2 is the value at the 50th percentile (median), and Q3 is the value at the 75th percentile. Then, these quartile thresholds are weighted and adjusted according to the variance of the data (reflecting the degree of data dispersion): for scenarios with larger variance (more volatile data), the interval between Q1 and Q3 is appropriately widened to avoid misjudgments caused by short-term fluctuations; for scenarios with smaller variance (more stable data), the interval is narrowed to improve the sensitivity of the grading.

[0119] Based on the revised quartile thresholds, the prediction results are divided into three benefit levels: Those above the Q3 weighted threshold are classified as Good, representing excellent resource efficiency, high PRB utilization and traffic matching, and efficient resource use; those between the Q1 and Q3 weighted thresholds are classified as Normal, representing average resource efficiency, reasonable PRB utilization and traffic matching, but with room for optimization; and those below the Q1 weighted threshold are classified as Bad, representing poor resource efficiency, low PRB utilization and traffic matching, and potential resource waste or shortage. This weighted classification method, which incorporates the discrete characteristics of the data, can more accurately adapt to the dynamic fluctuations of different service scenarios in 5G networks, ensuring that the classification level truly reflects the resource efficiency status of each scenario, thereby obtaining the specific classification level of each cell in the target area under different three-dimensional feature scenarios.

[0120] Step 302: Calculate the health index of the target area based on the classification level to obtain the evaluation result of the target area; wherein, the proportion of the health index is different for different classification levels.

[0121] The process of calculating the health index of the target area based on the classification levels (Good, Normal, Bad) obtained in step 301, and then obtaining the evaluation results, is as follows: First, for each cell within the target area, the number of cells at the Good, Normal, and Bad levels is counted under different three-dimensional feature scenarios (i.e., the B_iM_jU_k scenario) across all time periods. The statistical time period for each cell is determined based on a 15-minute spatiotemporal coding (CXTy), covering all monitoring time periods for that cell to ensure data comprehensiveness.

[0122] Calculate the proportion of each grade in the corresponding scenario: the Good proportion is the ratio of the number of Good grades to the total number of grades in the scenario (Good + Normal + Bad). The Normal and Bad proportions are calculated in the same way. These proportions reflect the distribution of each grade in the overall resource benefits of the community.

[0123] The health index of each community is calculated using the Multidimensional Health Index (MHI) formula: MHI = 0.68 × Good percentage + 0.24 × Normal percentage - 0.08 × Bad percentage. The weighting of the health index varies depending on the classification level. The Good level, representing excellent resource efficiency, is given the highest weight (0.68), the Normal level, representing average resource efficiency, is given a medium weight (0.24), and the Bad level, representing poor resource efficiency, is given a negative weight (-0.08). This differentiated weighting highlights the contribution of high-quality benefits and suppresses the negative impact of low-quality benefits, thus more accurately quantifying the resource efficiency level of a community.

[0124] By combining the MHI values ​​of each cell, a three-level benefit ratio heatmap of "cell-bad scenario-dimensional" is output. This heatmap intuitively displays the distribution of health index of each cell in the target area, as well as the scenarios and dimensions where the Bad level is concentrated, thus forming the evaluation results of the target area and providing data-driven decision support for the precise optimization of 5G network.

[0125] In some embodiments, after classifying the prediction results into benefit levels based on a preset classification algorithm to obtain the evaluation results of the target area, the method further includes:

[0126] A three-level benefit ratio heatmap is generated based on the health index ratio and displayed on a preset screen.

[0127] The process of generating a three-level benefit ratio heatmap based on the health index ratio and displaying it on a preset screen is as follows: After obtaining the health index (MHI) of each cell in the target area and the proportion of each classification level (Good, Normal, Bad), a three-level benefit ratio heatmap is generated with "cell-bad scenario-dimension" as the core structure. Among them, the "cell" dimension corresponds to the specific cell identifier (such as CX) in the target area, the "bad scenario" dimension points to the three-dimensional feature scenario of "business type-coverage quality-user density" that is classified as Bad (i.e., the combination marked as Bad in B_iM_jU_k), and the "dimension" is refined to the specific influencing factors that cause the Bad scenario (such as the transaction-intensive characteristics of the business type dimension, the low RSRP level of the coverage quality dimension, the high density fluctuation of the user density dimension, etc.).

