Cell service load prediction method, device, equipment, storage medium and product

Abstract

The invention discloses a cell service load prediction method and device, equipment, a storage medium and a product, and relates to the technical field of wireless, and the method comprises the steps: obtaining a target cell, a historical index state and a prediction date; obtaining a target cell utilization rate prediction model according to the historical index state, and obtaining a cell utilization rate according to the target cell and the prediction date; fusing the cell utilization rate, the minimum drive test data and the measurement report optimization data features to obtain fused features; determining a target random tree model to predict the fusion feature load based on the utilization rate granularity, and obtaining the service load of the target cell; according to the method, the target cell utilization rate prediction model is selected, the cell utilization rate is predicted according to whether the data is a special date or not, the cell utilization rate and other related characteristics are fused under different granularities, the comprehensiveness and accuracy of the model are improved, different time scale requirements are met, rich data are provided for a network planner, refined decision making is assisted, and the network planning efficiency is improved. And the network operation quality is improved.

Description

Technical Field

[0001] The present application relates to the field of wireless technology, and in particular to methods, devices, equipment, storage media and products for predicting cell service load. Background Art

[0002] In mobile communication systems, the logical unit on the base station side that carries user services is called a cell. The service load status of a cell is directly related to the user's access experience, Internet speed, call quality and other perception dimensions. By analyzing the historical load data of the cell to predict the future load situation, it is possible to warn of high load and indicate low load, and then guide operators to adjust the network capacity in advance to achieve the goal of improving network load and enhancing user perception.

[0003] In mobile communication systems, a cell is a basic component of a wireless network. It consists of one or more base stations (BS) and is responsible for providing access services to users in a specific geographical area. Each cell has its coverage area, within which user equipment (UE) such as mobile phones can communicate with the base station to implement services such as voice calls and data transmission. With the popularization of mobile Internet and the growth of application demand, the service load status of the cell has become increasingly important because it is directly related to the user experience. Future load forecasting is a very important link in modern mobile communication systems and plays a key role in ensuring a good user experience and service quality. Summary of the invention

[0004] The main purpose of this application is to provide a cell service load prediction method, device, equipment, storage medium and product, aiming to solve the technical problem of inaccurate cell load prediction due to lack of real-time cell communication load data when predicting cell service load.

[0005] To achieve the above purpose, the present application proposes a method for predicting cell traffic load, the method comprising:

[0006] Obtaining a target cell, a historical indicator status of the target cell, and a predicted date;

[0007] Obtain a target cell utilization prediction model according to the historical indicator status, and obtain a cell utilization rate according to the target cell and the prediction date based on the target cell utilization prediction model;

[0008] Performing feature fusion on the cell utilization, the minimized drive test data, and the measurement report optimization data to obtain a fusion feature;

[0009] A target random tree model is determined based on the utilization granularity of the fused feature, and a load forecast is performed on the fused feature based on the target random tree model to obtain a service load of a target cell.

[0010] In one embodiment, the historical indicator status includes a normal status and a missing status;

[0011] The step of obtaining a target cell utilization prediction model according to the historical indicator status, and obtaining a cell utilization according to the target cell and the prediction date based on the target cell utilization prediction model, includes:

[0012] When the historical indicator state is normal, using a type of utilization prediction model as a target cell utilization prediction model;

[0013] When the historical indicator state is in a missing state, the second type of utilization prediction model is used as the target cell utilization prediction model;

[0014] The cell utilization is obtained according to the target cell and the predicted date based on the first type of utilization prediction model or the second type of utilization prediction model.

[0015] In one embodiment, obtaining the cell utilization rate according to the target cell and the predicted date based on the first type of utilization prediction model or the second type of utilization prediction model includes:

[0016] Obtaining a basic indicator according to the annual utilization rate of the target cell through the first type of utilization rate prediction model;

[0017] When the predicted date is a special date, the special date utilization rate and the target holiday utilization rate of the target cell are obtained by using the first type of utilization prediction model, the annual utilization rate increase ratio is obtained according to the special date utilization rate, and the cell utilization rate is obtained according to the annual utilization rate increase ratio and the target holiday utilization rate;

[0018] When the predicted date is not a special date, a plurality of reference utilizations of the same attribute are obtained according to the week attribute of the predicted date through the utilization prediction model, and an annual reference utilization and a monthly reference utilization are obtained according to the plurality of reference utilizations; and the cell utilization is obtained according to the annual reference utilization, the monthly reference utilization and the basic indicators.

[0019] In one embodiment, obtaining the cell utilization rate according to the target cell and the predicted date based on the first type of utilization prediction model or the second type of utilization prediction model further includes:

[0020] Determine each subsequent coverage cell according to the target cell by using the second type utilization prediction model, and obtain the basic index of the target cell according to the annual utilization rate of each subsequent coverage cell;

[0021] When the predicted date is a special date, the distance allocation weight and the target holiday utilization rate are obtained according to the distance between each of the subsequent coverage cells and the target cell through the two-type utilization prediction model, and the cell utilization rate of the target cell is obtained according to the distance allocation weight, the target holiday utilization rate and the annual utilization rate of each of the subsequent coverage cells;

[0022] When the predicted date is not a special date, the second-class utilization prediction model is used to obtain multiple reference utilizations of the same attribute according to the week attribute of the predicted date, and the annual reference utilization and the monthly reference utilization are obtained according to the multiple reference utilizations; the cell utilization is obtained according to the annual reference utilization, the monthly reference utilization and the basic indicators.

[0023] In one embodiment, the fusion features include day granularity fusion features and hour granularity fusion features;

[0024] The step of fusing the cell utilization rate, the minimization of drive test data, and the measurement report optimization data to obtain a fusion feature includes:

[0025] Divide the cell utilization rate, the minimized drive test data, and the measurement report optimization data by hour to obtain an hourly granularity fusion feature, where the hourly granularity fusion feature includes the hourly granularity minimized drive test data, the hourly granularity cell utilization rate, and the hourly granularity measurement report optimization data;

[0026] The cell utilization rate, the minimized drive test data and the measurement report optimization data are divided by day to obtain a daily granularity fusion feature, wherein the daily granularity fusion feature includes daily granularity minimized drive test data, daily granularity cell utilization rate and daily granularity measurement report optimization data.

[0027] In one embodiment, the target random tree model includes a day granularity random tree model and an hour granularity random tree model;

[0028] The determining a target random tree model based on the utilization granularity of the fusion feature, and performing load prediction on the fusion feature based on the target random tree model to obtain a service load of a target cell includes:

[0029] When the utilization rate granularity is a day granularity, the day granularity random tree model is used as a target random tree model;

[0030] When the utilization rate granularity is an hourly granularity, the hourly granularity random tree model is used as a target random tree model;

[0031] Performing load prediction on the fusion feature through each random tree of the target random tree model to obtain a plurality of initial prediction results, wherein the initial prediction results include a first load state and a second load state;

[0032] Performing statistics on the multiple initial prediction results to obtain a first load state ratio and a second load state ratio;

[0033] The business load of the target cell is obtained by comparing the first load state proportion and the second load state proportion.

[0034] In addition, to achieve the above purpose, the present application also proposes a cell traffic load prediction device, the cell traffic load prediction device comprising:

[0035] An acquisition module, used to acquire a target cell, a historical indicator status of the target cell, and a predicted date;

[0036] A utilization prediction module, used to obtain a target cell utilization prediction model according to the historical indicator status, and obtain a cell utilization according to the target cell and the prediction date based on the target cell utilization prediction model;

[0037] A service load prediction module, used for fusing the cell utilization rate, the minimized drive test data, and the measurement report optimization data to obtain a fusion feature;

[0038] The service load prediction module is further used to determine a target random tree model based on the utilization granularity of the fused feature, and perform load prediction on the fused feature based on the target random tree model to obtain the service load of the target cell.

[0039] In addition, to achieve the above-mentioned purpose, the present application also proposes a cell traffic load prediction device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, and the computer program is configured to implement the steps of the cell traffic load prediction method as described above.

[0040] In addition, to achieve the above-mentioned purpose, the present application also proposes a storage medium, which is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the cell service load prediction method as described above are implemented.

[0041] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, the steps of the cell service load prediction method as described above are implemented.

[0042] One or more technical solutions proposed in the present application have at least the following technical effects: selecting a target cell utilization prediction model through historical indicator status and utilization granularity, predicting more accurate cell utilization from special dates and non-special dates according to the target cell utilization prediction model, being able to cope with daily business volume fluctuations and drastic changes in business volume during special periods, performing feature fusion of cell utilization and other related features at different granularities, improving the comprehensiveness and accuracy of the prediction model, enhancing the reliability of the prediction results, performing load forecasting at different granularities based on feature fusion, meeting different time scale requirements, providing rich data for network planners, facilitating refined decision-making, and improving network operation quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0044] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0045] Figure 1 A flowchart of the first embodiment of the method for predicting the traffic load of a cell in the present application is provided;

[0046] Figure 2 A schematic diagram of the overall structure of the cell service load prediction method in practical application provided in Embodiment 1 of the cell service load prediction method of the present application;

[0047] Figure 3 A flowchart of the second embodiment of the method for predicting the traffic load of a cell in the present application is provided;

[0048] Figure 4 A first utilization rate prediction model logic diagram provided for the second embodiment of the cell traffic load prediction method of the present application;

[0049] Figure 5 A third utilization rate prediction model logic diagram provided for the second embodiment of the method for predicting the traffic load of a cell in the present application;

[0050] Figure 6 A second utilization rate prediction model logic diagram provided for the second embodiment of the cell traffic load prediction method of the present application;

[0051] Figure 7 A fourth utilization rate prediction model logic diagram provided for the second embodiment of the method for predicting the traffic load of a cell of the present application;

[0052] Figure 8 This is a schematic diagram of the module structure of the cell traffic load prediction device according to an embodiment of the present application;

[0053] Fig. 9 This is a schematic diagram of the device structure of the hardware operating environment involved in the cell service load prediction method in the embodiment of the present application.

[0054] The purpose, features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0055] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.

[0056] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0057] The main solution of the embodiment of the present application is: obtaining the target cell, the historical indicator status of the target cell and the predicted date; obtaining the target cell utilization prediction model according to the historical indicator status, and obtaining the cell utilization according to the target cell and the predicted date based on the target cell utilization prediction model; performing feature fusion on the cell utilization, minimized drive test data and measurement report optimization data to obtain a fused feature; determining the target random tree model based on the utilization granularity of the fused feature, and performing load prediction on the fused feature based on the target random tree model to obtain the service load of the target cell.

[0058] As existing technologies advance with big data technology and machine learning algorithms, operators can identify patterns and trends by analyzing historical cell load data, thereby predicting future load conditions, which helps to plan and adjust network capacity in advance, such as dynamically allocating spectrum resources, optimizing antenna configurations, or launching new cells to divert users. The ultimate goal is to improve user experience and ensure high-quality services. Through effective management and optimization of cell loads, Internet access speeds can be increased, call quality can be guaranteed to be stable, and other aspects of perception can be enhanced, such as reducing waiting time and reducing drop rates.

