Cell resource reconfiguration method, system, computer device and storage medium

Abstract

The application provides a cell resource reconfiguration method, system, computer device and storage medium, and relates to the technical field of communication, wherein the method comprises the following steps: constructing a cell user behavior aggregation vector based on the user behaviors of all users in the cell; training a cell load prediction model based on a cell historical load index vector, the cell user behavior aggregation vector and a time context vector, and outputting a cell load prediction value through the trained model; calculating a cell resource adaptation score value through a preset cell resource adaptation score function based on the cell load prediction value, a cell resource capacity, a cell load fluctuation degree and a current cell user group structure change degree; and triggering cell resource reconfiguration, generating and executing a resource adjustment strategy in response to the cell resource adaptation score value being greater than a preset reconfiguration threshold. The technical scheme provided by the application can dynamically perceive network demand according to user behavior characteristics, predict load changes in advance, and realize adaptive optimization of cell resources.

Description

Technical Field

[0001] This invention relates to the field of communication technology, and in particular to a cell resource reconfiguration method, a cell resource reconfiguration system, a computer device, and a computer-readable storage medium. Background Technology

[0002] With the rapid development of 5G (5th Generation Mobile Communication Technology) mobile communication networks, cellular network structures are becoming increasingly complex, and network traffic is showing a clear trend of diversification and high volatility. In cellular systems, the cell, as the basic unit of resource scheduling and management, faces unprecedented challenges.

[0003] Currently, the strategy for allocating and adjusting community resources is as follows: Static configuration: Fixed settings during the initial deployment phase; Periodic adjustments: Adjustments are made based on daily / weekly load reports; Manual intervention: Relies on network engineers to manually analyze data and implement changes.

[0004] However, the above strategies ignore the multidimensional evolution of user behavior; they allocate cell resources based on current load or historical load trend predictions, lacking advance awareness of future load fluctuations; and they are difficult to respond in a timely manner in hotspot areas or sudden scenarios (such as concerts, public emergencies, etc.), leading to congestion, packet loss, or an increase in user complaints. Summary of the Invention

[0005] This invention was developed to at least partially address the technical problems of existing community resource allocation strategies, such as ignoring the multidimensional evolution of user behavior, lacking early perception of future load fluctuations, and being unable to respond promptly to sudden traffic growth and business demands.

[0006] According to one aspect of the present invention, a cell resource reconfiguration method is provided, comprising: Based on the user behavior of all users in the community, a community user behavior aggregation vector is constructed, wherein the user behavior is represented by time activity, spatial mobility, application preference and connection frequency; A cell load prediction model is trained based on the cell historical load index vector, cell user behavior aggregation vector, and time context vector, and the cell load prediction value is output through the trained cell load prediction model. Based on the predicted cell load, cell resource capacity, cell load fluctuation, and the degree of change in the current user group structure within the cell, the cell resource adaptation score is calculated using a preset cell resource adaptation score function. In response to the cell resource adaptation score being greater than a preset reconfiguration threshold, cell resource reconfiguration is triggered, and a resource adjustment strategy is generated and executed.

[0007] Optionally, constructing a cell user behavior aggregation vector based on the user behavior of all users in the cell includes: The user behavior of all users in the community is profiled to obtain a user behavior profile vector for each user; The user behavior profile vectors of each user are grouped using a clustering algorithm to generate user group category labels for each user. Based on the user group category labels of all users in the community, a community user behavior aggregation vector is constructed, where each feature in the community user behavior aggregation vector represents the number of users corresponding to each category label.

[0008] Optionally, the user behavior profile vector for each user is defined as: U i =[T i , L i A i , F i ]; Among them, U i T is the user behavior profile vector of the i-th user in the community; i L is the time-active feature vector of the i-th user in the cell; i Let A be the spatial movement feature vector of the i-th user in the cell; i Let F be the application preference feature vector of the i-th user in the community; i Let be the connection frequency feature vector of the i-th user in the cell.

[0009] Optionally, the cell load prediction model is derived in the following manner: A cell load prediction model is constructed using an attention-enhanced LSTM network. cell (t+1), and cell (t+1) = f(LSTM Attn ([Load t M cell Time t ])); in, cell (t+1) represents the predicted cell load value at the next time step; f(LSTM) Attn ( )) represents the mapping function of the LSTM neural network with fused attention mechanism; Load t M is a vector of historical load metrics for the community. cellFor aggregated vectors of user behavior in the community; Time t This is the time context vector.

[0010] Optionally, the cell resource adaptation scoring function adopts the following calculation formula: S=α1×( cell (t+1) / R cell )+α2×(σ cell )+α3×(Δ user_type ); Where S is the community resource adaptation score; cell (t+1) represents the predicted cell load value at the next time step; R cell For the community's resource capacity; σ cell For the load fluctuation of the community; Δ user_type α1 represents the degree of change in the user group structure within the current community; α2 and α3 are weighting coefficients.

[0011] Optionally, the degree of change in the user group structure within the current cell is calculated using the following formula: ; Where, Δ user_type This represents the degree of change in the current user group structure within the community; || ||1 represents the L1 norm; This is the aggregation vector of user behavior in the current community; This is the aggregated vector of user behavior in the community for the previous preset time period.

[0012] Optionally, the method further includes: After executing the resource adjustment strategy, obtain the average PRB utilization rate change, user disconnection rate change, and user satisfaction change within a preset time period before and after the execution of the resource adjustment strategy. In response to the changes in average PRB utilization, user dropout rate, and user satisfaction exceeding their respective indicator thresholds, the resource adjustment strategy is determined to be effective, and the relevant data involved in the resource adjustment strategy is used as sample data to train the cell load prediction model. If the change in average PRB utilization, user disconnection rate, or user satisfaction does not exceed the corresponding threshold, the resource adjustment strategy is deemed invalid, and a new resource adjustment strategy is generated and executed until the new resource adjustment strategy is deemed valid.

[0013] According to another aspect of the present invention, a cell resource reconfiguration system is provided, comprising: The user behavior aggregation module is configured to construct a cell user behavior aggregation vector based on the user behavior of all users in the cell. The user behavior is characterized by time activity, spatial mobility, application preference, and connection frequency. The cell load prediction module is configured to train a cell load prediction model based on the cell historical load index vector, the cell user behavior aggregation vector, and the time context vector, and output the cell load prediction value through the trained cell load prediction model. The resource adaptation scoring module is set to calculate the cell resource adaptation score based on the cell load prediction value, cell resource capacity, cell load fluctuation and the degree of change of the user group structure in the current cell through a preset cell resource adaptation scoring function. The resource reconfiguration module is configured to trigger cell resource reconfiguration and generate and execute a resource adjustment strategy in response to the cell resource adaptation score value being greater than a preset reconfiguration threshold.

[0014] According to another aspect of the present invention, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor executes the aforementioned cell resource reconfiguration method.

[0015] According to another aspect of the present invention, a computer-readable storage medium is provided having a computer program stored thereon, wherein when the computer program is executed by a processor, the processor performs the aforementioned cell resource reconfiguration method.

