Network prediction method, system and device and storage medium

Abstract

The invention discloses a network prediction method, system and device and a storage medium. The method comprises the following steps: acquiring user side data, base station data and topographic data in a target area; the user side data comprises user side moving time and user side longitude and latitude; the base station data comprises a base station load and base station longitude and latitude; the user side data, the base station data and the topographic data are input into the network prediction model, the network prediction result output by the network prediction model is obtained, network changes can be accurately predicted according to network historical data, real-time dynamic resource allocation is achieved according to the prediction result, and then the network prediction precision, the real-time performance and the resource utilization rate are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of communication, in particular to a network prediction method, system, device and storage medium. BACKGROUND

[0002] With the large-scale deployment of 5G networks and the deep integration of intelligent terminal applications, network traffic is growing exponentially, users' demand for service quality is increasingly stringent, and network service scenarios are becoming increasingly complex and dynamic. High-density user access, differentiated business needs, and multi-physical environment coupling have higher requirements for network prediction accuracy, resource allocation real-time performance, and environmental adaptability. Traditional network prediction techniques based on fixed rules and static models have been difficult to meet the new generation of communication requirements for low latency, high reliability, and self-adaptation.

[0003] In the field of network prediction and optimization, existing conventional techniques usually rely on statistical analysis of single-dimensional network indicators, resulting in one-sided network state perception, mismatch between resource allocation and actual scene demand, and difficulty in capturing nonlinear time-varying characteristics in complex environments. It cannot support millisecond-level dynamic response in burst traffic scenarios, resulting in real-time prediction, significantly reduced prediction reliability and network resource utilization efficiency. SUMMARY

[0004] The following is a summary of the subject matter described in detail herein. This summary is not intended to limit the scope of the claims.

[0005] The main purpose of the embodiments of the present disclosure is to provide a network prediction method, system, device and storage medium, which can accurately predict network changes according to network historical data, so as to realize real-time dynamic resource allocation according to the prediction results, thereby significantly improving network prediction accuracy, real-time performance and resource utilization.

[0006] The first aspect of the embodiments of the present application provides a network prediction method for a central controller, the method comprising:

[0007] Obtaining user terminal data, base station data and terrain data in a target area; the user terminal data includes user terminal moving time and user terminal latitude and longitude; the base station data includes base station load and base station latitude and longitude;

[0008] Inputting the user terminal data, the base station data and the terrain data into a network prediction model to obtain a network prediction result output by the network prediction model;

[0009] The prediction process of the network prediction model comprises:

[0010] Interpolating the base station data to obtain first base station data;

[0011] According to the first base station data, the user terminal data is time-aligned and space-fused to obtain first user terminal data;

[0012] Based on the user terminal moving time and the user terminal latitude and longitude, the user terminal is matched with the base station to obtain a matching result;

[0013] According to the matching result, the first user terminal data, the first base station data and the terrain data are fused to obtain corresponding fusion data;

[0014] According to the fusion data, a network prediction result of each base station is calculated.

[0015] In some embodiments of the present application, the training process of the network prediction model comprises:

[0016] An initial network prediction model is preset;

[0017] Training user terminal data, training base station data and training terrain data are obtained; the training user terminal data comprises training user terminal moving time and training user terminal latitude and longitude;

[0018] The training base station data is interpolated to obtain first training base station data;

[0019] According to the first training base station data, the training user terminal data is time-aligned and space-fused to obtain first training user terminal data;

[0020] Based on the training user terminal moving time and the training user terminal latitude and longitude, the training user terminal is matched with the training base station to obtain a training matching result;

[0021] According to the training matching result, the first training user terminal data, the first training base station data and the training terrain data are fused to obtain corresponding training fusion data;

[0022] A training simplified feature matrix is extracted from the training fusion data;

[0023] According to the training fusion data and the training simplified feature matrix, the initial network prediction model is optimized to obtain the network prediction model.