[0059] The present application provides a solution, which includes obtaining a target cell, a historical indicator status, and a predicted date; obtaining a target cell utilization prediction model according to the historical indicator status, and obtaining a cell utilization according to the target cell and the predicted date; fusing the cell utilization with minimized drive test data and measurement report optimization data features to obtain a fusion feature; determining a target random tree model based on the utilization granularity to predict the fusion feature load, and obtaining the service load of the target cell; the method selects a target cell utilization prediction model, predicts the cell utilization based on whether it is a special date, and fuses the cell utilization and other related features at different granularities, thereby improving the comprehensiveness and accuracy of the model, meeting the requirements of different time scales, providing network planners with rich data, facilitating refined decision-making, and improving network operation quality.

[0060] It should be noted that the execution subject of this embodiment may be a computing service device with data processing, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or an electronic device capable of realizing the above functions, a cell service load prediction device, etc. The following takes the cell service load prediction device as an example to illustrate this embodiment and the following embodiments.

[0061] Based on this, the embodiment of the present application provides a method for predicting cell service load, referring to Figure 1 , Figure 1 This is a flowchart of the first embodiment of the cell traffic load prediction method of the present application.

[0062] In this embodiment, the cell service load prediction method includes steps S10 to S40:

[0063] Step S10: Acquire the target cell, the historical indicator status of the target cell, and the predicted date.

[0064] It is understandable that the target cell and the predicted date may be pre-selected or input by the user; the historical indicator status of the target cell may be retrieved from the storage space storing the historical indicators of each cell after the target cell is determined.

[0065] It should be noted that the historical indicator status includes normal historical indicators and missing historical indicators.

[0066] It should be noted that business load refers to the amount of data or number of transactions processed by a cell per unit time. When the business load is too high, it may cause network congestion, manifested as increased connection failure rate, slower download speed, video playback freezes and other problems; on the contrary, if the business load is too low, it may mean that network resources are not fully utilized, resulting in waste.

[0067] Step S20: obtaining a target cell utilization prediction model according to the historical indicator status, and obtaining a cell utilization according to the target cell and the prediction date based on the target cell utilization prediction model.

[0068] It should be understood that each model may be pre-set on how to perform cell utilization prediction based on the prediction date and the target cell to be predicted.

[0069] It should be noted that the cell utilization prediction model can be composed of two types of models, one type can be a cell utilization prediction model with a normal historical indicator status, and the other type can be a cell utilization prediction model with a missing historical indicator status.

[0070] Furthermore, the two types of prediction models are further divided into two prediction models according to hourly granularity and daily granularity, which can be simply understood as having four prediction models for cell utilization prediction.

[0071] It is understandable that the target cell utilization prediction model can be selected from two types of models according to the historical indicator status, and the specific cell utilization prediction model can be further determined according to the granularity. The cell utilization can be the cell utilization of the target cell to be predicted.

[0072] Step S30: performing feature fusion on the cell utilization, the minimization of drive test data, and the measurement report optimization data to obtain fusion features.

[0073] It is understandable that the minimization of drive test data and the measurement report optimization data are relevant data of the target cell.

[0074] It should be noted that the historical utilization index of the cell, 5G cell engineering parameters, MRO source data, and MDT source data can be obtained through Figure 2 The underlying server module of the wireless communication network is collected through multiple channels.

[0075] It should be emphasized that the collected minimum drive test data and measurement report optimization data need to be further cleaned to eliminate noise and errors in the data; data cleaning can be done through Figure 2 Data cleaning and feature engineering are implemented in the data. Data cleaning can specifically include missing value processing, outlier detection, duplicate record identification and consistency checking.

[0076] It is further explained that Minimization of Drive Tests (MDT) can automatically collect network performance data without human intervention. Its sampling points cover a wide and comprehensive range, and the indicators involved are rich and diverse, covering but not limited to signal strength, signal quality, channel quality indicator (CQI), reference signal received power (RSRP), reference signal received quality (RSRQ), etc. Through these data, we can have a deep insight into the actual operation of the network. When predicting the future load of the cell, in addition to the cell utilization rate, feature engineering results based on MDT data are also introduced to assist in the prediction. The collected MDT sampling point data have varying degrees of missing or unreasonable data. Therefore, after completing the data collection, data cleaning must be carried out.

[0077] The cleaning rules for the collected minimization of drive test data can be:

[0078] First, duplicate record identification. To ensure the accuracy of MDT data analysis, it is crucial to eliminate the impact of duplicate records. A secure hashing technology based on SHA-256 is adopted, which has strong anti-collision properties and can generate hash identifiers with high distinctiveness. Specifically, by concatenating all field values ​​of MDT in a fixed field order and applying the SHA-256 hash function, the data is mapped to a unique 256-bit hash code. This can greatly reduce the risk of hash collisions, ensure the accuracy and completeness of the data set, and thus improve the overall data quality and the reliability of the analysis results. Define the MDT original data set D, each MDT data r∈D has constituent elements (v1, v2, ···, v n ), where v n Indicates the value of the nth field. Define two records r i =(v i1 , v i2 ,···,v in ) and r j =(v j1 , v j2 ,···,v jn ) is repeated if and only if for all 1≤k≤n, there is v ik =v jk Create an empty non-duplicate dataset D', and for each record r in the dataset D, calculate the hash value h(r). If there is no record r' in D' such that h(r') = h(r), then add r to D', and the resulting D' is the MDT dataset with duplicate records removed.

[0079] Second, sampling points that lack basic cell information are removed. The cell basic information set B includes key information such as cell ID, cell name, and geographic location. In the original MDT sample data, records that lack this basic information cannot be accurately mapped to specific cells, thus losing their effective value and constituting data noise. In order to purify the data set and maintain the effectiveness of the analysis, we will implement a preprocessing step to specifically filter out such entries that lack key information. Assume that D represents the sampling point data set, x i The data vector used to represent the i-th sampling point, B = {b1, b2, ···, b n} is a collection of basic information of the cell. Define a function f B (x i ) to determine whether the sampling point contains all the basic information. The function is as follows:

[0080]

[0081] The data cleaning operation W1 is expressed as: W1(D)={x i ∈D|f B (x i )=1}.

[0082] Third, the sampling points with missing key indicator values ​​are removed. First, a set of key indicators including the longitude, latitude, RSRP, RSRQ, SINR, etc. of the sampling points is constructed, K = {k1, k2, ···, k m}. These indicators are crucial for subsequent analysis and prediction. Define a function f x (x i ) to determine whether the sampling point contains all key indicator values. The function is as follows:

[0083]

[0084] The data cleaning operation W2 is expressed as: W1(D)={x i ∈D|f x (x i )=1}.

[0085] Fourth, the sampling points whose key indicator values ​​are beyond the reasonable range are removed. j ∈K, determine a reasonable value range R based on relevant standards and experience j =[a j , b j ], for example, the RSRP value should be in the range of [-140dBm, -47dBm], and the SINR value should be in the range of [-10dB, 40dB]. Define a function f Rj (x i (k j)) to determine whether the value of the indicator at the sampling point is within a reasonable range. The function is as follows:

[0086]

[0087] The data cleaning operation W3 is expressed as: W1(D)={x i ∈D|f Rj (x i (k j ))=1}.

[0088] It should be further emphasized that the MRO data in the measurement report optimization (MRO) data cleaning and feature engineering contains coverage indicators such as RSRP and RSRQ. These indicators can intuitively reflect the signal strength, signal quality and coverage of the cell, and are the key basis for evaluating network quality. For example, RSRP stands for reference signal received power, which reflects the signal strength received by the mobile phone. Good coverage is crucial to ensure the stability of network performance and improve user experience. Therefore, when predicting the future load of the cell, in addition to the cell utilization rate, these characteristic indicators of the MRO data are also incorporated to assist in the prediction.

[0089] Among them, the original MRO data collected has different degrees of missing or unreasonable data. Therefore, after the data collection is completed, it is necessary to perform data cleaning and feature engineering to improve data quality and availability. The data cleaning method is consistent with that of MDT data.

[0090] In a feasible implementation, step S30 may include steps A31 to A32:

[0091] Step A31, divide the cell utilization, the minimized drive test data and the measurement report optimization data by hour to obtain hourly granularity fusion features, wherein the hourly granularity fusion features include hourly granularity minimized drive test data, hourly granularity cell utilization and hourly granularity measurement report optimization data.

[0092] It should be noted that the hourly granularity minimization drive test data includes basic cell information, signal reception power and coverage index parameters every hour; the hourly granularity measurement report optimization data includes handover statistics, handover parameters, path loss and interference information every hour.

[0093] It should be noted that in order to comprehensively evaluate the cell status and provide accurate data input for subsequent cell traffic load prediction, the focus is on the fusion of predicted cell utilization, MDT and MRO data. Considering that random tree prediction models with hourly and daily granularity will be constructed in the future, and the two granularity models use different data granularity when constructing feature sets, data fusion work needs to be carried out for these two granularities respectively.

[0094] First, hourly granularity data integration. Based on the timestamp and cell location information, the predicted cell utilization is accurately matched with the MDT and MRO data. In this process, all indicator data involved in the MDT and MRO data will be integrated to the hourly granularity level.

[0095] For the hourly granularity data fusion of MDT data: the basic information of the cell is directly matched by hour, and the basic information such as the identification and type of the cell remains unchanged within each hour. RSRP (reference signal received power): the average RSRP of all sampling points within each hour is calculated as the RSRP value of the hour, which reflects the signal receiving power level of the cell within the hour. Other coverage indicators (such as SINR, etc.): the sampling data within each hour is averaged and integrated into the indicator value of hourly granularity.

[0096] For the hourly granularity data fusion of MRO data: Coverage indicators (such as RSRP): The processing method is the same as the corresponding indicators in the MDT data, and the average value is calculated by hour, and these coverage indicators are integrated into the hourly granularity to fully present the coverage characteristics of each hour's cell. A complete and interrelated set of data will be obtained every hour, which accurately reflects the relevant characteristics of the cell's operating status within that hour, providing detailed and timely data support for the construction of the hourly granularity random tree model.

[0097] It should be emphasized that there are some challenges in the process of hourly granular data fusion:

[0098] First, the timestamp of the data may have certain errors or inconsistencies. The time calibration of different data sources may be biased, which may cause misalignment when matching data by hour, affecting the accuracy of the data. To solve this problem, we adopted a time synchronization calibration mechanism. By introducing a high-precision time server, the timestamps of different data sources are calibrated and aligned. At the same time, a timestamp difference detection algorithm is established to mark and correct data points whose deviation exceeds a certain threshold to ensure accurate matching of data in the time dimension.