[0016] The technical solution provided by this invention may include the following beneficial effects: The cell resource reconfiguration method and system provided by this invention perceives the evolution of multi-dimensional user characteristics by constructing a user behavior aggregation vector, introduces a load prediction model to achieve early perception of future load fluctuations, and designs a resource adaptation scoring function and triggering mechanism. This solves the technical problems of existing technologies that ignore the evolution of multi-dimensional user behavior characteristics, lack early perception of future load fluctuations, and are difficult to respond to sudden traffic growth and service demands in a timely manner. It realizes dynamic perception of network demands based on user behavior characteristics, early prediction of load changes, proactive resource planning, and intelligent generation and execution of resource adjustment strategies, thereby improving cell resource utilization efficiency, enhancing user experience, and reducing network maintenance costs and complaint rates.

[0017] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the description, claims and drawings. Attached Figure Description

[0018] The accompanying drawings are provided to further understand the technical solutions of the present invention and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the technical solutions of the present invention, and do not constitute a limitation on the technical solutions of the present invention.

[0019] Figure 1 A flowchart illustrating a cell resource reconfiguration method provided in an embodiment of the present invention; Figure 2 A flowchart illustrating another cell resource reconfiguration method provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of a cell resource reconfiguration system provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the specific implementation methods of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific implementation methods described herein are for illustration and explanation only and are not intended to limit the present invention.

[0021] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a set order or sequence; furthermore, in the absence of conflict, the embodiments and features in the embodiments of this invention can be arbitrarily combined with each other. In the following description, the use of suffixes such as "module," "component," or "unit" to represent elements is only for the convenience of the description of this invention and has no inherent meaning. Therefore, "module," "component," or "unit" can be used interchangeably.

[0022] Current mobile communication networks have the following characteristics: 1) Increased differences in user behavior.

[0023] Different users exhibit significant differences in access time, application type, data traffic requirements, and activity area. For example, white-collar users are often active during weekdays and prefer office and instant messaging applications; student users are active after class or on holidays and prefer high-traffic applications such as video and games; mobile users (such as those in subway and bus scenarios) change locations frequently and have high requirements for handover and continuous coverage.

[0024] However, existing community resource allocation strategies generally adopt the "homogeneous user assumption," treating all users as average actors and ignoring individual differences and group evolution trends, resulting in insufficient targeting of resource allocation strategies.

[0025] 2) Enhanced network load dynamism.

[0026] In real-world applications, the load on cellular cells exhibits significant temporal fluctuations and spatial heterogeneity: high loads in work areas during the day and high loads in residential areas at night; special events (such as concerts, holiday travel, and public emergencies) can instantly trigger localized traffic surges; with the rise of high-bandwidth services such as high-definition video and AR / VR, the load growth rate in hotspot areas far exceeds the planned capacity.

[0027] However, existing community resource allocation strategies are mainly based on average statistics or periodic rules (such as hourly / daily adjustments) of historical load data, which cannot adapt to sudden fluctuations in real time and are prone to problems such as uneven resource allocation, hotspot congestion, and decreased user experience.

[0028] To address the aforementioned problems, this invention provides an intelligent reconfiguration scheme for cell resources based on user behavior profiling and dynamic load prediction, aiming to improve cell resource utilization efficiency, enhance user experience, and reduce network maintenance costs and complaint rates. Specific embodiments are described in detail below.

[0029] Figure 1 This is a flowchart illustrating a cell resource reconfiguration method provided in an embodiment of the present invention. Figure 1 As shown, the method includes the following steps S101 to S104.

[0030] S101. Based on the user behavior of all users in the cell, construct a cell user behavior aggregation vector, wherein the user behavior is characterized by temporal activity, spatial mobility, application preference and connection frequency; S102. Train a cell load prediction model based on the cell historical load index vector, cell user behavior aggregation vector, and time context vector, and output the cell load prediction value through the trained cell load prediction model; S103. Based on the predicted cell load, cell resource capacity, cell load fluctuation, and the degree of change in the current user group structure within the cell, the cell resource adaptation score is calculated using a preset cell resource adaptation score function. S104. In response to the cell resource adaptation score being greater than the preset reconfiguration threshold, cell resource reconfiguration is triggered, and a resource adjustment strategy is generated and executed.

[0031] In this embodiment, by constructing a user behavior aggregation vector to perceive the evolution of multi-dimensional user characteristics, a load prediction model is introduced to achieve early perception of future load fluctuations. A resource adaptation scoring function and triggering mechanism are designed to solve the technical problems of existing technologies that ignore the evolution of multi-dimensional user behavior characteristics, lack early perception of future load fluctuations, and have difficulty responding to sudden traffic growth and business demands in a timely manner. This enables dynamic perception of network demands based on user behavior characteristics, early prediction of load changes, proactive resource planning, and intelligent generation and execution of resource adjustment strategies, thereby improving the efficiency of cell resource utilization, enhancing user experience, and reducing network maintenance costs and complaint rates.

[0032] In one specific implementation, the resource adjustment strategy may employ any one or a combination of the following strategies: 1) Spectrum resource allocation strategy.

[0033] Definition: Changing the total amount of resources in a cell by adjusting its spectrum bandwidth or frequency band allocation.

[0034] Specific measures: Spectrum expansion: Allocating additional spectrum bandwidth from the network resource pool to this cell increases the total number of available PRBs (Physical Resource Blocks); Spectrum refarming: reallocating spectrum from other standards (such as 4G) to 5G cells to increase resource capacity; Dynamic spectrum sharing: The spectrum allocation ratio between adjacent cells is dynamically adjusted according to business needs.

[0035] Applicable scenarios: When the predicted load is consistently higher than the resource capacity, a fundamental improvement in the cell's service capabilities is required.

[0036] 2) PRB resource allocation strategy.

[0037] Definition: Optimize resource utilization by adjusting the allocation method of physical resource blocks.

[0038] Specific measures: PRB quota adjustment: Increase the PRB quota for this community to reduce resource shortage; Resource pool borrowing: Borrowing idle PRB resources from the network resource pool to temporarily relieve load pressure; Inter-cell resource borrowing: Based on the collaborative relationship matrix, PRB resources are borrowed from neighboring cells with lower load; Resource reservation adjustment: Dynamically adjust the resource reservation ratio for high-priority services to balance the needs of various services.

[0039] Applicable scenarios: Short-term load peaks that require temporary resource replenishment.

[0040] 3) Scheduling weight adjustment strategy.

[0041] Definition: Optimize the order of resource allocation by adjusting the scheduling priorities of different users or services.

[0042] Specific measures: User-level weight adjustment: Increase the scheduling weight of high-value users or users with sensitive complaints to ensure their service quality; Business-level priority adjustment: Dynamically adjust the scheduling priority of different applications such as video, games, and office based on the real-time business type; Fairness-efficiency balance: When resources are scarce, appropriately reduce the scheduling opportunities for low-priority services to ensure the experience of critical services.

[0043] Applicable scenarios: Limited resources but diverse user needs, requiring differentiated support.

[0044] 4) Power control adjustment strategy.

[0045] Definition: Optimizing coverage and capacity by adjusting the cell's transmit power or power allocation.