[0024] In some embodiments of the present application, the training base station data is interpolated to obtain first training base station data, comprising:

[0025] The training base station data is processed by cubic spline interpolation to obtain the first training base station data;

[0026] The calculation formula of the interpolation calculation comprises:

[0027] L interp (t) = a(t - t i ) 3 +b(t - t i ) 2 +c(t - t i )+d(t i ≤t<t i+1 );

[0028] wherein, L interp (t) is an interpolation base station load at time t, t i is a time point of the i-th base station load, t i+1 is a time point of the i+1-th base station load, t i and t i+1 are adjacent time points, a, b, c and d are all preset spline interpolation coefficients;

[0029] The training user end data is time-aligned and space-fused according to the first training base station data, to obtain first training user end data, which comprises:

[0030] aligning the training user end data to a preset space-time scale to generate space-time aligned data;

[0031] constructing a corresponding Thiessen polygon according to the first training base station data;

[0032] spatially connecting the space-time aligned data and the Thiessen polygon based on the first training base station data, to obtain the first training user end data.

[0033] In some embodiments of the present application, the first user end data, the first base station data and the terrain data are fused according to the matching result to obtain corresponding fusion data, which comprises:

[0034] based on the terrain data, the first user end data and the first base station data are processed to obtain second user end data and second base station data; the data processing at least includes interpolation processing, local outlier factor abnormal processing and wavelet threshold denoising processing;

[0035] the second user end data, the second base station data and the terrain data are fused to obtain the fusion data.

[0036] In some embodiments of the present application, the training simplified feature matrix is extracted from the training fusion data, which comprises:

[0037] the capacity demand feature weight is calculated according to the training fusion data;

[0038] a training initial feature matrix is extracted from the training fusion data;

[0039] mapping the training initial feature matrix into a query matrix, a key matrix and a value matrix;

[0040] calculating a signal quality feature weight according to the query matrix, the key matrix and the value matrix;

[0041] calculating a fusion weight according to the capacity requirement feature weight and the signal quality feature weight;

[0042] performing feature screening on the training initial feature matrix according to the fusion weight to obtain a training simplified feature matrix.

[0043] In some embodiments of the present application, the optimizing the initial network prediction model according to the training fusion data and the training simplified feature matrix to obtain the network prediction model comprises:

[0044] optimizing the initial network prediction model according to the training simplified feature matrix to obtain a first network prediction model;

[0045] inputting the training fusion data into the first network prediction model to obtain a first network prediction result output by the first network prediction model;

[0046] calculating a loss function according to an actual network value of the training fusion data and the first network prediction result to obtain a corresponding loss error;

[0047] optimizing the first network prediction model according to the loss error to obtain the network prediction model.

[0048] In some embodiments of the present application, after the inputting the user end data, the base station data and the terrain data into the network prediction model to obtain the network prediction result output by the network prediction model, the method further comprises:

[0049] calculating a network optimization strategy of the target area according to the network prediction result of each base station;

[0050] performing network optimization on the target area according to the network optimization strategy to obtain an optimized target area network.

[0051] To achieve the above object, a second aspect of the embodiments of the present application provides a network prediction system, which comprises:

[0052] an acquisition module configured to acquire user end data, base station data and terrain data in a target area; the user end data comprises user end moving time and user end latitude and longitude; the base station data comprises base station load and base station latitude and longitude;

[0053] a prediction module configured to input the user terminal data, the base station data, and the terrain data into a network prediction model to obtain a network prediction result output by the network prediction model;

[0054] The prediction process of the network prediction model comprises:

[0055] interpolating the base station data to obtain first base station data;

[0056] aligning the user terminal data in time and fusing the user terminal data in space according to the first base station data to obtain first user terminal data;

[0057] matching the user terminal and the base station based on the user terminal moving time and the user terminal latitude and longitude to obtain a matching result;

[0058] fusing the first user terminal data, the first base station data, and the terrain data according to the matching result to obtain corresponding fused data;

[0059] calculating the network prediction result of each base station according to the fused data.

[0060] To achieve the above object, a third aspect of the embodiments of the present application provides an electronic device, comprising: at least one control processor and a memory in communication connection with the at least one control processor; the memory stores instructions executable by the at least one control processor, and the instructions are executed by the at least one control processor to enable the at least one control processor to execute the network prediction method.

[0061] To achieve the above object, a fourth aspect of the embodiments of the present application provides a computer readable storage medium, which stores computer executable instructions for enabling a computer to execute the network prediction method.