[0099] Secondly, the difference in data sampling frequency is also a problem. The sampling frequencies of MDT and MRO data may be different. When integrating them to the hourly granularity, it is necessary to solve the problem of how to reasonably handle data of different frequencies to avoid information loss or data distortion. To this end, we used interpolation algorithms and data resampling techniques. For data with a lower sampling frequency, according to its data characteristics and the relationship between adjacent data points, use appropriate interpolation algorithms (such as linear interpolation, spline interpolation, etc.) to fill the data so that it is aligned with other data on the time scale. At the same time, for high-frequency data, the data resampling technology is used to reduce its frequency to the hourly granularity to ensure the consistency of data from different sources when integrated.

[0100] In addition, the huge amount of data also puts pressure on computing resources and processing efficiency. When calculating the average value of indicators every hour, a large amount of computing resources and time may be required. To meet this challenge, we use a distributed computing framework and data compression technology. The distributed computing framework is used to decompose data processing tasks to multiple computing nodes for parallel execution, which greatly improves computing efficiency. At the same time, targeted data compression is performed on the original data to remove redundant information in the data, reduce the burden of data storage and processing, and improve processing speed while ensuring data quality.

[0101] Step A32, divide the cell utilization, the minimized drive test data and the measurement report optimization data by day to obtain a daily granularity fusion feature, which includes daily granularity minimized drive test data, daily granularity cell utilization and daily granularity measurement report optimization data.

[0102] It should be noted that the daily granularity minimized drive test data includes daily basic cell information, signal reception power and coverage index parameters; the daily granularity measurement report optimization data includes daily switching statistics, switching parameters, path loss and interference information.

[0103] It is understandable that the principle is basically the same as "hourly granularity data fusion". However, the difference is that the indicator data involved in MDT and MRO data needs to be integrated to the daily granularity.

[0104] For data fusion of MDT data: the basic information of the cell is in units of days, and the basic information such as the cell identification and type remain unchanged, representing the basic attributes of the cell on that day. RSRP (reference signal received power): calculate the average RSRP value of all hours in a day, and use it as the RSRP value of the day, reflecting the overall level of signal received power of the cell on that day from a macro perspective. Other coverage indicators (such as SINR, etc.): average the sampled data of each hour in a day to form an indicator value with daily granularity, which comprehensively represents the coverage-related characteristics of the cell on that day.

[0105] For data fusion of MRO data: Coverage indicators (such as RSRP): The processing method is the same as the corresponding indicators in MDT data. The average value is calculated on a daily basis and these coverage indicators are integrated into the daily granularity to comprehensively present the coverage characteristics of the cell every day.

[0106] It should be emphasized that there are also some problems to be overcome in the fusion of daily granularity data: on the one hand, due to the integration of a whole day's data, some abnormal situations or important information in a short period of time may be concealed. For example, serious signal interference may occur in a certain hour of a day, but it is smoothed out in the calculation of the average value of the daily granularity, which may affect the accuracy of subsequent model predictions. In order to solve this problem, we have added additional abnormal data marking and statistical features to the daily granularity data. During the data integration process, anomaly detection is performed on the data of each hour. When the data of a certain hour deviates from the normal range (determined by historical data and statistical models), the anomaly is marked in the daily granularity data. At the same time, in addition to the average value, some statistical features are also calculated, such as standard deviation, maximum value, minimum value, etc. These features can reflect the fluctuation of the data of the day to a certain extent, retain some abnormal information in a short period of time, and provide richer information for subsequent models.

[0107] On the other hand, there may be large changes in data features between different days, such as differences in usage patterns between communities on weekdays and weekends, which increases the complexity of data integration and analysis, and requires more sophisticated processing methods to ensure that day-granularity data can accurately reflect the overall status of the community. In response to this situation, we introduced classification features to distinguish different types of days (such as weekdays, weekends, holidays, etc.). In the data integration process, these classification features are integrated into day-granularity data together with other indicator data. In addition, data models are established according to different types of days, and their respective data characteristics and change patterns are analyzed. In subsequent model training and prediction, the appropriate model can be selected according to the type of day or the model can be adjusted accordingly to better adapt to the data differences between different days.

[0108] In this implementation, by performing data fusion operations on two granularities, a solid foundation is laid for the subsequent construction of a random tree model, creating favorable conditions for accurately predicting cell load conditions, so that the model can make full use of data features of different granularities and more accurately capture the laws of cell load changes.

[0109] The above are only feasible implementations of step S30 provided in this embodiment, and this embodiment does not specifically limit the specific implementation of step S30.

[0110] Step S40: determining a target random tree model based on the utilization granularity of the fused feature, and performing load prediction on the fused feature based on the target random tree model to obtain a service load of a target cell.

[0111] It is understandable that the utilization granularity of the fused features may be a day granularity or an hour granularity.

[0112] It should be noted that in the process of final cell service load prediction, an advanced “random forest” algorithm is introduced to construct a model for cell load prediction.

[0113] It is worth noting that random forest integrates multiple decision trees, which can capture the nonlinear and interactive relationships between factors related to cell load. Cell load is affected by many factors, such as utilization, special dates, signal coverage and quality, etc. The model can more accurately reflect the actual load status and improve prediction accuracy. The model remains stable and enhances its coping capabilities when facing complex and uncertain cell service data (such as noise, outliers, and changes in data distribution). The cell service data has high dimensions, and the hourly granularity features in the present invention reach 15. Random forest does not require complex dimensionality reduction, can mine high-dimensional features and load-related information, avoid information loss, and provide rich basis for prediction. Analyzing decision trees can understand the impact of factors on load, which helps network operators understand the causes of load changes, optimize resource allocation, and formulate strategies.

[0114] It should be understood that the prediction of cell utilization takes into account different time granularities, specifically hourly granularity and daily granularity. Based on this, we construct two types of random tree models for predicting cell load conditions, namely hourly granularity random tree model and daily granularity random tree model.

[0115] It should be emphasized that the random tree model can be trained and optimized before load forecasting is performed through the random tree model. The construction of the random tree model covers seven key steps: first, comprehensive and accurate data preparation is carried out to lay the data foundation for subsequent processes; then a feature set closely related to the cell load forecast is constructed to fully mine the effective information in the data; then the constructed features are processed in a targeted manner to ensure that the data format meets the model requirements; then the high-load label definition is completed based on the actual operation of the cell and the load characteristics, and the model prediction target is clarified; on this basis, the model construction work is carried out to determine the model structure and parameters; then the constructed model is comprehensively evaluated to test its performance, and finally the model quality is further improved through the model optimization step to ensure the accuracy and reliability of the cell load forecast.

[0116] It should be emphasized that the construction process of random tree models with hourly granularity and daily granularity is similar. However, when constructing a random tree model with daily granularity to predict cell load, some steps can refer to the method of constructing the hourly granularity model, and some steps need to be adjusted according to the characteristics of daily granularity. The following takes the construction of a random tree with hourly granularity as an example. The construction steps of a random tree with hourly granularity are as follows:

[0117] Step 1: Data preparation. Based on the fused feature data at hourly granularity, in order to ensure the training effect of the model, the prepared data is divided into training set P train and the test set P test Considering that the amount of data is relatively sufficient, the data ratio of the training set and the test set is 70:30, that is, 70% of the data is used for model training and 30% of the data is used for testing. The data ratio division can provide enough data for the model to learn, and at the same time ensure that there is a certain regular test set to evaluate the generalization ability of the model. Assuming that the total number of samples in the data set is N, the number of samples in the training set is N train =N*P train , the number of test set samples N test =N*P test .

[0118] Step 2: Feature set construction. Feature set construction is the key foundation of the cell load forecasting model. It can provide the model with comprehensive and detailed data dimensions to capture the cell load pattern and is an important basis for mining the relationship between features and forecasting targets. This system construction method ensures the accuracy of model training and forecasting information, improves forecasting accuracy and reliability, and the hourly granularity feature set diagram can be referred to Table 1 and Table 2 below:

[0119] Table 1

[0120]

[0121] Table 2

[0122]

[0123] Step 3: Feature processing. In order to enable the model to learn the relationship between features and targets more effectively, the constructed features need to be processed. In the present invention, it is divided into numerical feature standardization and classification feature encoding, as follows:

[0124] First, standardize the numerical features. For all numerical features (the features with "serial numbers" 1 to 11 in the hourly granularity feature set diagram are all numerical features), use the Z-score standardization method to standardize the features. That is, for each numerical feature, the calculation of the mean μ and standard deviation σ can refer to the following formula:

[0125]

[0126]

[0127] Where n is the number of samples, x i is the eigenvalue of the i-th sample.

[0128] For each sample’s eigenvalue x, perform a standardized transformation:

[0129] x'=(x-μ) / σ

[0130] Secondly, the classification features are encoded. For classification features (the features of "serial numbers" 12 to 15 in the hourly granularity feature set diagram of the present invention are all classification features), the One-Hot Encoding method is used. Taking Is_Special_Date as an example, if its original value is 0 (indicating a non-special date) and 1 (indicating a special date), after one-hot encoding, it will become a two-dimensional vector, such as [1,0] for a non-special date, and [0,1] for a special date. For classification features with more category values, encoding is performed in the same way.

[0131] Step 4: High load label definition. High load label definition is a key step to clarify the model prediction target. When the cell is in a high load state, it is marked as 1, otherwise it is 0. The specific judgment conditions are as follows:

[0132] 1. High utilization leads to high load.

[0133] When Utilization_predicted>70%, the cell is in a high-load state. High utilization directly indicates that the cell resources are heavily occupied and the existing resources are under great pressure.

[0134] 2. Special dates and coverage issues together lead to high loads.

[0135] When Utilization_predicted>60% and Is_Special_Date=1 and Is_Weak_Coverage=1, the weak coverage on special dates will increase the cell load. User activities increase on special dates, and weak coverage will lead to increased power consumption of user equipment and increased number of retransmissions, increasing the burden on the cell.

[0136] When Utilization_predicted>60% and Is_Special_Date=1 and Is_Overlap_Coverage=1, the overlapping coverage problem on special dates will cause the cell to be under high load. Overlapping coverage is prone to cause signal interference. When the traffic volume on special dates increases, the interference will further affect the communication quality and resource utilization efficiency, and increase the cell load.

[0137] 3. Special dates and signal quality issues together lead to high load.

[0138] When Utilization_predicted>60% and Is_Special_Date=1 and RSRP_std>5dBm, the larger RSRP standard deviation in the special date means that the signal strength fluctuates violently, which will cause the user equipment to frequently adjust the receiving parameters and increase retransmissions, which will increase the cell resource consumption and load.

[0139] When Utilization_predicted>60% and Is_Special_Date=1 and SINR_std>3dB, the larger SINR standard deviation in the special date indicates that the signal interference is serious, which will reduce the communication quality and data transmission efficiency, and the cell needs to consume more resources to maintain communication, thus being in a high load state.

