Base station expansion method and device, computer device and storage medium

Abstract

This invention provides a base station expansion method, apparatus, computer equipment, and storage medium, relating to the field of network optimization technology. The method includes: predicting future load indicators of the base station based on historical operating indicators using a trained load prediction model; calculating a base station expansion risk index based on the future load indicators; calculating a corrected expansion risk index based on the base station expansion risk index, the prediction confidence of the load prediction model, and the degree of anomaly of the historical operating indicators; forming a corresponding expansion strategy based on the corrected expansion risk index; and executing the expansion strategy. The technical solution provided by this invention, through proactive prediction of future base station load, can dynamically adjust expansion decisions before congestion occurs, effectively avoiding the lag caused by threshold triggering.

Description

Technical Field

[0001] This invention relates to the field of network optimization technology, and in particular to a base station expansion method, a base station expansion device, a computer device, and a computer-readable storage medium. Background Technology

[0002] With the development of mobile communication technology and the rapid growth in the number of users, the load pressure on telecommunications operators' cell networks continues to increase. This is because users' network demands have shifted from simple voice communication to high-bandwidth, high-speed applications such as video streaming, online games, and virtual reality, causing a sharp increase in network traffic. Moreover, users' expectations for internet speeds are also rising, and prolonged network latency and buffering are becoming unacceptable. Simultaneously, against the backdrop of widespread internet access and the rapid development of mobile devices, the number of mobile internet users has surged, leading to an explosive growth in data traffic.

[0003] To ensure users have access to stable high-speed internet services, telecommunications operators take a series of measures, such as increasing the number of cells and bandwidth, and adding base stations, to expand the data traffic processing capacity within base stations, meet the needs of more people online at the same time, and thus improve the overall network experience. This method is called base station expansion, which increases the capacity of mobile networks to adapt to the growing user demand.

[0004] However, current base station expansion strategies in mobile communication networks mainly rely on static threshold triggering mechanisms, but this mechanism has a significant lag, which may lead to network congestion and a decline in user experience. Summary of the Invention

[0005] This invention was completed to at least partially address the technical problem of significant lag in existing base station expansion strategies.

[0006] According to one aspect of the present invention, a base station expansion method is provided, comprising: Based on historical operating indicators of base stations, the future load indicators of base stations are predicted through a trained load prediction model. Based on the future load indicators of the base station, calculate the base station expansion risk index; Based on the base station expansion risk index, the prediction confidence of the load prediction model, and the degree of anomaly of the historical operating indicators, a corrected expansion risk index is calculated. Based on the revised expansion risk index, a corresponding expansion strategy is formed; Execute the expansion strategy.

[0007] Optionally, the historical operating indicators of the base station include: historical PRB utilization rate, historical RRC connected users and other historical operating indicators, wherein the other historical operating indicators include one or more of the following: historical downlink average user rate, historical access failure rate, historical average user mobility speed, historical service type ratio, historical holiday identifiers and historical weather condition identifiers; the future load indicators of the base station include future PRB utilization rate and future RRC connected users.

[0008] Optionally, calculating the base station expansion risk index based on the future load indicators of the base station includes: The first and second differences of the future PRB utilization rate are calculated based on the predicted future PRB utilization rate. Based on the predicted future PRB utilization rate, the predicted future number of RRC connected users, and the first-order and second-order differences, the base station expansion risk index is calculated.

[0009] Optionally, the base station expansion risk index is calculated using the following formula: ; in, Risk index for base station expansion; The predicted PRB utilization rate for the i-th future day; Set the PRB expansion threshold; The predicted number of RRC connected users on the i-th future day; RRC expansion threshold; The first difference of the PRB utilization rate on the i-th day in the future; The second difference of the PRB utilization rate on the i-th day in the future; , , , These are the weighting coefficients; N is the total number of days for the forecast.

[0010] Optionally, the modified capacity expansion risk index is calculated using the following formula: ; in, To adjust the expansion risk index; The prediction confidence level of the load prediction model; Risk index for base station expansion; Anomaly scores are assigned to historical operating indicators to characterize the degree of abnormality of those indicators. The weighting is based on the percentage of abnormal scores.

[0011] Optionally, the prediction confidence of the load prediction model is calculated based on the MC Dropout algorithm or a Bayesian neural network algorithm.

[0012] Optionally, the step of forming a corresponding expansion strategy based on the modified expansion risk index includes: In response to the modified expansion risk index being less than or equal to a preset first risk threshold, the corresponding expansion strategy is to not expand. In response to the modified expansion risk index being greater than the first risk threshold and less than or equal to the preset second risk threshold, a corresponding expansion strategy is formed as a first-level expansion strategy, and the first-level expansion strategy is a preventive soft expansion strategy. In response to the modified expansion risk index being greater than the second risk threshold, a corresponding expansion strategy is formed as a secondary expansion strategy, which is an emergency hard expansion strategy.

[0013] Optionally, the primary capacity expansion strategy includes at least one of the following: dynamically allocating idle licensed resources from other low-load base stations, increasing the working bandwidth of the cell through software configuration, dynamically increasing the scheduling priority of users who meet preset conditions, adjusting cell access parameters to balance inter-cell load, and adjusting inter-cell handover offset to balance inter-cell load; and / or, The secondary capacity expansion strategy includes at least one of the following: dynamically activating secondary cells to aggregate multi-carrier resources, adaptively adjusting the uplink and downlink subframe configuration ratio to match service direction, and triggering resource allocation and configuration distribution of temporary radio nodes based on a preset emergency response plan.

[0014] Optionally, after implementing the expansion strategy, the base station expansion method further includes: Obtain the PRB utilization rate and RRC connected user count for each prediction within the preset verification period, and the base station expansion risk index corresponding to each prediction. and revised expansion risk index And the corresponding scaling strategy executed for each prediction, the actual PRB utilization and the actual number of RRC connected users after executing the scaling strategy; Implement at least one of the following correction mechanisms: If the deviation between the predicted PRB utilization rate and the actual PRB utilization rate exceeds a first deviation threshold, and / or the deviation between the predicted number of RRC connected users and the actual number of RRC connected users exceeds a second deviation threshold, the corresponding historical operating indicators of the base station, as well as the actual PRB utilization rate and the actual number of RRC connected users, are used as supplementary training data to fine-tune the parameters of the load prediction model; wherein, the corresponding verification samples include the historical operating indicators of the base station and the corresponding predicted future actual PRB utilization rate and the actual number of RRC connected users. When the resulting expansion strategy does not match the actual overload results, a logistic regression model is established using sample data within a preset verification period, with the corresponding base station expansion risk index. The four sub-items are used as input variables, and the actual overload label is used as the target variable for fitting. The base station expansion risk index is then adjusted based on the fitting results. The four weighting coefficients , , The actual overload result includes overload and non-overload. If the actual PRB utilization rate exceeds the PRB expansion threshold or the actual number of RRC connected users exceeds the RRC expansion threshold, it is determined to be overloaded. If the actual PRB utilization rate does not exceed the PRB expansion threshold and the actual number of RRC connected users does not exceed the RRC expansion threshold, it is determined to be non-overloaded. If the resulting capacity expansion strategy does not match the actual overload results, the capacity expansion risk index is adjusted based on sample data within a preset verification period. Prediction confidence The correlation between overload prediction accuracy and prediction confidence. Adjustments will be made; the expansion risk index will be revised based on sample data within the preset verification period. Abnormal scoring The correlation with actual overload results and the weighting of the anomaly score proportion. Adjustments are made; wherein, the overload prediction accuracy characterizes the consistency between the predicted overload state and the actual overload result; The contribution of each historical operating indicator to the prediction result output by the load prediction model is calculated based on a preset feature importance analysis algorithm. Indicators with a contribution value lower than a preset contribution value threshold are deleted, new indicators related to the prediction are added, and the load prediction model is retrained based on the adjusted indicators.