[0128] The heatmap is generated based on the distribution density of Bad levels in each cell and the corresponding health index ratio: the distribution of Bad levels in each cell is presented intuitively through different color depths or color block intensities. The darker the color, the higher the proportion of Bad scenarios and the lower the health index in that cell, and the more prominent the resource efficiency problem. At the same time, the heatmap displays the specific dimensional information that leads to the Bad level. For example, if a cell has an excessively high proportion of Bad in the "real-time interactive services - low coverage quality - high density users" scenario, the heatmap will mark the scenario and the main influencing dimensions (such as a low RSRP value in the coverage quality dimension) at the corresponding position.

[0129] The generated three-level benefit ratio heatmap is displayed on preset screens (such as the display interface of the network operation and maintenance management platform, the large screen of the monitoring center, etc.), enabling operation and maintenance personnel to intuitively and quickly locate cells with poor resource efficiency, specific bad scenarios and key influencing dimensions in the target area. This provides a visualized decision-making basis for the precise optimization of 5G networks (such as targeted adjustment of coverage quality, optimization of user density scheduling, and resource allocation adapted to service types, etc.), and ultimately forms a complete target area assessment result display process.

[0130] In some embodiments, classifying the prediction results into benefit levels based on the variance-weighted quartile method to obtain the classification level of the target area includes:

[0131] The measurement reports and physical uplink shared channel values ​​in the coverage quality dataset are reconstructed into a continuous probability density function through probability density estimation, and the cumulative distribution function is calculated.

[0132] Based on the cumulative distribution function curve, a preset quantile value is extracted as the coverage quality feature threshold;

[0133] Based on the distribution characteristics of the upstream and downstream business feature datasets, the four-quadrant method is used to classify the coverage quality levels.

[0134] The reference received power (RSRP) values ​​of the Measurement Report (MR) and Physical Uplink Shared Channel (PUSCH) in the coverage quality dataset CXTyMz are reconstructed into continuous probability density functions through probability density estimation, and the cumulative distribution function (CDF) is calculated. The coverage quality dataset CXTyMz contains the occurrence frequency of different RSRP values ​​(MR RSRP from -60dBm to -128dBm, PUSCH RSRP from -60dBm to -140dBm) for each cell in 15-minute granularities. The probability density estimation fits these discrete occurrence frequencies to transform them into a continuous probability density function that reflects the distribution pattern of RSRP values. The cumulative distribution function (CDF) is further calculated to represent the cumulative probability of an RSRP value less than or equal to a certain value. For example, the CDF of a cell with an MR RSRP of -88dBm in a specific time period is 68.80%, meaning that the cumulative probability of an RSRP value of 68.80% or higher is 68.80%.

[0135] Preset quantile values ​​are extracted from the cumulative distribution function curve as coverage quality characteristic thresholds: the preset quantile value is selected as the 80th quantile of the CDF curve. This quantile value can effectively reflect the overall level of coverage quality because it represents the signal strength achieved by 80% of the RSRP measurements, and has strong statistical representativeness. For example, setting CDF80th ≥ -90dBm for MR RSRP and CDF80th ≥ -125dBm for PUSCH RSRP as characteristic thresholds for good coverage quality, where CDF80th for MR RSRP reflects the overall strength of the reference signal received by the terminal, and CDF80th for PUSCH RSRP reflects the signal transmission quality of the uplink channel.

[0136] Based on the distribution characteristics of the uplink and downlink service feature datasets, a four-quadrant method is adopted to classify coverage quality levels. This method uses the 80th quantile of MR RSRP and PUSCH RSRP as coordinate axes to divide coverage quality into four levels (CXTyMi, i = 1, 2, 3, 4). Specifically, when both MR RSRP and PUSCH RSRP's CDF80th values ​​are above their corresponding thresholds, it is level 1 (excellent coverage); when MR RSRP is above the threshold but PUSCH RSRP is below, it is level 2 (uplink coverage needs optimization); when MR RSRP is below the threshold but PUSCH RSRP is above, it is level 3 (downlink coverage needs optimization); and when both are below the threshold, it is level 4 (poor overall coverage). This four-quadrant division, combining uplink and downlink service feature distributions, can accurately quantify coverage quality levels in different scenarios, providing accurate feature support for subsequent benefit level classification based on variance-weighted quartiles, ultimately yielding the coverage quality classification level for the target area.