[0140] 4. Changes in signal quality lead to high load.

[0141] When Utilization_predicted>60% and RSRP_change_rate>0.2, the signal strength change rate is high, indicating that the signal environment in the cell is unstable. This instability will cause frequent cell resource allocation, increase additional overhead, and lead to high load.

[0142] When Utilization_predicted>60% and RSRP_stability_index>3dBm, RSRP stability is poor, which means that the signal fluctuates frequently. This will cause problems such as large changes in signal strength, such as frequent adjustments of user equipment and frequent allocation of resources, which will increase the cell load.

[0143] Step 5: Model Building

[0144] 1. Initialize the random tree model. Select the random tree algorithm as the basis of the prediction model. Initialize the random tree model object and set the key parameters as follows:

[0145] The number of trees (n e stimators): Set to 100. This parameter determines the number of decision trees in the model. A larger number of trees can improve the stability and generalization ability of the model, but it will also increase the computational cost. The choice of 100 as the initial value is based on experience and the balance between computing resources and model performance.

[0146] The number of features that each tree considers when splitting a node (max f eatures): Set to (P) 1 / 2, where p is the total number of features after processing, which is 15 in the present invention. This setting is a common practice method, which avoids overfitting while ensuring that each decision tree has sufficient feature selection diversity.

[0147] The maximum depth of the tree (max d epth): Set to 10. The depth of the tree limits the complexity of the decision tree. Shallow trees may not be able to fully learn complex relationships in the data, while too deep trees are prone to overfitting. By setting it to 10, the complexity and generalization ability of the model can be controlled to a certain extent.

[0148] The minimum number of samples required for internal node repartitioning (min s amples a plit): Set to 5. This parameter specifies the minimum number of samples required to split an internal node of the decision tree. Smaller values ​​may make the tree too complex and overfit, while larger values ​​may make the tree too simple and not fit the data well.

[0149] Other parameters, such as random-state, are set to specific values ​​(such as 42) to ensure the reproducibility of model training results.

[0150] 2. Model training and voting mechanism. Use the processed training data set to train the random tree model. The training process is an iterative process. For each decision tree: randomly extract samples with replacement from the training data set (this is a source of randomness in the random forest) and construct a sub-dataset.

[0151] When constructing each node of each tree, randomly select the max from all features. f eatures features, and then calculates the best split point based on these features and sub-datasets (such as splitting criteria based on information gain, Gini impurity, etc.), and splits the node into two sub-nodes.

[0152] Repeat the above steps until the maximum depth (max_depth) is reached or the number of samples in the node is less than min_samples_split.

[0153] Through multiple iterations, n e stimators to form a random tree model. During the training process, the model automatically learns the complex relationship between features and target variables (whether the cell is in a high-load state), which is reflected in the structure of the decision tree and the node splitting conditions.

[0154] Step 6: Model evaluation.

[0155] 1. Prediction and evaluation index calculation. Input the preprocessed test data set into the trained random tree model to obtain the model's prediction results for each test sample. Then calculate various evaluation indicators:

[0156] Accuracy: Accuracy reflects the proportion of samples that are correctly predicted by the model. The calculation of accuracy can refer to the following formula:

[0157] Accuracy=(TP+TN) / (TP+TN+FP+FN)

[0158] Among them, TP (True Positive) is the number of true positive examples, that is, the number of samples predicted by the model to be high-loaded and actually are high-loaded; TN (True Negative) is the number of true negative examples, that is, the number of samples predicted by the model to be non-high-loaded and actually are not high-loaded; FP (False Positive) is the number of false positive examples, that is, the number of samples predicted by the model to be high-loaded but actually are not high-loaded; FN (False Negative) is the number of false negative examples, that is, the number of samples predicted by the model to be non-high-loaded but actually are high-loaded.

[0159] Precision: Precision measures the proportion of samples predicted by the model to be high-load that are actually high-load, focusing on the accuracy of the model's predictions. The calculation of precision is based on the following formula:

[0160] Precision=TP / (TP+FP)

[0161] Recall: The recall rate indicates the proportion of samples that are actually high-load samples that are correctly predicted by the model as high-load samples. It focuses on the model's ability to capture positive examples. The calculation of the recall rate can refer to the following formula:

[0162] Recall=TP / (TP+FN)

[0163] F1-score: The F1-score is the harmonic mean of precision and recall, which takes into account the accuracy and recall of the model. The specific calculation can be referred to the following formula:

[0164] F1-score=2*(Precision*Recall) / (Precision+Recall)

[0165] Among them, Precision is the precision rate and Recall is the recall rate.

[0166] 2. Confusion matrix drawing and analysis. When evaluating the performance of the random tree model proposed in this invention to predict the cell load situation, the confusion matrix is ​​an important tool. The confusion matrix is ​​a matrix structure used to visualize the model prediction results and the actual situation. Its dimension depends on the number of categories of the prediction target. In the cell load prediction scenario, it mainly involves two categories: high load and non-high load, so it is a 2×2 matrix. The following is an explanation of the cell load prediction results:

[0167] True Positive (TP): In the cell load prediction scenario, it represents the number of samples that the model accurately predicts as being in a high-load state and the actual cell is also in a high-load state. This value reflects the model's ability to correctly identify high-load cells. For example, in actual applications, the situation where a large number of high-load cells are correctly marked by the model will be reflected by the TP value. The higher the value, the more accurate the model is in capturing high-load situations.

[0168] True Negative (TN): refers to the number of samples that the model correctly predicts to be in a non-high-load state and the actual cell is also in a non-high-load state. It reflects the model's ability to accurately judge non-high-load cells, which is important for avoiding unnecessary resource allocation or false alarms. For example, accurately identifying cells that are operating normally and have low loads helps maintain stable network management.

[0169] False Positive (FP): indicates the number of cells that the model incorrectly predicts to be high-load cells when they are not actually high-load cells. A high FP value may lead to unnecessary allocation of resources. For example, network operators may perform unnecessary resource expansion or optimization operations on cells due to such misjudgment of the model, thereby increasing costs.

[0170] False Negative (FN): The model incorrectly predicts the number of cells with actual high load as non-high load cells. This may cause the cells to fail to adjust or optimize resources in a timely manner under high load conditions, thus affecting user experience, such as network congestion and reduced service quality.

[0171] Through the confusion matrix, a variety of evaluation indicators can be calculated to measure model performance. For example, Accuracy reflects the proportion of samples predicted correctly by the model in the total samples, and is an indicator for comprehensively evaluating the overall performance of the model. Precision focuses on the proportion of samples predicted by the model as high-load that are truly high-loaded, and can be used to evaluate the accuracy of the model's high-load predictions. Recall indicates the proportion of actual high-load cells correctly predicted by the model, which can reflect the model's ability to capture high-load situations. The F1 value (F1-score) is the harmonic mean of precision and recall, which more comprehensively evaluates the balance of model performance in different aspects. These indicators will be used to deeply analyze model performance when actual prediction data is available in the future.

[0172] Step 7: Model optimization.

[0173] 1. Hyperparameter adjustment. In the process of building a random tree model to predict cell load, the hyperparameter selection of the model has a crucial impact on its performance. Hyperparameters determine the structure and learning ability of the random tree model. Different hyperparameter settings may cause the model to show different accuracy and generalization ability when predicting cell load.

[0174] The present invention will use a suitable hyperparameter adjustment method to optimize the model. For example, a grid search or a randomized search method may be considered. Taking grid search as an example, a reasonable search range for key hyperparameters will be determined. For the number of trees (n e stimators), and preliminarily set its search range based on relevant field experience and similar model applications. Because too few trees may not be able to fully learn the complex relationship between cell load-related characteristics and prediction targets, and too many trees may lead to overfitting and increase computing resource consumption, the search range can be set to a suitable interval (such as [50,200]). d epth), too shallow a depth will make it difficult for the model to capture the deep relationships in the data, while too deep a depth will easily make the model too complex and overfitting, so the search can be conducted in the range of [3,15]. The minimum number of samples required for internal node re-division (min s amples a plit) also affects the growth and complexity of the tree, and its search range can be set to [2,10]. After determining the search range, different hyperparameter combinations are generated with a certain step size. For each combination, a suitable cross-validation method (such as k-fold cross-validation, k can be selected according to the data scale and computing resources, such as 5 or 10) is used to train the model on the training data set, and the model performance is evaluated. Finally, the hyperparameter combination with the best performance is selected.

[0175] 2. Feature selection and optimization. In cell load prediction, feature selection and optimization are key links to improve model performance. The present invention will use the property of the random tree model itself to evaluate feature importance. Feature importance reflects the contribution of each feature to the model's prediction of cell load results. By accurately evaluating feature importance, key features that have a significant impact on model predictions can be identified, as well as features that may interfere with the model or contribute less.

[0176] Based on the feature importance evaluation results, the present invention will adopt a step-by-step selection method. First, all features are sorted according to the feature importance score. For features with lower scores, consider deleting them step by step. After deleting one or a group of low-importance features each time, retrain the model and use appropriate evaluation indicators (such as the accuracy rate, F1 value, etc. mentioned above, calculated when there is actual data) to evaluate the model performance on the validation set. If the model performance does not decrease significantly (judged by setting a reasonable performance change threshold), continue to delete other low-importance features. On the contrary, if the model performance is significantly reduced after deleting a certain feature, retain the feature. In addition, the possibility of combining or transforming features will be explored. For example, for features related to cell signals (such as RSRP_mean and RSRP_std), you can try to create new features (such as the ratio or difference between RSRP_mean and RSRP_std, etc.), add the new features to the model, retrain and evaluate, and observe whether the model can improve the prediction ability of the cell load, thereby optimizing the model structure and improving its prediction performance and generalization ability.

[0177] Finally, it should be emphasized that when constructing a daily granularity random tree model to predict cell load, some steps can refer to the method of constructing an hourly granularity model, and some steps need to be adjusted according to the characteristics of daily granularity. The steps that can refer to the hourly granularity include:

[0178] In the data preparation step (step 1), the idea of ​​data collection and preliminary cleaning can be used, but the data integration needs to be converted from hourly level to daily level;

[0179] The feature processing step (step 3) is basically the same for numerical and categorical features (such as standardization and one-hot encoding).

[0180] Model building step (step 5), the basic process of model initialization and training and the concept of parameter setting can refer to the hourly granularity, including random tree algorithm selection, model parameter meaning, sample extraction in training, and node splitting criteria;

[0181] The model evaluation step (step 6) uses the same evaluation metrics (such as accuracy, precision, recall, F1 value, etc.) and evaluation strategies (such as data set partitioning and confusion matrix analysis);

[0182] In the model optimization step (step 7), the overall optimization strategy (such as hyperparameter tuning and feature selection optimization methods) remains unchanged.