[0015] According to another aspect of the present invention, a base station expansion device is provided, comprising: The load prediction module is configured to predict future load indicators of the base station based on the base station's historical operating indicators and through a trained load prediction model. The first calculation module is configured to calculate the base station expansion risk index based on the future load indicators of the base station. The second calculation module is configured to calculate the corrected expansion risk index based on the base station expansion risk index, the prediction confidence of the load prediction model, and the degree of anomaly of the historical operating indicators. The strategy generation module is configured to generate a corresponding expansion strategy based on the modified expansion risk index; The strategy execution module is configured to execute the expansion strategy.

[0016] 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 base station expansion method.

[0017] According to another aspect of the present invention, a computer-readable storage medium is provided, on which a computer program is stored, wherein when the computer program is executed by a processor, the processor performs the aforementioned base station expansion method.

[0018] The technical solution provided by this invention may include the following beneficial effects: The base station expansion method and apparatus provided by this invention obtain future load indicators of the base station through load prediction, calculate an expansion risk index based on this, and then correct the index using prediction confidence and the degree of anomaly in historical data. Based on the correction results, an expansion strategy is dynamically formed, realizing proactive prediction of the future load of the base station. This effectively avoids the lag caused by threshold triggering and can dynamically adjust expansion decisions before congestion occurs, thereby significantly reducing network congestion risk, improving user experience, and increasing the utilization rate of wireless resources.

[0019] 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

[0020] 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.

[0021] Figure 1 This is a flowchart illustrating a base station capacity expansion method provided in an embodiment of the present invention; Figure 2 This is a flowchart illustrating another base station expansion method provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the base station expansion device 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

[0022] 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.

[0023] 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.

[0024] In related technologies, base station expansion strategies in mobile communication networks mainly rely on static threshold triggering mechanisms. For example, the system only triggers base station expansion operations when the PRB (Physical Resource Block) resource utilization rate of a 4G base station reaches 80% and the number of RRC (Radio Resource Control) connected users exceeds 120. This approach has a significant lag, which may lead to network congestion and a decline in user experience.

[0025] To address the aforementioned problems, this invention provides a base station expansion method capable of predicting expansion needs in advance. Specific embodiments are described in detail below.

[0026] Figure 1 This is a flowchart illustrating a base station capacity expansion method provided in an embodiment of the present invention. Figure 1 As shown, the method includes the following steps S101 to S105.

[0027] S101. Based on the historical operating indicators of the base station, predict the future load indicators of the base station through the trained load prediction model; S102. Calculate the base station expansion risk index based on the future load indicators of the base station; S103. Calculate the corrected expansion risk index based on the base station expansion risk index, the prediction confidence of the load prediction model, and the degree of anomaly of the historical operating indicators; S104. Based on the aforementioned modified expansion risk index, a corresponding expansion strategy is formed; S105. Execute the expansion strategy.

[0028] In this embodiment, the future load indicators of the base station are obtained through load prediction, and the expansion risk index is calculated accordingly. The index is then corrected using the prediction confidence level and the degree of anomaly in historical data. Based on the correction results, an expansion strategy is dynamically formed, which realizes the proactive prediction of the future load of the base station, effectively avoids the lag caused by threshold triggering, and can dynamically adjust the expansion decision before congestion occurs, thereby significantly reducing the risk of network congestion, improving user experience, and increasing the utilization rate of wireless resources.

[0029] In one specific implementation, the historical operating indicators of the base station include: historical PRB utilization rate, historical RRC connected users, and other historical operating indicators. The other historical operating indicators include one or more of the following: historical downlink average user rate, historical access failure rate, historical average user mobility speed, historical service type ratio, historical holiday identifiers, and historical weather condition identifiers. The future load indicators of the base station include future PRB utilization rate and future RRC connected users.

[0030] It should be noted that the historical operating indicators of the base station need to be converted into a multi-dimensional time series first. Then, the features in the multi-dimensional time series need to be standardized or normalized (i.e., preprocessed) to obtain a historical feature vector sequence. This sequence is then input into the load prediction model to predict the future load indicators of the base station. Of course, the future load indicators of the base station output by the model are also a feature vector sequence.

[0031] The preferred load prediction model is the LSTM (Long Short-Term Memory) model.

[0032] In this embodiment, by defining the specific components of historical operating indicators (including PRB utilization, RRC connected users and rate, failure rate, mobile speed, service type ratio, holidays, weather and other multi-dimensional features), and by specifying future load indicators as PRB utilization and RRC connected users, the prediction model can capture multi-source information that affects load changes, thereby improving the comprehensiveness and accuracy of future load state prediction and providing a reliable data foundation for subsequent capacity expansion decisions.

[0033] In one specific embodiment, step S102 specifically includes the following steps S1021 and S1022.

[0034] S1021. Calculate the first and second differences of the future PRB utilization rate based on the predicted future PRB utilization rate; S1022. Calculate the base station expansion risk index based on the predicted future PRB utilization rate, the predicted future RRC connected user number, and the first-order difference and second-order difference.

[0035] In this embodiment, the first and second differences of the future PRB utilization rate are calculated, and the expansion risk index is constructed by combining the future PRB utilization rate and the predicted value of the number of RRC connected users. This index not only reflects the absolute level of the load, but also characterizes the rate and acceleration of load change, thereby enabling early identification of the trend of rapid load increase or accelerated deterioration. Compared with the static threshold method that only relies on the current load value, it has a stronger risk warning capability.

[0036] In one specific implementation, in step S1022, the base station expansion risk index is calculated using the following formula: .

[0037] in, Risk index for base station expansion. Let PRB be the predicted utilization rate for the i-th future day. Set the PRB expansion threshold. Let be the predicted number of RRC connected users on the i-th future day. This is the threshold for RRC expansion. Let PRB be the first difference of the utilization rate on the i-th day in the future. Specifically, when i=1, ,in For the predicted PRB utilization rate on the first day of the future, Let be the actual PRB utilization rate at the t-th sampling time, i.e., the current actual PRB utilization rate (which belongs to the historical operation index of the base station); when i≥2, ,in To predict the PRB utilization rate for the i-th future day, Let PRB be the predicted utilization rate for the next (i-1)th day. This is the second-order difference of the PRB utilization rate on the i-th day in the future. Specifically, when i=1, ,in , For the predicted PRB utilization rate on the first day of the future, The actual PRB utilization rate at the t-th sampling time is the current actual PRB utilization rate (which belongs to the historical operation indicators of the base station). ,and The actual PRB utilization rate at the (t-1)th sampling time is the actual PRB utilization rate of the previous sampling time (which belongs to the historical operation index of the base station); when i≥2, ,in , To predict the PRB utilization rate for the i-th future day, The predicted PRB utilization rate for the next (i-1)th day; , The predicted PRB utilization rate for the next (i-2)th day; , , , These are the weighting coefficients; N is the total number of days for the forecast.

