Network quality dynamic optimization method and device, equipment, medium and program product

Abstract

The invention provides a network quality dynamic optimization method and device, electronic equipment, a storage medium and a program product. The method comprises the following steps: S1, acquiring network quality related information of a to-be-tested area acquired by an unmanned aerial vehicle; and S2, obtaining prediction data of a plurality of signal quality parameters. And S3, determining a multi-objective function. And S4, determining the optimization quantity of each parameter in the at least one parameter of the to-be-tested base station. And S5, adjusting each parameter in the at least one parameter of the base station to be tested. According to the embodiment of the invention, the low-altitude signal data can be collected in real time through the unmanned aerial vehicle, the signal quality is predicted in combination with the environment information, and the base station parameters are dynamically adjusted to realize automatic optimization.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of network quality optimization, in particular to a network quality dynamic optimization method and device, electronic equipment, computer readable storage medium, computer program product. BACKGROUND

[0002] With the deployment of 5G networks, low-altitude applications such as unmanned aerial vehicle logistics and aerial monitoring are increasing. Since the low-altitude area is unobstructed, the signal propagates far, which leads to easy interference. This unstable signal quality affects low-altitude applications such as unmanned aerial vehicle communication based on 5G networks, and there is an urgent need to evaluate and optimize low-altitude 5G network coverage.

[0003] Currently, there are some 5G network coverage test schemes. The existing schemes usually adopt a fixed radius flight path, and collect signal strength (Reference Signal Receiving Power, RSRP) and signal to interference plus noise ratio (Signal to Interference plus Noise Ratio, SNR) and other parameters by uniform flight of the unmanned aerial vehicle, and transmit the data to the test terminal to analyze the coverage.

[0004] However, in the optimization method of the current existing scheme, manual adjustment of base station parameters (such as antenna tilt angle and power) is still needed after artificial analysis of data, which leads to long optimization period, poor real-time performance, and inability to adapt to dynamic environmental changes (such as weather and obstacle movement). SUMMARY

[0005] The present disclosure provides a network quality dynamic optimization method and device, electronic equipment, computer readable storage medium, computer program product.

[0006] Firstly, this disclosure provides a method for dynamic optimization of network quality, comprising: S1, acquiring network quality-related information of the area to be tested collected by a UAV; S2, preprocessing the network quality-related information to obtain preprocessed network quality-related information, and predicting signal quality parameters based on the preprocessed network quality-related information to obtain predicted data for multiple signal quality parameters; S3, acquiring the transmit power of the base station to be tested, and determining a multi-objective function based on the transmit power of the base station to be tested, the predicted data for multiple signal quality parameters, and the preprocessed network quality-related information; S4, determining the optimization amount of each parameter in at least one parameter of the base station to be tested based on the multi-objective function; and S5, adjusting each parameter in at least one parameter of the base station to be tested based on the optimization amount of each parameter in at least one parameter of the base station to be tested. The area to be tested is determined based on the base station to be tested. The network quality-related information includes the current acquisition time, geographical information of the current flight position, signal data of the current flight position, and environmental data of the current flight position. The geographical information includes the actual flight altitude and horizontal coordinate position. The signal data includes reference signal received power parameters, signal-to-interference-plus-noise ratio parameters, and transmission delay parameters. Environmental data includes obstacle image data, wind speed data, and temperature data. A multi-objective function is used to balance three objectives: signal coverage, base station power consumption, and signal interference. At least one parameter of the base station under test includes at least one of the following: antenna tilt angle, antenna azimuth angle, and transmit power.

[0007] Optionally, S1 includes: S11, setting the test area based on the location of the base station to be tested; S12, determining the center of the UAV's flight range based on the location of the base station to be tested, and determining multiple flight altitudes and corresponding flight radii based on the height of ground obstacles; S13, controlling the UAV to fly within its flight range and acquiring network quality-related information of the test area collected by the UAV at preset intervals.

[0008] Optionally, S2 includes: S21, removing invalid data from the reference signal received power parameter, signal-to-interference-plus-noise ratio parameter, and transmission delay parameter, and performing interpolation processing to obtain the interpolated reference signal received power parameter, interpolated signal-to-interference-plus-noise ratio parameter, and interpolated transmission delay parameter. S22, filtering the interpolated reference signal received power parameter and interpolated signal-to-interference-plus-noise ratio parameter to obtain the filtered reference signal received power parameter and filtered signal-to-interference-plus-noise ratio parameter. S23, extracting the obstacle index from the obstacle image data, and preprocessing the wind speed data and temperature data to obtain the preprocessed wind speed data and preprocessed temperature data. The preprocessing includes removing invalid data, interpolation processing, and normalization processing. The preprocessed network quality-related information includes the filtered reference signal received power parameter, the filtered signal-to-interference-plus-noise ratio parameter, the interpolated transmission delay parameter, the obstacle index, the preprocessed wind speed data, and the preprocessed temperature data. S24. A structured data table is created using the acquisition time point, geographical information of the current flight position, filtered reference signal received power parameters, filtered signal-to-interference-plus-noise ratio parameters, and interpolated transmission delay parameters. An environmental feature vector is created using the obstacle index, preprocessed wind speed data, and preprocessed temperature data. S25. Based on the structured data table and environmental feature vectors from the current acquisition time point and the previous three acquisition time points, the signal quality parameters for the next acquisition time point are predicted, resulting in predicted data for the signal quality parameters at the next acquisition time point. The predicted data for the signal quality parameters at the next acquisition time point includes predicted data for the signal-to-interference-plus-noise ratio parameters. Predicted data for multiple signal quality parameters includes predicted data for signal quality parameters from multiple acquisition time points.

[0009] Optionally, in S23, according to formula Extracting obstacle index from obstacle image data .in, It is the number of occluded pixels in the obstacle image data. It is the total number of pixels in the obstacle image data.

[0010] Optionally, S25 includes: S251, constructing a prediction model based on a Long Short-Term Memory (LSTM) network, wherein the prediction model architecture includes two LSTM layers and one fully connected output layer. S252, acquiring historical data from a structured data table and historical data from environmental feature vectors, and training the prediction model based on the historical data from the structured data table and historical data from the environmental feature vectors to obtain a prediction model for the quality parameters. S253, according to formula... Prediction is performed to obtain predicted data for the signal quality parameters at the next acquisition time point. Among these, It is the predicted data of signal quality parameters at the next acquisition time point. For the prediction model of quality parameters, For input data, This includes structured data tables and environmental feature vectors from the current data collection time point and the previous three data collection time points. These are the weight parameters of the prediction model for quality parameters.

[0011] Optionally, S3 includes: S31, acquiring the transmit power of the base station under test; S32, selecting signal quality parameter prediction data for M preset acquisition locations from the prediction data of multiple signal quality parameters. M is an integer, and S33, Obtain the interference index for M preset sampling locations based on the received power parameters of the filtered reference signal. The interference index represents the strength of interference signals from other base stations, which are the base stations other than the base station under test among those that can receive signals at the current location. S34, According to formula... Determine the multi-objective function J. Where M is the number of preset acquisition locations. It is the predicted data of the signal quality parameters of the i-th preset acquisition position among M preset acquisition positions. It is the threshold of the signal quality parameter. Let be the transmit power of the j-th base station. Let B be the total number of base stations to be tested. B is an integer, and... It is the interference index of the i-th preset acquisition position among M preset acquisition positions. These are the weighting coefficients.

[0012] Optionally, S4 includes: S41, using... Calculate the partial derivatives of each parameter in at least one parameter of the multi-objective function J for the base station under test. Wherein, This represents any one of at least one parameters of the base station under test. For parameters The applied small perturbation, Indicates parameters Given the current value, the value of the multi-objective function J is... Indicates parameters Adding small perturbations Then, the values ​​of the multi-objective function J are determined. S42, the hyperparameters are determined. and according to the formula Calculate the optimization amount of each parameter in at least one parameter of the base station under test. .

[0013] Optionally, S42 includes: acquiring historical adjustment data of each parameter in at least one parameter of the base station under test, and determining hyperparameters based on the historical adjustment data of each parameter in at least one parameter of the base station under test. .

[0014] Optionally, S5 includes: determining the constraint configuration of the base station under test, and adjusting each parameter of at least one parameter of the base station under test according to the constraint configuration of the base station under test and the optimization amount of each parameter in at least one parameter of the base station under test. The constraint configuration of the base station under test includes the transmit power range, the adjustable range of the antenna tilt angle, and the adjustable range of the antenna azimuth angle.

[0015] Secondly, this disclosure provides a network quality dynamic optimization device, which includes an information acquisition module, a prediction module, a multi-objective function determination module, an optimization quantity acquisition module, and a parameter adjustment module. The information acquisition module acquires network quality-related information of the test area collected by the UAV. The test area is determined based on the base station to be tested. The network quality-related information includes the current acquisition time, geographical information of the current flight position, signal data of the current flight position, and environmental data of the current flight position. The geographical information includes the actual flight altitude and horizontal coordinates. The signal data includes reference signal received power parameters, signal-to-interference-plus-noise ratio parameters, and transmission delay parameters. The environmental data includes obstacle image data, wind speed data, and temperature data. The prediction module preprocesses the network quality-related information to obtain preprocessed network quality-related information, and predicts signal quality parameters based on the preprocessed network quality-related information, obtaining predicted data for multiple signal quality parameters. The multi-objective function determination module acquires the transmission power of the base station to be tested and determines a multi-objective function based on the transmission power of the base station to be tested, the predicted data of multiple signal quality parameters, and the preprocessed network quality-related information. The multi-objective function balances three objectives: signal coverage, base station power consumption, and signal interference. The optimization quantity acquisition module determines the optimized quantity of each parameter in at least one parameter of the base station under test based on the multi-objective function. The at least one parameter of the base station under test includes at least one of the following: antenna tilt angle, antenna azimuth angle, and transmit power. The parameter adjustment module adjusts each parameter of the at least one parameter of the base station under test based on the optimized quantity of each parameter.

[0016] Thirdly, this disclosure provides an electronic device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores one or more computer programs executable by the at least one processor, the one or more computer programs being executed by the at least one processor to enable the at least one processor to perform the above-described network quality dynamic optimization method.

[0017] Fourthly, this disclosure provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the above-described dynamic network quality optimization method.

[0018] Fifthly, this disclosure provides a computer program product that includes computer-readable code or a non-volatile computer-readable storage medium carrying computer-readable code, wherein when the computer-readable code is run in a processor of an electronic device, the processor in the electronic device executes the above-described dynamic network quality optimization method.