[0046] Specific measures: Total transmit power adjustment: Increase the maximum transmit power of the cell to improve coverage and channel quality; Power spectral density adjustment: Dynamically allocate power in the frequency domain, prioritizing high-demand frequency bands; Downlink power allocation optimization: Adjust the power allocation ratio of each PRB according to the user channel quality; Inter-cell power coordination: Coordinate with neighboring cells to adjust power, reduce interference, and improve overall network performance.

[0047] Applicable scenarios: Areas with limited coverage or severe interference.

[0048] 5) User guidance and load balancing strategies.

[0049] Definition: By redirecting some users to neighboring cells, the load pressure on the current cell can be reduced.

[0050] Specific measures: Cell handover offset adjustment: Modify cell selection / reselection parameters to allow some users to access neighboring cells first; Load balancing trigger: Actively switch some users in this cell to neighboring cells with lower load; Traffic diversion: Directing specific types of traffic (such as high-bandwidth downloads) to Wi-Fi or small base stations; Access control adjustment: Temporarily restrict access requests from low-priority users to ensure service quality for already connected users.

[0051] Applicable scenario: Resources in this community are fully utilized, but there are still vacant resources in neighboring communities.

[0052] 6) QoS policy adjustment (service compression).

[0053] Definition: By adjusting the service quality parameters of different services, critical services can be guaranteed when resources are scarce.

[0054] Specific measures: Rate limiting: Temporarily limit the rate of high-bandwidth services such as video streaming and downloads; Service degradation: Temporarily lower the QoS (Quality of Service) level of non-real-time services to free up resources for real-time services; Cache management: Adjust buffer size and discarding policy to prioritize low-latency services; Encoding format adjustment: Guide video services to switch to a lower bitrate encoding format.

[0055] Applicable scenario: When all available resources have been exhausted, it is still necessary to ensure the core business experience.

[0056] In this embodiment, adaptive optimization of cell resources is achieved through the aforementioned resource adjustment strategies (single or synergistic). Specifically, the spectrum resource allocation strategy, by dynamically adjusting bandwidth and frequency band allocation, fundamentally improves cell capacity to meet long-term load growth demands; the PRB resource allocation strategy, through resource pool borrowing and inter-cell coordination, flexibly addresses short-term load peaks and alleviates instantaneous congestion; the scheduling weight adjustment strategy, through differentiated priority guarantees, improves the experience quality for high-value users and latency-sensitive services; the power control adjustment strategy, through dynamic balancing of coverage and capacity, improves edge user performance and inter-cell interference coordination; the user guidance and load balancing strategy, through cross-cell service offloading, achieves balanced utilization of network resources and avoids localized overload; and the QoS policy adjustment, under resource limits, ensures the continuity and stability of core services through service tiered protection. These resource adjustment strategies enable the network to shift from passive response to proactive prevention, significantly improving resource utilization efficiency, user satisfaction, and operational automation levels, while reducing complaint rates and the frequency of manual intervention.

[0057] In one specific embodiment, step S101 includes the following steps S1011 to S1013.

[0058] S1011. Create user profiles for all users in the community to obtain a user behavior profile vector for each user; S1012. Use a clustering algorithm to group the user behavior profile vectors of each user and generate user group category labels for each user; S1013. Based on the user group category labels of all users in the community, construct a community user behavior aggregation vector, wherein each feature in the community user behavior aggregation vector represents the number of users corresponding to each category label.

[0059] The process of outputting user tags through clustering algorithms is essentially a user grouping mechanism under unsupervised learning. Its core idea is to treat each user's pre-constructed multi-dimensional behavioral profile vector (which has already been normalized) as a point in a high-dimensional space, and to divide these points into N clusters using clustering algorithms (such as K-Means++), so that users within the same cluster are as similar as possible in terms of behavioral characteristics, and users between different clusters are as different as possible.

[0060] The clustering algorithm first initializes cluster centers using the K-Means++ strategy to ensure diversity in initial selection. Then, during iteration, it continuously calculates the Euclidean distance from each user to each cluster center and assigns them to the nearest cluster. The cluster centers are then updated based on the mean of all users within each cluster until the cluster center changes stabilize. Finally, each user is assigned a unique cluster number and corresponding output label, thus achieving a categorical abstraction of user behavior patterns and providing a foundation for subsequent cell-level user structure analysis and load prediction.

[0061] In this embodiment, by profiling user behavior, generating user group category labels through clustering, and constructing a community user behavior aggregation vector, a quantitative representation of the distribution characteristics of community user groups is achieved. This provides enhanced features for the subsequent load prediction model, enabling the prediction model to perceive the impact of changes in user composition on load and improving the accuracy and personalization of load prediction.

[0062] In one specific implementation, in step S1011, the user behavior profile vector for each user is defined as: U i =[T i , L i A i , F i ].

[0063] Among them, U i T is the user behavior profile vector of the i-th user in the community; i L is the time-active feature vector of the i-th user in the cell; i Let A be the spatial movement feature vector of the i-th user in the cell; i Let F be the application preference feature vector of the i-th user in the community; i Let be the connection frequency feature vector of the i-th user in the cell.

[0064] Specifically, the time-active feature vector T iThe calculation formula can be: Ti = [h0, h1, h2, ..., h 23 ] / max(h0, h1, ..., h 23 ), where h j This represents the average user activity at time j (0-23), a 24-dimensional vector. Normalization is performed by dividing by the maximum value, ensuring that values ​​for each time period fall within the range [0,1]. Activity can be defined in various ways: connection count, bandwidth usage, PRB consumption, etc.

[0065] Spatial movement feature vector L i The calculation formula can be: Li = [p1, p2, ..., p m ] / sum(p1, p2,..., p m ), where p j This represents the dwell time or number of times a user appears in the j-th region (e.g., a grid), where m is the total number of regions in the cell. The trajectory density distribution is formed by normalizing the distribution by dividing by the total time / total number of occurrences.

[0066] Application preference feature vector A i The calculation formula can be: Ai = [a1, a2, ..., a k / sum(a1, a2,..., a k ), where a j This represents the user's data usage or usage time in application category j, where k is the total number of application categories. Application categories can be divided into several types, such as video, games, social networking, and office applications.

[0067] Connection frequency eigenvector F i The calculation formula can be: Fi = [f conn , f disc , interval mean ,interval std ], or Fi = [f conn , f disc , act ratio conn density ]. Where, f conn f represents the number of network connections per unit of time (e.g., times / day). disc The interval is the number of times the network is disconnected per unit of time (e.g., times / day). mean The average connection interval (in minutes). std The standard deviation (minutes) of the connection interval; act ratio The percentage of active time (ratio 0-1), conn densityFor connection density (e.g., times / hour). Then for F i Each component is normalized (e.g., using Z-score for standardization), but for components that are already ratios (actual components)... ratio It can be used directly without further normalization.

[0068] In this embodiment, by defining the user behavior profile vector as a combination of time-activity features, spatial mobility features, application preference features, and connection frequency features, a comprehensive characterization of multi-dimensional user behavior features is achieved. This uncovers deep user characteristics in terms of time, space, and behavior, providing a standardized and personalized basis for subsequent user group classification and load prediction.