[0038] In this embodiment, the ERPI is calculated using a specific formula that includes square terms, first-order differences, and second-order differences, and adjustable weight coefficients α1 to α4 are introduced. This allows the expansion risk index to flexibly adapt to the differences in the importance of various factors under different scenarios. At the same time, the cumulative risk over the next N days is comprehensively assessed using a summation form, avoiding the one-sidedness of judgment at a single point in time and improving the scientificity and robustness of expansion decisions.

[0039] In one specific implementation, in S103, the corrected capacity expansion risk index is calculated using the following formula: .

[0040] in, To adjust the expansion risk index; The prediction confidence level of the load prediction model is used to quantify the uncertainty of the model output value (the higher the confidence level, the lower the uncertainty). Risk index for base station expansion; Anomaly scores are assigned to historical operating indicators to characterize the degree of abnormality of those indicators. The weighting is based on the percentage of abnormal scores.

[0041] Specifically, an autoencoder can be used to detect anomalies in the historical feature vector sequence input to the load prediction model. Specifically, the autoencoder is first trained using normal historical samples (historical samples without anomalies) to enable it to reconstruct normal patterns with high accuracy. For the historical sample to be detected, if its reconstruction error is significantly greater than that of the normal historical sample, it is judged as an anomaly. The anomaly score A is a quantitative indicator calculated based on the reconstruction error.

[0042] Anomaly score A can be calculated using the following formula:

[0043] Where A is the anomaly score, with a value range of (0,1). The closer the value is to 1, the higher the degree of anomaly of the historical sample to be detected. The closer the value is to 0, the closer the historical sample to be detected is to the normal pattern. E is the reconstruction error of the historical sample to be detected. The mean square error (MSE) is usually used, which is the average of the squares of the differences between the input sequence of the autoencoder and the reconstructed output sequence. This represents the mean of the reconstruction error for normal historical samples. The preset sensitivity parameters, and

[0044] Since the value of the anomaly score A is between [0,1], it can effectively prevent the anomaly score from overly dominating the expansion decision.

[0045] In this embodiment, by defining the modified expansion risk index as the product of the prediction confidence γ and the original ERPI plus the product of the anomaly score A and the weight β, independent control over prediction uncertainty and sudden abnormal events is achieved: when the model prediction is uncertain, only the contribution of normal risks is reduced, without suppressing the warning effect of abnormal events. Thus, while maintaining sensitivity to normal trends, it avoids missing sudden high load risks due to insufficient model confidence, significantly improving the adaptability and reliability of expansion decisions.

[0046] In one specific implementation, the prediction confidence of the load prediction model is calculated based on the MC Dropout algorithm or a Bayesian neural network algorithm.

[0047] The two algorithms are described in detail below: 1) MC Dropout (Monte-Carlo Dropout) algorithm: Dropout is used during model training and also during model prediction. The same input is forward propagated multiple times to obtain multiple predicted values. Then, the variance or standard deviation of these predicted values ​​is calculated. The larger the variance / standard deviation, the more uncertain the model is.

[0048] Specifically, during LSTM model training, Dropout is used to randomly discard some neurons to prevent overfitting. During LSTM model prediction, Dropout is also enabled, and forward propagation is performed multiple times on the same input sample (a sequence of historical feature vectors from the past T days), randomly discarding different neurons each time. This results in multiple different predictions, which form a distribution. If the LSTM model is highly deterministic, the predictions are concentrated with small variance; if the LSTM model is uncertain, the predictions are dispersed with large variance. Therefore, variance serves as a measure of prediction uncertainty: the larger the variance, the more uncertain the model, and the lower the confidence level.

[0049] Confidence of predictions from LSTM models The following formula can be used to calculate it:

[0050] Where N is the total number of days predicted; Let Variance be the variance of the predicted value obtained through multiple forward propagations using MC Dropout on the i-th future day. It is a zero-prevention factor to prevent division by zero.

[0051] 2) Bayesian Neural Network: Directly learns the distribution of weights, and the output has inherent uncertainty.

[0052] Specifically, a Bayesian Neural Network (BNN) treats the trainable parameters within an LSTM model as random variables and learns a probability distribution (usually a Gaussian distribution) for each trainable parameter. During prediction, multiple samples are taken from the posterior distribution of the trainable parameters to obtain multiple different model instances. Each instance provides a predicted value for the same input, thus obtaining a distribution of multiple predicted values. The variance (or standard deviation) of this distribution can be used to quantify the uncertainty of the prediction, and then the prediction confidence level can be calculated. .

[0053] Confidence of predictions from LSTM models The following formula can be used to calculate it:

[0054] Where N is the total number of days predicted; For the i-th future day, the variance of the predicted value obtained by sampling multiple times (M times in total) from the posterior distribution of the trainable parameters; It is a zero-prevention factor to prevent division by zero.

[0055]

[0056] in, Let be the predicted value for the i-th day in the future, obtained by sampling the m-th time from the posterior distribution of the trainable parameters; Let be the mean of the M predicted values ​​obtained from M samplings of the posterior distribution of the trainable parameters.

[0057] In this embodiment, by limiting the use of MC Dropout or Bayesian neural network to calculate the prediction confidence, two specific and feasible uncertainty quantification schemes are provided, enabling the load prediction model to output its own degree of certainty about the prediction results. This provides a calculable and statistically significant confidence parameter for subsequent correction of the capacity expansion risk index, enhancing the feasibility and flexibility of the method in engineering.

[0058] In one specific embodiment, step S104 specifically includes the following steps S1041 to S1043.

[0059] S1041. In response to the modified expansion risk index being less than or equal to a preset first risk threshold, a corresponding expansion strategy is formed as no expansion; S1042. In response to the modified expansion risk index being greater than the first risk threshold and less than or equal to the preset second risk threshold, a corresponding expansion strategy is formed as a first-level expansion strategy, and the first-level expansion strategy is a preventive soft expansion strategy. S1043. In response to the modified expansion risk index being greater than the second risk threshold, a corresponding expansion strategy is formed as a secondary expansion strategy, and the secondary expansion strategy is an emergency hard expansion strategy.

[0060] Specifically, the preventative soft capacity expansion strategy involves alleviating cell capacity pressure without adding physical resources by using non-hardware changes such as software configuration parameter optimization and load balancing, based on load forecasting and triggering in advance. The emergency hard capacity expansion strategy involves rapidly increasing the cell capacity limit in response to sudden high loads or emergency events by activating secondary cell carrier aggregation to adjust the uplink / downlink subframe ratio or deploying temporary radio nodes, through hardware upgrades or major configuration changes.

[0061] In this embodiment, by comparing the modified expansion risk index with the preset first and second risk thresholds, three different levels of response strategies are automatically generated: no expansion, Level 1 expansion (preventive soft expansion), or Level 2 expansion (emergency hard expansion). This achieves risk-level handling: mild risks only require soft adjustments, while severe risks trigger hardware-level expansion. This avoids the resource waste or insufficient response caused by traditional single strategies and improves the precision and cost efficiency of expansion operations.