[0137] In some embodiments, the two-stage density classification of the user density features based on K-means clustering and quantile regression models further includes:

[0138] The elbow rule is used to determine the optimal number of clusters, and the average number of users is divided into a preset number of density levels.

[0139] Based on the average number of users at each density level, the quantile regression model is constructed to quantify the maximum user number fluctuation threshold.

[0140] Determining the optimal number of clusters using the elbow rule and quantifying the maximum user number fluctuation threshold based on a quantile regression model are key steps in performing two-stage density grading of user density features. First, for the user density dataset CXTyUz (containing the maximum number of users and the average number of active users at a 15-minute granularity), the historical average user number sequence is extracted as input data for K-means clustering. Z-score standardization can be used to eliminate the influence of differences in user size across different cells. The sum of squared errors (ss) corresponding to different cluster numbers k (usually increasing from 2) is calculated, and a curve of ss versus k is plotted. The k value corresponding to the "elbow" (i.e., the point where the rate of ss decreases significantly) on the curve is the optimal number of clusters. This method typically divides the average number of users into three preset density levels: high density, medium density, and low density. The ranges of the average number of users corresponding to each level are μ_high±δ, μ_mid±δ, and μ_low±δ, respectively (μ is the center value, and δ is the cluster radius).

[0141] Based on the density levels, a quantile regression model is constructed to quantify the maximum user number fluctuation threshold: For each density level subset obtained from clustering, the average number of users (avg) is defined as the independent variable and the maximum number of users (u_max) is defined as the dependent variable. An appropriate quantile τ (reflecting the fluctuation boundary of the maximum number of users) is selected, and the regression problem is solved to obtain the regression equation u_max=β0+β1·avg for each level. The equation for high density level is u_max=β0_h+β1_h·avg, for medium density it is u_max=β0_m+β1_m·avg, and for low density it is u_max=β0_l+β1_l·avg (β0 and β1 are regression coefficients). This equation quantifies the fluctuation threshold of the maximum number of users as a function of the average number of users at each density level. Combined with the range of the average number of users, it forms a two-stage classification basis: first, the basic density level is determined by the cluster label of the average number of users, and then an anomaly is triggered based on whether the maximum number of users exceeds the regression threshold of the corresponding level. Finally, it achieves accurate classification of user density characteristics. The results include the range of the average number of users, the threshold of the maximum number of users, and the fluctuation range characteristics (such as high density corresponding to high volatility) for each level.

[0142] The resource benefit assessment method provided in this application is illustrated below with an example: Multidimensional feature data acquisition The multidimensional data was collected at a 15-minute granularity and labeled using a spatiotemporal coding method (cell ID + time series) to associate each data level, namely CXTySz (CX represents cell identifier, Ty represents time identifier, Sz represents each service feature dataset, and the subscript Z represents the data dimension or feature of different datasets). 1.1 Business Feature Dataset Abandoning traditional, extensive classification methods, and comprehensively considering factors such as service bandwidth requirements and latency sensitivity, all cell B-domain data was collected and correlated. Twenty-three service categories (such as live streaming, video, and payment) were extracted from the XDR data. Taking into account the uplink or downlink characteristics of these 23 service categories, the data was organized and labeled to form an uplink service feature dataset CXTyBzBu and a downlink service feature dataset CXTyBzd. For ease of computation and processing, The upstream business feature dataset is in CXTyBzu format (Z is 7) (covering 7 business types including live streaming, payment, games, navigation, instant messaging, cloud storage services, and VoIP). Please refer to Table 1, which is an upstream business feature dataset provided in this application embodiment. Table 1