[0019] The embodiments provided in this disclosure can collect low-altitude signal data in real time using drones, combine it with environmental information, predict signal quality, and dynamically adjust base station parameters to achieve automated optimization.

[0020] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0021] The accompanying drawings are provided to further illustrate the present disclosure and form part of the specification. They are used together with the embodiments of the present disclosure to explain the disclosure and do not constitute a limitation thereof. The above and other features and advantages will become more apparent to those skilled in the art from the detailed description of exemplary embodiments with reference to the accompanying drawings, in which: Figure 1 A flowchart of a network quality dynamic optimization method provided in this embodiment of the disclosure; Figure 2 A flowchart of another method for dynamic network quality optimization provided in this embodiment of the disclosure; Figure 3 A schematic diagram illustrating the flight range of a drone in a network quality dynamic optimization method provided in this embodiment of the present disclosure; Figure 4 A flowchart illustrating yet another method for dynamic network quality optimization provided in this embodiment of the disclosure; Figure 5 A flowchart illustrating yet another method for dynamic network quality optimization provided in this embodiment of the disclosure; Figure 6 A flowchart of yet another method for dynamic network quality optimization provided in this disclosure embodiment; Figure 7 A flowchart illustrating yet another method for dynamic network quality optimization provided in this embodiment of the disclosure; Figure 8A schematic diagram illustrating the interaction between a drone test terminal and a remote control device in an example of a network quality dynamic optimization method provided in this embodiment of the present disclosure. Figure 9 A block diagram of a network quality dynamic optimization device provided in this embodiment of the present disclosure; Figure 10 This is a block diagram of an electronic device provided in an embodiment of the present disclosure. Detailed Implementation

[0022] To enable those skilled in the art to better understand the technical solutions of this disclosure, exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments of this disclosure to aid understanding. These should be considered merely exemplary. Therefore, those skilled in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0023] Where there is no conflict, the various embodiments of this disclosure and the features thereof in the embodiments may be combined with each other.

[0024] As used herein, the term “and / or” includes any and all combinations of one or more related enumerated entries.

[0025] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit this disclosure. As used herein, the singular forms “a” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will also be understood that when the terms “comprising” and / or “made of” are used in this specification, the presence of the stated feature, integral, step, operation, element, and / or component is specified, but the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or groups thereof is not excluded. Words such as “connected” or “linked” are not limited to physical or mechanical connections but can include electrical connections, whether direct or indirect.

[0026] Unless otherwise specified, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art. It will also be understood that terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant art and this disclosure, and will not be interpreted as having an idealized or overly formal meaning, unless expressly so defined herein.

[0027] The network quality dynamic optimization method according to embodiments of this disclosure can be executed by an electronic device such as a terminal device (e.g., a test terminal) or a server. The terminal device can be an in-vehicle device, user equipment (UE), mobile device, user terminal, terminal, cellular phone, cordless phone, personal digital assistant (PDA), handheld device, computing device, wearable device, etc. The method can be implemented by a processor calling computer-readable program instructions stored in memory. Alternatively, the method can be executed by a server.

[0028] like Figure 1 As shown, the method includes steps S1 to S5.

[0029] Step S1: Obtain network quality-related information of the area to be tested collected by the drone.

[0030] In step S1, the area to be tested is determined based on the base station to be tested. Network quality-related information includes the current data collection time, geographical information of the current flight location, signal data of the current flight location, and environmental data of the current flight location. Geographical information includes the actual flight altitude and horizontal coordinates. Signal data includes reference signal received power parameters, signal-to-interference-plus-noise ratio parameters, and transmission delay parameters. Environmental data includes obstacle image data, wind speed data, and temperature data.

[0031] In some embodiments, such as Figure 2 As shown, step S1 includes steps S11 to S13.

[0032] Step S11: Set the test area according to the location of the base station to be tested.

[0033] For example, a test area can be set with the location of a base station to be tested as the center, and the test area can be a cylindrical area.

[0034] Step S12: Determine the center of the UAV's flight range based on the location of the base station to be tested, and determine multiple flight altitudes and corresponding flight radii based on the height of ground obstacles.

[0035] Generally, there is one base station to be tested, and the projection of the center of the drone's flight range onto the ground is the location of the base station. If there are multiple base stations to be tested, the projection of the center of the drone's flight range can be the center of the locations of multiple base stations (for example, the midpoint of the line connecting two base stations, or the center point of the triangle formed by three base stations).

[0036] For example, such as Figure 3As shown, the flight altitude can be set based on low-altitude environmental characteristics (such as obstacles like buildings and trees) to ensure coverage and determine the flight range. For example, flight altitudes can be set to 10 meters, 20 meters, 40 meters, and 60 meters. The flight altitude setting depends on the height of ground obstacles, the location of the base station, and actual application requirements.

[0037] The following is an example of the correspondence between flight altitude and radius: Flight altitude is 0-10 meters: suitable for low-altitude complex environments, with a short path, avoiding high obstacles, and usually planned with the base station as the center and a radius of 50 meters.

[0038] Flight altitude is 10-20 meters: At this altitude, the impact of obstacles is reduced, and the path can be slightly expanded, with a planning range of 150-200 meters.

[0039] Flight altitude is 20-40 meters: the signal is mainly transmitted by line of sight, with a path range of 200-500 meters.

[0040] Flight altitude of 40-60 meters: At this altitude, there are almost no obstacles to interfere with, and a path with a radius greater than 500 meters can be set.

[0041] Step S13: Control the drone to fly within its flight range and acquire network quality information related to the area under test collected by the drone at a preset cycle.

[0042] For example, after determining multiple flight altitudes and the corresponding flight radii for each flight altitude, the drone can be controlled to fly one circle at a constant speed each time, and the collected data can be processed before flying again at the same speed at another flight altitude and corresponding radius.

[0043] Understandably, network quality-related information includes the current data acquisition time, geographic information of the current flight location, signal data of the current flight location, and environmental data of the current flight location. Geographic information includes the actual flight altitude and horizontal coordinates. Signal data includes reference signal received power parameters, signal-to-interference-plus-noise ratio parameters, and transmission delay parameters. Environmental data includes obstacle image data, wind speed data, and temperature data.

[0044] The following is an example illustrating how drones can collect samples: The actual flight altitude can be measured in real time using the drone's altimeter (such as a barometer or laser rangefinder), indicating the drone's vertical position relative to the ground. At low altitudes (3-100 meters), the signal is easily blocked by obstacles, while at high altitudes (100-300 meters), the signal mainly propagates via line-of-sight.

[0045] The horizontal coordinate position can be obtained using the Global Positioning System (GPS) based on Real-Time Kinematic (RTK) technology, with an accuracy down to the centimeter level, for precisely identifying the geographical location of the UAV. The coordinate origin is usually set to the location of the base station being measured in order to calculate the relative distance.

[0046] The Reference Signal Receiving Power (RSRP, unit: dBm) parameter can be measured by the 5G communication module in the UAV test module, representing the received signal strength. Its value range is typically [value range missing]. The higher the value, the stronger the signal.

[0047] The Signal to Interference plus Noise Ratio (SINR, in dB) parameter can be measured by the 5G communication module in the UAV test module to evaluate signal quality. Typical range A higher value indicates better quality.

[0048] Transmission delay parameters include transmission delay data and jitter data. Transmission delay data (in milliseconds) can be measured using Ping tests or User Datagram Protocol (UDP) round-trip time, reflecting network response speed. Jitter data (in milliseconds) is used to view the rate of change of delay, and can be obtained by calculating the standard deviation of consecutive delay values, representing network stability.

[0049] Obstacle image data can be acquired using high-definition cameras (such as 4K resolution) mounted on drones for visual analysis of environmental obstacles. For example, the camera can record video at a rate of 30 frames per second and extract keyframes (such as 1 frame per second) for processing.

[0050] Wind speed and temperature data can be measured by weather sensors on the drone, with wind speed ranging from [missing information]. Temperature range These data can affect signal propagation (e.g., wind speed may cause slight antenna movement).

[0051] Understandably, the acquisition frequency of geographic information, signal data, and environmental data should be kept consistent to ensure data synchronization and time alignment. The camera can capture obstacle image data and transmit it in real time or temporarily store it in the drone's storage module (built-in SD card), then upload it in batches after testing to save bandwidth. Image transmission prioritizes lossy compression (such as JPEG, quality factor 80%) to balance size and detail.

[0052] For example, after the data collection is completed, a data quality report can be generated, such as a brief statistical report (e.g., mean, variance) and a data loss rate, to monitor the data collection process. If the data quality is below a threshold (e.g., loss rate > 5%), the data quality is considered poor and the data can be collected again.

[0053] Understandably, the test area can be further subdivided to ensure comprehensive and efficient base station signal testing. Specifically, the test area (assumed to be a rectangular or circular region) can be divided into multiple three-dimensional grids, each with a side length of 10 meters (horizontal XY plane) and 10 meters (vertical Z-axis). The smaller the spacing between the grids, the higher the testing accuracy, which is especially suitable for complex terrain or densely built-up areas. After the drone collects the signal, the data for each grid can be recorded to observe and display the signal situation, facilitating the observation of the signal strength of each grid.

[0054] Step S2: Preprocess the network quality-related information to obtain preprocessed network quality-related information, and predict the signal quality parameters based on the preprocessed network quality-related information to obtain prediction data for multiple signal quality parameters.

[0055] Understandably, all data obtained in step S1 can be packaged and transmitted in real time to the test terminal via the drone's communication module (e.g., via 5G), using the MQTT protocol to ensure low latency and reliability. The data packets are in JSON format and include timestamps, location, signal indicators, and environmental readings.

[0056] In some embodiments, such as Figure 4 As shown, step S2 includes steps S21 to S25.

[0057] Step S21: Remove invalid data from the reference signal received power parameter, signal-to-interference-plus-noise ratio parameter, and transmission delay parameter respectively, and perform interpolation processing to obtain the interpolated reference signal received power parameter, interpolated signal-to-interference-plus-noise ratio parameter, and interpolated transmission delay parameter.

[0058] Understandably, there may be acquisition errors during the data acquisition process. In order to avoid the impact of errors on subsequent processing, invalid data caused by errors can be removed, and interpolation can be used to fill in the removed data.