[0062] In one specific implementation, the primary capacity expansion strategy includes at least one of the following: dynamically allocating idle licensed resources from other low-load base stations, increasing the working bandwidth of the cell through software configuration, dynamically increasing the scheduling priority of users who meet preset conditions, adjusting cell access parameters to balance the inter-cell load, and adjusting the inter-cell handover offset to balance the inter-cell load.

[0063] The preset conditions refer to the filtering rules that are pre-set based on dimensions such as user level, service QCI (Quality of Service Class Identifier), and channel quality, which are used to identify user groups that need to be prioritized.

[0064] In this embodiment, by listing various specific methods of the primary capacity expansion strategy (permission allocation, increasing working bandwidth, increasing user scheduling priority, adjusting access parameters, and switching offset), a set of soft capacity expansion solutions that do not require the addition of new hardware is provided. These methods can be used individually or in combination, and can quickly respond to medium- and high-risk scenarios, alleviate cell capacity pressure without interrupting existing network services, and have the advantages of low cost, high timeliness and strong reversibility.

[0065] In one specific implementation, the secondary expansion strategy includes at least one of the following: dynamically activating secondary cells to aggregate multi-carrier resources, adaptively adjusting the uplink and downlink subframe configuration ratio to match service direction, and triggering resource allocation and configuration distribution of temporary radio nodes based on a preset emergency response plan.

[0066] In this embodiment, by listing various specific methods of the secondary capacity expansion strategy (carrier aggregation, subframe ratio adjustment, and temporary wireless node deployment), a hard capacity expansion scheme that can significantly increase the capacity limit is provided. It is suitable for extremely high-risk or sudden event scenarios, can quickly expand bandwidth or add coverage nodes, effectively cope with instantaneous traffic surges, and ensure user experience and network stability in extreme scenarios.

[0067] In one specific implementation, after step S105, the base station expansion method further includes the following steps S106 and S107.

[0068] S106. Obtain the PRB utilization rate and RRC connected user number for each prediction within the preset verification period, and the base station expansion risk index corresponding to each prediction. and revised expansion risk index And the corresponding expansion strategy (no expansion, first-level expansion, second-level expansion) executed for each prediction, as well as the actual PRB utilization and actual number of RRC connected users after executing the expansion strategy; S107. Perform at least one of the following correction mechanisms one through four.

[0069] Correction Mechanism 1: If the deviation between the predicted PRB utilization rate and the actual PRB utilization rate exceeds the first deviation threshold, and / or the deviation between the predicted number of RRC connected users and the actual number of RRC connected users exceeds the second deviation threshold, the corresponding historical operating indicators of the base station, as well as the actual PRB utilization rate and the actual number of RRC connected users, will be used as supplementary training data to fine-tune the parameters of the load prediction model.

[0070] The corresponding historical operating indicators of the base station refer to the historical operating indicators of the base station used to predict the future load indicators of the base station when the deviation between the predicted PRB utilization rate and the actual PRB utilization rate exceeds a first deviation threshold, and the historical operating indicators of the base station used to predict the future load indicators of the base station when the deviation between the predicted number of RRC connected users and the actual number of RRC connected users exceeds a second deviation threshold.

[0071] The parameters of the load prediction model can be fine-tuned as follows: freeze the shallow layer parameters of the network and update only the deep layer or output layer parameters. Specifically, keep the weights of several layers near the model input unchanged (these layers have learned common temporal features and long-term dependency patterns), and only retrain the fully connected output layer (or the last few layers) at the end of the model. This allows the model to adjust the mapping relationship according to the latest data distribution, thereby quickly adapting to recent changes in network behavior while avoiding overfitting and catastrophic forgetting.

[0072] Correction Mechanism Two: If the resulting expansion strategy does not match the actual overload results, a logistic regression model is established using sample data from a preset verification period, with the corresponding base station expansion risk index as the basis. The four sub-items (corresponding to four weight coefficients) are used as input variables, and the actual overload label is used as the target variable for fitting. The base station expansion risk index is then adjusted based on the fitting results. The four weighting coefficients , , .

[0073] The actual overload result includes overload and no overload. If the actual PRB utilization exceeds the PRB expansion threshold or the actual number of RRC connected users exceeds the RRC expansion threshold, it is determined to be overloaded. If the actual PRB utilization does not exceed the PRB expansion threshold and the actual number of RRC connected users does not exceed the RRC expansion threshold, it is determined to be no overloaded. The expansion strategy matching the actual overload result means that the expansion strategy is no expansion and the actual overload result is no overload, or the expansion strategy is a first-level or second-level expansion and the actual overload result is overloaded. The expansion strategy not matching the actual overload result means that the expansion strategy is no expansion but the actual overload result is overloaded, or the expansion strategy is a first-level or second-level expansion but the actual overload result is not overloaded.

[0074] Correction Mechanism 3: If the resulting expansion strategy does not match the actual overload results, the expansion risk index is corrected based on sample data within a preset verification period. Prediction confidence The correlation between overload prediction accuracy and prediction confidence. Adjustments will be made; the expansion risk index will be revised based on sample data within the preset verification period. Abnormal scoring The correlation with actual overload results and the weighting of the anomaly score proportion. Adjustments will be made.

[0075] The overload prediction accuracy characterizes the consistency between the predicted overload state and the actual overload result. Specifically, the predicted overload state is defined as follows: if the resulting expansion strategy is a first-level or second-level expansion, it is considered a predicted overload (i.e., the load prediction model believes an overload will occur in the future); if the resulting expansion strategy is no expansion, it is considered a predicted no-overload (i.e., the load prediction model believes there will be no overload in the future). The consistency between the predicted overload state and the actual overload result means that if both the predicted and actual overload results are overloaded, or both are predicted and are not overloaded, then they are consistent; if both are predicted and are not overloaded, then they are inconsistent.

[0076] Correction Mechanism 4: Calculate the contribution of each historical operating indicator to the prediction result output by the load prediction model based on a preset feature importance analysis algorithm, delete indicators whose contribution is lower than the preset contribution threshold, add new indicators related to the prediction, and retrain the load prediction model based on the adjusted indicators.

[0077] The feature importance analysis algorithm can employ the LSTM Attention algorithm, which is a class of algorithms used to measure and understand the influence of different input features on the model's prediction results. Specifically, by assigning dynamic weights to each input feature of the LSTM model, the influence (i.e., contribution) of each historical performance indicator on the prediction results is quantified, and the importance of each input feature is ranked according to its weight.

[0078] In this embodiment, a closed-loop intelligent optimization process is constructed by executing a verification and correction mechanism (including model fine-tuning, ERPI weight adjustment, γ / β optimization, and feature selection) after expansion. This process can utilize actual load feedback to correct the prediction model, risk index parameters, and input feature set, continuously improving the accuracy of expansion decisions, reducing false expansions and missed expansions, enabling the system to have adaptive evolution capabilities, and significantly reducing manual intervention costs while maintaining excellent expansion performance after long-term deployment.