[0150] The downlink service feature dataset is in CXTyBzd format (Z is 19) and includes 18 service types such as reading, microblogs, microblog communities, videos, music, app stores, animations, email, P2P services, MMS, browsing and downloading, finance, security and antivirus, shopping, travel, public traffic, instant messaging, and cloud storage services; please refer to Table 2, which is a downlink service feature dataset provided in the embodiments of this application:

[0151] Table 2

[0152]

[0153] 1.2 Coverage Quality Dataset

[0154] A coverage quality assessment dimension for the PUSCH channel is introduced, establishing a dual assessment system for uplink and downlink coverage quality. MR data and PUSCH RSRP data from all cells in the network are collected to construct a coverage quality dataset CXTyMz (where Z is 148). Please refer to Table 3, which presents a coverage quality dataset provided in an embodiment of this application.

[0155] Table 3

[0156]

[0157] 1.3 User Density Dataset

[0158] Data on the maximum and average number of users were collected at a granularity of 15 minutes to construct a user density dataset CXTyUz (where Z is 2). Please refer to Table 4, which provides one such user density dataset according to an embodiment of this application.

[0159] Table 4

[0160]

[0161] 1.4 Original Dataset for Community Benefit Ratio Evaluation

[0162] Using a granularity of 15 minutes, cell-level PRB utilization and service traffic metrics are synchronously acquired. A benefit assessment dataset CXTyQz (with Z = 4) is constructed through a decoupled uplink (UL) / downlink (DL) collection strategy. Please refer to Table 5, which provides one such benefit assessment dataset in this application embodiment.

[0163] Table 5

[0164]

[0165] 2. Multidimensional feature dataset processing and sub-model establishment

[0166] 2.1 Business Feature-Based Model

[0167] 2.1.1 Order Reduction Processing

[0168] Based on the characteristics of network requirements of the business, the uplink business feature dataset Bu and the downlink business feature dataset CXTyBzd are dimensionality reduced respectively. CXTyBzu is dimensionality reduced to three types of datasets CXTyBzu': real-time interaction, data transmission, and transaction processing; Bd is dimensionality reduced to five types of datasets CXTyBzd': real-time interaction, streaming media transmission, large file transmission, transaction processing, and lightweight information service.

[0169] CXTyBzu dataset network features

[0170] Real-time interactive (∑live streaming, gaming, instant messaging, VoIP): high real-time performance, low latency, primarily using UDP protocol.

[0171] Data transmission type (∑cloud storage): high bandwidth, TCP protocol, focusing on reliable transmission.

[0172] Transaction processing type (∑payment, navigation): low latency but not real-time, hybrid protocol, emphasizing security; CXTyBzu' dataset (Z is 3). Please refer to Table 6, which is a dimensionality-reduced uplink business feature dataset provided in the embodiments of this application, as shown in Table 6:

[0173] Table 6

[0174]

[0175] CXTyBzd dataset network features

[0176] Real-time interactive (instant messaging, microblogging communities): high real-time performance, low latency, primarily using UDP protocol.

[0177] Streaming media transmission type (∑video, music, animation): continuous high bandwidth

[0178] Large file transfer (∑app stores, cloud storage services, P2P services, browsing and downloading): Burst throughput transaction processing (∑shopping, payment, finance, security and antivirus): (Low latency but not real-time, hybrid protocols, emphasizing security)

[0179] Lightweight messaging services (∑reading, microblog, email, MMS, travel): intermittent access, low bandwidth. (Dataset CXTyBzd', Z = 5)

[0180] Please refer to Table 7, which is a dimensionality-reduced downlink service feature dataset provided in the embodiments of this application, as shown in Table 7:

[0181] Table 7

[0182]

[0183] 2.1.2 Data Standardization Processing

[0184] The traffic of each service in the CXTyBzu' and CXTyBzd' datasets is summed and normalized to map all values ​​to the [0,1] interval.

[0185]

[0186] Since the CXTyBzu' and CXTyBzd' datasets are built using the same method, differing only in their data dimensions, the following section will only present the CXTyBzu' construction process.

[0187] 2.1. Central Log-Ratio Transformation (CLR)

[0188] For the standardized data of CXTyBzu' and CXTyBzd', a business traffic ratio vector P = (P1, P2, P3, ..., Pn) is constructed respectively, with 3 for the CXTyBzu' dataset and 5 for the CXTyBzd' dataset. The geometric mean GM of the samples is obtained.