[0069] After obtaining the user behavior profile vector U for each user i Then, a clustering algorithm (such as K-Means++) is used to process the user behavior profile vectors of each user, generating a user group category label C for each user. i Specifically, the number of clusters N is set, and K-Means++ is used for initialization (to improve convergence quality). Then, the normalized user behavior profile vector U is... i Perform clustering to obtain the cluster number k for each user and output the label C. i = c k And C i ∈{c1, c2, ..., c N}

[0070] After obtaining the user group category label C of all users i Then, construct the community user behavior aggregation vector M. cell And M cell =[n c1 , n c2 , ..., n cN ]. Where, n ck Indicates the cth element in the community k The number of users of each category serves as a distribution characteristic of the "business behavior pressure" in this community.

[0071] The following example illustrates this: c1: High-frequency, high-traffic video users (active during evening peak hours).

[0072] The normalized behavioral profile vector U1 of a user is: The time-active feature vector T1 is [0.02, 0.01, 0.01, …, 0.85, 0.92, 0.65], with the peak occurring between 21:00 and 22:00 (evening period). Spatial movement feature vector L1: [0.75, 0.15, 0.10], with the main activity area being residential areas (accounting for 75%); Application preference feature vector A1: [0.60, 0.25, 0.10, 0.05], with video applications accounting for 60%, social applications for 25%, office applications for 10%, and games for 5%. Connection frequency feature vector F1: [8, 7, 180, 25], connects 8 times and disconnects 7 times per day, with an average connection interval of 180 minutes and a standard deviation of 25 minutes (few and regular connection times).

[0073] Clustering process: In the K-Means++ clustering algorithm, the Euclidean distance from the user to each cluster center is calculated. Assume that several initial centers are randomly selected during initialization, and after multiple rounds of iterative optimization, the user is assigned to the cluster containing cluster center μ1.

[0074] The characteristics of μ1 are: Peak activity time: evening rush hour (20:00-24:00); Spatial center: High concentration of residential areas; Application Center: Video applications account for over 50%; Frequency center: moderate number of connections, regular intervals.

[0075] The user is closest to μ1, so he / she is assigned cluster number 1 and output label c1.

[0076] Tag naming criteria: The common characteristics of this user cluster (number 1) are: concentrated use of video applications during evening hours, activity mainly in residential areas, and regular connection behavior. Therefore, it is named "High-frequency, high-traffic video users (active during evening peak hours)".

[0077] c2: Daytime mobile office users (multi-regional activities).

[0078] The normalized behavioral profile vector U2 of a user is: The time-active feature vector T2 is [0.65, 0.70, 0.45, …, 0.05, 0.03, 0.02], with the peak occurring between 9:00 and 17:00 (working hours). Spatial movement feature vector L2: [0.20, 0.70, 0.10], with the main activity area being the office area (accounting for 70%) and the residential area accounting for 20%; Application preference feature vector A2: [0.15, 0.30, 0.45, 0.10], with office applications accounting for 45%, social applications for 30%, video applications for 15%, and games for 10%. Connection frequency feature vector F2: [15, 14, 60, 15], connecting 15 times and disconnecting 14 times per day, with an average connection interval of 60 minutes and a standard deviation of 15 minutes (frequent and regular connections).

[0079] Clustering process: The user was assigned to the cluster where cluster center μ2 is located.

[0080] The characteristics of μ2 are: Time center: High activity level during daytime working hours (9:00-17:00); Space Center: The office area is highly concentrated, and it is also distributed to some extent in the residential area; App Center: Office and social applications dominate; Frequency center: Frequent connections, regular intervals.

[0081] The user is closest to μ2, so he / she is assigned cluster number 2 and output label c2.

[0082] Tag naming criteria: The common characteristics of this cluster of users (number 2) are: active during the daytime working hours, with activities spanning both office and residential areas (multi-regional movement), primarily using office and social applications, and exhibiting consistent connection behaviors. Therefore, they are named "Daytime Mobile Office Users (Multi-Regional Activity)".

[0083] c3: Users who prefer playing games at night.

[0084] The normalized behavioral profile vector U3 of a user is: The time-active feature vector T3 is [0.10, 0.05, 0.02, …, 0.75, 0.80, 0.40], with the peak occurring between 20:00 and 22:00 (evening). Spatial movement feature vector L3: [0.80, 0.10, 0.10], the main activity area is residential area (accounting for 80%); Application preference feature vector A3: [0.20, 0.05, 0.10, 0.65], with game applications accounting for 65% and video applications for 20%; Connection frequency feature vector F3: [25, 24, 45, 50], 25 connections and 24 disconnections per day, with an average connection interval of 45 minutes and a standard deviation of 50 minutes (frequent and highly volatile connections).

[0085] Clustering process: The user was assigned to the cluster where cluster center μ3 is located.

[0086] The characteristics of μ3 are: Time of day: High activity levels during the evening hours (20:00-24:00); Spatial center: High concentration of residential areas; App Center: Games account for over 60%; Frequency center: Frequent connections, large interval fluctuations.

[0087] The user is closest to μ3, so he / she is assigned cluster number 3 and output label c3.

[0088] Tag naming criteria: The common characteristics of this group of users (number 3) are: concentrated use of gaming applications during evening hours, activity mainly in residential areas, and frequent but fluctuating connection behavior. Therefore, they are named "Nighttime Gaming Preference Users".

[0089] c4: Users with large traffic fluctuations and potential high risk of complaints.

[0090] The normalized behavioral profile vector U4 of a user is: The time-active feature vector T4 is [0.30, 0.25, 0.20, …, 0.60, 0.55, 0.50], with active periods scattered and no obvious peaks. Spatial movement feature vector L4: [0.40, 0.35, 0.25], activity areas are dispersed (residential areas 40%, office areas 35%, commercial areas 25%). Application preference feature vector A4: [0.35, 0.30, 0.25, 0.10], indicating a relatively balanced use across various applications; Connection frequency feature vector F4: [40, 39, 15, 65], 40 connections and 39 disconnections per day, with an average connection interval of 15 minutes and an interval standard deviation of 65 minutes (extremely frequent connections and extremely volatile intervals).

[0091] Clustering process: The user was assigned to the cluster where cluster center μ4 is located.

[0092] The characteristics of μ4 are: Time center: Active periods are scattered and there is no obvious pattern; Space center: Dispersed activities in multiple areas; Application Center: Balanced usage across all application types; Frequency center: extremely frequent connections and extremely large interval fluctuations (obvious characteristics of sudden traffic bursts).

[0093] The user is closest to μ4, so he / she is assigned cluster number 4 and output label c4.

[0094] Tag naming criteria: The common characteristics of this cluster of users (number 4) are: highly irregular connection behavior (frequent connections with large intervals), sudden and strong traffic demands, and dispersed activity areas. This unstable behavior pattern can easily lead to network resource strain and service quality degradation, thereby triggering user complaints. Therefore, they are named "users with large traffic fluctuations and potential for high-risk complaints".

[0095] In one specific implementation, in step S102, the cell load prediction model is derived in the following manner: A cell load prediction model is constructed using an attention-enhanced LSTM (Long Short-Term Memory) network. cell (t+1), and cell (t+1) = f(LSTM Attn ([Load t M cell Time t ])).