[0059] Step S22: Filter the interpolated reference signal received power parameter and the interpolated signal-to-interference-plus-noise ratio parameter respectively to obtain the filtered reference signal received power parameter and the filtered signal-to-interference-plus-noise ratio parameter.

[0060] For example, for the RSRP and SINR parameters, a moving average filter can be used to smooth high-frequency noise. The filtered data reduces random errors and improves the stability of subsequent analysis.

[0061] Step S23: Extract obstacle index from obstacle image data, and preprocess wind speed data and temperature data respectively to obtain preprocessed wind speed data and preprocessed temperature data.

[0062] In step S23, preprocessing includes removing invalid data, interpolation, and normalization. The preprocessed network quality-related information includes the filtered reference signal received power parameter, the filtered signal-to-interference-plus-noise ratio parameter, the interpolated transmission delay parameter, the obstacle index, the preprocessed wind speed data, and the preprocessed temperature data.

[0063] In some embodiments, in step S23, obstacle indices are extracted from the obstacle image data according to equation (1). .

[0064] (1)

[0065] In equation (1), It is the number of occluded pixels in the obstacle image data. It is the total number of pixels in the obstacle image data.

[0066] Understandably, methods for extracting obstacle indices from image data can quantify the degree of environmental occlusion.

[0067] For example, image segmentation can be performed using a pre-trained Convolutional Neural Network (CNN) model (such as ResNet-18). The model weights can be obtained by training on a historical dataset (containing labeled low-altitude environment images). The training uses the cross-entropy loss function and the Adam optimizer. The model input is an RGB image (i.e., an obstacle image), and the output is a pixel-level classification (e.g., buildings, trees, sky). Occlusion pixels refer to pixels classified as obstacles (e.g., buildings, trees). The obstacle index can then be calculated using Equation (1). Understandably, the obstacle index Dimensionless, with a value range of [0,1].

[0068] Step S24: The geographical information of the acquisition time point, the current flight position, the filtered reference signal received power parameters, the filtered signal-to-interference-plus-noise ratio parameters, and the interpolated transmission delay parameters are made into a structured data table, and the obstacle index, the preprocessed wind speed data, and the preprocessed temperature data are made into an environmental feature vector.

[0069] For example, each row of the structured data table includes a timestamp (obtained from the time of collection), location coordinates (h,x,y) (obtained from the geographic information of the current flight location), filtered RSRP (in dBm), filtered SINR (in dB), and interpolated transmission delay parameter (milliseconds). The data is stored in CSV format for easy subsequent analysis.

[0070] Step S25: Based on the structured data table and environmental feature vector of the current acquisition time point and the previous three acquisition time points, predict the signal quality parameters of the next acquisition time point to obtain the predicted data of the signal quality parameters of the next acquisition time point.

[0071] In step S25, the predicted data for the signal quality parameters at the next acquisition time point includes the predicted data for the parameter signal-to-interference-plus-noise ratio parameter. The predicted data for multiple signal quality parameters includes the predicted data for the signal quality parameters at multiple acquisition time points.

[0072] In some embodiments, such as Figure 5 As shown, step S25 includes steps S251 to S253.

[0073] Step S251: Construct a prediction model based on the Long Short-Term Memory (LSTM) network.

[0074] In step S251, the architecture of the prediction model includes two LSTM layers and one fully connected output layer.

[0075] Understandably, Long Short-Term Memory (LSTM) networks are used because they can effectively handle long-term dependencies in sequential data, are suitable for the temporal correlation of signal data, and, when the model input is time series data, can output future... Time (for example, The predicted value of ).

[0076] Step S252: Obtain historical data from the structured data table and historical data from the environmental feature vectors, and train the prediction model based on the historical data from the structured data table and historical data from the environmental feature vectors to obtain the prediction model for the quality parameters.

[0077] Step S253: Make a prediction according to equation (2) to obtain the predicted data of the signal quality parameters at the next acquisition time point.

[0078] (2)

[0079] In equation (2), It is the predicted data of signal quality parameters at the next acquisition time point. For the prediction model of quality parameters, For input data, This includes structured data tables and environmental feature vectors from the current data collection time point and the previous three data collection time points. These are the weight parameters of the prediction model for quality parameters.

[0080] For example, in equation (1), the time interval can be 5 seconds, because in low-altitude environments, signal changes are mainly caused by UAV movement and environmental factors, and a 10-second window can cover typical change cycles while maintaining predictive practicality. The prediction model is trained by supervision, with training data from historical data of structured data tables and historical data of environmental feature vectors, using the mean squared error loss function as shown in equation (3).

[0081] (3)

[0082] In equation (3), This represents the mean squared error. i represents the current sample number, and N represents the total number of samples. These are the actual data of the signal quality parameters for the i-th sample. It is the predicted data of the signal quality parameters of the i-th sample.

[0083] The optimizer used is Adam, with a learning rate of 0.001 and a training cycle of 100 epochs (one epoch refers to a complete traversal and training process of the entire training dataset). The model architecture consists of two LSTM layers (128 units each) and one fully connected output layer. After training, the model weights W are fixed and deployed on the test terminal, and are incrementally updated monthly with new data to maintain accuracy.

[0084] Step S3: Obtain the transmit power of the base station under test, and determine a multi-objective function based on the transmit power of the base station under test, the predicted data of multiple signal quality parameters, and the preprocessed network quality related information.

[0085] In step S3, the multi-objective function is used to balance the three objective items: signal coverage, base station power consumption, and signal interference.

[0086] Understandably, defining a multi-objective function J is to quantify the optimization objectives and balance coverage, power consumption, and interference. The core purpose is to find a balance point among multiple conflicting objectives, ensuring good signal coverage, saving base station power consumption, and reducing signal interference.

[0087] In some embodiments, such as Figure 6 As shown, step S3 includes steps S31 to S34.

[0088] Step S31: Obtain the transmit power of the base station under test.

[0089] For example, the transmit power of the base station under test can be obtained from the configuration information of the base station under test.

[0090] Step S32: Select the predicted signal quality parameters for M preset acquisition locations from the predicted data of multiple signal quality parameters.

[0091] In step S32, M is an integer and M≥1.

[0092] Understandably, M represents the number of preset data collection points (e.g., 100) determined based on the drone's flight path, ensuring comprehensive coverage. Each monitoring point corresponds to a geographic coordinate (e.g., longitude, latitude, and altitude), determined based on the drone's real-time position while flying at low altitudes (3-300 meters). Monitoring points are not fixed physical devices, but rather sampling points dynamically generated through the drone's flight trajectory, used to comprehensively cover the tested area. The main function of the monitoring points is to collect signal data (e.g., RSRP, SINR) and interference information so that the system can calculate coverage quality and identify problem areas. The number of M can be set to 100; this is an empirical value based on drone flight efficiency and a balance of computational resources. Too many points (e.g., M>200) will increase processing latency, while too few points (e.g., M<50) may miss critical areas. The location of the monitoring points is determined by the drone's flight path: for example, when the drone flies along a layered circular path, the system will sample evenly along the path, determining M points from many sampling points. Each point represents a "test location," where signal data is collected in real-time by a test module on the drone. Assuming a drone flies in a circle with a radius of 200 meters around a base station at an altitude of 50 meters, the system could set up a monitoring point every 2 meters (approximately 100 points in total) to comprehensively assess coverage at that altitude. The monitoring point data helps identify weak signal areas (such as areas with a SINR below 5 dB) and guides antenna adjustments.

[0093] Step S33: Obtain the interference index of M preset acquisition locations based on the received power parameters of the filtered reference signal.

[0094] In step S33, the interference index represents the strength of interference signals from other base stations, which are base stations other than the base station under test that can receive signals at the current location.

[0095] Understandably, in order to maintain data synchronization, it is necessary to ensure that the interference index is also the data from the M preset acquisition locations. For example, the interference index of the i-th preset acquisition location among the M preset acquisition locations can be determined according to equation (4). .

[0096] (4)

[0097] In equation (4), This refers to the sum of all data in the filtered reference signal received power parameters at the i-th preset acquisition location out of M preset acquisition locations, excluding the reference signal received power of the base station under test. The j-th base station is the base station under test, and the k-th base station is any base station other than the base station under test that can receive signals at the i-th preset acquisition location out of the M preset acquisition locations (the number of k-th base stations can be one, multiple, or zero). This refers to the reference signal received power data from base station k in the filtered reference signal received power parameters at the i-th preset acquisition position out of M preset acquisition positions.

[0098] S34, determine the multi-objective function J according to equation (5).

[0099] (5)

[0100] In equation (5), It is the predicted data of the signal quality parameters of the i-th preset acquisition position among M preset acquisition positions. It is the threshold of the signal quality parameter. Let be the transmit power of the j-th base station. Let B be the total number of base stations to be tested. B is an integer, and... . It is the interference index of the i-th preset acquisition position among M preset acquisition positions. These are the weighting coefficients.

[0101] Understandably, in equation (5), the first term on the right-hand side measures whether the signal quality is sufficient, the second part measures how much power the base station consumes, and the third part measures the magnitude of signal interference. The smaller the value of the multi-objective function J, the better the overall network condition. The goal of the optimization algorithm is to minimize J, thereby automatically adjusting the parameters of the base station under test.

[0102] In equation (5), the first term on the right side of the equal sign, In This means only positive values ​​are considered; if X is a negative number, In other words, the first part is only penalized (because its value will increase) when the predicted value is below the threshold. If the predicted value is above the threshold, it indicates good coverage, and the penalty is 0. For example, the threshold... It can be set to 5 dB if the predicted value A reading below 5 dB indicates insufficient coverage, and users may experience lag or disconnections in this area.

[0103] In equation (5), the second term on the right side of the equal sign, This represents the transmit power (in dBm) of the j-th base station. A higher value indicates higher power consumption and operating costs. The second term on the right-hand side of the equation represents the summation over all base stations (B). If B=1, it means optimizing only a single base station, based on a common scenario where drone testing often revolves around a core base station. If B>1, it indicates collaborative optimization of multiple base stations to reduce interference or improve coverage efficiency. For example, if the drone's flight area involves overlapping coverage from multiple base stations, the system will include these base stations in the optimization (B=2 or 3) to calculate total power consumption and interference. In most scenarios, B=1 (single base station optimization), but this disclosure also supports multi-base station scenarios. B represents the number of base stations participating in the optimization process, which are the monitored and adjusted base station entities. These base stations may include one or more devices, depending on the optimization scope.