[0183] The steps that need to be adjusted are mainly feature set construction (step 2) and high-load label definition (step 4). In feature set construction, the calculation method and data source of daily granularity features are different due to changes in time scales and need to be reconsidered. For example, utilization and signal-related features need to be calculated and generated from the perspective of a whole day. The definition of high-load labels also needs to be re-established based on the generation mechanism and characteristics of cell load at daily granularity. The thresholds and conditions of hourly granularity cannot be used directly. The feature set of daily granularity can refer to the following Tables 3 and 4:

[0184] Table 3

[0185]

[0186] Table 4

[0187]

[0188] The construction of feature sets with daily granularity (step 2) and the definition of high-load labels (step 4) are described in detail below:

[0189] Step 2: Feature set construction. The feature set of day granularity is different from that of hour granularity and needs to be reconstructed. Although the concepts of some features are similar, the calculation method and data source are different. The feature set of day granularity is detailed in Table 4 of the present invention.

[0190] Step 4: High load label definition. The high load label definition at the daily granularity needs to be redefined. Due to the change in time scale, the judgment criteria for cell load are different. When the cell is in a high load state on the day, it is marked as 1, otherwise it is marked as 0. The specific judgment conditions are as follows:

[0191] 1. High utilization leads to high load. When Day_Utilization_predicted>75%, the cell is in a high load state. High daily granularity utilization prediction value indicates that the cell resources are heavily used on that day and face greater load pressure.

[0192] 2. Special dates and coverage issues together lead to high loads.

[0193] When Day_Utilization_predicted>65% and Is_Special_Day=1 and Is_Weak_Coverage_Day=1, the weak coverage on a special day will increase the cell load on that day. Special days usually have increased user activities, and weak coverage will lead to increased power consumption of user equipment and more retransmissions, increasing the burden on the cell on that day.

[0194] When Day_Utilization_predicted>65% and Is_Special_Day=1 and Is_Overlap_Coverage_Day=1, the overlapping coverage problem on special days will cause the cell to be under high load. Overlapping coverage on special days is prone to cause signal interference, affecting communication quality and resource utilization efficiency, and increasing the cell load on that day.

[0195] 3. Special dates and signal quality issues together lead to high load.

[0196] When Day_Utilization_predicted>65% and Is_Special_Day=1 and Day_RSRP_std>8dBm, the larger daily granularity RSRP standard deviation on special days means that the signal strength fluctuates violently on that day. This will cause the user equipment to frequently adjust the receiving parameters and increase retransmissions, which will increase the cell resource consumption and load.

[0197] When Day_Utilization_predicted>65% and Is_Special_Day=1 and Day_SINR_std>5dBm, the larger daily SINR standard deviation in the special day indicates that the signal interference is serious on that day. This will reduce the communication quality and data transmission efficiency, and the cell needs to consume more resources to maintain communication, thus being in a high-load state.

[0198] 4. Changes in signal quality lead to high load.

[0199] When Day_Utilization_predicted>65% and Day_RSRP_change_rate>0.3, the daily granularity signal strength change rate is high, indicating that the signal environment of the cell is unstable within the day. This instability will cause frequent cell resource allocation, increase additional overhead, and lead to high load.

[0200] When Day_Utilization_predicted>65% and Day_RSRP_stability_index>4dBm, the daily RSRP stability is poor, which means that the signal fluctuates frequently on that day. This will cause problems such as large changes in signal strength, such as frequent adjustments of user equipment and frequent allocation of resources, which will increase the cell load.

[0201] In a feasible implementation, step S40 may include steps A41 to A45:

[0202] Step A41, when the utilization granularity is a day granularity, the day granularity random tree model is used as a target random tree model.

[0203] It should be understood that the prediction of cell utilization takes into account different time granularities, specifically hourly granularity and daily granularity. Based on this, two types of random tree models are constructed to predict cell load conditions, namely hourly granularity random tree model and daily granularity random tree model.

[0204] Step A42: When the utilization granularity is hourly granularity, the hourly granularity random tree model is used as the target random tree model.

[0205] It should be noted that both the hourly granularity random tree model and the daily granularity random tree model are trained, and a comprehensive evaluation is performed on the trained models to estimate the performance of the random tree model.

[0206] Step A43, performing load prediction on the fusion feature through each random tree of the target random tree model to obtain a plurality of initial prediction results, wherein the initial prediction results include a first load state and a second load state.

[0207] It is understandable that the target random tree model uses a voting mechanism to determine the final prediction result, and the initial prediction result can be understood as each random tree independently gives a prediction (yes or no) on whether the cell is in a high-load state based on the fusion features.

[0208] It should be understood that the initial prediction result includes a first load state and a second load state, ie, a high load and a low load.

[0209] Step A44, performing statistics on the multiple initial prediction results to obtain the first load state proportion and the second load state proportion.

[0210] It should be noted that the prediction results of all decision trees are counted. The number of decision trees predicted to be in a high load state is calculated according to the following formula:

[0211]

[0212] Among them, n e stimators is the number of random trees.

[0213] It is understandable that the first load state proportion and the second load proportion can be understood as the proportion of high load / low load prediction results in the total prediction results among the prediction results given by all random trees.

[0214] Step A45, comparing the first load state proportion and the second load state proportion to obtain the service load of the target cell.

[0215] It should be noted that if the number of decision trees predicting a cell to be in a high load state exceeds half, the prediction result output by the final model is high load; conversely, if the number of decision trees predicting a non-high load state exceeds half, the final result is non-high load.

[0216] In this implementation, the prediction results of multiple decision trees are fully utilized through the voting mechanism of random trees, which effectively reduces the erroneous predictions that may be caused by problems such as data noise or overfitting in a single decision tree, enhances the reliability and stability of the entire model for cell load prediction, and thus better leverages the advantages of the random forest model in dealing with such complex classification problems.

[0217] The above are only feasible implementations of step S40 provided in this embodiment, and this embodiment does not specifically limit the specific implementation of step S40.

[0218] It should be emphasized that with the random tree model as the core, based on a large number of 5G cell historical load indicators, the key parameters of Minimization of Drive Tests (MDT) and the indicators of Measurement Reporting Optimization (MRO) are integrated to build a flexible and accurate load prediction framework. In the process of building this framework, the ability of the random tree model to process complex data is utilized to organically combine MDT and MRO data with historical load indicators. By extracting and analyzing the features of these data, the random tree model can accurately learn the patterns and laws in the data, thereby achieving accurate prediction of cell load. Compared with traditional methods, this prediction method based on the random tree model has higher accuracy and flexibility in processing complex and changeable 5G cell load data.

[0219] It is further emphasized that the overall structure of the cell traffic load prediction method in practical application can be referred to Figure 2 , Figure 2 It mainly includes five modules: 1. Wireless communication network underlying server, 2. Data cleaning and feature engineering, 3. Cell utilization prediction, 4. Multi-data feature fusion, 5. Cell business load prediction.

[0220] It should be noted that the cell service load prediction method based on the above application structure integrates the cell historical utilization record (PRB utilization, hereinafter referred to as "utilization"), MDT, and MRO indicators. Through in-depth analysis of historical data, the system can identify the periodic pattern and unexpected fluctuations of the cell load, and especially emphasizes the load characteristics of weekdays and holidays to improve the timeliness and accuracy of the prediction. Indicators such as RSRP, RSRQ, and SINR in MDT and MRO data are like a high-precision mirror, clearly reflecting the dynamic changes of cell signal strength and quality, and providing key real-time health monitoring information for the prediction model. At the same time, special attention is paid to the load difference between daily and special dates (such as holidays), and the comprehensive analysis of historical utilization data and MDT and MRO indicators is used to achieve a detailed prediction of the future cell load. This not only helps to reasonably allocate daily network resources, but also ensures that the network can still maintain stable operation during special periods to meet the communication needs of users in various scenarios.

[0221] This embodiment provides a method for predicting cell service load, which selects a target cell utilization prediction model through historical indicator status and utilization granularity, and predicts more accurate cell utilization from special dates and non-special dates according to the target cell utilization prediction model. The method can cope with daily business volume fluctuations and drastic changes in business volume during special periods, and performs feature fusion of cell utilization and other related features at different granularities to improve the comprehensiveness and accuracy of the prediction model and enhance the reliability of the prediction results. Load prediction at different granularities is performed based on feature fusion to meet different time scale requirements, provide rich data for network planners, help make refined decisions, and improve network operation quality.

[0222] Based on the first embodiment of the present application, in the second embodiment of the present application, the same or similar contents as those in the above-mentioned embodiment 1 can be referred to the above introduction, and will not be repeated in the following. Figure 3 , step S20 also includes steps S21 to S23:

[0223] Step S21: when the historical indicator state is normal, a type of utilization prediction model is used as a target cell utilization prediction model.

[0224] It should be noted that the utilization prediction model includes four prediction models. The first type of prediction model includes a first utilization prediction model and a third utilization prediction model. Among them, the first utilization prediction model is a cell utilization prediction model with a daily granularity and normal utilization historical indicators. The second utilization prediction model is a cell utilization prediction model with an hourly granularity and normal utilization historical indicators.

[0225] Step S22: When the historical indicator state is missing, the second type of utilization prediction model is used as the target cell utilization prediction model.

[0226] It should be understood that the two types of utilization prediction models may include a second utilization prediction model and a fourth utilization prediction model. The second utilization prediction model is a cell utilization prediction model with a daily granularity and missing utilization history indicators. The fourth utilization prediction model is a cell utilization prediction model with an hourly granularity and missing utilization history indicators.

[0227] Step S23: obtaining the cell utilization rate according to the target cell and the predicted date based on the first type of utilization rate prediction model or the second type of utilization rate prediction model.

[0228] It is understandable that the first type of utilization prediction model or the second type of utilization prediction model is determined according to the historical utilization index of the target cell, and after the determination, a specific model is selected according to the granularity required for prediction.

[0229] In a feasible implementation, step S23 may include steps A231 to A233:

[0230] Step A231, obtaining basic indicators according to the annual utilization rate of the target cell through the first type of utilization rate prediction model.

[0231] It should be noted that when the first type of utilization prediction model is the first utilization prediction model, the first utilization prediction model logic diagram can refer to Figure 4 , Figure 4 First, the utilization index of the target cell in the past year is obtained, and the minimum value of the index in the past year is taken as the basic index a. The calculation of the minimum value of the index is when calculating the minimum value index in a specific time period.

[0232] If we simply select the minimum value of all indicator data, we will seriously ignore the impact of recent data and it will be difficult to accurately capture the dynamic changes of data. In view of this, we introduced the exponentially weighted moving minimum (EWMM) method to give significantly higher weight to recent data. The details are as follows:

[0233] First, set the first indicator data as the initial value EWMM1=x1; then define an EWMM update rule, and calculate the subsequent EWMM values ​​according to the update rule. The goal is to maintain a minimum value estimate that is smoothly updated over time, while giving higher weights to the most recent data points, and the following formula is obtained:

[0234]

[0235] The smoothing factor α is still set to 0.6, so that EWMM can respond to the latest data points more quickly, and can also be adjusted dynamically according to usage requirements. It should be emphasized that the subsequent minimum value calculation is the same as here, so it will not be repeated here.