[0079] The base station expansion method provided in this invention achieves proactive prediction and dynamic decision-making regarding cell load by integrating deep learning prediction, multi-dimensional risk index construction, and closed-loop feedback optimization. It uses historical operational indicators to predict future load and introduces first- / second-order differences to represent load change trends, enabling expansion decisions to respond to congestion risks in advance. Simultaneously, it combines prediction confidence and anomaly scoring to correct the risk index, effectively addressing model uncertainties and sudden traffic events, avoiding misjudgments and omissions. Through a tiered expansion strategy (preventative soft expansion and emergency hard expansion) and a multi-dimensional model self-correction mechanism (parameter fine-tuning, weight optimization, and feature selection), it significantly improves the accuracy, timeliness, and resource utilization efficiency of expansion decisions, reduces manual intervention costs, and ensures user experience and network stability.

[0080] Figure 2 This is a flowchart illustrating another base station expansion method provided in an embodiment of the present invention. Figure 2 As shown, the method includes the following steps S201 to S205.

[0081] S201. Construct an expanded feature vector sequence.

[0082] The purpose of this step is to transform the historical operating indicators of the base station into a multi-dimensional time series input to the LSTM model.

[0083] First, construct the following feature vectors: .

[0084] The descriptions of each feature in the above feature vector are shown in Table 1 below: Table 1

[0085] Sampling is performed once a day to obtain a historical feature vector sequence for T consecutive days (where the t-th sampling time is the current sampling time): X=[X t-T+1 , X t-T+2 , ..., X t ].

[0086] S202. LSTM Model Training and Prediction.

[0087] The purpose of this step is to predict the base station PRB utilization and the number of RRC connected users for the next N days and analyze their growth trends. Specifically, a multi-layer LSTM network is used for time series modeling to output a feature vector sequence of predicted load indicators for the next N days.

[0088] Input the historical feature vector sequence X into the LSTM to obtain the predicted values ​​of future load indicators: ; .

[0089] Output metrics: Predict PRB utilization rate for the i-th future day, in % (%). Predict the number of RRC connected users for the i-th future day, in units.

[0090] S203. Calculate the dynamic expansion risk index and the modified expansion risk index.

[0091] This step not only determines whether the threshold is exceeded, but also considers the load growth trend and rate.

[0092] First, construct the expansion risk index: .

[0093] The parameters in the above formula are described in Table 2 below: Table 2

[0094] Reconstruction and expansion risk index: .

[0095] The parameters in the above formula are described in Table 3 below: Table 3

[0096] Because LSTM predictions are inherently uncertain, even if the model fits well, real-world business scenarios still present risks such as low confidence levels, large data fluctuations, and training data distribution drift. Directly using LSTM model predictions as the basis for subsequent scaling can easily lead to misjudgments or omissions. Furthermore, unpredictable events such as holidays, accidents, and large-scale activities can cause sudden changes in user behavior, such as a short-term surge in traffic. The LSTM model might underestimate scaling needs because it "hasn't seen this pattern before." Therefore, a modified scaling risk index is needed.

[0097] Adjusted expansion risk index It includes two key ingredients: 1) Introduction of prediction confidence γ.

[0098] Uncertainty estimation is performed on the output of each LSTM model using methods such as MC Dropout or Bayesian neural networks; each predicted value is labeled with a "confidence tag" for weight adjustment during scaling.

[0099] For example: γ = 0.9 → The model is very confident (the prediction results are highly reliable and can be used to trigger expansion), γ = 0.3 → The model is not confident (it may encounter unknown patterns or data drift).

[0100] 2) Introduction of anomaly score A.

[0101] The autoencoder method is used to detect "abnormal behavior" in the input historical feature vector sequence to determine whether a sudden high-load scenario is likely to occur.

[0102] Example: A sudden surge in user numbers on Saturday → This never occurred in the training data, so the prediction was underestimated.

[0103] S204. Expansion strategy generation and execution.

[0104] The purpose of this step is to This is used to trigger different levels of scaling behavior, distinguish different risk levels, and refine scaling strategies.

[0105] The specific expansion strategies are shown in Table 4 below: Table 4

[0106] Level 1 expansion definition: When If the value is higher than the threshold θ1, it indicates that "the risk of expansion is high but has not yet occurred", and the first-level expansion mechanism is triggered at this time.

[0107] Level 1 expansion employs a preventative soft expansion strategy, including at least one of the following: License allocation in advance: Dynamically allocate idle license resources from other low-load base stations, such as transferring idle license resources from other low-load base stations.

[0108] Resource scaling: This involves increasing the total PRB resources of a cell by one level. Specifically, it involves adjusting the cell's working bandwidth from its current value to the next standard bandwidth tier via software configuration, thereby increasing the cell's working bandwidth. For example, increasing the cell's working bandwidth from 10MHz to 20MHz by enabling 10MHz software scaling increases the cell's total bandwidth by 10MHz, correspondingly increasing the number of available PRBs.

[0109] Scheduling pre-optimization: Dynamically increase the scheduling priority of users who meet preset conditions, i.e., temporarily increase the scheduling weight of users; adjust cell access parameters to balance the load between cells, i.e., fine-tune cell access parameters, such as temporary small adjustments: cell reselection parameters, used to guide idle users to preferentially camp on low-load cells; handover parameters (such as cell individual offset CIO, A3 event offset), used to control the handover timing of connected users between cells to achieve proactive load transfer; and access control parameters, used to appropriately increase the access threshold under high load to alleviate congestion pressure.

[0110] Cell offset control: Adjusting the inter-cell handover offset to balance the inter-cell load, i.e., appropriately adjusting the neighbor cell load sharing strategy. Specifically, by dynamically adjusting the relevant offset of inter-cell handover / reselection, the signal quality comparison threshold between the serving cell and neighboring cells is changed, so that edge users trigger handover earlier or later when the conditions are met. Thus, without changing the user's perception, some users are actively transferred from high-load cells to low-load neighboring cells, thereby balancing the inter-cell load.

[0111] Secondary expansion definition: when If the value exceeds the threshold θ2 and is accompanied by an “abnormal event” (such as a holiday, a meeting, or a gathering of people), the secondary expansion mechanism will be triggered immediately.

[0112] Secondary capacity expansion employs an emergency hard expansion strategy, including at least one of the following: Automatic configuration change triggering: Dynamically activate secondary cells to aggregate multi-carrier resources. Specifically, dynamically activate secondary cells, enable carrier aggregation (CA) function, bind the primary carrier with one or more secondary carriers to achieve joint scheduling of cross-carrier resources, thereby improving user peak rate and total cell throughput; Adaptively adjust uplink and downlink subframe configuration ratio to match service direction. Specifically, under TDD (Time Division Duplexing) system, adaptively adjust the TDD uplink and downlink subframe configuration ratio to match service direction according to real-time service requirements (increase the proportion of downlink subframes when downlink is congested, and increase the proportion of uplink subframes when uplink is congested), thereby improving capacity and alleviating congestion.

[0113] Emergency planning activation: Based on the preset emergency response plan, the resource allocation and configuration of temporary wireless nodes are automatically or manually triggered. This may include activating micro / pico cells, deploying fill-in sites, activating emergency communication vehicles or temporary sites, etc. By adding coverage and capacity nodes, the service pressure of high-load cells is relieved, and the user experience is guaranteed in emergency scenarios.