[0096] in, cell (t+1) represents the predicted cell load value at the next time step; f(LSTM) Attn ( )) represents the mapping function of the LSTM neural network with fused attention mechanism; Load t M is a vector of historical load metrics for the community. cell For aggregated vectors of user behavior in the community; Time t This is the time context vector.

[0097] Specifically, the historical load index vector of a cell can be defined as: ; Where t is the current time; T is the length of the historical time window (such as the past 24 hours, the past 7 days, etc.). The number of load metric dimensions at each time point; Let be the cell load index vector at time s, and s∈[t-T+1, t].

[0098] Cell load metric vector at each time step It can be defined as: ; in, Let be the cell load rate at time s; The utilization rate of physical resource blocks (PRBs) in the cell at time s is the percentage of PRBs used. The uplink data volume per unit time in the cell at time s, i.e., the uplink traffic. This represents the amount of downlink data per unit time in the cell at time s, i.e., the downlink traffic. Let s be the number of users simultaneously online in the cell at time s, i.e., the number of user connections.

[0099] Specifically, the time context vector can be defined as: ; in, This represents the feature vector at the current moment, used to characterize the current moment's position within the day, reflecting the intraday periodicity of the load, and One-hot encoding can be used to map 24 hours into a 24-dimensional binary vector; This represents a weekday / holiday feature vector, used to identify whether the current day is a weekday or a holiday, reflecting the weekly / holiday pattern of the workload. 0 represents a working day, and 1 represents a public holiday; This represents a feature vector indicating peak hours, used to identify whether the current period is during specific times such as morning or evening rush hours, reflecting the sudden and predictable patterns of load changes. 0 indicates that the current period is off-peak, and 1 indicates that the current period is peak (such as 7-9 am, 17-19 pm). This represents the event interval feature vector, used to characterize the time interval between the most recent hot event, reflecting the decay law of the event's impact, and The values ​​are numerical and normalized to the [0, 1] interval. For example, if the maximum effective time window for the event's impact is 72 hours, and there are 10 hours left until the concert ends, then... =[10 / 72=0.139]; This represents the monthly feature vector, used to characterize the current month and reflect the monthly pattern of the load. One-hot encoding can be used to map the 12 months into a 12-dimensional binary vector, or This represents the seasonal feature vector, used to characterize the current season and reflect the seasonal pattern of the load. One-hot encoding can be used to map the four seasons into a four-dimensional binary vector.

[0100] In this embodiment, a load prediction model is constructed by using an attention-enhanced LSTM network. The model is trained based on historical load indicators, user behavior aggregation vectors, and time context vectors. The trained cell load prediction model can achieve high-precision prediction of future load change trends, enabling the network to perceive resource pressure changes in advance and providing a basis for resource reallocation decisions.

[0101] In one specific implementation, in step S103, the cell resource adaptation scoring function adopts the following calculation formula: S=α1×( cell (t+1) / R cell )+α2×(σ cell )+α3×(Δ user_type ).

[0102] Where S is the community resource adaptation score; cell (t+1) represents the predicted cell load value at the next time step; R cell For the community's resource capacity; σ cell For the load fluctuation of the community; Δ user_type α1 represents the degree of change in the user group structure within the current community; α2, α3 are weighting coefficients. It should be noted that α1, α2, and α3 each contain implicit units matching their respective variables, such that α1 × ( cell (t+1) / R cell ), α2×(σ cell ), α3×(Δ user_type Since both are dimensionless, the community resource adaptation score S obtained after weighted summation is also dimensionless.

[0103] Specifically, the community resource capacity R cell R refers to the total amount of rated radio resources possessed by the cell under the current configuration. It is a static configuration indicator that measures the theoretical maximum service capacity of the cell. Its value can take, but is not limited to, any one of the following: the total number of allocated PRBs (e.g., 100 PRBs), the effective bandwidth (e.g., 20 MHz), the size of the scheduling resource pool (e.g., 80% resources for ordinary users, 20% resources for high-priority users), the maximum transmit power (e.g., 40W), and the number of MIMO (Multiple Input Multiple Output) layers (e.g., 4 layers), or a weighted sum of multiple parameters. In the cell resource adaptation scoring function, R... cell As the denominator, with the predicted load cell (t+1) together constitute the resource stress index, reflecting the pressure of the predicted load on the current configured resource capacity.

[0104] Community load fluctuation σ cell This indicates the degree of fluctuation in cell load within a certain time window, usually measured by the standard deviation or variance of the load rate. High volatility indicates unstable demand in the cell, necessitating the ability to dynamically adjust resources.

[0105] Current user group structure change Δ user_type It indicates the degree of change in the composition of different user groups within the community during the current time period, and is used to measure whether there has been a significant change in the community's service recipients.

[0106] In this embodiment, by designing a resource adaptation scoring function, the supply and demand pressure of predicted load and resource capacity, the degree of load fluctuation, and changes in user group structure are comprehensively considered to achieve intelligent judgment of resource reconfiguration needs. When the score exceeds the threshold, adjustment is triggered, so that resource configuration can dynamically respond to changes in network demand and avoid congestion and user experience degradation caused by response lag.

[0107] In one specific implementation, the degree of change in the user group structure within the current cell is calculated using the following formula: ; Where, Δ user_type This represents the degree of change in the current user group structure within the community; || ||1 represents the L1 norm; This is the aggregation vector of user behavior in the current community; This is the aggregated vector of user behavior in the community for the previous preset time period.

[0108] In this embodiment, by using the L1 norm to calculate the rate of change in user group distribution, the degree of change in the composition of user groups within the cell is quantified, enabling the scoring function to perceive significant changes in user structure. When user types change drastically, resource strategies are adjusted in a timely manner, improving the responsiveness of resource allocation to changes in business needs.

[0109] In one specific embodiment, the method further includes the following steps S105 to S107.

[0110] S105. After executing the resource adjustment strategy, obtain the average PRB utilization rate change, user disconnection rate change, and user satisfaction change within a preset time period before and after the execution of the resource adjustment strategy. S106. In response to the average PRB utilization change ΔPRB, the user drop rate change ΔDrop, and the user satisfaction change ΔSatis exceeding their respective corresponding indicator thresholds (e.g., ΔPRB≥5%, ΔDrop≤-2%, ΔSatis≥3%), the resource adjustment strategy is determined to be effective, and the relevant data involved in the resource adjustment strategy is used as sample data to train the cell load prediction model. S107. If the change in average PRB utilization, the change in user disconnection rate, or the change in user satisfaction does not exceed the corresponding indicator threshold, the resource adjustment strategy is determined to be invalid, and the resource adjustment strategy is regenerated and executed until the new resource adjustment strategy is determined to be valid.

[0111] In this embodiment, by verifying the adjustment effect after execution and feeding back effective samples to train the prediction model, a closed-loop optimization process of data-driven, prediction-guided, policy response, and effect verification is formed. This realizes the self-learning and continuous optimization of the model, ensuring that the network performance can be continuously improved after reconfiguration, and further improving resource utilization efficiency and user satisfaction.

