Coverage area classification method and device and electronic equipment

Abstract

The invention discloses a coverage area classification method and device and electronic equipment. The method comprises the steps that user plane flow data of sectors to be classified are acquired from a network management system of a base station, the sectors to be classified are sub-areas of a coverage area of the base station, and the user plane flow data comprise uplink and downlink total flow data; and classifying the user plane traffic data by adopting a classification model based on a time sequence to obtain the coverage area type of the sector to be classified. According to the method and the device, the technical problem that the accuracy of classification of the wireless sector coverage area is low due to the fact that a coverage scene identification method adopted in the related technology depends on physical parameters such as base station antenna information and geographical location information is solved.

Description

Technical Field

[0001] The present application relates to the field of communication technology, and in particular to a method, device and electronic device for classifying coverage areas. Background Art

[0002] Accurately classifying the coverage areas of base station sectors is of great significance in wireless communication network planning and network resource allocation optimization. However, the coverage scenario identification methods used in related technologies rely on physical parameters such as base station antenna information and geographic location information, resulting in low accuracy in wireless sector coverage area classification, making it difficult to accurately reflect the actual business needs of sector users.

[0003] To address the above-mentioned problems, no effective solutions have been proposed so far. Summary of the Invention

[0004] The embodiments of the present application provide a coverage area classification method, device and electronic device to at least solve the technical problem that the coverage scene recognition method adopted by the related technology relies on physical parameters such as base station antenna information and geographic location information, resulting in low accuracy in wireless sector coverage area classification.

[0005] According to one aspect of an embodiment of the present application, a method for classifying coverage areas is provided, comprising: obtaining user-plane traffic data of a sector to be classified from a base station network management system, wherein the sector to be classified is a sub-area of ​​the coverage area of ​​the base station, and the user-plane traffic data includes total uplink and downlink traffic data; and classifying the user-plane traffic data using a time series-based classification model to obtain the coverage area type of the sector to be classified.

[0006] In some embodiments of the present application, a time series-based classification model is used to classify user plane traffic data to obtain the coverage area type of the sector to be classified, including: using a convolution layer in the time series-based classification model to determine the time series corresponding to the user plane traffic data and the first-order difference sequence corresponding to the time series, wherein the length of the time series is greater than the length of the first-order difference sequence; the convolution layer uses different random convolution kernel sets to perform convolution operations on the time series and the first-order difference sequence respectively to obtain a target feature map, wherein the convolution kernel in the random convolution kernel set is a fixed convolution kernel and multiple convolution kernel instances obtained by deforming it with different expansion coefficients and bias values; the pooling layer in the time series-based classification model determines multiple pooling features corresponding to the target feature map output by each random convolution kernel, and determines a multidimensional feature vector corresponding to the target feature map based on the pooling features, wherein the pooling features are used to summarize the distribution characteristics of the target feature map; the output layer in the time series-based classification model determines the coverage area type based on the multidimensional feature vector.

[0007] In some embodiments of the present application, the output layer in the time series-based classification model determines the coverage area type based on the multidimensional feature vector, including: determining the weight vector of the ridge regression classifier, wherein the weight vector is used to quantify the importance of each feature in the multidimensional feature vector to the classification result, and the weight vector corresponds to the dimension of the multidimensional feature vector; performing a dot product between the weight vector and the multidimensional feature vector to obtain a predicted value of the multidimensional feature vector, and determining the coverage area type based on the predicted value.

[0008] In some embodiments of the present application, when the coverage area type includes two types, the coverage area type is determined based on the positive and negative signs of the predicted value; when the coverage area type includes more than two types, the weight vector matrix corresponding to multiple pending coverage area types is determined, and the predicted value corresponding to each pending coverage area type is determined based on the weight vector matrix, and the pending coverage area type corresponding to the maximum predicted value is determined as the coverage area type.

[0009] In some embodiments of the present application, a time series-based classification model is configured in the following manner: determining multiple fixed convolution kernels, wherein the multiple fixed convolution kernels have the same length and different fixed convolution kernels have different weight configurations, and the fixed convolution kernels are used to extract the time series characteristics of user-side traffic data; expanding each fixed convolution kernel using an expansion coefficient sequence, wherein the maximum value of the expansion coefficient sequence is determined based on the length of the traffic data; the bias corresponding to each combination of the fixed convolution kernel and the expansion coefficient is determined based on the output of each fixed convolution kernel for a random sample; the first combination of all combinations is padded with zeros, and the second combination is not padded, wherein the second combination is a combination of all combinations except the first combination, and the number of first combinations and second combinations is the same.

[0010] In some embodiments of the present application, the convolution layer uses different random convolution kernel sets to perform convolution operations on the time series and the first-order difference sequence respectively to obtain a target feature map, including: determining the maximum expansion coefficients corresponding to the time series and the first-order difference sequence respectively, and determining the expansion sequences corresponding to the maximum expansion coefficients respectively, wherein the maximum expansion coefficient of the time series is greater than the maximum expansion coefficient of the first-order difference sequence, and the expansion sequence is used to perform an expansion operation on the fixed convolution kernel; determining a first random convolution kernel set corresponding to the time series and a second random convolution kernel set corresponding to the first-order difference sequence according to the expansion sequence; using the random convolution kernel of the first random convolution kernel set to perform a convolution operation on the time series to obtain a first feature map, and using the random convolution kernel of the second random convolution kernel set to perform a convolution operation on the first-order difference sequence to obtain a second feature map; determining a target feature map corresponding to the first feature map and the second feature map.

[0011] In some embodiments of the present application, a time series-based classification model is trained in the following manner: historical user plane traffic data of multiple sectors of multiple base stations are obtained, wherein the historical user plane traffic data include historical user plane traffic data of base station sectors based on a fourth-generation mobile communication system and historical user plane traffic data of base station sectors based on a fifth-generation mobile communication system; an initial classification model is used to classify the historical user plane traffic data to obtain an initial coverage area type; if the initial coverage area type does not meet a preset condition, the parameters of the initial classification model are adjusted, and the initial coverage area type is updated, and the parameter updating is stopped until the updated initial coverage area type meets the preset condition, thereby obtaining a time series-based classification model.