[0104] Understandably, the weights α, β, and γ can be determined through grid search and experimentation: α emphasizes coverage (prioritizing SINR), β controls power consumption (reducing operating costs), and γ reduces interference (improving network stability). The weights can be dynamically adjusted based on network strategies (e.g., increasing β in power-saving mode).

[0105] Step S4: Determine the optimization amount of each parameter in at least one parameter of the base station under test according to the multi-objective function.

[0106] In step S4, at least one parameter of the base station under test includes at least one of the antenna tilt angle, antenna azimuth angle, and transmit power of the base station under test.

[0107] Understandably, among various parameter combinations, how to quickly find a direction to guide how to fine-tune the tilt angle, azimuth angle, and transmit power of the base station to improve the overall performance of the network? The specific method is to reduce the value of the objective function J. Based on the predicted value, the gradient descent method can be used to calculate the optimization amount of at least one parameter of the base station under test when the objective function J reaches its minimum value.

[0108] Understanding this, the gradient of a function is a vector pointing in the direction of the fastest increase in function value. Conversely, the opposite direction of the gradient points in the direction of the fastest decrease in function value. Therefore, by fine-tuning the parameters along the opposite direction of the gradient, the value of the multi-objective function J can be reduced. This is the basic idea behind gradient descent.

[0109] The multi-objective function J is a complex function. Its input variables are the base station parameters, and its output is a numerical value representing the overall network performance. It is difficult to directly write its mathematical expression and then find its derivative, but the partial derivatives can be approximated using the numerical difference method. To know the effect of changing a certain parameter on the multi-objective function J, make a slight change to it and see how much the value of the multi-objective function J changes.

[0110] The gradient descent method was chosen because it does not depend on the specific mathematical form of the multi-objective function J. Regardless of the complexity of the mathematical form of J, as long as its value at a certain point can be calculated, the gradient can be estimated using numerical differencing. This is highly suitable for complex, nonlinear systems like 5G networks. The entire process is a deterministic algorithm, easily implemented through coding, and enables fully automated optimization decisions without human intervention. The computational cost is relatively low. For three parameters, only four calculations of the multi-objective function J are needed (the current point and the points where the three parameters are increased by δ) to estimate the entire gradient, making it ideal for real-time or near-real-time optimization requirements.

[0111] In some embodiments, such as Figure 7 As shown, step S4 includes steps S41 to S42.

[0112] Step S41: Use equation (6) to calculate the partial derivatives of each parameter in at least one parameter of the multi-objective function J for the base station under test.

[0113] (6)

[0114] In equation (6), This represents any one of at least one parameters of the base station under test. For parameters The applied small perturbation, Indicates parameters Given the current value, the value of the multi-objective function J is... Indicates parameters Adding small perturbations Then, the value of the multi-objective function J is determined.

[0115] Understandably, For parameters The applied perturbation is a very small value (e.g., 0.1°). It cannot be too large (which would lose accuracy) nor too small (which would introduce numerical calculation errors). This value is an empirical value determined through numerous experiments. The entire fraction represents how much the objective function J changes for every unit increase in the inclination angle. If the result is positive, it means that increasing the inclination angle will increase J (the situation worsens); if it is negative, it means that increasing the inclination angle will decrease J (the situation improves).

[0116] Step S42, determine hyperparameters And according to equation (7), the optimization amount of each parameter in at least one parameter of the base station under test is calculated respectively. .

[0117] (7)

[0118] Understandably, calculating the partial derivatives of the three parameters according to equation (6) is equivalent to obtaining the gradient. Then, taking a small step in the opposite direction of the gradient, we can obtain the adjustment amount of the parameters. The negative sign on the right side of equation (7) is equivalent to adjusting in the opposite direction of the gradient to ensure that the value of the multi-objective function J decreases.

[0119] In some embodiments, the hyperparameters are determined in step S42. The method includes: obtaining historical adjustment data of each parameter in at least one parameter of the base station under test, and determining hyperparameters based on the historical adjustment data of each parameter in at least one parameter of the base station under test. .

[0120] Understandably, hyperparameters This is a very important hyperparameter (e.g., set to 0.1). It determines the size of each step. If the hyperparameter... If the hyperparameters are too large, the step size might be too large, potentially exceeding the minimum point and even causing the value of the multi-objective function J to oscillate and fail to converge. If the hyperparameters... If the learning rate is too small, the step size will be too small, resulting in very slow convergence and low optimization efficiency. The value of the learning rate η needs to be determined through multiple experiments and adjustments to achieve the best balance between convergence speed and stability.

[0121] For example, historical adjustment data of each parameter in at least one parameter of the base station under test can be recorded and stored in a database as historical optimization data. The historical adjustment data may include: past base station parameter adjustment records (such as historical adjustment data of tilt angle, azimuth angle, and power) and historical signal data and their corresponding optimization effects (such as SINR improvement and power consumption changes) for reference. The data covers the past 30 days to capture seasonal and environmental changes.

[0122] Step S5: Adjust each parameter of the base station under test according to the optimization amount of each parameter in at least one parameter of the base station under test.

[0123] In some embodiments, the implementation method of step S5 includes: determining the constraint configuration of the base station under test, and adjusting each parameter of at least one parameter of the base station under test according to the constraint configuration of the base station under test and the optimization amount of each parameter in at least one parameter of the base station under test. The constraint configuration of the base station under test includes the transmit power range, the adjustable range of the antenna tilt angle, and the adjustable range of the antenna azimuth angle.

[0124] Understandably, the constraint configuration of the base station under test can be retrieved from the network management system. The transmit power range of the base station under test includes the maximum transmit power (e.g., 46 dBm) and the minimum transmit power (e.g., 20 dBm). The adjustable range of the antenna tilt angle includes the interval [-15°, 15°], and the adjustable range of the antenna azimuth angle includes the interval [0°, 360°]. These constraints ensure that optimization adjustments are made within feasible limits, avoiding equipment damage or violations.

[0125] For example, the antenna tilt angle can be adjusted using equation (8).

[0126] (8)

[0127] In equation (8), It is the current data of each parameter in at least one parameter of the base station under test. Yes, at least one of the parameters is the optimized data for each parameter. In the test, the hyperparameters... When the value is 0.1, stable convergence can be achieved in most scenarios, avoiding oscillations (such as repeated adjustments). Therefore, the hyperparameter... It can be set to 0.1.

[0128] Understandably, the optimized data needs to meet constraints. Taking the transmission power as an example, this can be achieved through the function shown in equation (9).

[0129] (9)

[0130] In equation (9), This is the minimum transmit power of the base station under test. This represents the maximum transmit power of the base station under test.

[0131] In some embodiments, during the first preset time period after the adjustment, a drone can collect a real-time signal according to the current collection trajectory and determine whether the improvement of the signal quality parameter is greater than a preset threshold. If the improvement of the signal quality parameter is greater than the preset threshold, the current adjustment can be maintained; otherwise, it can be rolled back or further optimized.

[0132] The rollback mechanism includes: restoring the original parameters of the base station under test, marking the adjustment as a failure, and re-executing steps S1 to S5.

[0133] For example, based on actual measurements, a 0.5 dB change in a 5G network typically indicates a significant improvement, and the preset threshold can be set to 0.5 dB.

[0134] For example, the reduction in total transmit power can also be calculated to see the power savings after optimization. Equation (10) can also be used to calculate the interference reduction rate, thus providing a visual representation of the interference index. The changes.

[0135] (10)

[0136] In formula (10) It is the number of points where the current SINR is greater than the threshold. It is the number of points where the SINR is greater than the threshold after the last optimization.

[0137] Understandably, if any of the following parameters—signal quality, base station power consumption, and signal interference—deteriorates beyond a preset level, performance degradation can be identified. In this case, it is possible to check whether the environment has changed. For example, if the obstacle index increases by 0.2, the performance degradation may be due to environmental changes, and further adjustments can be made accordingly.

[0138] The network quality dynamic optimization method provided in this disclosure uses UAVs to collect low-altitude signal data in real time, combines it with environmental information, uses machine learning algorithms to predict signal quality, and dynamically adjusts base station parameters to achieve automated optimization. A Long Short-Term Memory (LSTM) network is used to predict the SINR value for future seconds. Input features include historical signal data, environmental indicators, and location information. The model is trained using historical data, and the weights are updated periodically. In 5G low-altitude optimization, LSTM is applied for real-time prediction, capturing the time-series characteristics of the signal, which is superior to traditional statistical methods. A multi-objective function J is used to balance coverage, power consumption, and interference, transforming the multi-objective optimization into a computable scalar. Dynamic trade-offs are achieved through weight coefficients, which are optimized based on grid search and measured data. The gradient descent method is used to minimize the value of the multi-objective function J, and the near-partial derivatives are calculated using numerical differencing to calculate the parameter optimization quantities.

[0139] The following example illustrates a method for dynamic optimization of network quality provided in this disclosure.

[0140] This example demonstrates a method and system for optimizing low-altitude coverage by dynamically adjusting 5G network parameters using intelligent algorithms based on real-time signal data collected by drones. This method is applicable to low-altitude scenarios (3-300 meters altitude) for cellular networks (such as 5G NR networks) and aims to address the problems of manual intervention, low efficiency, and high response latency in the post-test optimization process for low-altitude coverage. Details are as follows: Step 1: Flight Grid Planning Flight area grid division: The test area is finely divided to ensure comprehensive and efficient base station signal testing. Specifically, the area to be tested (assumed to be a rectangular or circular area) is divided into multiple three-dimensional grids, each with a side length of 10 meters (horizontal XY plane) and 10 meters (vertical Z-axis). The smaller the spacing between the grids, the higher the test accuracy, especially suitable for complex terrain or densely built-up areas. The flight altitude is set according to low-altitude environmental characteristics (e.g., buildings, trees, and other obstacles) to ensure coverage and determine the flight range.

[0141] Specifically, the flight altitude is set (e.g., 10 meters, 20 meters, 40 meters, and 60 meters). The flight altitude is set based on the height of ground obstacles, the location of the base station, and actual application requirements.

[0142] 0-10 meters: Suitable for complex low-altitude environments, with shorter paths, avoiding high obstacles, and usually planned with a radius of 50 meters around the base station.

[0143] 10-20 meters: At this point, the impact of obstacles is reduced, and the path can be slightly expanded, with a planning range of 150-200 meters.

[0144] 20-40 meters: The signal mainly propagates at line of sight, with a path range of 200-500 meters.