[0112] The cell resource reconfiguration method provided in this invention constructs user behavior profiles, aggregates user behavior features, predicts future load trends in conjunction with time context, and introduces a scoring function to intelligently determine resource reconfiguration needs. This enables dynamic perception of network needs based on user behavior features, advance prediction of load changes, proactive resource planning, intelligent generation and execution of resource adjustment strategies, and closed-loop optimization of network operation. As a result, it improves cell resource utilization efficiency, enhances user experience, and reduces network maintenance costs and complaint rates.

[0113] Figure 2 This is a flowchart illustrating another cell resource reconfiguration method provided in an embodiment of the present invention. Figure 2 As shown, the method includes the following steps S201 to S205.

[0114] S201. Construct a multi-dimensional user behavior profile model.

[0115] The purpose of this step is to uncover deeper user characteristics in terms of time, space, and behavior, providing personalized data for subsequent load forecasting and network optimization.

[0116] The inputs for this step are: terminal-reported data, including location, APP usage, service type, connection frequency, etc.; network-side data, such as MR data, cell dwell time, uplink and downlink traffic.

[0117] Construct user behavior profile vectors as follows: Let the user set of the current cell be U = {U1, U2, ..., U...} n}, for any user U i Its behavioral profile vector is defined as: U i =[T i , L i A i , F i ].

[0118] Among them, U i : The user behavior profile vector of the i-th user; T i : The time-active feature vector of the i-th user (e.g., the daily active time period distribution vector); L i : Spatial movement feature vector of the i-th user (e.g., trajectory density heatmap); A i : The application preference feature vector of the i-th user (such as the usage ratio of short videos / games / social media); F i : Connection frequency feature vector of the i-th user (frequency of network connection, number of times online and offline).

[0119] Each user is grouped into profiles using a clustering algorithm (such as K-Means++), generating typical user group category labels C. i ∈{c1, c2, ..., c N}

[0120] Clustering process: Set the number of clusters N, initialize with K-Means++ (to improve convergence quality), and process the normalized user behavior profile vector U. i Perform clustering to obtain the cluster number k for each user. Output label C. i ∈{c1, c2,..., c N}

[0121] C i For example (assuming N=4): c1: High-frequency, high-traffic video users (active during evening peak hours); c2: Daytime mobile office users (multi-regional activities); c3: Users who prefer playing games at night; c4: Users with large traffic fluctuations and potential high risk of complaints.

[0122] The output of this step is: the behavioral profile vector U for each user. i and user group category tag C i And C i =c k , where k∈[1, N].

[0123] S202. Construct a community user behavior aggregation model.

[0124] The purpose of this step is to quantify the distribution characteristics of various user groups in the community as a benchmark for resource allocation.

[0125] The input for this step is: User group category label C i .

[0126] Construct the community user behavior aggregation vector as follows: M cell =[n c1 , n c2 , ..., n cN ].

[0127] Where, n ck Indicates the cth element in the community k The number of users of each category serves as a distribution characteristic of the "business behavior pressure" in this community.

[0128] The output of this step is: the user behavior aggregation vector M for each cell. cell .

[0129] S203. Construct a cell load prediction model.

[0130] The purpose of this step is to anticipate changes in resource pressure trends and assist in making judgments regarding resource reallocation within the community.

[0131] The inputs for this step are: the cell's historical load metric vector: {load rate, PRB utilization, uplink / downlink traffic, number of connections}; the cell's user behavior aggregation vector; and the time context vector (such as the current hour, whether it is a holiday, etc.).

[0132] This step uses an attention-enhanced LSTM network to build a cell load prediction model: cell (t+1)= f(LSTM Attn ([Load t M cell Time t ])).

[0133] in, cell (t+1): The predicted load value of the cell at the next time step (which can be obtained by weighted summation of multiple predicted load indicators); Load t : Vector of historical load indicators for the community; M cell : Aggregated vector of user behavior in the community; Time t The time context vector typically includes the following: the current time; whether it is a weekday or holiday (0 / 1); whether it is a peak period (e.g., morning / evening rush hour, 0 / 1); the time interval since the last hot event (e.g., a concert); and the current season or month. This information is combined into a vector in numerical or one-hot encoded form and input into the model to better predict the trend of network load changes over time.

[0134] The output of this step is: the cell load prediction result for the next time step. cell (t+1).

[0135] S204. Decision-making on resource reallocation strategies.

[0136] The purpose of this step is to intelligently determine whether resource configuration needs to be adjusted based on the cell load prediction results and the current resource configuration of the cell, and to formulate corresponding strategies.

[0137] The inputs for this step are: current cell resource configuration parameters: frequency band allocation, PRB allocation, scheduling weight, power control, etc.; and cell load prediction values. cell (t+1); Remaining capacity of the network resource pool; Inter-cell collaboration relationships (e.g., cell A can borrow resources from cell B).

[0138] This step defines a resource adaptation scoring function: S=α1×( cell (t+1) / R cell )+α2×(σ cell )+α3×(Δ user_type ).

[0139] Among them, R cell This indicates the number of available wireless resources currently available in the cell (derived from the cell's resource configuration parameters) and is an indicator of the cell's service capability.

[0140] Common value formats include: the number of allocated PRBs (physical resource blocks, such as 100 PRBs); effective bandwidth (such as 20MHz); the proportion of allocated scheduling resources (such as 80% of resources for ordinary users and 20% for high-priority users); antenna transmit power or the number of MIMO layers can also be used as reference parameters. Any one of these can be selected, or a weighted sum of multiple parameters can be used.

[0141] This parameter reflects the current resource status of the cell in the network and is the basis for "whether expansion or release is needed".

[0142] σ cell This indicates the degree of fluctuation in cell load within a certain time window, usually measured by the standard deviation or variance of the load rate. High volatility indicates unstable demand in the cell, necessitating the ability to dynamically adjust resources.

[0143] Δ user_type This refers to the degree of change in the composition of different user groups within a residential community during the current time period, measuring whether there has been a significant change in the community's service recipients. A common calculation method is the rate of change of the user group distribution vector. .

[0144] For the above formula, the first parameter on the right side of the equals sign is the user behavior aggregation vector of the cell at the current time (also known as the cell user group distribution vector), such as [40, 30, 20, 10] representing the number of four types of users; the second parameter on the right side of the equals sign is the distribution of the previous time period.

[0145] The L1 norm (the sum of the absolute values ​​of vector differences) is used to represent the overall change in the user group structure. If the user type changes drastically (for example, a large number of video users suddenly gather), it indicates that the business pressure faced by the community may have changed, and resource strategies should be adjusted in a timely manner.

[0146] α1, α2, α3: Configurable weights.

[0147] When S>S thresh If the reconfiguration threshold is reached, the resource reconfiguration process will be triggered, and the corresponding resource adjustment strategy (spectrum bandwidth, PRB resources, scheduling weight, power allocation, etc.) will be generated.

[0148] The output of this step is: whether to trigger the resource reconfiguration process; and the recommended resource adjustment strategy type (spectrum, power, user guidance, etc.).

[0149] S205. Resource adjustments and verification feedback.

[0150] The purpose of this step is to ensure that network performance can be continuously improved after the reconfiguration of community resources, thus achieving closed-loop optimization.

[0151] The input for this step is the resource adjustment strategy output in the previous step, and the load and user perception metrics for the K time period before and after the resource strategy adjustment.