[0012] According to another aspect of an embodiment of the present application, a coverage area classification device is also provided, including: an acquisition module for obtaining user-plane traffic data of a sector to be classified from a base station network management system, wherein the sector to be classified is a sub-area of ​​the coverage area of ​​the base station, and the user-plane traffic data includes total uplink and downlink traffic data; a classification module for classifying the user-plane traffic data using a time series-based classification model to obtain the coverage area type of the sector to be classified.

[0013] According to another aspect of the embodiments of the present application, an electronic device is provided, including: a memory and a processor, the memory being used to store program instructions; the processor being connected to the memory and being used to execute the above-mentioned coverage area classification method.

[0014] According to another aspect of the embodiments of the present application, a non-volatile storage medium is provided, which includes a stored computer program, wherein the device where the non-volatile storage medium is located executes the above-mentioned coverage area classification method by running the computer program.

[0015] According to another aspect of the embodiments of the present application, a computer program product is provided, including computer instructions, which implement the above-mentioned coverage area classification method when executed by a processor.

[0016] In an embodiment of the present application, by collecting protocol layer traffic data of base station sectors, using a model to capture the time series characteristics in the traffic data, and performing feature extraction and classification on the sectors, the purpose of accurately identifying the type of coverage area of ​​the sector to be classified is achieved, thereby achieving the technical effect of improving the efficiency of wireless network resource allocation and optimizing network performance, and further solving the technical problem that the coverage scenario recognition method adopted by the related technology depends on physical parameters such as base station antenna information and geographic location information, resulting in low accuracy in the classification of wireless sector coverage areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0018] Figure 1 This is a hardware structure block diagram of a computer terminal according to a coverage area classification method according to an embodiment of the present application;

[0019] Figure 2 is a flow chart of a method for classifying coverage areas according to an embodiment of the present application;

[0020] Figure 3 It is a structural diagram of a device for classifying coverage areas according to an embodiment of the present application. DETAILED DESCRIPTION

[0021] In order to enable those skilled in the art to better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0022] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in a sequence other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0023] In order to better understand the embodiments of the present application, the technical terms involved in the embodiments of the present application are explained as follows:

[0024] 4G / 5G sector: A 4G sector refers to a unit of coverage area of ​​a base station that uses the fourth-generation mobile communication technology. Each 4G sector typically uses a specific antenna direction and frequency band to serve a geographical area within a certain angle. A 5G sector refers to a unit of coverage area of ​​a base station that uses the fifth-generation mobile communication technology. Compared to a 4G sector, a 5G sector can use a wider frequency band, higher frequency, and more advanced antenna technology to provide services.

[0025] PDCP (Packet Data Convergence Protocol) Layer: In 4G and 5G networks, the PDCP layer is responsible for processing data header compression, encryption, and reordering, ensuring efficient and secure data transmission across the wireless interface. In this embodiment, PDCP layer traffic data can be used to provide direct information about the characteristics of 4G sector data transmission, serving as the basis for building a time series classification model.

[0026] RLC layer (Radio Link Control Layer): Located below the PDCP layer, the RLC layer is primarily responsible for segmenting, reassembling, and detecting errors in upper-layer data, ensuring reliable data transmission over wireless channels. In this embodiment of the present application, traffic data from the RLC layer can be used to analyze the transmission efficiency and user behavior of 5G sectors.

[0027] MultiRocket model: A time series classification algorithm based on convolution and pooling operations that increases feature diversity through multiple pooling operators and transformations, thereby achieving fast and efficient time series classification. In an embodiment of the present application, the time series-based classification model includes a MultiRocket model (as a feature extractor). Through the pre-training and fine-tuning mechanism of the MultiRocket model, the model can quickly adapt to the classification requirements of new sectors, improving classification speed and accuracy.

[0028] Ridge Regression Classifier: This is a classification model based on an improved ridge regression. By introducing an L2 regularization term to constrain the weight coefficients, it addresses feature collinearity and improves the generalization capability of the classification task. In this embodiment of the present application, the time series-based classification model includes a ridge regression classifier, which is used to map the feature vectors generated by the MultiRocket model to coverage area categories, enabling the conversion from traffic data to scene classification.

[0029] The coverage area classification methods used in related technologies rely too heavily on signal strength measurements and geographic information, failing to fully consider the role of traffic data in classification. For example, classification based on base station antenna parameters (azimuth, downtilt) or drive test data relies on physical parameters, resulting in classification results that are constrained by physical environmental factors while ignoring the substantial impact of user behavior and dynamic network load on classification, limiting classification accuracy and service relevance. Furthermore, while related technologies employ deep learning models such as LSTM (Long Short-Term Memory) networks and Transformers for sector coverage area classification, while capable of processing time series data, their high computational complexity and long training times make them inefficient for real-time analysis and large-scale datasets, making them difficult to meet operators' needs for rapid optimization of network coverage and resource scheduling. Furthermore, related technologies struggle to effectively adapt to the complexities of diverse network environments and service distributions. In particular, in scenarios where 4G and 5G networks are converged, the lack of mechanisms to handle differences in multi-vendor equipment limits model generalization and makes it difficult to maintain stability and classification accuracy in a highly dynamic environment.

[0030] In order to solve the above technical problems, the embodiments of the present application provide corresponding solutions, which are described in detail below.

[0031] The coverage area classification method embodiment provided in the embodiment of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 1 FIG. 1 shows a hardware structure block diagram of a computer terminal for implementing a classification method for coverage areas. Figure 1 As shown, the computer terminal 10 may include one or more (illustrated by 102a, 102b, ..., 102n in the figure) processors (the processor may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 104 for storing data, and a transmission module 106 for communication functions connected via a wired and / or wireless network. In addition, it may also include: a display, a keyboard, a cursor control device, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the I / O interface), a network interface, and a BUS bus. It will be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components than shown, or with Figure 1 Different configurations shown.

[0032] It should be noted that the one or more processors and / or other data processing circuits described above may generally be referred to herein as "data processing circuitry." The data processing circuitry may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuitry may be a single, independent processing module, or may be incorporated in whole or in part into any of the other components of the computer terminal 10. As described in the embodiments of the present application, the data processing circuitry serves as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0033] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the coverage area classification method in the embodiment of the present application. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, implementing the above-mentioned coverage area classification method. The memory 104 may include a high-speed random access memory and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely located relative to the processor, and these remote memories may be connected to the computer terminal 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0034] The transmission module 106 is configured to receive or transmit data via a network. A specific example of the aforementioned network may include a wireless network provided by the communications provider of the computer terminal 10. In one embodiment, the transmission module 106 includes a network interface controller (NIC), which can be connected to other network devices via a base station to enable communication with the Internet. In another embodiment, the transmission module 106 may be a radio frequency (RF) module, which is configured to communicate with the Internet wirelessly.

