LTE base station antenna angle optimization method and device

Abstract

The present invention relates to a method for optimizing the antenna angle of an LTE base station, the method comprising: using an extreme gradient boosting (XGBoost) algorithm to respectively train an azimuth prediction model and a downtilt prediction model; obtaining first feature group data and second feature group data of a target base station; inputting the first feature group data into the azimuth prediction model to predict the azimuth; inputting the second feature group data and the predicted azimuth into the downtilt prediction model to predict the downtilt; and outputting a target base station antenna angle optimization list containing the predicted azimuth and downtilt. The present invention also relates to an LTE base station antenna angle optimization device, the device comprising a training unit, an acquisition unit, a first prediction unit, a second prediction unit, and an output unit. The method and device of the present invention improve computational efficiency and can effectively reduce the impact of network structure problems such as weak coverage, overlapping coverage, and cross-area coverage.

Description

Technical Field

[0001] The present invention relates to the fields of wireless networks and artificial intelligence, and in particular to a method and device for optimizing the angle of an LTE base station antenna. Background Art

[0002] Current methods for optimizing LTE base station antenna angles, such as patent application number 2021105243207, entitled "Method, Apparatus, and Computing Device for Optimizing LTE Base Station Azimuth Angles," and patent application number 2017114944125, entitled "Antenna Angle Adjustment Method, Apparatus, Equipment, and Medium," generally determine a new azimuth angle for the target base station based on relevant parameters of the target base station and its neighboring stations, and then optimize the target base station's azimuth angle based on this new azimuth angle. These current optimization methods are computationally complex and inefficient, and they only optimize the antenna azimuth angle, failing to address network structural issues such as weak coverage, overlapping coverage, and cross-area coverage. Summary of the Invention

[0003] In view of this, the main purpose of the present invention is to provide a method and device for optimizing the antenna angle of an LTE base station. The extreme gradient boosting (XGBoost) algorithm is used to predict the optimal antenna angle of an LTE base station in an overlapping environment from two aspects: azimuth angle and downtilt angle, thereby improving the computational efficiency and effectively reducing the impact of network structure problems such as weak coverage, overlapping coverage, and cross-area coverage.

[0004] To achieve the above objectives, on the one hand, the present application provides a method for optimizing the antenna angle of an LTE base station, comprising:

[0005] The XGBoost algorithm is used to train the azimuth angle prediction model and the downtilt angle prediction model respectively;

[0006] Acquire first feature group data and second feature group data of a target base station;

[0007] Inputting the first feature group data into the azimuth prediction model to predict the azimuth;

[0008] Input the second feature group data and the predicted azimuth angle into the downdip angle prediction model to predict the downdip angle;

[0009] Outputs a target base station antenna angle optimization list containing predicted azimuth and downtilt angles.

[0010] In a possible implementation, before the training to obtain the prediction model, the process further includes:

[0011] Obtain historical data of each base station, including azimuth, downtilt, longitude and latitude, as well as feature data to be screened;

[0012] Based on the obtained longitude and latitude, use Pandas to batch calculate the supplementary feature historical data of each base station.

[0013] In another possible implementation, the features to be screened include: coordinates, coverage radius, cell availability, average downlink cell leaf swallowing rate, uplink interference average value, average TA, TA<312m ratio, TA>624m ratio, station height, antenna hanging height, coverage rate, CPU occupancy rate exceeding the threshold of 85%, drop rate, LTE inter-station x2 same-frequency switching preparation success rate, average coverage distance, average QCI, average CPU occupancy rate, maximum number of users, eNodeB paging congestion rate, downlink weak coverage ratio, E-RAB establishment success rate, E-RAB drop rate, QCI=1 wireless connection rate, uplink PRB utilization rate.

[0014] In another possible implementation, the supplementary features of any base station include: supplementary feature 1, supplementary feature 2, and supplementary feature 3; wherein,

[0015] The six base stations closest to the base station are regarded as neighboring stations. The first supplementary feature is the distance between each neighboring station and the base station; the second supplementary feature is the coverage radius of each neighboring station; and the third supplementary feature is the angle between the line connecting each neighboring station and the base station and the north direction along the clockwise direction with the base station as the center.

[0016] In another possible implementation, the training to obtain the azimuth prediction model includes:

[0017] Analyze the correlation coefficient between each feature to be screened and the azimuth angle, and take the feature to be screened whose absolute value of the correlation coefficient is greater than a preset first threshold as the azimuth angle-related feature;

[0018] The azimuth-related features, supplementary features 1, 2 and 3 are grouped into the first feature group;

[0019] Take the data of each feature in the first feature group of the historical data and the supplementary feature historical data as the sample set;

[0020] Based on the sample set, the XGBoost algorithm model is trained to obtain the azimuth prediction model.