[0114] S205. Prediction Validation and Model Correction.

[0115] For the LSTM-based model designed in the aforementioned steps and The scaling prediction scheme based on the scoring mechanism may encounter the following problems after actual deployment: The predicted results are inconsistent with the actual network performance; Some base stations were predicted to be expanded, but were not actually overloaded → false expansion; Some base stations that were not planned for capacity expansion were actually overloaded → missed capacity expansion; As the system environment changes, the accuracy of the original LSTM model gradually decreases. It is necessary to review whether each prediction was correct and what areas for improvement exist.

[0116] Therefore, this step introduces mechanisms such as model validation, parameter correction, and feature optimization to form a closed-loop optimization process, specifically including the following steps S2051 to S2056.

[0117] S2051. Collect validation data and evaluate the accuracy of predictions.

[0118] This step is to determine the performance of the LSTM model over a past period. Whether the expansion predictions are accurate provides the basic evaluation data for subsequent "model correction" and "algorithm optimization".

[0119] Implementation process: 1) Select a verification period, such as the past 30 days or the most recent 100 capacity expansion predictions; 2) For each capacity expansion prediction, record the following information: Predicted PRB / RRC sequences; The ERPI calculated at that time and ; Has expansion been triggered (Level 1 / Level 2 / Not triggered)? Actual load value (PRB / RRC true value); 3) Construct a data comparison table based on the collected verification data (as shown in Table 5).

[0120] Table 5

[0121] S2052. Adjust LSTM model parameters (only when prediction bias is large).

[0122] If the predicted values ​​of PRB / RRC differ significantly from the actual values, fine-tune the LSTM model to improve prediction accuracy. This allows the LSTM model to better reflect the current network operating conditions and adapt to new situations (such as holidays, events, and changes in user structure).

[0123] Implementation process: The validation dataset from substep S2051 (especially the days with large errors) is used as supplementary training data. Instead of training from scratch, a "fine-tuning" approach is adopted: the first few layers of the LSTM model are frozen (i.e., preserving the learned long-term patterns); only the parameters of the model's output layer are updated (i.e., learning from recent behavior). Training requires only a small amount of data and a few epochs to avoid overfitting; the updated LSTM model is used for the next round of expansion prediction.

[0124] S2053. Modify the weighting coefficients α1 to α4 of ERPI.

[0125] The goal of this step is to ensure that if the LSTM prediction is good, but... If the expansion strategy corresponding to the score is consistently incorrect, it indicates that the expansion risk index needs to be adjusted. This scoring function needs adjustment. The decision logic for whether to trigger capacity expansion needs to be made more reasonable to avoid judgment errors.

[0126] Implementation process: 1) Count the number of times expansion is triggered each time. Value, corresponding to the actual load situation (i.e., whether it is overloaded); if it is frequently " "High score but not actually overloaded", or "High score but not actually overloaded" The statement "the score is low but the actual load is overloaded" indicates that there is a problem with the allocation of weight coefficients α1 to α4. 2) Use the following method to adjust the weights: Build a simple linear regression model with "overload" as the target variable and the four sub-items of ERPI as input variables; fit the model with the actual results to obtain the optimal combination of α values; the new combination of α values ​​will make the ERPI calculation more reasonable the next time it is used.

[0127] S2054. Correction The prediction confidence γ and the anomaly score weight β.

[0128] if If the model is too sensitive or too insensitive, the coefficients γ and β need to be adjusted. This allows for a more reasonable balance between the weights of "model confidence" and "sudden anomalies."

[0129] Implementation process: 1) During each prediction: To determine whether the model prediction confidence γ truly reflects the reliability of the predicted value, and whether the anomaly score A accurately increases when sudden anomalies occur in the historical feature vectors; 2) Adjustment method: If γ is very low multiple times, but the overload prediction results are accurate → the model confidence is too conservative → increase γ appropriately; If the anomaly score A is high, but there is actually no overload → β is set too high → reduce the anomalous amplification level; 3) Apply the adjusted γ and β values ​​to the next round of capacity expansion prediction.

[0130] S2055. Feature optimization mechanism (automatic adjustment of input data dimensions)

[0131] The goal of this step is to adjust the feature set if some features are not very helpful for model prediction, or if stronger features are introduced externally (such as CQI, switching failure rate, etc.). This simplifies the model, improves generalization ability, and reduces the training burden.

[0132] Implementation process: 1) Use a feature importance analysis algorithm (such as LSTM Attention) to calculate the contribution of each input feature to the output prediction; 2) Feature importance ranking: RRC connected users → Contribution: 42%; PRB utilization rate → contribution: 34%; CQI average → Contribution: 14%; Holiday sign → Contribution: 7%; Switchover failure rate → Contribution: 3%; 3) Delete fields with extremely low contribution (such as <3%) and try to add new and more effective features (such as neighboring cell load, mobility rate fluctuation). 4) Retrain the LSTM model.

[0133] S2056. Generate a backtracking report.

[0134] The goal of this step is to enable engineers, managers, and AI auditing systems to see whether each scaling-up trigger was accurate and why. This provides data support for continuous optimization, reviewing scaling-up strategies, and improving trust.

[0135] Implementation process: 1) Log each expansion prediction event is recorded, including: Comparison of predicted and actual trends (curve graph); Whether to trigger capacity expansion, and the trigger strategy level (Level 1 / Level 2); ERPI and Numerical changes; The relative roles of γ and A; 2) Automatically generate reports and provide them in PDF format; 3) Management personnel review the data weekly or monthly, taking into account the actual number of complaints and overload events, to analyze whether further optimization and expansion strategies are needed.

[0136] The base station expansion method provided in this invention is based on LSTM deep learning network for base station expansion prediction. Specifically, it uses historical base station operation index data to predict future load development trends and constructs and corrects an expansion risk index to achieve early expansion judgment, thereby effectively overcoming the lag defect of traditional static threshold triggering mechanisms. This method introduces the first and second order differences of PRB utilization and RRC connection count, enabling the expansion risk index to capture the speed and acceleration of load changes and identify rapid deterioration trends in advance. Simultaneously, it uses prediction confidence and anomaly scoring to correct the risk index, avoiding misjudgments and omissions caused by model uncertainty or sudden abnormal events. Based on this, a tiered expansion strategy is automatically matched according to the corrected risk index: maintaining the status quo for low risk, triggering preventative soft expansion (such as license allocation, bandwidth soft expansion, and parameter optimization) for medium and high risk, and initiating emergency hard expansion (such as carrier aggregation, subframe adjustment, and temporary site deployment) for extremely high risk or emergency events, achieving precise matching of resource investment and risk level. Furthermore, this method integrates closed-loop verification and self-correction mechanisms. By dynamically adjusting prediction bias, weight coefficients, confidence levels, and feature sets across multiple dimensions, it continuously improves the model's prediction accuracy and decision robustness, significantly reduces the cost of manual intervention, and ultimately achieves the comprehensive benefits of improving network capacity utilization, ensuring user experience, and reducing congestion.

[0137] Figure 3 This is a schematic diagram of the base station expansion device provided in an embodiment of the present invention. Figure 3 As shown, the device includes: a load prediction module 301, a first calculation module 302, a second calculation module 303, a strategy generation module 304, and a strategy execution module 305.