[0035] The display may be, for example, a touch screen liquid crystal display (LCD) that enables a user to interact with a user interface of the computer terminal 10 .

[0036] It should be noted that, in some optional embodiments, the above Figure 1 The computer terminal shown may include hardware elements (including circuits), software elements (including computer code stored on a computer-readable medium), or a combination of hardware elements and software elements. Figure 1 This is merely one example of a particular embodiment and is intended to illustrate the types of components that may be present in the computer terminal described above.

[0037] In the above-mentioned operating environment, an embodiment of the present application provides an embodiment of a classification method for coverage areas. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0038] Figure 2 is a flow chart of a method for classifying coverage areas according to an embodiment of the present application. Figure 2 As shown, the method includes the following steps:

[0039] Step S202: acquiring user plane traffic data of the to-be-classified sector from the base station network management system, wherein the to-be-classified sector is a sub-area of ​​the coverage area of ​​the base station, and the user plane traffic data includes total uplink and downlink traffic data.

[0040] In step S202, the base station network management system is a system used in wireless communication networks to manage base station equipment, monitor network performance, and configure network parameters. It directly monitors and records various base station operational data, including user-plane traffic information. The sectors to be classified refer to sub-areas within the base station coverage area that require scenario classification. Each sector uses a specific antenna and frequency band to cover a certain angle of the geographical area. The sectors to be classified are the direct objects of traffic data classification, and the classification results are used to guide network planning and optimization for the sectors to be classified. The user plane refers to the interface for direct data transmission between the base station and user equipment. User-plane traffic data includes total uplink and downlink traffic data transmitted through this interface, which is used to reflect user behavior and network load characteristics of the sectors to be classified. It can include traffic data based on the packet data convergence protocol layer of the fourth-generation mobile communication system or traffic data based on the radio link control layer of the fifth-generation mobile communication system.

[0041] Since the classification method used in related technologies relies on signal strength measurement or drive test data, the data acquisition cycle is long and susceptible to external interference. The user plane traffic data obtained directly from the base station network management system is more accurate and real-time, and can more accurately reflect the actual activities and data usage patterns of users in the sector under the current network status.

[0042] Unlike data collected from user terminals or third-party monitoring tools, the base station network management system covers the sector traffic data of the entire base station network, including 4G and 5G networks. This enables the base station network management system to provide deeper internal network information, including data transmission details at the protocol layer. This internal information contains more implicit characteristics about network load and user behavior.

[0043] In some embodiments of the present application, the classification model can be pre-trained using historical traffic data obtained from the base station network management system, that is, the classification model is trained on a larger and more diverse historical traffic data set to generate a general feature extractor (such as), and then the classifier is adjusted on the traffic data within a preset period of a specific sector (such as the traffic data of the past week) to adapt to the local characteristics of the specific sector, and the final time series-based classification model is obtained.

[0044] Specifically, a time series-based classification model can be trained in the following manner: obtaining historical user-plane traffic data of multiple sectors of multiple base stations, wherein the historical user-plane traffic data includes historical user-plane traffic data of base station sectors based on a fourth-generation mobile communication system and historical user-plane traffic data of base station sectors based on a fifth-generation mobile communication system; using an initial classification model to classify the historical user-plane traffic data to obtain an initial coverage area type; when the initial coverage area type does not meet a preset condition, adjusting the parameters of the initial classification model and updating the initial coverage area type, and stopping updating the parameters until the updated initial coverage area type meets the preset condition, thereby obtaining a time series-based classification model.

[0045] The training of the feature extractor in a time series-based classification model can include the following steps (taking the MultiRocket model as an example):

[0046] (1) Obtaining historical user-plane traffic data: Historical user-plane traffic data refers to the total uplink and downlink traffic records transmitted through the base station user plane over a period of time, including 4G (fourth-generation mobile communication system) and 5G (fifth-generation mobile communication system) traffic data, which is used to learn the characteristics of traffic patterns in different coverage areas. For example, historical user-plane traffic data of multiple 4G / 5G base station sectors can be extracted from the base station network management system or other relevant data storage, especially the hourly traffic data of the PDCP layer (4G) and RLC layer (5G). The data coverage period (preset period) can be one week, including working days and non-working days, so that the model can learn the daytime and nighttime changes of traffic patterns.

[0047] In some embodiments of the present application, historical user plane traffic data may be obtained by following the steps below:

[0048] Step 1: Traffic data collection and preprocessing. For 4G sectors, the hourly user plane uplink and downlink total traffic data of the PDCP layer for one week is extracted from the base station network management system; for 5G sectors, the hourly user plane uplink and downlink total traffic data of the RLC layer for one week is collected from the base station network management system. The traffic data of a sector for one week can be the data from 12:00 on Monday to 11:00 on the next Monday. It can be observed from the original user plane uplink and downlink total traffic that there is an obvious periodic pattern, with one traffic value every hour, so the data sample is of length l input =A time series of 147.

[0049] Step 2: Data normalization. Normalization is less sensitive to outliers and is used to eliminate differences in flow extremes between sectors, allowing the model to focus on the temporal characteristics of flow rather than the magnitude of flow. The flow data is normalized to a distribution with a mean of 0 and a standard deviation of 1, eliminating the impact of differences in equipment models. The process is as follows:

[0050] 1) Standardization formula:

[0051]

[0052] Where X is the original data, μ is the mean of the original data, and σ is the standard deviation of the original data.

[0053] 2) Calculate the mean:

[0054]

[0055] Among them, N is the number of user plane traffic data, X i is the value of the i-th user plane traffic data.

[0056] 3) Calculate the standard deviation:

[0057]

[0058] 4) Standardize each data:

[0059]

[0060] (2) Model initialization and preliminary classification: Use the untrained MultiRocket model (initial classification model) to classify historical user plane traffic data and output preliminary coverage area types.