[0021] In another possible implementation, the first threshold value ranges from 0.3 to 1.

[0022] In another possible implementation, the training to obtain the downtilt angle prediction model includes:

[0023] Analyze the correlation coefficient between each feature to be screened and the downtilt angle, and take the feature to be screened whose absolute value of the correlation coefficient is greater than a preset second threshold as a downtilt angle-related feature;

[0024] The downtilt angle related features, supplementary features 1 and 2 are classified into the second feature group;

[0025] Taking the data and azimuth of each feature in the second feature group in the historical data and the supplementary feature historical data as a sample set;

[0026] Based on the sample set, the XGBoost algorithm model is trained to obtain the downtilt angle prediction model.

[0027] In another possible implementation, the second threshold value ranges from 0.3 to 1.

[0028] On the one hand, the present application provides an LTE base station antenna angle optimization device, comprising: a training unit, an acquisition unit, a first prediction unit, a second prediction unit and an output unit; wherein,

[0029] A training unit is used to use the XGBoost algorithm to train an azimuth prediction model and a downtilt prediction model respectively;

[0030] An acquiring unit, configured to acquire first feature group data and second feature group data of a target base station;

[0031] A first prediction unit is configured to input the first feature group data into an azimuth prediction model to predict an azimuth;

[0032] The second prediction unit is configured to input the second feature group data and the predicted azimuth angle into a downtilt angle prediction model to predict the downtilt angle;

[0033] The output unit is used to output a target base station antenna angle optimization list including the predicted azimuth angle and downtilt angle. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 A schematic diagram of a flow chart of a method for optimizing the antenna angle of an LTE base station according to an embodiment of the present invention;

[0035] Figure 2 Schematic diagram of the angle between the connection between a base station and its neighboring station and the true north direction in an embodiment of the present invention. DETAILED DESCRIPTION

[0036] Specifically, the process of an LTE base station antenna angle optimization method according to an embodiment of the present invention is as follows: Figure 1 As shown, it includes steps 101 to 105.

[0037] Step 101: Use the XGBoost algorithm to train and obtain an azimuth angle prediction model and a downtilt angle prediction model.

[0038] Step 102: Acquire first feature group data and second feature group data of the target base station.

[0039] Step 103: Input the first feature group data into the azimuth prediction model to predict the azimuth.

[0040] Step 104: Input the second feature group data and the predicted azimuth angle into the downtilt angle prediction model to predict the downtilt angle.

[0041] Step 105: Output a target base station antenna angle optimization list including the predicted azimuth angle and downtilt angle.

[0042] In a possible implementation, steps A and B are further included before step 101 .

[0043] Step A: Obtain historical data of each base station, including azimuth, downtilt, longitude and latitude, and feature data to be screened.

[0044] The features to be screened include: coordinates, coverage radius, cell availability, average downlink cell leaf swallowing rate, uplink interference average value, average TA, TA<312m ratio, TA>624m ratio, site height, antenna hanging height, coverage rate, CPU occupancy rate exceeding the threshold of 85%, call drop rate, LTE inter-site x2 same-frequency handover preparation success rate, average coverage distance, average QCI, average CPU occupancy rate, maximum number of users, eNodeB paging congestion rate, downlink weak coverage ratio, E-RAB establishment success rate, E-RAB call drop rate, QCI=1 wireless connection rate, and uplink PRB utilization rate.

[0045] Here, it can also include filling in missing values ​​of historical data.

[0046] Step B: Based on the obtained longitude and latitude, use Pandas to batch calculate the supplementary feature historical data of each base station.

[0047] The supplementary features of any base station include: supplementary feature 1, supplementary feature 2 and supplementary feature 3; wherein,

[0048] The six base stations closest to the base station are regarded as neighboring stations. The first supplementary feature is the distance between each neighboring station and the base station; the second supplementary feature is the coverage radius of each neighboring station; the third supplementary feature is the angle between the line connecting each neighboring station and the base station and the north direction in the clockwise direction with the base station as the center, such as Figure 2 shown.

[0049] In another possible implementation, in step 101, the specific implementation of the training to obtain the azimuth angle prediction model includes steps a1 to d1.

[0050] Step a1: Analyze the correlation coefficient between each feature to be screened and the azimuth angle, and take the feature to be screened whose absolute value of the correlation coefficient is greater than a preset first threshold as the azimuth angle-related feature.

[0051] Here, the value range of the first threshold is 0.3 to 1.

[0052] When the first threshold is 0.3, the azimuth-related characteristics include: coverage radius, site height, antenna height, average coverage distance, average TA, coordinates, downlink weak coverage ratio, TA<312m ratio, TA>624m ratio, LTE inter-site X2 same-frequency handover preparation success rate, coverage rate, maximum number of users, average QCI, and eNodeB paging congestion rate.