[0145] 40-60 meters: At this distance, there are almost no obstacles to interfere with the path, and a path with a radius greater than 500 meters can be set.

[0146] Step 2: Data Acquisition and Preprocessing

[0147] Data acquisition and preprocessing are the foundational steps of the entire optimization process. The purpose is to acquire signal data (such as SINR) and environmental information (such as obstacles and weather conditions) in the low-altitude environment in real time and comprehensively through the UAV system to reflect the true state of network coverage. These raw data are then preliminarily processed to reduce measurement errors and environmental interference, thereby ensuring the accuracy and reliability of subsequent analysis and optimization.

[0148] The data acquisition process is as follows: The UAV autonomously flies along a predefined layered circular path, ensuring uniform flight speed (e.g., a default speed of 2 m / s) to maintain data consistency. During flight, the UAV test module continuously collects signal data, including: The drone's flight parameters include: Flight altitude h (unit: meters): This is measured in real time by the drone's altimeter (such as a barometer or laser rangefinder), indicating the drone's vertical position relative to the ground. At low altitudes (3-100 meters), the signal is easily blocked by obstacles, while at high altitudes (100-300 meters), the signal mainly propagates via line of sight.

[0149] Horizontal position coordinates (x, y): obtained based on an RTK (Real-Time Kinematic) GPS positioning system, with an accuracy down to the centimeter level, used to accurately identify the geographical location of the drone. The origin of the coordinates is usually set to the location of the base station being measured in order to calculate the relative distance.

[0150] These parameters are sampled once per second to ensure the continuity of spatial coverage.

[0151] Real-time signal data, including: Reference Signal Received Power (RSRP, unit: dBm): Measured by the 5G communication module in the UAV test module, it represents the received signal strength. The value typically ranges from -140 dBm to -40 dBm; a higher value indicates a stronger signal.

[0152] Signal-to-interference-plus-noise ratio (SINR, in dB): Used to evaluate signal quality. Typical range is -10 dB to 30 dB, with higher values ​​indicating better quality.

[0153] Transmission delay (unit: milliseconds): Measured by Ping test or UDP packet round-trip time, reflecting network response speed.

[0154] Jitter (unit: milliseconds): The rate of change of latency, obtained by calculating the standard deviation of consecutive latency values, and represents network stability.

[0155] These signal data are sampled at a rate of 100 milliseconds per sampling (i.e., 10 Hz) to capture rapidly changing signal characteristics.

[0156] Environmental data, including: Image data: Acquired via a high-definition camera (e.g., 4K resolution) mounted on the drone for visual analysis of environmental obstacles. The camera records video at 30 frames per second, and keyframes are extracted (e.g., 1 frame per second) for processing.

[0157] Wind speed and temperature (units: m / s and ℃): Measured by weather sensors on the drone, wind speed range 0-20 m / s, temperature range -20℃ to 60℃. These data affect signal propagation (e.g., wind speed may cause slight antenna movements).

[0158] The environmental data sampling rate is synchronized with the signal data to ensure time alignment. Environmental images are captured by the camera and transmitted in real time or temporarily stored in the drone's storage module (built-in SD card), and then uploaded in batches after testing to save bandwidth. Image transmission prioritizes lossy compression (such as JPEG, quality factor 80%) to balance size and detail.

[0159] Base station information, including: Base station location coordinates: obtained from the network operator's database, used to calculate the distance between the drone and the base station.

[0160] Current antenna parameters: including antenna tilt angle θ tilt (Unit: degrees, range -15° to +15°), azimuth angle azimuth (unit: degrees, range 0° to 360°) and transmit power Ptx (unit: dBm, range 20-46 dBm). These parameters are obtained from the base station network management platform.

[0161] All data is packaged and transmitted in real time to the test terminal via the drone's communication module (using 5G), with MQTT as the transmission protocol to ensure low latency and reliability. The data packets are in JSON format and include timestamps, location, signal strength, and environmental readings.

[0162] Next, the collected data will be preprocessed: For RSRP and SINR data, a moving average filter is used to smooth high-frequency noise. The filtered data reduces random errors and improves the stability of subsequent analysis.

[0163] Extracting the obstacle index O from image data index (Dimensionless, range 0-1), used to quantify the degree of environmental occlusion. The method is as follows: Image segmentation is performed using a pre-trained convolutional neural network (CNN) model (such as ResNet-18). The model input is an RGB image, and the output is a pixel-level classification (e.g., buildings, trees, sky). The calculation formula is shown in Equation (1). Here, occluded pixels refer to pixels classified as obstacles (e.g., buildings, trees). The model weights are obtained through training on a historical dataset (containing labeled low-altitude environmental images), using the cross-entropy loss function and the Adam optimizer.

[0164] Meanwhile, wind speed and temperature data are directly used as part of the environmental feature vector without additional processing, but unit normalization is performed to avoid numerical differences affecting the model.

[0165] The preprocessed output includes: The preprocessed signal dataset consists of a structured data table, with each row containing a timestamp, location coordinates (h, x, y), filtered RSRP (dBm), filtered SINR (dB), delay (milliseconds), and jitter (milliseconds). The data is stored in CSV format for easy subsequent analysis.

[0166] Environmental feature vector: includes obstacle index O index (Scalar), normalized wind speed (dimensionless), normalized temperature (dimensionless). These vectors are time-aligned with the signal data.

[0167] Data Quality Report: Generates brief statistics (such as mean, variance) and data loss rate to monitor the data collection process. If the data quality falls below a threshold (e.g., data loss rate > 5%), the system will automatically trigger a re-collection.

[0168] The output data is directly input into the prediction model in step two, without further processing.

[0169] Step 3: Signal Quality Prediction and Optimization Target Calculation

[0170] This is the core decision-making step in the entire optimization process. Using preprocessed data, machine learning models are used to predict future signal quality in order to anticipate network changes and avoid coverage blind spots or interference problems.

[0171] Define a multi-objective function to balance coverage, power consumption, and interference, quantify the optimization direction, provide quantitative targets for base station parameter adjustment, reduce manual intervention, and automatically generate optimization suggestions through algorithms to improve optimization speed and accuracy.

[0172] This step transforms data into decisions, ensuring that the optimization process is based on predictions and improving network efficiency.

[0173] The input data comes from the output of step two, historical optimization records, and network constraint configurations, and specifically includes the following parameters: The output of step two will not be described again.

[0174] Historical optimization records: Read from the central processing platform's database, including: Past base station parameter adjustment records (such as antenna tilt angle θ) tilt Antenna azimuth angle azimuth Power P tx (The history of change).

[0175] Historical signal data and its corresponding optimization effects (such as SINR improvement and power consumption changes) are used for model training and reference. The data covers the past 30 days to capture seasonal and environmental changes.

[0176] Network constraint configuration: Obtained from the network management system, including: Base station maximum transmit power Pmax (unit: dBm, default 46 dBm): hardware limitation.

[0177] Minimum base station transmit power Pmin (unit: dBm, default 20 dBm): Ensures basic coverage.

[0178] Antenna adjustable range: tilt angle θ tilt (Unit: degrees, range -15° to +15°), azimuth angle azimuth(Unit: degrees, range 0° to 360°)

[0179] These constraints ensure that optimization suggestions are within feasible limits and avoid equipment damage or violations.

[0180] Data flow involves transmission from the drone to the platform; its system architecture can be found in [reference needed]. Figure 8 This demonstrates how data is transmitted from the drone device to the test terminal (i.e., the central processing platform) via the remote controller.

[0181] Step three's method comprises two main parts: signal quality prediction using a machine learning model and optimization target calculation using a mathematical function. Specific implementation details are as follows: 1. Signal quality prediction: Long Short-Term Memory (LSTM) network is used for time series prediction. LSTM is chosen because it can effectively handle long-term dependencies in sequence data and is suitable for the temporal correlation of signal data. The model input is time series data, and the output is the SINR prediction value at a future time Δt (default Δt=5 seconds). The prediction formula is shown in Equation (2).

[0182] In equation (2), X(t) is the input feature vector, including the filtered RSRP, filtered SINR, delay, jitter, and O at the current time t. index Wind speed, temperature, and historical data from three time steps (t-1, t-2, t-3) are used to capture trends. LSTM This represents the LSTM model function, where W is the model weight parameter.

[0183] A 5-second time interval is appropriate because in low-altitude environments, signal changes are mainly caused by drone movement and environmental factors. A 10-second window can cover typical change cycles while maintaining predictive usability.

[0184] The LSTM model was trained using supervised learning. The training data came from a historical drone-collected dataset (containing millions of records), and the mean squared error (MSE) loss function was used as shown in Equation (3). The optimizer used was Adam with a learning rate of 0.001 and a training period of 100 epochs. The model architecture consisted of two LSTM layers (128 units per layer) and one fully connected output layer. After training, the model weights W were fixed and deployed on a central processing platform, and were updated monthly with new data increments to maintain accuracy.

[0185] 2. Optimize target calculation: A multi-objective function J is defined to quantify the optimization objectives and balance coverage, power consumption and interference. The core purpose is to find a balance point among multiple conflicting objectives: to ensure good signal coverage, save base station power consumption, and reduce signal interference. The function formula is shown in equation (5).

[0186] In equation (5), the first term on the right measures whether the signal quality is sufficient, the second part measures the power consumption of the base station, and the third part measures the magnitude of signal interference. The smaller the value of function J, the better the overall network condition. The goal of the optimization algorithm (gradient descent) is to minimize J, thereby automatically adjusting the base station parameters.

[0187] In equation (5), SINR pred,i It is the predicted future signal quality value (in dB), which is predicted by a machine learning model (LSTM), and represents the expected SINR at the i-th monitoring point.

[0188] In equation (5), (SINR) th -SINR pred,i ) + (x) + This means only positive values ​​are considered (if x is negative, it becomes 0). In other words, this portion is only penalized (the value increases) when the predicted SINR is below the threshold; if the SINR is above the threshold, it indicates good coverage, and the penalty is 0. th This is the SINR threshold, set to 5 dB. If the SINR is below 5 dB, users may experience lag or disconnections, indicating insufficient coverage.