[0061] (3) Parameter adjustment and optimization iteration: When the classification performance (such as accuracy) of the initial coverage area type fails to meet the preset conditions, the model parameters (such as convolution kernel size, bias value, expansion coefficient, etc.) are adjusted, and the classification is re-performed. The iteration is continued until the classification result meets the set performance threshold, and the pre-trained model (i.e., the time series-based classification model after the feature extractor is trained) is obtained.

[0062] After pre-training the classification model based on historical user-plane traffic data and generating a general feature extractor, the process of adjusting the model for a specific sector can include the following steps (using the MultiRocket model as an example): Migrating feature extraction-related parameters such as convolution kernel parameters, dilation coefficients, and bias values ​​from the pre-trained model to the model of the sector to be classified to accelerate convergence of the model; Adjusting the pre-trained model using traffic data from the sector to be classified within a preset period (e.g., the most recent week) and retraining the classifier weight parameters in the pre-trained model. It should be noted that at this stage, the feature extraction portion of the pre-trained model remains unchanged (i.e., the underlying parameters related to feature extraction are frozen), while only the classifier weights are updated. This is because the model has already learned a general traffic feature representation during the pre-training phase, and adaptability to the sector to be classified is primarily achieved by adjusting the classification boundaries (i.e., the classifier weights); Evaluating the model's performance on test data from the sector to be classified. Training is terminated when the classification accuracy of the adjusted pre-trained model on new data meets a preset threshold, resulting in a time series-based classification model.

[0063] Step S204: classify the user plane traffic data using a time series-based classification model to obtain the coverage area type of the sector to be classified.

[0064] In the above step S204, the coverage area type of the sector to be classified is the specific geographical or functional area type to which the sector belongs given by the time series-based classification model, such as dense urban area, suburban area, high-speed rail scenario, etc.

[0065] In some embodiments of the present application, the user plane traffic data of the sector to be analyzed can be used as input (such as the hourly data traffic in the past week, that is, a traffic value is determined for each hour in the past week); the MultiRocket model in the trained time series-based classification model is used to extract features of the input traffic data to generate a feature vector; the feature vector is input into the ridge regression classifier in the time series-based classification model, and the ridge regression classifier outputs the coverage area type according to the classification rules.

[0066] The time series-based classification model can be configured using the following strategy: determine multiple fixed convolution kernels, where the lengths of the multiple fixed convolution kernels are the same and the weight configurations of different fixed convolution kernels are different, and the fixed convolution kernels are used to extract the time series characteristics of the user plane traffic data; use an expansion coefficient sequence to expand each fixed convolution kernel, where the maximum value of the expansion coefficient sequence is determined according to the length of the traffic data; the bias corresponding to each combination of the fixed convolution kernel and the expansion coefficient is determined according to the output of each fixed convolution kernel for a random sample; the first combination of all combinations is padded with zeros, and the second combination is not padded, where the second combination is a combination of all combinations except the first combination, and the number of first combinations and second combinations is the same.

[0067] A fixed convolution kernel is a convolution kernel with a fixed weight configuration that is pre-set in the MultiRocket model and is used to extract features from the time series data corresponding to the user-side traffic data. Different convolution kernels of the same length have different weight configurations to capture the multi-scale features of the time series. In some embodiments of the present application, a set of fixed convolution kernels of length 9 can be defined, and the weight values ​​in each convolution kernel are different (such as one part has a weight of -1 and the other part has a weight of 2) to extract different time series features in the user-side traffic data. For example, one of the convolution kernels can be defined as {-1, -1, -1, -1, -1, 2, 2, 2, -1}. Based on this weight configuration, 84 fixed convolution kernels can be generated in the end. Through the design of fixed convolution kernels, local and global features in the time series can be effectively extracted, avoiding the problem of requiring a large number of parameter adjustments in traditional convolutional neural networks.

[0068] The expansion coefficient sequence is a parameter sequence that controls the weight interval in the convolution kernel, which is used to expand the receptive field so that the model can focus on long-distance dependencies in the time series. The maximum expansion coefficient (i.e., the maximum value of the expansion coefficient sequence) is determined according to the length of the traffic data to ensure the rationality of the convolution operation. By introducing the expansion coefficient, the model can take into account information of different time spans (such as the pattern and changes of user-side traffic data in different time windows) when extracting features. In some embodiments of the present application, each convolution kernel uses the same fixed expansion sequence, and the expansion coefficient value range is {2 0 ,…,2 max}, where the exponents are uniformly distributed in the interval [0, max], where max is calculated as:

[0069]

[0070] Among them, l input Indicates the input length of the time series (such as 147), l kernel is the convolution kernel length (e.g. fixed to 9).

[0071] It should be noted that the purpose of introducing the expansion factor is to expand the receptive field of the convolution kernel, allowing the model to capture both short- and long-term dependencies and cyclical characteristics in traffic data. For short-term dependencies, a small expansion factor allows the classification model to capture fluctuations in traffic data within an hour or even shorter timeframes. For example, these patterns correspond to specific user activities at a specific time, such as the momentary traffic surge caused by a large number of users accessing social media or video streaming services during the lunch rush hour. For medium-term dependencies, a medium expansion factor helps the model identify traffic trends within a day, such as peaks and valleys during the morning and evening rush hours, and differences in usage patterns between weekdays and weekends. These patterns reflect the rhythm of daily life and work activities within the coverage area of ​​the sector to be classified and are used to distinguish different types of coverage areas (such as office areas, residential areas, and commercial areas). For long-term dependencies, a larger expansion factor allows the model to focus on cyclical changes in traffic data over timescales of a week or even longer. For example, the difference in total traffic between weekdays and weekends, or monthly cyclical traffic peaks (such as the end-of-month top-up peak), can all be identified through analysis of long-term dependencies, thereby understanding the long-term usage patterns of the sector coverage area, such as seasonal tourist hotspots and holiday effects.

[0072] Bias refers to a constant bias value associated with each fixed convolution kernel and expansion coefficient combination, which is used to adjust the convolution output to better fit the target distribution. The bias value is determined by analyzing the output of the fixed convolution kernel on random samples, ensuring the robustness and adaptability of the model. In some embodiments of the present application, for each combination of fixed convolution kernel and expansion coefficient, a sample can be randomly selected from the training set, the convolution output of the combination on the sample is calculated, and the bias value is determined based on the output quantile (such as 25%, 50%, 75%, etc.).