[0189]

[0190] Where n = 3, the vector obtained after CLR transformation is... for

[0191]

[0192] 2.1.4 GMM Clustering Based on Mahalanobis Distance

[0193] Cluster analysis was performed using a Gaussian Mixture Model (GMM) based on Mahalanobis distance. The Bayesian Information Criterion (BIC) was also selected to automatically determine the optimal number of clusters.

[0194] 2.1.5 Clustering Results Illustration

[0195] The business type clustering results are automatically output according to sections 2.1.1-2.1.4 and labeled as CXTyBui (i is the cluster label). Please refer to Table 8, which is a schematic table of business type clustering results provided in an embodiment of this application, as shown in Table 8:

[0196] Table 8

[0197]

[0198] 2.2 Coverage Quality Level Model

[0199] 2.2.1 Constructing a continuous probability density function

[0200] The frequency of occurrences of discrete MR and PUSCH RSRP values ​​in the coverage quality dataset CXTyMz is reconstructed into continuous probability density functions using probability density estimation (CDF), and the cumulative distribution function (CDF) is calculated. Please refer to Table 9, which is a schematic table of business type clustering results provided in an embodiment of this application, as shown in Table 9:

[0201] Table 9

[0202]

[0203] 2.2.2 Constructing a Coverage Quality Model

[0204] By calculating the key quantile, the 80th quantile value in the CDF curve is extracted as the coverage feature threshold (e.g., CDF80th ≥ 90dBm in MR-RSRP, CDF80th ≥ -125dBm in PUSCH-RSRP). The coverage quality level is divided using the four-quadrant method and denoted as CXTyMi (i = 1, 2, 3, 4).

[0205] 2.3 User Density Number Model

[0206] The average number of users in CXTyUz is classified into density levels using K-means clustering, and the elbow rule is used to determine the optimal number of clusters. A two-stage density classification model based on quantile regression is established to quantify the maximum user fluctuation threshold within each density level, and to assess the maximum and average number of users.

[0207] The user density model classification is denoted as CXTyUk (k is 1, 2, 3); please refer to Table 10, which is a schematic table of user density model classification results provided in the embodiments of this application, as shown in Table 10:

[0208] Table 10

[0209]

[0210] μ_high, μ_mid, and μ_low represent the centers of high, medium, and low densities, respectively; β0_h, β0_m, and β0_l represent the intercepts of the high, medium, and low density quantile regression models, respectively; β1_h·avg, β1_m·avg, and β1_l·avg represent the average intercepts of the high, medium, and low density quantile regression models, respectively; and δ represents the cluster radius.

[0211] 3. Three-dimensional feature scene profile based on "business type - coverage quality - user density"

[0212] By linking the user density model, business feature model, and coverage quality model in series and parallel, a system is constructed. Two 3D feature scene classifications. Based on CXTy spatiotemporal coding, cell uplink PRB utilization and uplink traffic are mapped to... In each scenario profile; based on CXTy spatiotemporal coding, the cell downlink PRB utilization rate and downlink traffic are mapped to... Each scenario profile enables the classification of PRB utilization and traffic in different communities at different times.

[0213] 4. PRB utilization-flow random forest regression model

[0214] For different 3D feature scene profiles, a PRB utilization-traffic random forest regression model is constructed, and the quartile method is used to divide different cells and different time periods into three levels: good, normal, and bad.

[0215] 5. Community Benefit Assessment

[0216] Statistics on different scenarios in each community throughout the entire time period i M j U kThe number of good, normal, and bad grades in the profile is used to intelligently evaluate the community resource (PRB) benefit ratio using a multi-dimensional health index (MHI = 0.68 × Good percentage + 0.24 × Normal percentage - 0.08 × Bad percentage), and outputs a three-level benefit ratio heatmap of "community-bad scenario-dimensional".

[0217] According to C x T y Spatiotemporal coding reveals that: the number of cells is x, the number of time periods is y, and the three-dimensional feature scene is B. i M j U k .