[0236] It should be noted that when the first type of utilization prediction model is the third utilization prediction model, the third utilization prediction model logic diagram can refer to Figure 5 , Figure 5 First, take the hourly and daily utilization indicators for the past year.

[0237] The hourly average value for the whole day is calculated based on the hourly granularity index. The minimum value for the whole day is a1 according to the distance weight, and the minimum value for the day granularity is a2 according to the distance weight.

[0238] Among them, the utilization index of the successor coverage cell is calculated by the distance sharing weight. When the cell utilization index is missing, it is necessary to use the utilization index of the successor cell to make a judgment. In view of the difference in the distance between the successor cells, the impact on the cell is also different, and the attenuation caused by the increase in distance presents nonlinear characteristics, so the inverse square root of the distance is introduced to allocate weights for the successor coverage cells. The specific allocation formula is as follows:

[0239]

[0240] Among them, d j Indicates the distance of the consular district, w i Represents the normalized weight of the ith neighbor based on distance. Such a weight setting ensures that the neighbors with closer distances will have greater weights, while the neighbors with farther distances will have smaller weights.

[0241] It should be emphasized that the utilization index of the subsequent coverage cell can be obtained by calculating the distance sharing weight in the same way, and the present invention will not elaborate on it.

[0242] In the specific implementation, it is necessary to predict the load situation of cell B on 2024 / 7 / 23. When obtaining the indicators of cell B in the past three months, it is found that the data is missing, and it is necessary to match the continuous coverage cell.

[0243] Get the six connected coverage cells a / b / c / d / e / f / h of cell B, and get the distances between these six connected coverage cells and cell B, as well as the utilization rate indicators for the past three months.

[0244] According to the prediction algorithm process, when taking the minimum value of the indicator of the same week attribute of the continuous coverage cells in the past three months, it is necessary to distribute it according to the distance weight. The sharing weight is calculated as follows:

[0245] Continuation coverage cell number Distance(KM) <![CDATA[Sharing weight w i > Community A 1.2 0.292 Community B 1.8 0.243 Cell c 2.3 0.207 Cell D 2.9 0.176 Community e 3.1 0.163 Cell f 3.4 0.129

[0246] After calculating the minimum utilization index of each connected coverage area in the past three months, the sharing weight w calculated above is used. i ,Finally, the predicted utilization rate of cell B can be calculated, providing characteristic indicators for cell load prediction.

[0247] Here, the utilization index of the continuous coverage cells in the past month is combined with the above-calculated sharing weight to calculate the utilization index of cell B as shown in the following table (here it is assumed that the minimum utilization index of the six continuous coverage cells in the past three months has been calculated according to formula 4):

[0248]

[0249] According to the above sharing weight W i Calculate the utilization rate of cell B:

[0250] 0.292*21.93920287+0.243*17.8419374+0.207*14.95233432+0.176*12.31274913+0.163*16.67392363+0.129*5.423199549=19.42145737.

[0251] Therefore, according to the weight distribution calculation of the distance to the successive coverage cells, the utilization rate of cell B is approximately equal to 19.42.

[0252] Further, the size of a1 and a2 is determined. If a1>a2, the index value a1 is assigned a weight of 0.6, the index value a2 is assigned a weight of 0.4, and the weighted average calculation value a is used as the basic index;

[0253] If a1≤a2, the index value a1 is assigned a weight of 0.4, the index value a2 is assigned a weight of 0.6, and the weighted average calculation value a is the basic index.

[0254] Step A232, when the predicted date is a special date, the special date utilization and target holiday utilization of the target cell are obtained through the first type of utilization prediction model, the annual utilization increase ratio is obtained according to the special date utilization ratio, and the cell utilization is obtained according to the annual utilization increase ratio and the target holiday utilization ratio.

[0255] It should be noted that when the first type of utilization rate prediction model is the first utilization rate prediction model, whether the predicted date is a special date is determined according to the pre-defined predicted date and date attributes.

[0256] Furthermore, when the predicted date is a special date, the median b of all special date indicators of the year is taken, and the method of taking the median is a general method;

[0257] Take the index t1 of the most recent holiday as the reference value, that is, the target holiday utilization rate; calculate the increase ratio s1 = b / a; and the utilization rate of the predicted date = t1*s1.

[0258] In particular, when the first type of utilization prediction model is the third utilization prediction model, Figure 5 First determine the date attribute. The determination of the date attribute can refer to the following table 5:

[0259] Table 5

[0260]

[0261] The judgment of subsequent date attributes is similar and will not be repeated. If it is a special date, take the median b of the average hourly index of all special dates in the year, take the index t1 of the nearest holiday as the reference value, and calculate the increase ratio s1:b / a; the utilization rate of the predicted date = t1*s2.

[0262] Step A233, when the predicted date is not a special date, multiple reference utilizations of the same attribute are obtained according to the week attribute of the predicted date through the first type of utilization prediction model, and annual reference utilization and monthly reference utilization are obtained according to the multiple reference utilizations; the cell utilization is obtained according to the annual reference utilization, the monthly reference utilization and the basic indicators.

[0263] It should be noted that when the forecast date is not a special date, Figure 4 According to the week attribute of the predicted date, the judgment of the week attribute can refer to the following formula:

[0264]

[0265] Where q is the day of the month. m is the month. For January and February, they are considered to be the 13th and 14th months of the previous year, that is, m=13 or m=14. K is the last two digits of the year. J is the century, that is, the first two digits of the full year. The final h value (0-6) corresponds to the day of the week (Saturday-Friday). The subsequent day of the week attribute judgment is similar and will not be repeated here.

[0266] For example, to calculate the day of the week of July 23, 2024, we can apply Formula 2 to get: q = 23, m = 7, k = 24, j = 20, and the calculation is as follows:

[0267]

[0268] The final h value is 3. Combined with Table 2, it is determined that the weekday attribute of 2024 / 7 / 23 is Tuesday.

[0269] further, Figure 4Calculate the minimum value c of the indicator for the dates with the same week attribute in the past year, and the calculation of the minimum value of the indicator. Calculate the value of c / a as h1, calculate the indicator value d for the dates with the same week attribute in the past month, calculate the value of d / a as h2, and compare the magnitudes of h1 and h2.

[0270] If h1 > h2, assign a weight of 0.6 to h1 and a weight of 0.4 to h2, and calculate a value s through weighted average; for the calculation of weighted average, take the average value x of the utilization rate of the same week attribute in the past month as the reference point; the predicted utilization rate of the cell = x * s.

[0271] Among them, when calculating the average indicator value using the weighted average algorithm, it does not adopt the method of simply adding and then taking the average, but uses the weighted average method. Because the weighted average can fully consider the importance or weight differences of different data, making the calculation result more reasonable and accurate. Compared with the simple average, the weighted average can better reflect the actual influence of the data and provide a more valuable reference basis for decision-making. Specifically, it can refer to the following formula:

[0272] A = W1 * X1 + W2 * X2

[0273] Among them: W1 and W2 are the weights respectively, and x1 and x2 are the values respectively. Similarly, the subsequent weighted average algorithm can be known, and the present invention will not elaborate.

[0274] For example, when calculating the value of the basic indicator a using weighted average, a1 = 21.24, a2 = 17.28, and it is known that a1 > a2. Assign a weight of 0.6 to the indicator value a1 and a weight of 0.4 to the indicator value a2. The calculated value of the basic indicator a through weighted average = 0.6 * 21.24 + 0.4 * 17.28 = 19.66.

[0275] If h1 < h2, assign a weight of 0.4 to h1 and a weight of 0.6 to h2, calculate a value s through weighted average, and take the average value x of the utilization rate of the same week attribute in the past two weeks as the reference point; the predicted utilization rate of the cell = x * s.

[0276] It should be emphasized that when a certain utilization rate prediction model is the third utilization rate prediction model, reference can be made to Figure 5 First, perform the judgment of the week attribute, and take the minimum value c of the single-hour indicator for the whole day with the same week attribute in the past year. Calculate (c / a), and the value is h1. Take the minimum value d of the single-hour indicator for the whole day with the same week attribute in the past month, calculate (d / a), and the value is h2, and judge the magnitudes of h1 and h2.

[0277] Furthermore, when h1 > h2, assign a weight of 0.6 to h1 and a weight of 0.4 to h2, and calculate a value s through weighted average. Take the average value x of the single-hour indicator for the whole day with the same week attribute in the past month as the reference point. The predicted utilization rate of the cell on the predicted date = x * s.

[0278] Furthermore, when h1≤h2, h1 is assigned a weight of 0.4, h2 is assigned a weight of 0.6, and the weighted average is calculated as a value s. The average value x of the same week attribute in the past two weeks is taken as the reference point, and the cell utilization rate on the predicted date = x*s.

[0279] In this implementation, the utilization rate of cells with normal utilization rate historical indicators is predicted through two granularity utilization rate prediction models. Considering whether the prediction date is a special date, utilization rate predictions are performed separately, so as to obtain more accurate cell utilization rate.

[0280] In another feasible implementation, step S23 may include steps B231 to A233:

[0281] Step B231, determining each subsequent coverage cell according to the target cell through the second type utilization prediction model, and obtaining the basic index of the target cell according to the annual utilization of each subsequent coverage cell.

[0282] It can be understood that the successive coverage cell may be an associated neighboring cell.

[0283] It should be noted that when the utilization history indicator of the target cell is missing, the second utilization prediction model is used for prediction based on the day granularity. The logic diagram of the second utilization prediction model can be found in Figure 6 , Figure 6 The model can obtain the basic index of the target cell according to the annual utilization rate of each subsequent coverage cell by associating 5G engineering parameters, matching the subsequent coverage cell, and taking the utilization index of the subsequent coverage cell in the past year. The minimum value of the one-year index of the subsequent cell is taken as the basic index a according to the weight.

[0284] Among them, first, the working parameter information of all 5G cells is extracted from the 5G network management system, including basic information such as cell ID, frequency, base station identification, and the longitude and latitude coordinates of the cell. At the same time, the associated neighboring cell information is obtained, covering the cell ID, basic attributes and longitude and latitude coordinates of the neighboring cells. These data are maintained and updated by operators during network planning and operation, and serve as the basic data source for subsequent analysis. Then, according to the ID of the predicted cell, the basic information of the predicted cell and the basic information of the neighboring cells are matched from the obtained full 5G cell working parameters. Finally, based on the obtained longitude and latitude of the cell and the longitude and latitude of each neighboring cell, the distance between each neighboring cell and the predicted cell is calculated. This distance information will be used as the key parameter for the subsequent calculation of the neighboring cell utilization rate according to the distance weight. The calculation of the distance between each neighboring cell and the predicted cell can refer to the following formula:

[0285]

[0286] The longitude and latitude of the two points are assumed to be A1(ψ1, λ1) and A2(ψ2, λ2), and the average radius of the earth is R = 6371 km. ψ1 and ψ2 are the latitudes of the two points, λ1 and λ2 are the longitudes of the two points, and the angles must be in radians.