[0073] Zero padding refers to adding zero values ​​at both ends of the time series to keep the length of the time series unchanged before and after the convolution operation, and to avoid the reduction of the sequence length caused by the convolution operation. In the MultiRocket model, zero padding is used for the first combination of all combinations, while no padding is used for the second combination, thereby introducing a variant of data enhancement and increasing the generalization ability of the model. This means that for all combinations of convolution kernels and expansion coefficients, zero padding is used for the first combination to maintain the length of the output sequence; the second combination (the remaining combinations of all combinations excluding the first combination) is not padded, allowing the convolution kernel to slide freely to produce output features of variable length. This alternating padding method increases the adaptability of the model to different types of time series data. Zero padding can maintain the sequence length, while no padding can introduce more changes, allowing the model to learn richer feature representations.

[0074] In some embodiments of the present application, user plane traffic data can be classified through the following steps to obtain the coverage area type of the sector to be classified, specifically: using the convolution layer in the time series-based classification model to determine the time series corresponding to the user plane traffic data and the first-order difference sequence corresponding to the time series, wherein the length of the time series is greater than the length of the first-order difference sequence; the convolution layer uses different random convolution kernel sets to perform convolution operations on the time series and the first-order difference sequence respectively to obtain a target feature map, wherein the convolution kernel in the random convolution kernel set is a fixed convolution kernel deformed by different expansion coefficients and bias values. The convolution kernel instance is obtained; the pooling layer in the time series-based classification model determines multiple pooling features corresponding to the target feature map output by each random convolution kernel, and determines a multidimensional feature vector corresponding to the target feature map based on the pooling features, wherein the pooling features are used to summarize the distribution characteristics of the target feature map; the output layer in the time series-based classification model determines the coverage area type based on the multidimensional feature vector.

[0075] A first-order difference series is a new series derived from performing a first-order difference operation on the original time series data (i.e., subtracting the value at the previous time point from the current time point). This is used to eliminate trend components in the time series and highlight the rate of change within the series, thereby identifying sudden traffic events or cyclical changes. The introduction of a first-order difference series helps eliminate long-term trends, allowing the model to focus more on sudden changes and cyclical characteristics in traffic, and is beneficial for capturing instantaneous traffic fluctuations or cyclical patterns in different coverage areas (i.e., sectors).

[0076] Specifically, user plane traffic data is extracted from the base station network management system to form a time series. A first-order difference operation is performed on the time series to obtain a first-order difference sequence. Since the first-order difference sequence is one data point shorter than the original time series, the time series length is greater than the first-order difference sequence length. The convolution layer uses multiple different sets of random convolution kernels to perform convolution operations on the time series and its first-order difference sequence (each using the MultiRocket model for convolution operations) to generate their respective target feature maps. It should be noted that by applying a set of random convolution kernels, the model can extract traffic data features from multiple angles and scales. The pooling layer calculates multiple pooling features (also called pooling operators, such as PPV, MPV, MIPV, and LSPV) for the target feature map output by each convolution kernel and merges all pooling features into a multidimensional feature vector. The output layer uses a trained classifier (such as a ridge regression classifier) ​​based on the multidimensional feature vector to determine the coverage area type. By integrating all extracted features, the classifier can comprehensively consider the time dependence, sudden changes, and periodic patterns of the traffic data to accurately determine the coverage area type to which the sector belongs.

[0077] It should be noted that in the above process, after the convolution operation, MultiRocket can calculate four features (called pooling operators) for the output Z (i.e., the target feature map, length n) of each random convolution kernel to summarize the distribution characteristics of Z. The explanation of the four pooling operators is shown in Table 1:

[0078] Table 1: Overview of pooling operators.

[0079] Pooling operator describe Positive Proportion (PPV) The proportion of positive values ​​in Z to the total value. Mean Positive Value (MPV) The arithmetic mean of all positive values ​​in Z. Mean Positive Position Value (MIPV) The positive values ​​in Z correspond to the mean of the index positions (such as the 1st, 3rd, and 5th points). Longest consecutive positive value length (LSPV) Length of the longest segment of consecutive positive values ​​in Z.

[0080] The four pooling operators, namely proportion of positive values ​​(PPV), mean positive values ​​(MPV), mean positive position values ​​(MIPV) and longest continuous positive value length (LSPV), are used to extract key information that helps to classify the wireless sector coverage area from the feature map Z obtained by the convolution operation. Specifically:

[0081] (1) PPV can reflect the activity of a sector in different time periods and can be used to identify areas with stable or sporadic traffic demand. For example, high PPV may indicate a dense commercial area, while low PPV may correspond to nighttime or low population density areas.

[0082] (2) MPV helps to quantify the intensity of sector traffic activity and is particularly useful for distinguishing between high traffic load areas (such as large shopping malls and stadiums) and low traffic load areas (such as remote suburbs).

[0083] (3) MIPV provides information on the temporal distribution of traffic activity, which can assist the model in determining whether a sector follows a typical weekday or weekend traffic pattern, which is crucial for identifying specific coverage areas (such as office areas and leisure areas).

[0084] (4) LSPV emphasizes the continuity of traffic activities and is very helpful in identifying areas that experience long-term high-intensity traffic activities (such as airport waiting halls and train station platforms). It is also an important indicator for monitoring network congestion and resource pressure.

[0085] In order to ensure that the classification of wireless network coverage areas is not only based on stable long-term trends but also takes into account short-term traffic fluctuations, the target feature map can be determined by the following steps: determining the maximum expansion coefficient corresponding to the time series and the first-order difference sequence respectively, and determining the expansion sequences corresponding to the maximum expansion coefficients respectively, wherein the maximum expansion coefficient of the time series is greater than the maximum expansion coefficient of the first-order difference sequence, and the expansion sequence is used to perform an expansion operation on the fixed convolution kernel; according to the expansion sequence, determining a first random convolution kernel set corresponding to the time series and a second random convolution kernel set corresponding to the first-order difference sequence respectively; using the random convolution kernel of the first random convolution kernel set to convolve the time series to obtain a first feature map, and using the random convolution kernel of the second random convolution kernel set to convolve the first-order difference sequence to obtain a second feature map; and determining the target feature map corresponding to the first feature map and the second feature map.