[0218] Count the number of Good, Normal, and Bad.

[0219] Each community C, in scenario B i M j U k In this scenario, the number of Good grades is G. c,ijk The number of Normal grades is N. C,ijk The Bad rating is B. C,ijk .

[0220] Calculate the percentage of each level

[0221] Good percentage

[0222] Normal percentage

[0223] Bad percentage

[0224] Multidimensional Health Index (MHI) Calculation

[0225] MHI(C x ) = 0.68 * P G,C,ijk +0.24*P G,C,ijk -0.08* G,C,ijk .

[0226] Corresponding to the resource benefit assessment method described above, this invention also proposes a resource benefit assessment device. Since the device embodiments of this invention correspond to the method embodiments described above, details not disclosed in the device embodiments can be referred to in the method embodiments described above, and will not be repeated here.

[0227] Figure 4 This is a schematic diagram of the structure of a resource benefit assessment device provided in an embodiment of the present disclosure, as shown below. Figure 4 As shown, it includes:

[0228] The acquisition unit 41 is used to acquire multi-dimensional feature data of the target area and construct a three-dimensional feature space based on the multi-dimensional feature data; wherein, the three-dimensional feature space includes service type features, coverage quality features and user density features;

[0229] Input unit 42 is used to input the three-dimensional feature space into the dynamic mapping model to obtain the prediction results of resource efficiency under different business scenarios output by the dynamic mapping model;

[0230] Evaluation unit 43 is used to classify the prediction results into benefit levels based on a preset level classification algorithm to obtain the evaluation results of the target area.

[0231] Furthermore, in one possible implementation of the embodiments of this disclosure, such as Figure 5 As shown, the device further includes:

[0232] Establishment unit 44 is used to establish the dynamic mapping model between physical resource block utilization and traffic under different scenarios based on random forest regression algorithm before the input unit 42 inputs the three-dimensional feature space into the dynamic mapping model and obtains the prediction results of resource benefits under different business scenarios output by the dynamic mapping model.

[0233] Furthermore, in one possible implementation of this disclosure, the acquisition unit 41 is further configured to:

[0234] The business type features are clustered and dimensionality reduced using a Gaussian mixture model based on Mahalanobis distance.

[0235] The coverage quality characteristics are dynamically classified based on the measurement reports of the target area and the cumulative distribution function of the physical uplink shared channel;

[0236] The user density features are classified in a two-stage density classification based on K-means clustering and quantile regression models.

[0237] Furthermore, in one possible implementation of this disclosure, the acquisition unit 41 is further configured to:

[0238] The business type features are dimensionality reduced, reducing the uplink business feature dataset to three business types: real-time interaction, data transmission, and transaction processing; and reducing the downlink business feature dataset to five business types: real-time interaction, streaming media transmission, large file transmission, transaction processing, and lightweight information service. The business type features include both the uplink and downlink business feature datasets.

[0239] The traffic of the uplink business feature dataset and the downlink business feature dataset in the business feature dataset is summed and normalized;

[0240] The normalized flow rate is subjected to feature transformation based on the central logarithmic ratio transform.

[0241] Furthermore, in one possible implementation of the embodiments of this disclosure, such as Figure 5 As shown, the evaluation unit 43 is further used for:

[0242] The prediction results are classified into benefit levels based on the variance-weighted quartile method to obtain the classification level of the target area.

[0243] The health index of the target area is calculated based on the classification level to obtain the evaluation result of the target area; wherein, the proportion of the health index is different for different classification levels.

[0244] Furthermore, in one possible implementation of this disclosure, the apparatus further includes:

[0245] The display unit 45 is used to generate a three-level benefit ratio heat map based on the health index ratio after the evaluation unit 43 classifies the prediction results into benefit levels based on the preset level classification algorithm and obtains the evaluation results of the target area, and then displays it on a preset screen.

[0246] Furthermore, in one possible implementation of this disclosure embodiment, the evaluation unit 43 is further configured to:

[0247] The measurement reports and physical uplink shared channel values ​​in the coverage quality dataset are reconstructed into a continuous probability density function through probability density estimation, and the cumulative distribution function is calculated.