[0189] In equation (5), M is the number of points determined based on the UAV's flight position (default 100) to ensure comprehensive coverage of the area. Each monitoring point corresponds to a geographic coordinate (such as longitude, latitude, and altitude), determined based on the real-time position of the UAV flying at low altitude (3-300 meters). Monitoring points are not fixed physical devices, but sampling points dynamically generated through the UAV's flight trajectory, used to comprehensively cover the tested area. The main function of the monitoring points is to collect signal data (such as RSRP, SINR) and interference information so that the system can calculate coverage quality and identify problem areas. The number of M is set to 100 by default, which is an empirical value based on the balance between UAV flight efficiency and computing resources. Too many points (such as M>200) will increase processing latency, while too few points (such as M<50) may miss key areas. The location of the monitoring points is determined by the UAV's flight path: for example, when the UAV flies along a layered circular path, the system will sample evenly along the path, determining M points from many sampling points. Each point represents a "test position," where signal data is collected in real time by the test module on the UAV. Assuming a drone flies in a circle with a radius of 200 meters around a base station at an altitude of 50 meters, the system could set up a monitoring point every 2 meters (approximately 100 points in total) to comprehensively assess coverage at that altitude. The monitoring point data helps identify weak signal areas (such as areas with a SINR below 5 dB) and guides antenna adjustments.

[0190] In equation (5), P tx,,jThe first term represents the transmit power (in dBm) of base station j. A higher value indicates higher power consumption and operating costs. The second term represents the summation over all base stations (B base stations). If B=1, it means optimizing only a single base station, based on common scenarios where drone testing often revolves around a core base station. If B>1, it means coordinating the optimization of multiple base stations to reduce interference or improve coverage efficiency. For example, if the drone's flight area involves overlapping coverage from multiple base stations, the system will include these base stations in the optimization (B=2 or 3) to calculate total power consumption and interference. In most scenarios, B=1 (single base station optimization), but the system supports multiple base stations. B represents the number of base stations participating in the optimization process; these are the monitored and adjusted base station entities, which may include one or more devices, depending on the optimization scope.

[0191] In equation (5), It is the interference index of point i, which represents the sum of interference signal strengths from other base stations at the i-th monitoring point. The calculation method is as shown in equation (4).

[0192] In equation (5), α, β, and γ are weighting coefficients, which are determined through grid search and experimentation: α emphasizes coverage (prioritizing SINR), β controls power consumption (reducing operating costs), and γ reduces interference (improving network stability). The weights can be dynamically adjusted based on network strategies (e.g., increasing β in power-saving mode).

[0193] Among various parameter combinations, how can we quickly find a direction, and how should we fine-tune the base station's tilt angle, azimuth angle, and transmit power to improve the overall network performance? The specific method is to reduce the value of the objective function J. The central processing platform uses predicted SINR... pred The value is automatically calculated using the gradient descent method, determining the parameter adjustment Δθ required by the base station to minimize the objective function J. tilt Δ azimuth ΔP tx .

[0194] The gradient of a function is a vector that points in the direction in which the function value increases the most. Therefore, the opposite direction of the gradient is the direction in which the function value decreases the most. Thus, by fine-tuning the parameters along the opposite direction of the gradient, the value of J can be reduced. This is the basic idea behind gradient descent.

[0195] The objective function J is a complex function. Its input variables are the parameters of the base station, and its output is a numerical value representing the overall performance of the network. It is difficult to directly write out its mathematical expression and then find its derivative. Therefore, the numerical difference method shown in equation (6) is used to approximate the partial derivative. If we want to know the effect of changing a certain parameter on J, we can slightly change it and see how much J will change.

[0196] δ is a very small perturbation value (e.g., 0.1°). It cannot be too large (which would lose accuracy) nor too small (which would introduce numerical calculation errors). This value is an empirical value determined through a large number of experiments. The meaning of the whole fraction is how much the objective function J will change for every unit increase in the inclination angle. If the result is positive, it means that increasing the inclination angle will make J larger (the situation worsens); if it is negative, it means that increasing the inclination angle will make J smaller (the situation improves). After calculating the partial derivatives of the three parameters, the gradient is obtained. Then, by taking a small step in the opposite direction of the gradient, the adjustment amount of the parameters can be obtained according to equation (7).

[0197] The negative sign (-) indicates adjustment in the opposite direction of the gradient to ensure that the J value decreases.

[0198] η (learning rate, or hyperparameter) is a very important hyperparameter (e.g., set to 0.1). It determines the size of each step. If η is too large, the step size is too large, which may cause the minimum point to be exceeded, or even lead to oscillations in the J value and failure to converge. If η is too small, the step size is too small, resulting in very slow convergence and low optimization efficiency. The value of the learning rate η needs to be determined through multiple experiments and adjustments to achieve the best balance between convergence speed and stability.

[0199] Here's an explanation of why gradient descent was chosen: it doesn't depend on the specific mathematical form of the objective function J. No matter how complex J is, as long as its value at a certain point can be calculated, the gradient can be estimated using numerical differencing. This is highly suitable for complex, nonlinear systems like 5G networks. The entire process is a deterministic algorithm that can be easily coded and implemented in a central processing platform, enabling fully automated optimization decisions without human intervention. The computational load is relatively small. For three parameters, only four calculations of J are needed (the current point and the points where the three parameters are increased by δ) to estimate the entire gradient, making it ideal for real-time or near-real-time optimization needs.

[0200] The output of this step includes: Predicted signal quality data: Predicted SINR values ​​(in dB) for the next Δt time interval, stored as an array with one value for each monitoring point. For example, the output is SINR. pred=[s1,s2,…,sM] , where sM is the predicted SINR of point i.

[0201] Optimization target value: Scalar J (dimensionless), representing the optimization requirement of the current network state. The lower the value, the better the network.

[0202] Recommended base station parameter adjustments include: tilt angle adjustment Δθ tilt Azimuth adjustment amount Δ azimuth Power adjustment amount ΔP tx .

[0203] Step 4: Dynamic Parameter Adjustment and Execution

[0204] This step automatically and in real-time adjusts base station parameters (such as antenna tilt angle, azimuth angle, and transmit power) based on the optimization suggestions generated in step three to improve low-altitude coverage quality and verify the adjustment effect, ensuring the reliability and efficiency of the optimization closed loop. This step emphasizes automated execution and rapid response, reducing manual intervention and adapting to dynamic environmental changes.

[0205] Specifically, based on optimization suggestions, the system automatically calculates and applies adjustments to base station parameters to achieve real-time updates of network configuration.

[0206] Immediately after adjustments, signal changes are monitored to confirm the effectiveness of the optimization; otherwise, a rollback or further adjustments are triggered to ensure reliability. Through execution and feedback, an optimization loop is formed, enabling the system to continuously adapt to changes in network conditions and improve overall network performance. The core of this step lies in translating algorithm output into physical actions, reducing latency through automated control (from decision to execution is typically completed within seconds), ensuring timely optimization, and avoiding the response delays (usually minutes or longer) of traditional manual methods.

[0207] The input data comes from the output tilt adjustment Δθ from step three. tilt Azimuth adjustment amount Δ azimuth Power adjustment amount ΔP tx .

[0208] Current base station status: Read in real time from the base station controller, including: Current antenna tilt angle θ tilt Current antenna azimuth angle azimuth Current transmit power P tx The health status of the base station (such as temperature and load rate) is used to ensure safe adjustments.

[0209] Here, maximum / minimum power limits and antenna adjustable range are specified to ensure adjustments are within feasible limits.

[0210] Short-term signal data (such as adjusted RSRP and SINR within 1 second) continuously flowing in from drones and base stations is used for rapid verification. Input data is received via the control interface of the central processing platform, with a processing latency requirement of less than 20 milliseconds to maintain real-time performance. The data flow involves interactions between system components; its system architecture can be found in [reference needed]. Figure 1 The demonstration showed the communication path between the drone device, the remote controller device, and the test terminal (central processing platform), where the central processing platform is responsible for sending control commands to the base station control unit.

[0211] Step four includes parameter adjustment calculation, command execution and verification, and the specific implementation details are as follows: 1. Parameter adjustment calculation: Adjustment Algorithm: A variant of gradient descent is used to calculate the final adjustment amount to ensure a smooth transition and avoid over-adjustment. For each parameter, the adjustment amount is fine-tuned based on the suggestion in step three and the current state, as shown in equation (8).

[0212] The adjusted parameters must satisfy the network constraints. For example, if the calculated P... tx, new More than P max (46 dBm), then the cutoff is P. max Similarly, the tilt and azimuth angles, after adjustment, need to be within the range of -15° to +15° and 0° to 360°. This is achieved through the function in equation (9).

[0213] 2. Send execution command: The central processing platform generates control commands in XML or JSON format, containing adjusted parameter values, execution timestamps, and command IDs. For example: { "command":"adjust_parameters", "timestamp":"2023-10-05T14:30:00Z", "parameters":{ "tilt":5.2, "azimuth":90.5, "power": 42.0 } } The command is sent to the base station control unit. After receiving the command, the base station control unit drives the stepper motor to adjust the antenna angle or modify the power amplifier settings. The adjustment time depends on the hardware: tilt adjustment usually takes 1-2 seconds, azimuth adjustment takes 2-5 seconds, and power adjustment is completed instantaneously (millisecond level). During the execution, the base station sends back a confirmation signal, including the actual adjustment value and status code. After the adjustment, the system immediately starts short-term monitoring (the default duration is 5 seconds), and collects signal data (such as SINR) in each grid in real time through the UAV, comparing the values ​​before and after the adjustment. The verification formula is as shown in equation (11).

[0214] (11)

[0215] In equation (11), This adjusts the average value over xx seconds (xx is a preset number). This is the average value over yy seconds after adjustment (yy is a preset number). If ΔSINR > 0.5 dB (indicating improvement), the adjustment is maintained; otherwise, a rollback or further optimization is triggered. Rollback mechanism: Restore the original parameters, mark the adjustment as failed, and return to recalculation. The 0.5 dB threshold is based on actual measurements: In 5G networks, a 0.5 dB change typically indicates a significant improvement. All executed operations are recorded in the database, including adjustment time, parameter values, performance metrics, and any error codes, for auditing and subsequent analysis.

[0216] The output of this step includes: Adjusted base station parameters: Parameter values ​​used in actual applications, including the new tilt angle θ. tilt, new (Unit: degrees), new azimuth azimuth , new (unit: degree), new power P tx, new (Unit: dBm) These values ​​are returned to the central processing platform in the form of acknowledgment messages.

[0217] Execution confirmation signals include a status code (e.g., 1 for success, 0 for failure), adjustment time, and a summary of results (e.g., ΔSINR value). The data format is JSON, used for real-time monitoring.