[0138] The load prediction module 301 is configured to predict future load indicators of the base station based on historical operating indicators of the base station using a trained load prediction model; the first calculation module 302 is configured to calculate the base station expansion risk index based on the future load indicators of the base station; the second calculation module 303 is configured to calculate a corrected expansion risk index based on the base station expansion risk index, the prediction confidence of the load prediction model, and the degree of anomaly of the historical operating indicators; the strategy generation module 304 is configured to form a corresponding expansion strategy based on the corrected expansion risk index; and the strategy execution module 305 is configured to execute the expansion strategy.

[0139] In one specific implementation, the historical operating indicators of the base station include: historical PRB utilization rate, historical RRC connected users, and other historical operating indicators. The other historical operating indicators include one or more of the following: historical downlink average user rate, historical access failure rate, historical average user mobility speed, historical service type ratio, historical holiday identifiers, and historical weather condition identifiers. The future load indicators of the base station include future PRB utilization rate and future RRC connected users.

[0140] In one specific implementation, the first calculation module 302 is specifically configured to: calculate the first-order difference and the second-order difference of the future PRB utilization rate based on the predicted future PRB utilization rate; and calculate the base station expansion risk index based on the predicted future PRB utilization rate, the predicted future number of RRC connected users, and the first-order difference and the second-order difference.

[0141] In one specific implementation, the first calculation module 302 calculates the base station expansion risk index using the following formula: ; in, Risk index for base station expansion; The predicted PRB utilization rate for the i-th future day; Set the PRB expansion threshold; The predicted number of RRC connected users on the i-th future day; RRC expansion threshold; The first difference of the PRB utilization rate on the i-th day in the future; The second difference of the PRB utilization rate on the i-th day in the future; , , , These are the weighting coefficients; N is the total number of days for the forecast.

[0142] In one specific implementation, the second calculation module 303 calculates the corrected expansion risk index using the following formula: ; in, To adjust the expansion risk index; The prediction confidence level of the load prediction model; Risk index for base station expansion; Anomaly scores are assigned to historical operating indicators to characterize the degree of abnormality of those indicators. The weighting is based on the percentage of abnormal scores.

[0143] In one specific implementation, the second calculation module 303 calculates the prediction confidence of the load prediction model based on the MC Dropout algorithm or the Bayesian neural network algorithm.

[0144] In one specific implementation, the strategy generation module 304 is specifically configured as follows: In response to the modified expansion risk index being less than or equal to a preset first risk threshold, the corresponding expansion strategy is to not expand. In response to the modified expansion risk index being greater than the first risk threshold and less than or equal to the preset second risk threshold, a corresponding expansion strategy is formed as a first-level expansion strategy, and the first-level expansion strategy is a preventive soft expansion strategy. In response to the modified expansion risk index being greater than the second risk threshold, a corresponding expansion strategy is formed as a secondary expansion strategy, which is an emergency hard expansion strategy.

[0145] In one specific implementation, the primary capacity expansion strategy includes at least one of the following: dynamically allocating idle licensed resources from other low-load base stations, increasing the working bandwidth of the cell through software configuration, dynamically increasing the scheduling priority of users who meet preset conditions, adjusting cell access parameters to balance the inter-cell load, and adjusting the inter-cell handover offset to balance the inter-cell load.

[0146] In one specific implementation, the secondary expansion strategy includes at least one of the following: dynamically activating secondary cells to aggregate multi-carrier resources, adaptively adjusting the uplink and downlink subframe configuration ratio to match service direction, and triggering resource allocation and configuration distribution of temporary radio nodes based on a preset emergency response plan.

[0147] In one specific embodiment, the base station expansion device further includes a data collection module and a correction module.

[0148] The data collection module is configured to acquire the PRB utilization rate and RRC connected user number for each prediction within a preset verification period, as well as the base station expansion risk index corresponding to each prediction. and revised expansion risk index And the corresponding scaling strategy executed for each prediction, the actual PRB utilization and the actual number of RRC connected users after executing the scaling strategy; The correction module is configured to perform at least one of the following correction mechanisms: If the deviation between the predicted PRB utilization rate and the actual PRB utilization rate exceeds a first deviation threshold, and / or the deviation between the predicted number of RRC connected users and the actual number of RRC connected users exceeds a second deviation threshold, the corresponding historical operating indicators of the base station, as well as the actual PRB utilization rate and the actual number of RRC connected users, are used as supplementary training data to fine-tune the parameters of the load prediction model; wherein, the corresponding verification samples include the historical operating indicators of the base station and the corresponding predicted future actual PRB utilization rate and the actual number of RRC connected users. When the resulting expansion strategy does not match the actual overload results, a logistic regression model is established using sample data within a preset verification period, with the corresponding base station expansion risk index. The four sub-items are used as input variables, and the actual overload label is used as the target variable for fitting. The base station expansion risk index is then adjusted based on the fitting results. The four weighting coefficients , , The actual overload result includes overload and non-overload. If the actual PRB utilization rate exceeds the PRB expansion threshold or the actual number of RRC connected users exceeds the RRC expansion threshold, it is determined to be overloaded. If the actual PRB utilization rate does not exceed the PRB expansion threshold and the actual number of RRC connected users does not exceed the RRC expansion threshold, it is determined to be non-overloaded. If the resulting capacity expansion strategy does not match the actual overload results, the capacity expansion risk index is adjusted based on sample data within a preset verification period. Prediction confidence The correlation between overload prediction accuracy and prediction confidence. Adjustments will be made; the expansion risk index will be revised based on sample data within the preset verification period. Abnormal scoring The correlation with actual overload results and the weighting of the anomaly score proportion. Adjustments are made; wherein, the overload prediction accuracy characterizes the consistency between the predicted overload state and the actual overload result; The contribution of each historical operating indicator to the prediction result output by the load prediction model is calculated based on a preset feature importance analysis algorithm. Indicators with a contribution value lower than a preset contribution value threshold are deleted, new indicators related to the prediction are added, and the load prediction model is retrained based on the adjusted indicators.

[0149] The base station expansion device provided in this invention achieves proactive prediction and dynamic decision-making regarding cell load by integrating deep learning prediction, multi-dimensional risk index construction, and closed-loop feedback optimization. It uses historical operational indicators to predict future load and introduces first- / second-order differences to represent load change trends, enabling expansion decisions to respond to congestion risks in advance. Simultaneously, it combines prediction confidence and anomaly scoring to correct the risk index, effectively addressing model uncertainties and sudden traffic events, avoiding misjudgments and omissions. Through a tiered expansion strategy (preventative soft expansion and emergency hard expansion) and a multi-dimensional model self-correction mechanism (parameter fine-tuning, weight optimization, and feature selection), it significantly improves the accuracy, timeliness, and resource utilization efficiency of expansion decisions, reduces manual intervention costs, and ensures user experience and network stability.

[0150] 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 base station expansion method.

[0151] Based on the same technical concept, this embodiment of the invention also provides 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 base station expansion method.