[0248] Based on the cumulative distribution function curve, a preset quantile value is extracted as the coverage quality feature threshold;

[0249] Based on the distribution characteristics of the upstream and downstream business feature datasets, the four-quadrant method is used to classify the coverage quality levels.

[0250] Furthermore, in one possible implementation of this disclosure, the acquisition unit 41 is further configured to:

[0251] The elbow rule is used to determine the optimal number of clusters, and the average number of users is divided into a preset number of density levels.

[0252] Based on the average number of users at each density level, the quantile regression model is constructed to quantify the maximum user number fluctuation threshold.

[0253] It should be noted that the foregoing explanation of the method embodiments also applies to the apparatus of the embodiments of this disclosure, and the principle is the same. Therefore, the embodiments of this disclosure are not limited thereto.

[0254] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0255] Figure 6 A schematic block diagram of an example electronic device 500 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0256] like Figure 6 As shown, device 500 includes a computing unit 501, which can perform various appropriate actions and processes based on a computer program stored in ROM (Read-Only Memory) 502 or a computer program loaded from storage unit 508 into RAM (Random Access Memory) 503. RAM 503 can also store various programs and data required for the operation of device 500. The computing unit 501, ROM 502, and RAM 503 are interconnected via bus 504. I / O (Input / Output) interface 505 is also connected to bus 504.

[0257] Multiple components in device 500 are connected to I / O interface 505, including: input unit 506, such as keyboard, mouse, etc.; output unit 507, such as various types of monitors, speakers, etc.; storage unit 508, such as disk, optical disk, etc.; and communication unit 509, such as network card, modem, wireless transceiver, etc. Communication unit 509 allows device 500 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0258] The computing unit 501 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 501 include, but are not limited to, CPUs (Central Processing Units), GPUs (Graphics Processing Units), various special-purpose AI (Artificial Intelligence) computing chips, various computing units running machine learning model algorithms, DSPs (Digital Signal Processors), and any suitable processor, controller, microcontroller, etc. The computing unit 501 performs the various methods and processes described above, such as resource efficiency assessment methods. For example, in some embodiments, the resource efficiency assessment method may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 508. In some embodiments, part or all of the computer program may be loaded and / or installed on device 500 via ROM 502 and / or communication unit 509. When the computer program is loaded into RAM 503 and executed by the computing unit 501, one or more steps of the methods described above may be performed. Alternatively, in other embodiments, the computing unit 501 may be configured to perform the aforementioned resource benefit assessment method by any other suitable means (e.g., by means of firmware).

[0259] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, FPGAs (Field Programmable Gate Arrays), ASICs (Application-Specific Integrated Circuits), ASSPs (Application-Specific Standard Products), SOCs (System-on-Chips), CPLDs (Complex Programmable Logic Devices), computer hardware, firmware, software, and / or combinations thereof. These various implementations may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0260] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0261] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, RAM, ROM, EPROM (Electrically Programmable Read-Only Memory) or flash memory, optical fiber, CD-ROM (Compact Disc Read-Only Memory), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0262] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (Cathode-Ray Tube) or LCD (Liquid Crystal Display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0263] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include LANs (Local Area Networks), WANs (Wide Area Networks), the Internet, and blockchain networks.

[0264] Computer systems can include clients and servers. Clients and servers are generally geographically separated and typically interact via communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. A server can be a cloud server, also known as a cloud computing server or cloud host, a hosting product within the cloud computing service ecosystem, addressing the shortcomings of traditional physical hosts and VPS (Virtual Private Server, or simply "VPS") services, such as high management difficulty and weak business scalability. Servers can also be servers for distributed systems or servers incorporating blockchain technology.

[0265] It's important to note that artificial intelligence (AI) is the study of enabling computers to simulate certain human thought processes and intelligent behaviors (such as learning, reasoning, thinking, and planning). It encompasses both hardware and software technologies. AI hardware technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, and big data processing. AI software technologies primarily include computer vision, speech recognition, natural language processing, machine learning / deep learning, big data processing, and knowledge graph technologies.