[0086] In the above process, the maximum expansion coefficient is determined based on the length of the time series. Since the time series contains the original data points, its maximum expansion coefficient is greater than the maximum expansion coefficient of the first-order difference sequence. For example, if the time series length is 147, the maximum expansion coefficient may be set to a value close to log2(147). The length of the first-order difference sequence is reduced by 1, so its maximum expansion coefficient is correspondingly reduced. Based on the determined expansion sequence, different expansion coefficients and bias values ​​are applied to each set of fixed convolution kernels to create a first random convolution kernel set and a second random convolution kernel set. The original time series is convolved using the first random convolution kernel set to obtain a first feature map. The first-order difference sequence is convolved using the second random convolution kernel set to obtain a second feature map. By using the first feature map and the second feature map, the model can take into account the stability and variability of traffic data when extracting features, which is particularly important for identifying coverage areas (such as business districts and transportation hubs) where traffic patterns have both regular trends and sudden changes. The first feature map and the second feature map are integrated, such as by splicing, adding, etc., to generate the final target feature map.

[0087] It should be noted that in the process of integrating the first feature map and the second feature map, in order to make up for the lack of long-term dependency information in the first-order difference sequence, a complementary strategy can be adopted, such as the attention mechanism or feature weighting in deep learning technology, to ensure that even when the sequence length changes or some information is lost, the model can still use the complementarity between different feature maps to make accurate coverage area classification. Specifically: add a position code Epos (a predefined position vector) to the first feature map Z1 and the second feature map Z2 respectively, and use sine and cosine functions to encode the position information of the time point to enhance the model's sensitivity to the position in the time series; use a small neural network to learn the position-aware weight Wp, which accepts the combination of the position code Epos and the feature map Z (Z1 and / or Z2) as input and outputs a weight vector to reflect the importance of each time point in the feature map based on position; perform weighted summation of the position-aware weight Wp with the corresponding feature map Z (Z1 and / or Z2) to obtain the position-aware attention feature vector Zp; obtain the target feature map based on the position-aware attention feature vector.

[0088] In the process of learning location-aware weights in a small neural network, (1) if the goal is to enhance the model's sensitivity to the location of time points in the original time series features (i.e., Z1), then the input for learning location-aware weights will be a combination of the location code Epos and Z1. In this way, the learned Wp will highlight the importance of each time point in the original traffic data based on location, helping the model better understand long-term trends or cyclical patterns.

[0089] (2) On the contrary, if the purpose is to emphasize the importance of the time point position in the first-order difference series feature (i.e., Z2), then the input should be a combination of Epos and Z2. In this case, Wp will be used to evaluate the relevance of the flow rate change at a specific time point, which is particularly useful for capturing sudden flow surges or drops, especially when it is necessary to identify areas where the flow pattern changes suddenly.

[0090] (3) Z can also be a fusion of Z1 and Z2, such as a feature map generated by summation, splicing, etc. This fusion takes into account the stability and variability of traffic data, so that the location-aware weight Wp can simultaneously reflect the location importance of time points from two different perspectives, thereby providing the model with a more comprehensive understanding of traffic data.

[0091] The location-aware attention mechanism focuses on the temporal characteristics of traffic data. Traffic data at different points in time has different meanings depending on its position in the time series. For example, traffic data during peak hours in the morning and evening is crucial for distinguishing commercial and residential areas, while traffic data at night may be more valuable for identifying industrial or remote areas.

[0092] In some embodiments of the present application, the output layer in the time series-based classification model can perform the following steps: determine the weight vector of the ridge regression classifier, wherein the weight vector is used to quantify the importance of each feature in the multidimensional feature vector to the classification result, and the weight vector corresponds to the dimension of the multidimensional feature vector; perform a dot product between the weight vector and the multidimensional feature vector to obtain a predicted value of the multidimensional feature vector, and determine the coverage area type based on the predicted value.

[0093] In the above process, the multidimensional feature vector includes a combination of features extracted based on the MultiRocket model, such as the feature graph of the original time series, the feature graph of the first-order difference sequence, and the feature statistics generated after the pooling operation of these feature graphs (such as PPV, MPV, MIPV and LSPV, etc.).

[0094] The weight vector is a parameter in the ridge regression classifier, which is used to quantify the importance of each feature in the multidimensional feature vector to the classification result. The weight vector has the same dimension as the multidimensional feature vector, and each element corresponds to the coefficient of the corresponding feature in the multidimensional feature vector. A larger weight indicates that the feature has higher discriminative power and importance in the classification task.

[0095] During the model training phase, the weight vector of the ridge regression classifier is obtained through a supervised learning process, where the weights are estimated by minimizing a loss function with an L2 regularization term. Specifically, for a given set of training samples, each sample contains a multidimensional feature vector and its corresponding coverage area type label. The ridge regression classifier uses iterative optimization to find the weight vector that minimizes the prediction error. For each sector to be classified, its corresponding multidimensional feature vector is dot-producted with the weight vector of the ridge regression classifier. The result of the dot-product operation can be considered a comprehensive score, representing the likelihood that the sector belongs to a specific coverage area type.

[0096] In some embodiments of the present application, MultiRocket generates approximately 50,000 feature vectors, and a RidgeClassifier is used to perform supervised learning on the feature vectors, outputting coverage category labels (e.g., dining area, shopping area, dense urban area, suburban area, high-speed rail scene, highway scene, etc.). Specifically:

[0097] (1) The ridge regression classifier is a linear classification method based on L2 regularization. Its core idea is to alleviate the multicollinearity problem between features by introducing regularization terms, while improving the generalization ability of the model. Its mathematical expression is similar to that of ridge regression (regression task), but the target variable is processed accordingly in the classification task.

[0098] (2) The basic principle of classification is to minimize the following loss function with regularization:

[0099]

[0100] in, It is the least ordinary squares loss function, which measures the difference between the predicted value and the true value;

[0101] is the L2 regularization term, which penalizes the sparse absolute value sum of squares to prevent overfitting. α is the regularization strength parameter, which controls the weight of the regularization term.

[0102] (3) Parameter estimation: By solving the optimization problem, the parameter β (i.e., weight vector) of the ridge regression classifier can be obtained by the following closed-form solution:

[0103] β=(X T X+αI) -1 X T Y

[0104] Where X is the input feature matrix, Y is the target variable (classification label), and I is the identity matrix to ensure that the matrix is ​​invertible.