[0287] It should be emphasized that when the historical utilization indicators of the target cell are missing, the fourth utilization prediction model is used for prediction based on hourly granularity. The target cell is determined according to each subsequent coverage cell. The basic indicator of the target cell obtained according to the annual utilization rate of each subsequent coverage cell can be associated with 5G engineering parameters, matching the subsequent coverage cell, taking the hourly granularity and daily granularity utilization indicators of the past year, and calculating the hourly average value of the whole day according to the hourly granularity indicator. The minimum value of the whole day is a1 according to the distance weight. The minimum value of the daily granularity is a2 according to the distance weight, and the size of a1 and a2 is determined. If a1>a2, the indicator value a1 is assigned a weight of 0.6, the indicator value a2 is assigned a weight of 0.4, and the weighted average calculation value a is the basic indicator; if a1≤a2, the indicator value a1 is assigned a weight of 0.4, the indicator value a2 is assigned a weight of 0.6, and the weighted average calculation value a is the basic indicator.

[0288] It should be understood that 5G engineering parameters may refer to engineering parameters for 5G networks, which may include but are not limited to: Frequency bandwidth: 5G supports multiple frequency bands, each with different bandwidths. Cell ID: used to distinguish different base station coverage areas. PCI (Physical Cell ID): Physical cell identification, which is a unique identifier for each 5G NR cell. TA (Timing Advance): Timing advance, used to compensate for the difference in uplink signal transmission time. RSRP / RSRQ / SINR: Wireless environment indicators such as reference signal received power, reference signal received quality, signal-to-noise ratio plus interference. MIMO configuration: The specific configuration of multiple-input multiple-output technology affects data transmission rate and capacity. Beamforming parameters: used for directional transmission to improve spectrum efficiency.

[0289] Step B232, when the predicted date is a special date, the distance allocation weight and the target holiday utilization rate are obtained according to the distance between each subsequent coverage cell and the target cell through the second type of utilization prediction model, and the cell utilization rate of the target cell is obtained according to the distance allocation weight, the target holiday utilization rate and the annual utilization rate of each subsequent coverage cell.

[0290] It should be noted that when the utilization rate prediction model is the second utilization rate prediction model, further reference may be made to Figure 6Determine the date attribute of the predicted date. When the predicted date is a special date, the distance allocation weight and target holiday utilization rate can be obtained based on the distance between each of the subsequent coverage cells and the target cell. The median of all special date indicators of the subsequent coverage cell in that year is taken, and a comprehensive indicator b is calculated according to the distance allocation weight. Calculate the increase ratio s1=b / a, take the median of the holiday indicators of the subsequent coverage cell on the most recent day, calculate a comprehensive reference value indicator x according to the distance allocation weight, and estimate the utilization rate of the cell = x*s1.

[0291] It should be emphasized that when the utilization prediction model is the fourth utilization prediction model, the logic diagram of the first utilization prediction model can refer to Figure 7 , Figure 7 First determine the date attribute of the predicted date, and take the median of the average hourly index of all special dates in the year for the subsequent coverage area. Calculate a comprehensive index b according to the distance distribution weight. Calculate the increase ratio s1: b / a, take the index x of the most recent holiday as the reference value, and estimate the utilization rate of the cell = x*s1.

[0292] Step B233, when the predicted date is not a special date, multiple reference utilizations of the same attribute are obtained according to the week attribute of the predicted date through the second-category utilization prediction model, and annual reference utilization and monthly reference utilization are obtained according to the multiple reference utilizations; the cell utilization is obtained according to the annual reference utilization, the monthly reference utilization and the basic indicators.

[0293] It should be noted that the reference Figure 6 The specific prediction step may be that when the predicted date is not a special date, multiple reference utilizations of the same attribute are obtained according to the week attribute of the predicted date through the second utilization prediction model, and the annual reference utilization and the monthly reference utilization are obtained according to the multiple reference utilizations. The week attribute of the predicted date is first determined, and the minimum utilization c of the same week attribute of the continuous coverage cells in the past year is obtained, and the value of c / a is calculated as h1, and the minimum value d of the same week attribute in the past month is taken. The value of d / a is calculated as h2; and the size of h1 and h2 is determined.

[0294] Furthermore, if h1>h2, h1 is assigned a weight of 0.6, h2 is assigned a weight of 0.4, and the weighted average is calculated as a value s; the average utilization rate x of the same weekly attribute in the past month is taken as the reference point. The estimated utilization rate of the cell = x*s.

[0295] If h1≤h2, h1 is assigned a weight of 0.4, h2 is assigned a weight of 0.6, and the weighted average is calculated as a value s. The average utilization value x of the same weekly attribute in the past two weeks is taken as the reference point, and the estimated cell utilization = x*s.

[0296] It should be emphasized that when the utilization rate prediction model is the fourth utilization rate prediction model, you can refer to Figure 7 As described in the method, first determine the attribute of the week, take the minimum value c of the single-hour index of the same attribute of the week for the continuous coverage cells in the past year, and take the minimum value d of the single-hour index of the same attribute of the week for the past month; calculate (c / a), the value is h1, calculate (d / a), the value is h2, and determine the size of h1 and h2.

[0297] If h1>h2, h1 is assigned a weight of 0.6, h2 is assigned a weight of 0.4, and the weighted average is calculated as a value s. Take the average value x of the hourly index of the same weekly attribute in the past month as the reference point, and estimate the utilization rate of the cell = x*s1.

[0298] If h1≤h2, h1 is assigned a weight of 0.4, h2 is assigned a weight of 0.6, and the weighted average is calculated as a value s. Take the average value x of the hourly index of the same week attribute in the past two weeks as a reference, and estimate the utilization rate of the cell = x*s1.

[0299] In this implementation, the utilization rate of the cell with missing utilization rate history indicators is predicted by using two granularity utilization rate prediction models, and whether the prediction date is a special date is considered, and utilization rate predictions are performed separately, so as to obtain more accurate cell utilization rate.

[0300] The above are only two feasible implementations of step S23 provided in this embodiment, and this embodiment does not specifically limit the specific implementation of step S23.

[0301] This embodiment provides a method for predicting cell service load. By integrating MDT data and MRO data with cell utilization indicators, a prediction model is established based on a random tree to predict cell load, breaking through traditional limitations, mining more information, improving the comprehensiveness and accuracy of the prediction model, and enhancing the reliability of prediction results, laying a foundation for decision-making; in response to the lack of historical cell utilization data and prediction requirements of different granularities, different load prediction models are set to meet different time scale requirements, provide rich data for network planners, help refine decisions, and improve network operation quality.

[0302] It should be noted that the above examples are only used to understand the present application and do not constitute a limitation on the cell service load prediction method of the present application. More simple transformations based on this technical concept are all within the scope of protection of the present application.

[0303] This application also provides a cell service load prediction device, please refer to Figure 8 , the cell traffic load prediction device comprises:

[0304] An acquisition module 10 is used to acquire a target cell, a historical indicator status of the target cell, and a predicted date;

[0305] A utilization prediction module 20, configured to obtain a target cell utilization prediction model according to the historical indicator status, and obtain a cell utilization according to the target cell and the prediction date based on the target cell utilization prediction model;

[0306] The service load prediction module 30 is used to perform feature fusion on the cell utilization, the minimization of drive test data, and the measurement report optimization data to obtain a fusion feature;

[0307] The service load prediction module 30 is further used to determine a target random tree model based on the utilization granularity of the fused feature, and perform load prediction on the fused feature based on the target random tree model to obtain the service load of the target cell.

[0308] The cell traffic load prediction device provided by the present application adopts the cell traffic load prediction method in the above embodiment, which can solve the technical problem that the cell load prediction is inaccurate due to the lack of real-time communication load data of the cell when predicting the cell traffic load. Compared with the prior art, the beneficial effects of the cell traffic load prediction device provided by the present application are the same as the beneficial effects of the cell traffic load prediction method provided by the above embodiment, and the other technical features in the cell traffic load prediction device are the same as the features disclosed in the above embodiment method, which will not be repeated here.

[0309] In one embodiment, the utilization prediction module 20 is further configured to use a type of utilization prediction model as a target cell utilization prediction model when the historical indicator state is normal;

[0310] When the historical indicator state is in a missing state, the second type of utilization prediction model is used as the target cell utilization prediction model;

[0311] The cell utilization is obtained according to the target cell and the predicted date based on the first type of utilization prediction model or the second type of utilization prediction model.

[0312] In one embodiment, the utilization rate prediction module 20 is further used to obtain a basic index according to the annual utilization rate of the target cell through the first type of utilization rate prediction model;

[0313] When the predicted date is a special date, the special date utilization rate and the target holiday utilization rate of the target cell are obtained by using the first type of utilization prediction model, the annual utilization rate increase ratio is obtained according to the special date utilization rate, and the cell utilization rate is obtained according to the annual utilization rate increase ratio and the target holiday utilization rate;

[0314] When the predicted date is not a special date, a plurality of reference utilizations of the same attribute are obtained according to the week attribute of the predicted date through the utilization prediction model, and an annual reference utilization and a monthly reference utilization are obtained according to the plurality of reference utilizations; and the cell utilization is obtained according to the annual reference utilization, the monthly reference utilization and the basic indicators.

[0315] In one embodiment, the utilization prediction module 20 is further used to determine each subsequent coverage cell according to the target cell through the second type of utilization prediction model, and obtain the basic index of the target cell according to the annual utilization of each subsequent coverage cell;

[0316] When the predicted date is a special date, the distance allocation weight and the target holiday utilization rate are obtained according to the distance between each of the subsequent coverage cells and the target cell through the two-type utilization prediction model, and the cell utilization rate of the target cell is obtained according to the distance allocation weight, the target holiday utilization rate and the annual utilization rate of each of the subsequent coverage cells;

[0317] When the predicted date is not a special date, the second-class utilization prediction model is used to obtain multiple reference utilizations of the same attribute according to the week attribute of the predicted date, and the annual reference utilization and the monthly reference utilization are obtained according to the multiple reference utilizations; the cell utilization is obtained according to the annual reference utilization, the monthly reference utilization and the basic indicators.

[0318] In one embodiment, the service load prediction module 30 is further used to divide the cell utilization, the minimized drive test data, and the measurement report optimization data by hour to obtain an hourly granularity fusion feature, wherein the hourly granularity fusion feature includes hourly granularity minimized drive test data, hourly granularity cell utilization, and hourly granularity measurement report optimization data;

[0319] The cell utilization rate, the minimized drive test data and the measurement report optimization data are divided by day to obtain a daily granularity fusion feature, wherein the daily granularity fusion feature includes daily granularity minimized drive test data, daily granularity cell utilization rate and daily granularity measurement report optimization data.