[0152] In summary, the base station expansion method, apparatus, computer equipment, and storage medium provided in this embodiment of the invention can predict base station load change trends in advance, effectively overcome the lag defects of traditional static threshold triggering, and thus proactively implement expansion decisions before network congestion occurs. At the same time, it can significantly reduce false expansion and missed expansion, improve the efficiency of wireless resource utilization, and adaptively match preventive or emergency expansion strategies according to different risk levels. Ultimately, it continuously improves decision accuracy through closed-loop optimization, reduces manual intervention costs, and ensures user service quality and network stability.

[0153] 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.

[0154] 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 expanding base station capacity, characterized in that, include: Based on historical operating indicators of base stations, the future load indicators of base stations are predicted through a trained load prediction model. Based on the future load indicators of the base station, calculate the base station expansion risk index; Based on the base station expansion risk index, the prediction confidence of the load prediction model, and the degree of anomaly of the historical operating indicators, a corrected expansion risk index is calculated. Based on the revised expansion risk index, a corresponding expansion strategy is formed; Execute the expansion strategy.

2. The base station expansion method according to claim 1, characterized in that, The historical operating indicators of the base station include: historical PRB utilization rate, historical RRC connected users, and other historical operating indicators. The other historical operating indicators include one or more of the following: historical downlink average user rate, historical access failure rate, historical average user mobility speed, historical service type ratio, historical holiday identifiers, and historical weather condition identifiers. The future load indicators of the base station include future PRB utilization rate and future RRC connected users.

3. The base station expansion method according to claim 2, characterized in that, The calculation of the base station expansion risk index based on the future load indicators of the base station includes: The first and second differences of the future PRB utilization rate are calculated based on the predicted future PRB utilization rate. Based on the predicted future PRB utilization rate, the predicted future number of RRC connected users, and the first-order and second-order differences, the base station expansion risk index is calculated.

4. The base station expansion method according to claim 3, characterized in that, The base station expansion risk index is calculated using the following formula: ； in, Risk index for base station expansion; The predicted PRB utilization rate for the i-th future day; Set the PRB expansion threshold; The predicted number of RRC connected users on the i-th future day; RRC expansion threshold; The first difference of the PRB utilization rate on the i-th day in the future; The second difference of the PRB utilization rate on the i-th day in the future; , , , These represent the weighting coefficients; N is the total number of days in the forecast.

5. The base station expansion method according to claim 4, characterized in that, The revised expansion risk index is calculated using the following formula: ； in, To adjust the expansion risk index; The prediction confidence level of the load prediction model; Risk index for base station expansion; Anomaly scores are assigned to historical operating indicators to characterize the degree of abnormality of those indicators. The weighting is based on the percentage of abnormal scores.

6. The base station expansion method according to claim 5, characterized in that, The prediction confidence of the load prediction model is calculated based on the MC Dropout algorithm or the Bayesian neural network algorithm.

7. The base station expansion method according to any one of claims 1-6, characterized in that, The process of formulating a corresponding expansion strategy based on the modified expansion risk index includes: In response to the modified expansion risk index being less than or equal to a preset first risk threshold, the corresponding expansion strategy is to not expand. In response to the modified expansion risk index being greater than the first risk threshold and less than or equal to the preset second risk threshold, a corresponding expansion strategy is formed as a first-level expansion strategy, and the first-level expansion strategy is a preventive soft expansion strategy. In response to the modified expansion risk index being greater than the second risk threshold, a corresponding expansion strategy is formed as a secondary expansion strategy, which is an emergency hard expansion strategy.

8. The base station expansion method according to claim 7, characterized in that, The first-level capacity expansion strategy includes at least one of the following: dynamically allocating idle licensed resources from other low-load base stations, increasing the working bandwidth of the cell through software configuration, dynamically increasing the scheduling priority of users who meet preset conditions, adjusting cell access parameters to balance the inter-cell load, and adjusting the inter-cell handover offset to balance the inter-cell load. And / or, The secondary capacity expansion strategy includes at least one of the following: dynamically activating secondary cells to aggregate multi-carrier resources, adaptively adjusting the uplink and downlink subframe configuration ratio to match service direction, and triggering resource allocation and configuration distribution of temporary radio nodes based on a preset emergency response plan.

9. The base station expansion method according to claim 5, characterized in that, After implementing the expansion strategy, the base station expansion method further includes: Obtain the PRB utilization rate and RRC connected user count for each prediction within the preset verification period, and the base station expansion risk index corresponding to each prediction. and revised expansion risk index And the corresponding scaling strategy executed for each prediction, the actual PRB utilization and the actual number of RRC connected users after executing the scaling strategy; Implement at least one of the following correction mechanisms: If the deviation between the predicted PRB utilization rate and the actual PRB utilization rate exceeds a first deviation threshold, and / or the deviation between the predicted number of RRC connected users and the actual number of RRC connected users exceeds a second deviation threshold, the corresponding historical operating indicators of the base station, as well as the actual PRB utilization rate and the actual number of RRC connected users, are used as supplementary training data to fine-tune the parameters of the load prediction model; wherein, the corresponding verification samples include the historical operating indicators of the base station and the corresponding predicted future actual PRB utilization rate and the actual number of RRC connected users. When the resulting expansion strategy does not match the actual overload results, a logistic regression model is established using sample data within a preset verification period, with the corresponding base station expansion risk index. The four sub-items are used as input variables, and the actual overload label is used as the target variable for fitting. The base station expansion risk index is then adjusted based on the fitting results. The four weighting coefficients , , The actual overload result includes overload and non-overload. If the actual PRB utilization rate exceeds the PRB expansion threshold or the actual number of RRC connected users exceeds the RRC expansion threshold, it is determined to be overloaded. If the actual PRB utilization rate does not exceed the PRB expansion threshold and the actual number of RRC connected users does not exceed the RRC expansion threshold, it is determined to be non-overloaded. If the resulting capacity expansion strategy does not match the actual overload results, the capacity expansion risk index is adjusted based on sample data within a preset verification period. Prediction confidence The correlation between overload prediction accuracy and prediction confidence. Adjustments will be made; the expansion risk index will be revised based on sample data within the preset verification period. Abnormal rating The correlation with actual overload results and the weighting of the abnormal score proportion. Adjustments are made; wherein, the overload prediction accuracy characterizes the consistency between the predicted overload state and the actual overload result; The contribution of each historical operating indicator to the prediction result output by the load prediction model is calculated based on a preset feature importance analysis algorithm. Indicators with a contribution value lower than a preset contribution value threshold are deleted, new indicators related to the prediction are added, and the load prediction model is retrained based on the adjusted indicators.

10. A base station expansion device, characterized in that, include: The load prediction module is configured to predict future load indicators of the base station based on the base station's historical operating indicators and through a trained load prediction model. The first calculation module is configured to calculate the base station expansion risk index based on the future load indicators of the base station. The second calculation module is configured to calculate the corrected expansion risk index based on the base station expansion risk index, the prediction confidence of the load prediction model, and the degree of anomaly of the historical operating indicators. The strategy generation module is configured to generate a corresponding expansion strategy based on the modified expansion risk index; The strategy execution module is configured to execute the expansion strategy.

11. A computer device, characterized in that, The system 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 base station expansion method according to any one of claims 1 to 9.

12. 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 base station expansion method according to any one of claims 1 to 9.

Images