[0218] Verification Report: Generates a brief report on the optimization effect, such as "Adjustment successful, SINR improved by 0.8 dB" or "Requires re-optimization." The report is stored in the database and triggers log updates.

[0219] The output data is used in step four. If verification fails, the system immediately triggers a re-optimization loop. The overall transmission latency is less than 100 milliseconds, ensuring efficient closed-loop operation.

[0220] Step 5: Closed-loop monitoring and iterative optimization

[0221] This is a crucial step in ensuring the long-term effectiveness and adaptability of the system. Its purpose is to maintain optimal low-altitude coverage performance by regularly collecting network performance data, continuously monitoring optimization effects, periodically reassessing network status, and iteratively updating models and parameters to dynamically adapt to environmental changes (such as weather and obstacle movement) and network load fluctuations. This step emphasizes automation, periodicity, and data-driven approaches, forming a complete optimization loop, reducing human intervention, and improving network robustness.

[0222] The core of this step lies in forming a closed-loop control system, which continuously cycles from data acquisition to optimization execution and monitoring feedback to ensure the network is always in an optimal state. Compared to traditional static optimization, it provides adaptive capabilities, enabling it to handle unforeseen changes such as sudden disturbances or equipment aging.

[0223] Input data comes from multiple real-time and historical sources to ensure comprehensive monitoring: 1. Real-time monitoring data: Continuously flowing in from drone devices and base station controllers, including: Signal metrics: RSRP (dBm), SINR (dB), delay (milliseconds), jitter (milliseconds), sampling rate 10 Hz (once every 100 milliseconds), collected by the UAV test module.

[0224] Environmental data: obstacle index, wind speed (m / s), temperature (°C), acquired from drone sensors.

[0225] Base station status: Current parameters (tilt angle, azimuth angle, power) and load rate (%) are read in real time via SNMP protocol.

[0226] 2. Historical optimization records: Read from the database of the central processing platform, including: past optimization events: adjustment time, parameter changes, effect indicators (such as SINR improvement, power consumption changes), stored as time series data, covering the most recent 30 days.

[0227] Model training data: historical signal and environmental datasets, used for model updates.

[0228] 3. Performance Metric Thresholds: Predefined thresholds used to trigger actions. SINR falling threshold Δ th : 1 dB by default (if the SINR drops below this value, it indicates a performance degradation).

[0229] Environmental change threshold: such as increasing O index 0.2 (indicating significant occlusion change).

[0230] These thresholds are set based on network strategies and empirical research.

[0231] System configuration: monitoring interval Tmonitor (default 5 minutes), model update cycle Tupdate (default 30 days), read from configuration file.

[0232] Input data is integrated through the data aggregation module of the central processing platform, with a processing latency requirement of less than 100 milliseconds to ensure real-time performance.

[0233] This step involves monitoring data collection, performance evaluation, decision triggering, and iterative updates. Specific implementation details are as follows: 1. Monitoring data collection: Periodic re-flight tests: The drone automatically performs flight missions at a preset interval of Tmonitor=5 minutes, following a similar layered circular path to the one in step one, but with adjustments to altitude and radius based on historical data. For example, if a specific altitude was found to be problematic during the last optimization, that area will be tested first. During flight, the drone collects signal and environmental data in real time and transmits it to the central processing platform via a 5G link. The data format is the same as in step one, including filtered RSRP, SINR, etc.

[0234] 2. Performance Evaluation: Compare current data with historical optimized data, using Key Performance Indicators (KPIs): SINR improvement rate Rcover: Calculated as SINR above the threshold SINR th =5dB of point percentage change.

[0235] Power saving ΔP: Calculated as the reduction in total transmit power (in dBm).

[0236] Interference reduction rate (Rinterference): based on the average change of the interference index.

[0237] As shown in equation (10), in equation (10) It is the number of points where the current SINR is greater than the threshold. This is the number of points where the SINR after the last optimization is greater than the threshold. If any KPI deteriorates beyond the threshold, performance degradation is flagged. Simultaneously, environmental changes are checked, such as an increase in Oindex of 0.2.

[0238] Decision trigger: The central processing platform runs decision-making algorithms based on a rules engine. If performance degrades or environmental changes O index If the value exceeds a certain threshold, a re-optimization will be triggered immediately (returning to the previous steps).

[0239] If the performance is stable, but the model update cycle Tupdate=30 days has arrived, then a model update will be triggered.

[0240] Automated triggering reduces latency, the rule engine is simple and effective, and the 30-day model update cycle ensures that the model does not drift, while avoiding the overhead of frequent retraining.

[0241] 4. Iterative updates: Model retraining: For the LSTM prediction model, incremental training is performed monthly using newly collected data. The training process is similar to step three, but online learning techniques are used. Data: Combining historical data and the latest 30-day data.

[0242] Loss function: Mean squared error (MSE), optimizer Adam, learning rate 0.001.

[0243] Training time: Running on the GPU cluster of the central processing platform takes about 1 hour, with minimal impact on the system.

[0244] As shown.

[0245] (12)

[0246] In equation (12), It is an MSE loss. It is a regularization parameter to prevent overfitting.

[0247] If a re-optimization is triggered, the system returns to the previous steps and uses the updated model to calculate the new optimization objective.

[0248] Incremental training maintains model accuracy, regularization parameters Cross-validation was used to determine the weights of the old and new data.

[0249] The entire method runs on a central processing platform, with computational resource requirements: CPU utilization is <5% during the monitoring phase and peaks up to 20% during model updates.

[0250] Here, the output of step five includes: Monitoring Reports: Regularly generated reports include performance metrics (such as Rcover, ΔP), summaries of environmental changes, and trigger event logs. Reports are stored in PDF or JSON format for operational analysis.

[0251] Trigger signal: such as a re-optimization instruction or model update command, sent to step three.

[0252] Updated model weights W new Stored in the database for subsequent predictions.

[0253] System status log: Records monitoring time and evaluation results in detail for auditing and debugging.

[0254] The output data is used to maintain the optimization loop. If re-optimization is triggered, the system seamlessly transitions to step three.

[0255] Step five, through automated monitoring and iterative updates, ensures the long-term effectiveness of 5G low-altitude coverage optimization. This method innovatively combines periodic testing, performance evaluation, and model retraining to form an adaptive closed-loop system. The entire process emphasizes data-driven and automated approaches, reducing labor costs, improving network resilience, and providing reliable assurance for 5G low-altitude applications.

[0256] In summary, this example uses a drone to collect low-altitude signal data in real time, combines this data with environmental information, employs machine learning algorithms to predict signal quality, and dynamically adjusts base station parameters (such as beamforming angle and transmit power) to achieve automated optimization. The system in this example includes a drone device, a base station control unit, and a central processing platform, realizing closed-loop control of data acquisition, analysis, decision-making, and execution.

[0257] Beneficial effects: Real-time data collection and algorithm processing via drones enable optimization adjustments to be completed within seconds, far faster than manual methods (which typically take hours). This reduces human intervention, lowers operating costs, and improves optimization accuracy. Utilizing environmental data and historical records allows for dynamic responses to changes (such as the addition of obstacles), enhancing network robustness. Through algorithm optimization, base station power consumption is minimized while maximizing coverage, reducing interference.

[0258] It is understood that the various method embodiments mentioned above in this disclosure can be combined with each other to form combined embodiments without violating the principle and logic. Due to space limitations, this disclosure will not elaborate further. Those skilled in the art will understand that in the above methods of specific implementation, the specific execution order of each step should be determined by its function and possible internal logic.

[0259] In addition, this disclosure also provides a network quality dynamic optimization device, electronic device, and computer-readable storage medium, all of which can be used to implement any of the network quality dynamic optimization devices provided in this disclosure. The corresponding technical solutions and descriptions are described in the corresponding descriptions in the method section, and will not be repeated here.

[0260] like Figure 9 As shown, this disclosure provides a network quality dynamic optimization device 900, which includes: The information acquisition module 901 is used to acquire network quality-related information of the test area collected by the UAV. The test area is determined based on the base station to be tested. The network quality-related information includes the current acquisition time, the geographic information of the current flight position, the signal data of the current flight position, and the environmental data of the current flight position. Among them, the geographic information includes the actual flight altitude and horizontal coordinate position. The signal data includes reference signal received power parameters, signal-to-interference-plus-noise ratio parameters, and transmission delay parameters. The environmental data includes obstacle image data, wind speed data, and temperature data.

[0261] The prediction module 902 is used to preprocess network quality-related information to obtain preprocessed network quality-related information, and predict signal quality parameters based on the preprocessed network quality-related information to obtain prediction data for multiple signal quality parameters.

[0262] The multi-objective function determination module 903 is used to obtain the transmit power of the base station under test and determine a multi-objective function based on the transmit power of the base station under test, predicted data of multiple signal quality parameters, and preprocessed network quality-related information. The multi-objective function balances three objective items: signal coverage, base station power consumption, and signal interference.

[0263] The optimization quantity acquisition module 904 is used to determine the optimization quantity of each parameter in at least one parameter of the base station under test according to a multi-objective function. The at least one parameter of the base station under test includes at least one of the antenna tilt angle, antenna azimuth angle, and transmit power of the base station under test.

[0264] The parameter adjustment module 905 is used to adjust each parameter of at least one parameter of the base station under test according to the optimization amount of each parameter in at least one parameter of the base station under test.

[0265] like Figure 10 As shown, this disclosure provides an electronic device, which includes: at least one processor 1001; at least one memory 1002; and one or more I / O interfaces 1003 connected between the processor 1001 and the memory 1002; wherein the memory 1002 stores one or more computer programs that can be executed by the at least one processor 1001, and the one or more computer programs are executed by the at least one processor 1001 to enable the at least one processor 1001 to perform the above-described network quality dynamic optimization method.

[0266] This disclosure also provides a computer-readable storage medium storing a computer program thereon, wherein the computer program, when executed by a processor, implements the above-described dynamic network quality optimization method. The computer-readable storage medium may be volatile or non-volatile.

[0267] This disclosure also provides a computer program product, including computer-readable code, or a non-volatile computer-readable storage medium carrying computer-readable code, wherein when the computer-readable code is run in a processor of an electronic device, the processor in the electronic device executes the above-described network quality dynamic optimization method.