[0320] In one embodiment, the business load prediction module 30 is further configured to use the random tree model with the daily granularity as the target random tree model when the utilization granularity is the daily granularity;

[0321] When the utilization rate granularity is an hourly granularity, the hourly granularity random tree model is used as a target random tree model;

[0322] Performing load prediction on the fusion feature through each random tree of the target random tree model to obtain a plurality of initial prediction results, wherein the initial prediction results include a first load state and a second load state;

[0323] Performing statistics on the multiple initial prediction results to obtain a first load state ratio and a second load state ratio;

[0324] The business load of the target cell is obtained by comparing the first load state proportion and the second load state proportion.

[0325] The present application provides a cell traffic load prediction device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the cell traffic load prediction method in the above-mentioned embodiment 1.

[0326] Reference below Fig. 9 , which shows a schematic diagram of the structure of a cell traffic load prediction device suitable for implementing the embodiment of the present application. The cell traffic load prediction device in the embodiment of the present application may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Fig. 9 The cell traffic load prediction device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0327] like Fig. 9As shown, the cell traffic load prediction device may include a processing device 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1002 or a program loaded from a storage device 1003 to a random access memory (RAM: Random Access Memory) 1004. Various programs and data required for the operation of the cell traffic load prediction device are also stored in RAM1004. The processing device 1001, ROM1002, and RAM1004 are connected to each other via a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to the I / O interface 1006: input devices 1007 including, for example, a touch screen, a touchpad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 1003 including, for example, a magnetic tape, a hard disk, etc.; and communication devices 1009. The communication device 1009 can allow the cell traffic load prediction device to communicate wirelessly or wired with other devices to exchange data. Although the figure shows a cell traffic load prediction device with various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems may be implemented or have alternatively.

[0328] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.

[0329] The cell traffic load prediction device provided by the present application adopts the cell traffic load prediction method in the above embodiment, which can solve the technical problem of inaccurate cell load prediction due to lack of real-time cell communication load data when predicting cell traffic load. Compared with the prior art, the beneficial effects of the cell traffic load prediction device provided by the present application are the same as the beneficial effects of the cell traffic load prediction method provided by the above embodiment, and other technical features in the cell traffic load prediction device are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.

[0330] It should be understood that the various parts disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0331] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0332] The present application provides a computer-readable storage medium having computer-readable program instructions (ie, computer programs) stored thereon, and the computer-readable program instructions are used to execute the cell service load prediction method in the above-mentioned embodiment.

[0333] The computer-readable storage medium provided in the present application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems or devices, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM: Random Access Memory), a read-only memory (ROM: Read Only Memory), an erasable programmable read-only memory (EPROM: Erasable Programmable Read Only Memory or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM: CD-Read Only Memory), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, system or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency: Radio Frequency), etc., or any suitable combination of the above.

[0334] The computer-readable storage medium may be included in the cell traffic load prediction device; or may exist independently without being assembled into the cell traffic load prediction device.

[0335] The computer-readable storage medium carries one or more programs. When the one or more programs are executed by the cell service load prediction device, the cell service load prediction device: obtains the target cell, the historical indicator status of the target cell and the prediction date; obtains the target cell utilization prediction model according to the historical indicator status, and obtains the cell utilization according to the target cell and the prediction date based on the target cell utilization prediction model; performs feature fusion on the cell utilization, minimized drive test data and measurement report optimization data to obtain a fusion feature; determines the target random tree model based on the utilization granularity of the fusion feature, performs load prediction on the fusion feature based on the target random tree model, and obtains the service load of the target cell.

[0336] The computer program code for performing the operation of the present application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as an independent software package, partially on the user's computer and partially on the remote computer, or entirely on the remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (for example, using an Internet service provider to connect through the Internet).

[0337] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0338] The modules involved in the embodiments described in this application may be implemented by software or hardware, wherein the name of the module does not constitute a limitation on the unit itself in some cases.

[0339] The readable storage medium provided by the present application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned cell traffic load prediction method, and can solve the technical problem of inaccurate cell load prediction due to lack of cell real-time communication load data when predicting cell traffic load. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided by the present application are the same as the beneficial effects of the cell traffic load prediction method provided by the above-mentioned embodiment, and will not be repeated here.

[0340] The present application also provides a computer program product, including a computer program, which implements the steps of the above-mentioned cell service load prediction method when executed by a processor.

[0341] The computer program product provided by the present application can solve the technical problem that the cell load prediction is inaccurate due to the lack of real-time cell communication load data when predicting the cell traffic load. Compared with the prior art, the beneficial effects of the computer program product provided by the present application are the same as the beneficial effects of the cell traffic load prediction method provided by the above embodiment, which will not be repeated here.

[0342] The above descriptions are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A method for predicting cell traffic load, characterized in that: The method comprises: Obtaining a target cell, a historical indicator status of the target cell, and a predicted date; Obtain a target cell utilization prediction model according to the historical indicator status, and obtain a cell utilization rate according to the target cell and the prediction date based on the target cell utilization prediction model; Performing feature fusion on the cell utilization, the minimized drive test data, and the measurement report optimization data to obtain a fusion feature; A target random tree model is determined based on the utilization granularity of the fused feature, and a load prediction is performed on the fused feature based on the target random tree model to obtain a service load of a target cell.

2. The method according to claim 1, characterized in that The historical indicator status includes a normal status and a missing status; The step of obtaining a target cell utilization prediction model according to the historical indicator status, and obtaining a cell utilization according to the target cell and the prediction date based on the target cell utilization prediction model, includes: When the historical indicator state is normal, using a type of utilization prediction model as a target cell utilization prediction model; When the historical indicator state is in a missing state, the second type of utilization prediction model is used as the target cell utilization prediction model; The cell utilization is obtained according to the target cell and the predicted date based on the first type of utilization prediction model or the second type of utilization prediction model.

3. The method according to claim 2, characterized in that The obtaining the cell utilization rate according to the target cell and the predicted date based on the first type of utilization rate prediction model or the second type of utilization rate prediction model includes: Obtaining a basic indicator according to the annual utilization rate of the target cell through the first type of utilization rate prediction model; When the predicted date is a special date, the special date utilization rate and the target holiday utilization rate of the target cell are obtained by using the first type of utilization prediction model, the annual utilization rate increase ratio is obtained according to the special date utilization rate, and the cell utilization rate is obtained according to the annual utilization rate increase ratio and the target holiday utilization rate; When the predicted date is not a special date, a plurality of reference utilizations of the same attribute are obtained according to the week attribute of the predicted date through the utilization prediction model, and an annual reference utilization and a monthly reference utilization are obtained according to the plurality of reference utilizations; and the cell utilization is obtained according to the annual reference utilization, the monthly reference utilization and the basic indicators.

4. The method according to claim 2, characterized in that The obtaining of the cell utilization rate according to the target cell and the predicted date based on the first type of utilization rate prediction model or the second type of utilization rate prediction model further includes: Determine each subsequent coverage cell according to the target cell by using the second type utilization prediction model, and obtain the basic index of the target cell according to the annual utilization rate of each subsequent coverage cell; When the predicted date is a special date, the distance allocation weight and the target holiday utilization rate are obtained according to the distance between each of the subsequent coverage cells and the target cell through the two-type utilization prediction model, and the cell utilization rate of the target cell is obtained according to the distance allocation weight, the target holiday utilization rate and the annual utilization rate of each of the subsequent coverage cells; When the predicted date is not a special date, the second-class utilization prediction model is used to obtain multiple reference utilizations of the same attribute according to the week attribute of the predicted date, and the annual reference utilization and the monthly reference utilization are obtained according to the multiple reference utilizations; the cell utilization is obtained according to the annual reference utilization, the monthly reference utilization and the basic indicators.

5. The method according to claim 1, characterized in that The fusion features include day granularity fusion features and hour granularity fusion features; The step of fusing the cell utilization rate, the minimization of drive test data, and the measurement report optimization data to obtain a fusion feature includes: Divide the cell utilization rate, the minimized drive test data, and the measurement report optimization data by hour to obtain an hourly granularity fusion feature, where the hourly granularity fusion feature includes the hourly granularity minimized drive test data, the hourly granularity cell utilization rate, and the hourly granularity measurement report optimization data; The cell utilization rate, the minimized drive test data and the measurement report optimization data are divided by day to obtain a daily granularity fusion feature, wherein the daily granularity fusion feature includes daily granularity minimized drive test data, daily granularity cell utilization rate and daily granularity measurement report optimization data.

6. The method according to claim 1, characterized in that The target random tree model includes a day granularity random tree model and an hour granularity random tree model; The determining a target random tree model based on the utilization granularity of the fusion feature, and performing load prediction on the fusion feature based on the target random tree model to obtain a service load of a target cell includes: When the utilization rate granularity is a day granularity, the day granularity random tree model is used as a target random tree model; When the utilization rate granularity is an hourly granularity, the hourly granularity random tree model is used as a target random tree model; Performing load prediction on the fusion feature through each random tree of the target random tree model to obtain a plurality of initial prediction results, wherein the initial prediction results include a first load state and a second load state; Performing statistics on the multiple initial prediction results to obtain a first load state ratio and a second load state ratio; The business load of the target cell is obtained by comparing the first load state proportion and the second load state proportion.

7. A cell traffic load prediction device, characterized in that: The cell traffic load prediction device comprises: An acquisition module, used to acquire a target cell, a historical indicator status of the target cell, and a predicted date; A utilization prediction module, used to obtain a target cell utilization prediction model according to the historical indicator status, and obtain a cell utilization according to the target cell and the prediction date based on the target cell utilization prediction model; A service load prediction module, used for fusing the cell utilization rate, the minimized drive test data, and the measurement report optimization data to obtain a fusion feature; The service load prediction module is further used to determine a target random tree model based on the utilization granularity of the fused feature, and perform load prediction on the fused feature based on the target random tree model to obtain the service load of the target cell.

8. A cell traffic load prediction device, characterized in that: The cell traffic load prediction device comprises: a memory, a processor, and a cell traffic load prediction program stored in the memory and executable on the processor, wherein the cell traffic load prediction program is configured to implement the cell traffic load prediction method according to any one of claims 1 to 6.

9. A storage medium, characterized in that: The storage medium stores a cell traffic load prediction program, which, when executed by a processor, implements the cell traffic load prediction method according to any one of claims 1 to 6.

10. A computer program product, characterized in that The computer program product comprises a cell traffic load prediction program, and when the cell traffic load prediction program is executed by a processor, the steps of the cell traffic load prediction method according to any one of claims 1 to 6 are implemented.

Images