[0105] In the case where the coverage area type includes two types, the coverage area type is determined based on the positive and negative signs of the predicted value; in the case where the coverage area type includes more than two types, a weight vector matrix corresponding to multiple pending coverage area types is determined, and the predicted value corresponding to each pending coverage area type is determined based on the weight vector matrix, and the pending coverage area type corresponding to the maximum predicted value is determined as the coverage area type. Specifically:

[0106] (1) Binary classification: The target variable Y can be converted into a binary form {-1, 1} (for example, when the original label is {0, 1}, it is converted into -1 and 1). Therefore, the problem is regarded as a regression problem, and the ridge regression model is used to fit the parameter β to obtain the predicted value:

[0107]

[0108] in, is the predicted value, X is the input feature matrix, and β is the ridge regression model fitting parameter.

[0109] Classify based on the sign of the regression prediction value:

[0110]

[0111] (2) Multi-classification: The multi-classification problem can be viewed as a multi-output regression problem, where each category corresponds to an independent regression task. Assuming there are K categories, the model will learn a parameter vector β for each category k. k , and finally form a parameter matrix β. For each sample, calculate its regression output for each category:

[0112]

[0113] According to the classification results, the sample is assigned to the class with the largest predicted value:

[0114]

[0115] Through the above steps S202 to S204, by collecting the protocol layer traffic data of the base station sector, using the model to capture the time series characteristics in the traffic data, and extracting and classifying the sectors, the purpose of accurately identifying the type of coverage area of ​​the sector to be classified is achieved, thereby achieving the technical effect of improving the efficiency of wireless network resource allocation and optimizing network performance, and further solving the technical problem that the coverage scenario recognition method adopted by the related technology depends on physical parameters such as base station antenna information and geographic location information, resulting in low accuracy in the classification of wireless sector coverage areas.

[0116] In some embodiments of this application, in order to improve the efficiency and accuracy of wireless sector coverage area classification, the MultiRocket model is further optimized and adapted for actual application scenarios in 4G / 5G networks to ensure its stability and effectiveness in complex network environments. This includes but is not limited to the following points:

[0117] (1) Optimization of the model architecture. The basic architecture of the MultiRocket model is optimized to process traffic data of 4G / 5G networks more efficiently. For example, a fixed-length convolution kernel group (such as a fixed convolution kernel of length 9) and a specific expansion coefficient and bias strategy are used to improve feature diversity while controlling the complexity of the model to ensure fast response on large-scale data sets. By defining a fixed convolution kernel, using a specific expansion coefficient sequence, determining the bias value based on quantiles, and alternating between zero padding and no padding strategies, the MultiRocket model in this application is able to fully capture the patterns and changes of traffic data at different time scales while maintaining computational efficiency.

[0118] (2) The introduction of a pre-training mechanism aims to utilize a large amount of historical traffic data so that the model can already process and classify a variety of coverage scenarios before starting to classify specific sectors. During the pre-training phase, the model is trained on a variety of coverage scenario data to learn generally effective traffic feature extraction capabilities. Before being applied to a specific sector (the sector to be classified), online fine-tuning is performed, updating only the weights of the classifier while keeping the convolution kernel parameters unchanged. This not only maintains the generalization ability of the model, but also ensures that the model can quickly adapt to the traffic patterns of new sectors, improving the accuracy and efficiency of classification.

[0119] (3) Regarding the processing of data at different layers of 4G / 5G, and taking into account the differences between the 4G and 5G network protocol layers, this application specifically configures the MultiRocket model so that it can process traffic data at the 4G PDCP layer and the 5G RLC layer respectively, thereby ensuring that the model maintains consistent performance in applications across different network types. By adjusting the number of convolution kernels, the range of the expansion coefficient, and the setting of the pooling operator, the model can automatically adapt to the characteristics of data at different layers. For example, due to the shortened length of the first-order difference time series, its maximum expansion coefficient is adjusted accordingly to match the requirements of feature extraction. This adaptive processing method enhances the adaptability and robustness of the model in different wireless environments.

[0120] This application uses 4G / 5G protocol layer traffic data (PDCP / RLC) as the core feature of classification input, establishes the association between protocol layer behavior and coverage scenarios, and provides a more accurate data basis for classification. The use of traffic data can eliminate the differences caused by different equipment manufacturers. By adjusting model parameters and data preprocessing methods, it can be applied to various complex network scenarios. In addition, the lightweight time series classification framework based on MultiRocket supports hourly data input and millisecond-level decision-making, with faster computing speed. Its convolution kernel group configuration and feature statistics calculation logic can effectively extract traffic data features and improve classification accuracy and efficiency. In addition, the model input data complies with the PM format defined by the 3GPP TS 32.425 standard and can be directly connected to the operator's OSS system, which is convenient for application and promotion in the existing network management system.

[0121] Figure 3 is a structural diagram of a classification device for coverage areas according to an embodiment of the present application, such as Figure 3 As shown, the device includes:

[0122] An acquisition module 302 is configured to acquire user plane traffic data of a sector to be classified from a base station network management system, wherein the sector to be classified is a sub-area of ​​a coverage area of ​​a base station, and the user plane traffic data includes total uplink and downlink traffic data;

[0123] The classification module 304 is configured to classify the user plane traffic data using a time series-based classification model to obtain the coverage area type of the sector to be classified.

[0124] It should be noted that Figure 3 The classification device of the coverage area shown is used to perform Figure 2 The classification method of the coverage area shown is therefore Figure 2 The explanations in the classification of coverage areas in the Figure 3 The classification device of the coverage area shown is not described in detail here.

[0125] An embodiment of the present application further provides an electronic device, which includes a memory and a processor, wherein the memory is used to store program instructions; the processor is connected to the memory and is used to execute the steps of the coverage area classification method in each embodiment of the present application.

[0126] For example, the processor performs the following functions by executing program instructions stored in the memory: obtaining user-plane traffic data of the sector to be classified from the base station network management system, wherein the sector to be classified is a sub-area of ​​the coverage area of ​​the base station, and the user-plane traffic data includes total uplink and downlink traffic data; using a time series-based classification model to classify the user-plane traffic data to obtain the coverage area type of the sector to be classified.

[0127] An embodiment of the present application further provides a non-volatile storage medium, which includes a stored computer program, wherein the device where the non-volatile storage medium is located executes the steps of the coverage area classification method in each embodiment of the present application by running the computer program.