[0053] Step b1: classify the azimuth-related features, supplementary feature 1, supplementary feature 2, and supplementary feature 3 into the first feature group.

[0054] Correspondingly, the first feature group data of the target base station in step 102 is the data of each feature of the first feature group among the collected parameters of the target base station, as well as the supplementary feature 1, supplementary feature 2 and supplementary feature 3 of the target base station.

[0055] Step c1: Take the data of each feature in the first feature group in the historical data and the supplementary feature historical data as a sample set.

[0056] Step d1: Based on the sample set, train the XGBoost algorithm model to obtain the azimuth prediction model.

[0057] Here, the specific implementation of obtaining the azimuth angle prediction model based on the sample set includes steps d11 to d14.

[0058] Step d11: Standardize the data in the sample set.

[0059] Step d12: Using the cross-validation method, the standardized sample set is divided into a training set and a validation set. The training set is used to train the XGBoost algorithm model to obtain the initial azimuth prediction model; the validation set is used to score the performance of the initial azimuth prediction model.

[0060] Step d13: Based on the performance score of the initial azimuth angle prediction model, the initial azimuth angle prediction model is iteratively optimized until the model score reaches a preset score.

[0061] Step d14: Export and save the model that reaches the preset score as an azimuth prediction model.

[0062] Here, the performance of the model is scored using the Adjusted R-Square; the value range of the preset score is 0.8 to 1.

[0063] The iterative optimization of the initial azimuth prediction model includes:

[0064] Based on the performance score of the previously optimized model, a grid search method is used to optimize the parameters;

[0065] Increase the training set data and train the model obtained by grid search;

[0066] Score the performance of the trained model.

[0067] In another possible implementation, in step 101, the specific implementation of training the downtilt angle prediction model includes steps a2 to d2.

[0068] Step a2: Analyze the correlation coefficient between each feature to be screened and the downtilt angle, and select the feature to be screened whose absolute value of the correlation coefficient is greater than the preset second threshold as the downtilt angle-related feature.

[0069] Here, the value range of the second threshold is 0.3 to 1.

[0070] When the second threshold is 0.3, the azimuth-related characteristics include: cell availability, downlink cell average leaf swallowing rate, uplink interference average, drop rate, E-RAB establishment success rate, E-RAB drop rate, wireless connection rate (QCI = 1), uplink PRB utilization, CPU average occupancy, and the proportion of CPU occupancy exceeding the threshold (85%).

[0071] Step b2: Classify the downtilt angle related features, the supplementary feature 1, and the supplementary feature 2 into the second feature group.

[0072] Correspondingly, the second feature group data of the target base station in step 102 is the data of each feature of the second feature group among the collected parameters of the target base station, as well as the supplementary feature 1 and the supplementary feature 2 of the target base station.

[0073] Step c2: taking the data and azimuth of each feature in the second feature group in the historical data and the supplementary feature historical data as a sample set;

[0074] Step d2: Based on the sample set, train the XGBoost algorithm model to obtain a downtilt angle prediction model.

[0075] Here, the performance of the model is scored using Adjusted R-Square; the value range of the preset score is 0.8 to 1.

[0076] The specific implementation of obtaining the downtilt angle prediction model based on the sample set includes steps d21 to d24.

[0077] Step d21: Standardize the data in the sample set.

[0078] Step d22: Using the cross-validation method, the standardized sample set is divided into a training set and a validation set. The training set is used to train the XGBoost algorithm model to obtain the initial downtilt angle prediction model; the validation set is used to score the performance of the initial downtilt angle prediction model.

[0079] Step d23: Based on the performance score of the initial downtilt angle prediction model, the initial downtilt angle prediction model is iteratively optimized until the model score reaches a preset score.

[0080] Step d24: Export and save the model that reaches the preset score as the downtilt angle prediction model.

[0081] The iterative optimization of the initial downtilt angle prediction model includes:

[0082] Based on the performance score of the previously optimized model, a grid search method is used to optimize the parameters;

[0083] Increase the training set data and train the model obtained by grid search;

[0084] Score the performance of the trained model.

[0085] The present invention provides an LTE base station antenna angle optimization device, comprising: a training unit, an acquisition unit, a first prediction unit, a second prediction unit, and an output unit.

[0086] A training unit is used to use the XGBoost algorithm to train an azimuth prediction model and a downtilt prediction model respectively;

[0087] An acquiring unit, configured to acquire first feature group data and second feature group data of a target base station;

[0088] A first prediction unit is configured to input the first feature group data into an azimuth prediction model to predict an azimuth;

[0089] The second prediction unit is configured to input the second feature group data and the predicted azimuth angle into a downtilt angle prediction model to predict the downtilt angle;

[0090] The output unit is used to output a target base station antenna angle optimization list including the predicted azimuth angle and downtilt angle.