[0266] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0267] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A resource benefit evaluation method, characterized in that, include: Acquire multidimensional feature data of the target area, and construct a three-dimensional feature space based on the multidimensional feature data; wherein, the three-dimensional feature space includes service type features, coverage quality features, and user density features; The three-dimensional feature space is input into the dynamic mapping model to obtain the prediction results of resource efficiency under different business scenarios output by the dynamic mapping model. The prediction results are classified into benefit levels based on a preset level classification algorithm to obtain the evaluation results of the target area.

2. The method according to claim 1, characterized in that, Before inputting the three-dimensional feature space into the dynamic mapping model to obtain the prediction results of resource efficiency under different business scenarios output by the dynamic mapping model, the method further includes: A dynamic mapping model between physical resource block utilization and traffic under different scenarios is established based on the random forest regression algorithm.

3. The method according to claim 1, characterized in that, The step of acquiring multidimensional feature data of the target region and constructing a three-dimensional feature space based on the multidimensional feature data further includes: The business type features are clustered and dimensionality reduced using a Gaussian mixture model based on Mahalanobis distance. The coverage quality characteristics are dynamically classified based on the measurement reports of the target area and the cumulative distribution function of the physical uplink shared channel; The user density features are classified in a two-stage density classification based on K-means clustering and quantile regression models.

4. The method according to claim 3, characterized in that, The step of acquiring multidimensional feature data of the target region and constructing a three-dimensional feature space based on the multidimensional feature data further includes: The business type features are dimensionality reduced, reducing the uplink business feature dataset to three business types: real-time interaction, data transmission, and transaction processing; and reducing the downlink business feature dataset to five business types: real-time interaction, streaming media transmission, large file transmission, transaction processing, and lightweight information service. The business type features include both the uplink and downlink business feature datasets. The traffic of the uplink business feature dataset and the downlink business feature dataset in the business feature dataset is summed and normalized; The normalized flow rate is subjected to feature transformation based on the central logarithmic ratio transform.

5. The method according to claim 1, characterized in that, The step of classifying the prediction results into benefit levels based on a preset level classification algorithm to obtain the evaluation results of the target area includes: The prediction results are classified into benefit levels based on the variance-weighted quartile method to obtain the classification level of the target area. The health index of the target area is calculated based on the classification level to obtain the evaluation result of the target area; wherein, the proportion of the health index is different for different classification levels.

6. The method according to any one of claims 1-5, characterized in that, After classifying the prediction results into benefit levels based on a preset classification algorithm to obtain the evaluation results of the target area, the method further includes: A three-level benefit ratio heatmap is generated based on the health index ratio and displayed on a preset screen.

7. The method according to claim 5, characterized in that, The benefit level classification of the prediction results based on the variance-weighted quartile method, resulting in the classification level of the target area, includes: The measurement reports and physical uplink shared channel values ​​in the coverage quality dataset are reconstructed into a continuous probability density function through probability density estimation, and the cumulative distribution function is calculated. Based on the cumulative distribution function curve, a preset quantile value is extracted as the coverage quality feature threshold; Based on the distribution characteristics of the upstream and downstream business feature datasets, the four-quadrant method is used to classify the coverage quality levels.

8. The method as described in claim 3, characterized in that, The two-stage density classification of the user density features based on the K-means clustering and quantile regression model also includes: The elbow rule is used to determine the optimal number of clusters, and the average number of users is divided into a preset number of density levels. Based on the average number of users at each density level, the quantile regression model is constructed to quantify the maximum user number fluctuation threshold.

9. A resource benefit assessment device, characterized in that, include: The acquisition unit is used to acquire multi-dimensional feature data of the target area and construct a three-dimensional feature space based on the multi-dimensional feature data; wherein, the three-dimensional feature space includes service type features, coverage quality features, and user density features; The input unit is used to input the three-dimensional feature space into the dynamic mapping model to obtain the prediction results of resource efficiency under different business scenarios output by the dynamic mapping model. The evaluation unit is used to classify the prediction results into benefit levels based on a preset level classification algorithm to obtain the evaluation results of the target area.

10. An electronic device, characterized in that, include: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-8.

11. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-8.

12. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method according to any one of claims 1-8.

Images