[0128] An embodiment of the present application further provides a computer program product, including computer instructions, which, when executed by a processor, implement the steps of the coverage area classification method in each embodiment of the present application.

[0129] The embodiments of the present application further provide a computer program, which, when executed by a processor, implements the steps of the coverage area classification method in each embodiment of the present application.

[0130] The serial numbers of the above embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0131] In the above embodiments of the present application, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0132] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only exemplary. For example, the division of the units can be a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0133] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0134] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0135] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for enabling a computer device (which can be a personal computer, a server or a network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk.

[0136] The above is only a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for classifying coverage areas, characterized in that: include: Acquire user plane traffic data of a sector to be classified from a base station network management system, wherein the sector to be classified is a sub-area of ​​a coverage area of ​​a base station, and the user plane traffic data includes total uplink and downlink traffic data; The user plane traffic data is classified using a time series-based classification model to obtain the coverage area type of the sector to be classified.

2. The method according to claim 1, characterized in that The user plane traffic data is classified using a time series-based classification model to obtain the coverage area type of the sector to be classified, including: A convolution layer in a time series-based classification model is used to determine a time series corresponding to the user plane traffic data and a first-order difference sequence corresponding to the time series, wherein the length of the time series is greater than the length of the first-order difference sequence; the convolution layer uses different random convolution kernel sets to perform convolution operations on the time series and the first-order difference sequence, respectively, to obtain a target feature map, wherein the convolution kernels in the random convolution kernel set are multiple convolution kernel instances obtained by deforming a fixed convolution kernel using different expansion coefficients and bias values; The pooling layer in the time series-based classification model determines a plurality of pooling features corresponding to the target feature map output by each random convolution kernel, and determines a multidimensional feature vector corresponding to the target feature map based on the pooling features, wherein the pooling features are used to summarize the distribution characteristics of the target feature map; An output layer in the time series-based classification model determines the coverage area type according to the multidimensional feature vector.

3. The method according to claim 2, characterized in that The output layer in the time series-based classification model determines the coverage area type according to the multidimensional feature vector, including: Determining a weight vector of a ridge regression classifier, wherein the weight vector is used to quantitatively represent the importance of each feature in the multidimensional feature vector to the classification result, and the weight vector corresponds to the dimension of the multidimensional feature vector; Performing a dot product between the weight vector and the multidimensional feature vector to obtain a predicted value of the multidimensional feature vector, and determining the coverage area type according to the predicted value.

4. The method according to claim 3, characterized in that Determining the coverage area type according to the predicted value includes: In a case where the coverage area type includes two types, determining the coverage area type according to a positive or negative sign of the predicted value; In the case where the coverage area type includes more than two types, determine the weight vector matrix corresponding to multiple pending coverage area types, determine the prediction value corresponding to each pending coverage area type based on the weight vector matrix, and determine the pending coverage area type corresponding to the maximum prediction value as the coverage area type.

5. The method according to claim 1, wherein The time series-based classification model is configured in the following way: Determining a plurality of fixed convolution kernels, wherein the plurality of fixed convolution kernels have the same length and different fixed convolution kernels have different weight configurations, and the fixed convolution kernels are used to extract time series features of the user plane traffic data; dilating each of the fixed convolution kernels using a dilation coefficient sequence, wherein a maximum value of the dilation coefficient sequence is determined according to the length of the flow data; The bias corresponding to each combination of the fixed convolution kernel and the dilation coefficient is determined according to the output of each fixed convolution kernel for a random sample; The first combination among all the combinations is padded with zeros, and the second combination is not padded, wherein the second combination is the combination among all the combinations except the first combination, and the number of the first combination and the number of the second combination are the same.

6. The method according to claim 2, characterized in that The convolution layer uses different random convolution kernel sets to perform convolution operations on the time series and the first-order difference series respectively to obtain a target feature map, including: Determining maximum dilation coefficients corresponding to the time series and the first-order difference sequence, respectively, and determining dilation sequences corresponding to the maximum dilation coefficients, respectively, wherein the maximum dilation coefficient of the time series is greater than the maximum dilation coefficient of the first-order difference sequence, and the dilation sequences are used to perform a dilation operation on the fixed convolution kernel; Determining a first random convolution kernel set corresponding to the time series and a second random convolution kernel set corresponding to the first-order difference sequence according to the dilation sequence; Performing a convolution operation on the time series using a random convolution kernel of the first random convolution kernel set to obtain a first feature map, and performing a convolution operation on the first-order difference sequence using a random convolution kernel of the second random convolution kernel set to obtain a second feature map; Determine the target feature map corresponding to the first feature map and the second feature map.

7. The method according to claim 1, characterized in that The time series-based classification model is trained in the following way: Acquire historical user plane traffic data of multiple sectors of multiple base stations, wherein the historical user plane traffic data includes historical user plane traffic data of sectors of base stations based on a fourth generation mobile communication system and historical user plane traffic data of sectors of base stations based on a fifth generation mobile communication system; Using an initial classification model to classify the historical user plane traffic data to obtain an initial coverage area type; When the initial coverage area type does not meet the preset conditions, the parameters of the initial classification model are adjusted and the initial coverage area type is updated until the updated initial coverage area type meets the preset conditions, and the updating of the parameters is stopped to obtain the time series-based classification model.

8. A classification device for a coverage area, characterized in that: include: An acquisition module, configured to acquire user plane traffic data of a sector to be classified from a base station network management system, wherein the sector to be classified is a sub-area of ​​a coverage area of ​​a base station, and the user plane traffic data includes total uplink and downlink traffic data; The classification module is used to classify the user plane traffic data using a time series-based classification model to obtain the coverage area type of the sector to be classified.

9. An electronic device, characterized in that: include: A memory and a processor, wherein the memory is used to store program instructions; the processor is connected to the memory and is used to execute the coverage area classification method according to any one of claims 1 to 7.

10. A non-volatile storage medium, characterized in that: The non-volatile storage medium includes a stored computer program, wherein the device where the non-volatile storage medium is located executes the coverage area classification method according to any one of claims 1 to 7 by running the computer program.

11. A computer program product comprising computer instructions, characterized in that When the computer instructions are executed by a processor, the coverage area classification method according to any one of claims 1 to 7 is implemented.

Images