[0152] Compare the following indicators within the K time period before and after the resource strategy adjustment: change in average PRB utilization ΔPRB, change in user drop rate ΔDrop, and change in user satisfaction (such as NPS or complaint rate) ΔSatis.

[0153] If the following closed-loop indicators are met: ΔPRB≥5%, ΔDrop≤-2%, ΔSatis≥3%, then the resource adjustment is considered effective, and the current cell user behavior profile-cell load prediction model-cell resource adjustment strategy is recorded as a sample to train the next round of model.

[0154] The output of this step is the verification result of the resource adjustment effect, which is used to update the adjustment strategy sample library.

[0155] The above five steps constitute a closed-loop optimization process of data-driven, prediction-guided, strategy-responsive, and effect-verified approaches, with the specific relationships as follows: Step S201 establishes multi-dimensional behavioral characteristics for each user, forming a user profile, which is the basic input for the entire system. It provides input data (profile classification) for step S202.

[0156] Step S202 aggregates the user profile classification results from step 1 to the cell level, reflecting the current user composition and business behavior pressure distribution of the cell, providing enhanced features (M) for prediction in step S203. cell ).

[0157] Step S203 combines historical load data and user aggregation information from step S202 to predict the future load change trend of the cell, providing a basis for decision-making in step S204.

[0158] Step S204 calculates the resource adaptation scoring function based on the predicted load results, changes in user structure, current resource configuration, and other factors, determines whether resource reconfiguration is needed, and outputs a resource adjustment strategy to provide strategy output for the execution of step S205.

[0159] Step S205 executes the resource adjustment strategy recommended in step S204, and verifies the optimization effect by comparing network metrics before and after the adjustment, thus updating the model and strategy library in a closed loop. This feeds back into steps S201-S203, forming a self-learning closed loop.

[0160] The entire solution is forward-looking, dynamic, adaptive, and self-learning, enabling truly intelligent cellular resource scheduling.

[0161] The cell resource reconfiguration method provided in this invention incorporates user behavior profiles into the cell load prediction model when predicting future cell load indicators, rather than solely relying on historical load data, thus improving prediction accuracy and personalization. It dynamically combines user behavior aggregation vectors and the cell load prediction model with the cell user structure and temporal context for resource optimization. A resource adaptation scoring function S and a threshold triggering mechanism are established, integrating prediction, fluctuations, and user structure differences to intelligently trigger resource adjustments. This method can be directly applied to the automatic optimization system of existing 5G / 4G cellular networks, and is particularly suitable for the following scenarios: high-speed changing scenarios (concerts, subways, commercial districts); areas with high complaint rates (network congestion); and intelligent scheduling of network resources (in conjunction with MEC, slicing, etc.). It effectively solves the problems in related technologies where user behavior differences are ignored, cell resource allocation strategies lack personalization, cell load fluctuates significantly, resource allocation response is lagging, and resource allocation mechanisms lack intelligent judgment and rapid adjustment capabilities, relying excessively on manual or periodic adjustments.

[0162] Figure 3 This is a schematic diagram of the structure of a cell resource reconfiguration system provided in an embodiment of the present invention. Figure 3As shown, the system includes: a user behavior aggregation module 301, a cell load prediction module 302, a resource adaptation scoring module 303, and a resource reconfiguration module 304.

[0163] The user behavior aggregation module 301 is configured to construct a cell user behavior aggregation vector based on the user behavior of all users in the cell, wherein the user behavior is characterized by temporal activity, spatial mobility, application preference, and connection frequency; the cell load prediction module 302 is configured to train a cell load prediction model based on the cell historical load index vector, the cell user behavior aggregation vector, and the temporal context vector, and output the cell load prediction value through the trained cell load prediction model; the resource adaptation scoring module 303 is configured to calculate the cell resource adaptation score value based on the cell load prediction value, cell resource capacity, cell load fluctuation, and the degree of change in the current user group structure in the cell through a preset cell resource adaptation scoring function; and the resource reconfiguration module 304 is configured to trigger cell resource reconfiguration and generate and execute a resource adjustment strategy in response to the cell resource adaptation score value being greater than a preset reconfiguration threshold.

[0164] In one specific implementation, the user behavior aggregation module 301 includes: a profile unit, a grouping unit, and an aggregation unit.

[0165] The profiling unit is configured to profile the user behavior of all users in the community, and obtain a user behavior profile vector for each user; the grouping unit is configured to group the user behavior profile vectors of each user using a clustering algorithm, and generate a user group category label for each user; the aggregation unit is configured to construct a community user behavior aggregation vector based on the user group category labels of all users in the community, wherein each feature in the community user behavior aggregation vector is the number of users corresponding to each category label.

[0166] In one specific implementation, the user behavior profile vector obtained by the profiling unit for each user is defined as: U i =[T i , L i A i , F i ]; Among them, U i T is the user behavior profile vector of the i-th user in the community; i L is the time-active feature vector of the i-th user in the cell; i Let A be the spatial movement feature vector of the i-th user in the cell; i Let F be the application preference feature vector of the i-th user in the community; i Let be the connection frequency feature vector of the i-th user in the cell.

[0167] In one specific implementation, the cell load prediction module 302 obtains the cell load prediction model in the following manner: A cell load prediction model is constructed using an attention-enhanced LSTM network. cell (t+1), and cell (t+1)= f(LSTM Attn ([Load t M cell Time t ])); in, cell (t+1) represents the predicted cell load value at the next time step; f(LSTM) Attn ( )) represents the mapping function of the LSTM neural network with fused attention mechanism; Load t M is a vector of historical load metrics for the community. cell For aggregated vectors of user behavior in the community; Time t This is the time context vector.

[0168] In one specific implementation, the resource adaptation scoring module 303 uses the following formula to derive the cell resource adaptation scoring function: S=α1×( cell (t+1) / R cell )+α2×(σ cell )+α3×(Δ user_type ); Where S is the community resource adaptation score; cell (t+1) represents the predicted cell load value at the next time step; R cell For the community's resource capacity; σ cell For the load fluctuation of the community; Δ user_type α1 represents the degree of change in the user group structure within the current community; α2 and α3 are weighting coefficients.

[0169] In one specific implementation, the degree of change in the user group structure within the current cell is calculated using the following formula: ; Where, Δ user_type This represents the degree of change in the current user group structure within the community; || ||1 represents the L1 norm; This is the aggregation vector of user behavior in the current community; This is the aggregated vector of user behavior in the community for the previous preset time period.

[0170] In one specific implementation, the system further includes: an indicator acquisition module, an indicator judgment module, and a model training module.

[0171] The indicator acquisition module is configured to acquire the average PRB utilization rate change, user dropout rate change, and user satisfaction change within a preset time period before and after the execution of the resource adjustment strategy. The indicator judgment module is configured to determine whether the average PRB utilization rate change, user dropout rate change, and user satisfaction change exceed their respective indicator thresholds. If the average PRB utilization rate change, user dropout rate change, and user satisfaction change exceed their respective indicator thresholds, the resource adjustment strategy is determined to be effective. If the average PRB utilization rate change, user dropout rate change, or user satisfaction change does not exceed their respective indicator thresholds, the resource adjustment strategy is determined to be invalid. The model training module is configured to use the relevant data involved in the resource adjustment strategy as sample data to train the cell load prediction model if the resource adjustment strategy is effective. The resource reconfiguration module 304 is further configured to regenerate and execute the resource adjustment strategy if the resource adjustment strategy is invalid, until the new resource adjustment strategy is determined to be effective.