[0091] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention.

Claims

1. A method for optimizing the antenna angle of an LTE base station, characterized in that: include: The extreme gradient boosting XGBoost algorithm is used to train the azimuth prediction model and the downtilt prediction model respectively; Obtain first feature group data and second feature group data of the target base station; the first feature group data includes the inter-station distance between each neighboring station of the target base station and the base station, the coverage radius of each neighboring station, and the angle between the line connecting each neighboring station and the base station and the north direction along the clockwise direction with the base station as the center; the second feature group data includes the inter-station distance between each neighboring station of the target base station and the base station, and the coverage radius of each neighboring station; the neighboring stations are the six base stations closest to the base station; Inputting the first feature group data into the azimuth prediction model to predict the azimuth; Input the second feature group data and the predicted azimuth angle into the downdip angle prediction model to predict the downdip angle; Outputs a target base station antenna angle optimization list containing predicted azimuth and downtilt angles.

2. The method according to claim 1, characterized in that Before the training to obtain the prediction model, the following steps are also included: Obtain historical data of each base station, including azimuth, downtilt, longitude and latitude, as well as feature data to be screened; Based on the obtained longitude and latitude, use Pandas to batch calculate the supplementary feature historical data of each base station.

3. The method according to claim 2, characterized in that The features to be screened include: coordinates, coverage radius, cell availability, average downlink cell leaf swallowing rate, uplink interference average value, average TA, TA<312m ratio, TA>624m ratio, site height, antenna hanging height, coverage rate, CPU occupancy rate exceeding the threshold of 85%, call drop rate, LTE inter-site x2 same-frequency handover preparation success rate, average coverage distance, average QCI, average CPU occupancy rate, maximum number of users, eNodeB paging congestion rate, downlink weak coverage ratio, E-RAB establishment success rate, E-RAB call drop rate, QCI=1 wireless connection rate, and uplink PRB utilization rate.

4. The method according to claim 3, characterized in that The supplementary features of any base station include: supplementary feature 1, supplementary feature 2 and supplementary feature 3; wherein, The six base stations closest to the base station are regarded as neighboring stations. The first supplementary feature is the distance between each neighboring station and the base station; the second supplementary feature is the coverage radius of each neighboring station; and the third supplementary feature is the angle between the line connecting each neighboring station and the base station and the north direction along the clockwise direction with the base station as the center.

5. The method according to claim 4, characterized in that The training to obtain the azimuth prediction model includes: Analyze the correlation coefficient between each feature to be screened and the azimuth angle, and take the feature to be screened whose absolute value of the correlation coefficient is greater than a preset first threshold as the azimuth angle-related feature; The azimuth-related features, supplementary features 1, 2 and 3 are grouped into the first feature group; Take the data of each feature in the first feature group of the historical data and the supplementary feature historical data as the sample set; Based on the sample set, the XGBoost algorithm model is trained to obtain the azimuth prediction model.

6. The method according to claim 5, characterized in that The first threshold value ranges from 0.3 to 1.

7. The method according to claim 4, characterized in that The training to obtain the downtilt angle prediction model includes: Analyze the correlation coefficient between each feature to be screened and the downtilt angle, and take the feature to be screened whose absolute value of the correlation coefficient is greater than a preset second threshold as a downtilt angle-related feature; The downtilt angle related features, supplementary features 1 and 2 are classified into the second feature group; Taking the data and azimuth of each feature in the second feature group in the historical data and the supplementary feature historical data as a sample set; Based on the sample set, the XGBoost algorithm model is trained to obtain the downtilt angle prediction model.

8. The method according to claim 7, characterized in that The second threshold value ranges from 0.3 to 1.

9. An LTE base station antenna angle optimization device, characterized in that: include: training unit, acquisition unit, first prediction unit, second prediction unit and output unit; wherein, A training unit is used to use the XGBoost algorithm to train an azimuth prediction model and a downtilt prediction model respectively; An acquisition unit is configured to acquire first feature group data and second feature group data of a target base station; the first feature group data includes the distance between each neighboring station of the target base station and the base station, the coverage radius of each neighboring station, and the angle between the line connecting each neighboring station and the base station and the north direction along the clockwise direction with the base station as the center; the second feature group data includes the distance between each neighboring station of the target base station and the base station, and the coverage radius of each neighboring station; the neighboring stations are the six base stations closest to the base station; A first prediction unit is configured to input the first feature group data into an azimuth prediction model to predict an azimuth; The second prediction unit is configured to input the second feature group data and the predicted azimuth angle into a downtilt angle prediction model to predict the downtilt angle; The output unit is used to output a target base station antenna angle optimization list including the predicted azimuth angle and downtilt angle.

Images