[0268] Those skilled in the art will understand that all or some of the steps, systems, and apparatuses disclosed above, and their functional modules / units, can be implemented as software, firmware, hardware, or suitable combinations thereof. In hardware implementations, the division between functional modules / units mentioned above does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed collaboratively by several physical components. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit (ASIC). Such software can be distributed on a computer-readable storage medium, which may include computer storage media (or non-transitory media) and communication media (or transient media).

[0269] As is known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information, such as computer-readable program instructions, data structures, program modules, or other data. Computer storage media includes, but is not limited to, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), static random access memory (SRAM), flash memory or other memory technologies, portable compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, it is known to those skilled in the art that communication media typically contain computer-readable program instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0270] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0271] Computer program instructions used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C++, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing the status information of the computer-readable program instructions to implement various aspects of this disclosure.

[0272] The computer program product described herein can be implemented specifically through hardware, software, or a combination thereof. In one alternative embodiment, the computer program product is specifically embodied in a computer storage medium; in another alternative embodiment, the computer program product is specifically embodied in a software product, such as a software development kit (SDK), etc.

[0273] Various aspects of this disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0274] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that, when executed by the processor of the computer or other programmable data processing apparatus, they create means for implementing the functions / actions specified in one or more blocks of the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner; thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing aspects of the functions / actions specified in one or more blocks of the flowchart and / or block diagram.

[0275] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0276] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0277] Example embodiments have been disclosed herein, and while specific terminology has been used, it is for illustrative purposes only and should be construed as such, and is not intended to be limiting. In some instances, it will be apparent to those skilled in the art that features, characteristics, and / or elements described in connection with particular embodiments may be used alone, or in combination with features, characteristics, and / or elements described in connection with other embodiments, unless otherwise expressly indicated. Therefore, those skilled in the art will understand that various changes in form and detail may be made without departing from the scope of this disclosure as set forth by the appended claims.

Claims

1. A method for dynamic optimization of network quality, characterized in that, include: S1, acquire network quality-related information of the area to be tested collected by the drone; The area to be tested is determined based on the base station to be tested; the network quality-related information includes the current data collection time, the geographical information of the current flight position, the signal data of the current flight position, and the environmental data of the current flight position; wherein, the geographical information includes the actual flight altitude and horizontal coordinates; the signal data includes reference signal received power parameters, signal-to-interference-plus-noise ratio parameters, and transmission delay parameters; the environmental data includes obstacle image data, wind speed data, and temperature data; S2, preprocess the network quality related information to obtain preprocessed network quality related information, predict signal quality parameters based on the preprocessed network quality related information, and obtain prediction data for multiple signal quality parameters; S3, obtain the transmit power of the base station under test, and determine a multi-objective function based on the transmit power of the base station under test, the predicted data of the multiple signal quality parameters, and the preprocessed network quality related information; the multi-objective function is used to balance the three objective items of signal coverage, base station power consumption, and signal interference. S4, determine the optimization amount of each parameter in at least one parameter of the base station under test according to the multi-objective function; the at least one parameter of the base station under test includes at least one of the antenna tilt angle, the antenna azimuth angle, and the transmit power of the base station under test. S5, adjust each parameter of the base station under test according to the optimization amount of each parameter in at least one parameter of the base station under test.

2. The network quality dynamic optimization method according to claim 1, characterized in that, S1 includes: S11, The test area is set according to the location of the base station to be tested; S12, determine the center of the UAV's flight range based on the location of the base station to be tested, and determine multiple flight altitudes and the corresponding flight radii based on the height of ground obstacles; S13, control the drone to fly within its flight range and acquire network quality-related information of the area to be tested collected by the drone at a preset period.

3. The network quality dynamic optimization method according to claim 1, characterized in that, S2 includes: S21, invalid data in the reference signal received power parameter, the signal-to-interference-plus-noise ratio parameter, and the transmission delay parameter are removed respectively, and interpolation processing is performed to obtain the interpolated reference signal received power parameter, the interpolated signal-to-interference-plus-noise ratio parameter, and the interpolated transmission delay parameter. S22, filter the interpolated reference signal received power parameters and the interpolated signal-to-interference-plus-noise ratio parameters respectively to obtain the filtered reference signal received power parameters and the filtered signal-to-interference-plus-noise ratio parameters; S23, extract the obstacle index from the obstacle image data, and preprocess the wind speed data and temperature data respectively to obtain preprocessed wind speed data and preprocessed temperature data; the preprocessing includes removing invalid data, interpolation processing and normalization processing; the preprocessed network quality related information includes the filtered reference signal received power parameter, the filtered signal-to-interference-plus-noise ratio parameter, the interpolated transmission delay parameter, the obstacle index, the preprocessed wind speed data, and the preprocessed temperature data; S24, the geographical information of the acquisition time point, the current flight position, the filtered reference signal received power parameter, the filtered signal-to-interference-plus-noise ratio parameter, and the interpolated transmission delay parameter are made into a structured data table; the obstacle index, the preprocessed wind speed data, and the preprocessed temperature data are made into an environmental feature vector; S25, based on the structured data table and environmental feature vector of the current acquisition time point and the previous three acquisition time points, predict the signal quality parameters of the next acquisition time point to obtain the predicted data of the signal quality parameters of the next acquisition time point; the predicted data of the signal quality parameters of the next acquisition time point includes the predicted data of the parameter signal and interference plus noise ratio parameter; the predicted data of multiple signal quality parameters includes the predicted data of the signal quality parameters of multiple acquisition time points.

4. The network quality dynamic optimization method according to claim 3, characterized in that, In S23, according to formula Extract obstacle index from the obstacle image data ;in, It is the number of occluded pixels in the obstacle image data. It is the total number of pixels in the obstacle image data.

5. The network quality dynamic optimization method according to claim 3, characterized in that, S25 includes: S251, a prediction model is built based on the Long Short-Term Memory (LSTM) network. The architecture of the prediction model includes two LSTM layers and one fully connected output layer. S252, acquire historical data from the structured data table and historical data from the environmental feature vector, and train the prediction model based on the historical data from the structured data table and historical data from the environmental feature vector to obtain a prediction model for the quality parameters; S253, according to formula Prediction is performed to obtain predicted data for the signal quality parameters at the next acquisition time point; among which, It is the predicted data of signal quality parameters at the next acquisition time point; For the prediction model of quality parameters, For input data, This includes structured data tables and environmental feature vectors from the current data collection time point and the previous three data collection time points. These are the weight parameters of the prediction model for quality parameters.

6. The network quality dynamic optimization method according to claim 3, characterized in that, S3 includes: S31, Obtain the transmit power of the base station under test; S32, select M predicted signal quality parameter data for preset acquisition locations from the predicted data of the plurality of signal quality parameters; M is an integer, and ; S33, obtain the interference index of M preset sampling locations based on the received power parameters of the filtered reference signal; the interference index represents the strength of interference signals from other base stations, and the other base stations are base stations other than the base station under test among the base stations that can receive signals at the current location. S34, according to formula Determine the multi-objective function J; where M is the number of preset acquisition locations; It is the predicted data of the signal quality parameters of the i-th preset acquisition position among M preset acquisition positions; It is the threshold of the signal quality parameter; Let B be the transmit power of the j-th base station; B is the total number of base stations to be tested; B is an integer, and ; It is the interference index of the i-th preset acquisition position among M preset acquisition positions; These are the weighting coefficients.

7. The network quality dynamic optimization method according to claim 1, characterized in that, S4 includes: S41, Usage Calculate the partial derivatives of each parameter in at least one parameter of the multi-objective function J for the base station under test; where, This represents any one of at least one parameters of the base station under test. For parameters The applied small perturbation, Indicates parameters Given the current value, the value of the multi-objective function J is... Indicates parameters Adding small perturbations The subsequent values ​​of the multi-objective function J; S42, Determine hyperparameters and according to the formula Calculate the optimization amount of each parameter in at least one parameter of the base station under test. .

8. The network quality dynamic optimization method according to claim 7, characterized in that, S42 includes: acquiring historical adjustment data of each parameter in at least one parameter of the base station under test, and determining hyperparameters based on the historical adjustment data of each parameter in at least one parameter of the base station under test. .

9. The network quality dynamic optimization method according to claim 1, characterized in that, S5 includes: determining the constraint configuration of the base station under test, and adjusting each parameter of at least one parameter of the base station under test according to the constraint configuration of the base station under test and the optimization amount of each parameter in at least one parameter of the base station under test; the constraint configuration of the base station under test includes the transmit power range, the adjustable range of antenna tilt angle, and the adjustable range of antenna azimuth angle of the base station under test.

10. A network quality dynamic optimization device, characterized in that, include: The information acquisition module is used to acquire network quality-related information of the area under test collected by the drone; The area to be tested is determined based on the base station to be tested; the network quality-related information includes the current data collection time, the geographical information of the current flight position, the signal data of the current flight position, and the environmental data of the current flight position; wherein, the geographical information includes the actual flight altitude and horizontal coordinates; the signal data includes reference signal received power parameters, signal-to-interference-plus-noise ratio parameters, and transmission delay parameters; the environmental data includes obstacle image data, wind speed data, and temperature data; The prediction module is used to preprocess the network quality-related information to obtain preprocessed network quality-related information, and predict signal quality parameters based on the preprocessed network quality-related information to obtain prediction data for multiple signal quality parameters. The multi-objective function determination module is used to obtain the transmit power of the base station under test, and determine a multi-objective function based on the transmit power of the base station under test, the predicted data of multiple signal quality parameters, and the preprocessed network quality related information; the multi-objective function is used to balance three objective items: signal coverage, base station power consumption, and signal interference. The optimization quantity acquisition module is used to determine the optimization quantity of each parameter in at least one parameter of the base station under test according to the multi-objective function; the at least one parameter of the base station under test includes at least one of the antenna tilt angle, the antenna azimuth angle, and the transmit power of the base station under test. The parameter adjustment module is used to adjust each parameter of at least one parameter of the base station under test according to the optimization amount of each parameter in at least one parameter of the base station under test.

11. An electronic device, characterized in that, include: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores one or more computer programs that can be executed by the at least one processor, and the one or more computer programs are executed by the at least one processor to enable the at least one processor to perform the network quality dynamic optimization method as described in any one of claims 1-9.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the network quality dynamic optimization method as described in any one of claims 1-9.

13. A computer program product, characterized in that, Includes computer-readable code, or a non-volatile computer-readable storage medium carrying computer-readable code, wherein when the computer-readable code is run in a processor of an electronic device, the processor in the electronic device performs the network quality dynamic optimization method as described in any one of claims 1-9.

Images