[0172] The cell resource reconfiguration system provided in this invention constructs user behavior profiles, aggregates user behavior features, predicts future load trends by combining time context, and introduces a scoring function to intelligently determine resource reconfiguration needs. This enables the system to dynamically perceive network needs based on user behavior features, predict load changes in advance, proactively plan resources, intelligently generate and execute resource adjustment strategies, and optimize network operation in a closed loop. This improves cell resource utilization efficiency, enhances user experience, and reduces network maintenance costs and complaint rates.

[0173] Based on the same technical concept, embodiments of the present invention also provide a computer device, such as... Figure 4 As shown, the computer device includes a memory 401 and a processor 402. The memory 401 stores a computer program. When the processor 402 runs the computer program stored in the memory 401, the processor 402 executes the aforementioned cell resource reconfiguration method.

[0174] Based on the same technical concept, embodiments of the present invention also provide a computer-readable storage medium storing a computer program thereon, wherein when the computer program is executed by a processor, the processor executes the aforementioned cell resource reconfiguration method.

[0175] In summary, the cell resource reconfiguration method, system, computer equipment, and storage medium provided in this embodiment of the invention, by constructing user behavior profiles, aggregating user behavior features, predicting future load trends in conjunction with time context, and introducing a scoring function to intelligently determine resource reconfiguration needs, achieves dynamic perception of network demand based on user behavior characteristics, predicts load changes in advance, proactively plans resources, and intelligently generates and executes resource adjustment strategies. This enables the early prediction of high-load cells and proactive resource adjustment, effectively alleviating user-perceived problems such as congestion and lag, and reducing complaints. It achieves user-perceived resource allocation, improves the user experience for key users or those with high traffic, reduces the frequency of manual intervention and repetitive fault handling work orders, and improves network operation and maintenance efficiency.

[0176] It will be understood by those skilled in the art that all or some of the steps, systems, or apparatuses disclosed above, and their functional modules / units, can be implemented as software, firmware, hardware, or suitable combinations thereof. In hardware implementations, the division between functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed collaboratively by several physical components. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit (ASIC). Such software may be distributed on a computer-readable medium, which may include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and can be accessed by a computer. Furthermore, it is well known to those skilled in the art that communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0177] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for reallocating community resources, characterized in that, include: Based on the user behavior of all users in the community, a community user behavior aggregation vector is constructed, wherein the user behavior is represented by time activity, spatial mobility, application preference and connection frequency; A cell load prediction model is trained based on the cell historical load index vector, cell user behavior aggregation vector, and time context vector, and the cell load prediction value is output through the trained cell load prediction model. Based on the predicted cell load, cell resource capacity, cell load fluctuation, and the degree of change in the current user group structure within the cell, the cell resource adaptation score is calculated using a preset cell resource adaptation score function. In response to the cell resource adaptation score being greater than a preset reconfiguration threshold, cell resource reconfiguration is triggered, and a resource adjustment strategy is generated and executed.

2. The method according to claim 1, characterized in that, The construction of a cell user behavior aggregation vector based on the user behavior of all users in the cell includes: The user behavior of all users in the community is profiled to obtain a user behavior profile vector for each user; The user behavior profile vectors of each user are grouped using a clustering algorithm to generate user group category labels for each user. Based on the user group category labels of all users in the community, a community user behavior aggregation vector is constructed, where each feature in the community user behavior aggregation vector represents the number of users corresponding to each category label.

3. The method according to claim 2, characterized in that, The user behavior profile vector for each user is defined as: U i =[T i , L i A i , F i ]; Among them, U i T is the user behavior profile vector of the i-th user in the community; i L is the time-active feature vector of the i-th user in the cell; i Let A be the spatial movement feature vector of the i-th user in the cell; i F is the application preference feature vector of the i-th user in the community; i Let be the connection frequency feature vector of the i-th user in the cell.

4. The method according to claim 1, characterized in that, The cell load prediction model is derived in the following way: A cell load prediction model is constructed using an attention-enhanced LSTM network. cell (t+1), and cell (t+1) =f(LSTM Attn ([Load t M cell Time t ])); in, cell (t+1) represents the predicted cell load value at the next time step; f(LSTM) Attn ( )) represents the mapping function of the LSTM neural network with fused attention mechanism; Load t M is a vector of historical load metrics for the community. cell For aggregated vectors of user behavior in the community; Time t This is the time context vector.

5. The method according to claim 1, characterized in that, The community resource adaptation scoring function uses the following calculation formula: S=α1×( cell (t+1) / R cell )+α2×(σ cell )+α3×(D user_type ); Where S is the community resource adaptation score; cell (t+1) represents the predicted cell load value at the next time step; R cell For the community's resource capacity; σ cell For the load fluctuation of the community; Δ user_type α1 represents the degree of change in the user group structure within the current community; α2 and α3 are weighting coefficients.

6. The method according to claim 5, characterized in that, The degree of change in the current user group structure within the community is calculated using the following formula: ； Where, Δ user_type This represents the degree of change in the current user group structure within the community; || ||1 represents the L1 norm; This is the aggregation vector of user behavior in the current community; This is the aggregated vector of user behavior in the community for the previous preset time period.

7. The method according to any one of claims 1-6, characterized in that, Also includes: After executing the resource adjustment strategy, obtain the average PRB utilization rate change, user disconnection rate change, and user satisfaction change within a preset time period before and after the execution of the resource adjustment strategy. In response to the changes in average PRB utilization, user dropout rate, and user satisfaction exceeding their respective indicator thresholds, the resource adjustment strategy is determined to be effective, and the relevant data involved in the resource adjustment strategy is used as sample data to train the cell load prediction model. If the change in average PRB utilization, user disconnection rate, or user satisfaction does not exceed the corresponding threshold, the resource adjustment strategy is deemed invalid, and a new resource adjustment strategy is generated and executed until the new resource adjustment strategy is deemed valid.

8. A community resource reallocation system, characterized in that, include: The user behavior aggregation module is configured to construct a cell user behavior aggregation vector based on the user behavior of all users in the cell. The user behavior is characterized by time activity, spatial mobility, application preference, and connection frequency. The cell load prediction module is configured to train a cell load prediction model based on the cell historical load index vector, the cell user behavior aggregation vector, and the time context vector, and output the cell load prediction value through the trained cell load prediction model. The resource adaptation scoring module is configured to calculate the cell resource adaptation score based on the cell load prediction value, cell resource capacity, cell load fluctuation and the degree of change in the current user group structure in the cell, through a preset cell resource adaptation scoring function. The resource reconfiguration module is configured to trigger cell resource reconfiguration and generate and execute a resource adjustment strategy in response to the cell resource adaptation score value being greater than a preset reconfiguration threshold.

9. A computer device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor executes the cell resource reconfiguration method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, the processor performs the cell resource reconfiguration method according to any one of claims 1 to 7.

Images