[0062] The embodiments of the present application provide a network prediction method, which obtains user terminal data, base station data, and terrain data in a target region; the user terminal data comprises user terminal moving time and user terminal latitude and longitude; the base station data comprises base station load and base station latitude and longitude; the user terminal data, the base station data, and the terrain data are input into a network prediction model to obtain a network prediction result output by the network prediction model, which can accurately predict network changes according to network historical data, so as to realize real-time dynamic resource allocation according to the prediction result, and further significantly improve network prediction accuracy, real-time performance, and resource utilization.

[0063] It can be understood that the beneficial effects of the second aspect to the fourth aspect and related technologies compared with the first aspect and related technologies are the same as the beneficial effects of the first aspect and related technologies, which can be referred to the related description in the first aspect and will not be repeated here. BRIEF DESCRIPTION OF DRAWINGS

[0064] The above and / or additional aspects and advantages of the present application will become apparent and be readily appreciated from the following description, including the references to the figures, in which:

[0065] Figure 1 is a flow diagram of a network prediction method provided by an embodiment of the present application;

[0066] Figure 2 is a structural diagram of a network prediction training system provided by an embodiment of the present application;

[0067] Figure 3 is a hardware structure diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0068] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, in which the same or similar notations represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be understood as a limitation of the present application.

[0069] In the description of the present application, if there is a description of first, second, etc., it is only for the purpose of distinguishing technical features, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features or the order of indicated technical features.

[0070] In the description of the present application, it should be understood that the position description, such as the position or location relationship indicated by up, down, etc., is based on the position or location relationship shown in the drawings, and is only for the convenience of describing the present application and simplifying the description, and therefore cannot be understood as indicating or implying that the device or element indicated must have a particular position, be constructed and operated in a particular position, and therefore cannot be understood as a limitation of the present application.

[0071] In the description of the present application, it should be noted that, unless otherwise explicitly limited, the words such as setting, installing, connecting, etc. should be broadly understood, and the person skilled in the art can reasonably determine the specific meaning of the above words in the present application in combination with the specific content of the technical solution.

[0072] With the large-scale deployment of 5G networks and the deep integration of intelligent terminal applications, network traffic is growing exponentially, users' demand for service quality is increasingly stringent, and network service scenarios are becoming increasingly complex and dynamic. High-density user access, differentiated service requirements, and multi-physical environment coupling have higher requirements for network prediction accuracy, resource allocation real-time performance, and environmental adaptability. Traditional network prediction techniques based on fixed rules and static models have been difficult to meet the new generation of communication requirements for low latency, high reliability, and self-adaptation.

[0073] In the field of network prediction and optimization, existing conventional techniques usually rely on statistical analysis of single-dimensional network indicators, resulting in one-sided network state perception, mismatch between resource allocation and actual scene requirements, and difficulty in capturing nonlinear time-varying characteristics in complex environments, which cannot support millisecond-level dynamic response in burst traffic scenarios, resulting in real-time prediction, significantly reduced prediction reliability, and network resource utilization efficiency.

[0074] Based on this, the embodiments of the present application provide a network prediction method, system, electronic device and storage medium, aiming to accurately predict network changes according to network historical data, to realize real-time dynamic resource allocation according to the prediction results, and to significantly improve network prediction accuracy, real-time performance and resource utilization.

[0075] The network prediction method, system, electronic device and storage medium provided by the embodiments of the present application are specifically explained by the following embodiments. First, the network prediction method in the embodiments of the present application is described.

[0076] The embodiments of the present application can acquire and process related data based on artificial intelligence technology. Among them, artificial intelligence (AI) is to use digital computers or digital computer controlled machines to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. Theory, method, technology and application system.

[0077] The basic technology of artificial intelligence generally includes technologies such as sensors, special artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction system, mechatronics, etc. Artificial intelligence software technology mainly includes computer vision technology, robot technology, biometric technology, speech processing technology, natural language processing technology, and machine learning / deep learning, etc. Several major directions.

[0078] The network prediction method provided in the embodiments of the present application relates to the technical field of communication. The network prediction method provided in the embodiments of the present application can be applied to a terminal, can be applied to a server end, and can also be software running in the terminal or the server end. In some embodiments, the terminal can be a smart phone, a tablet computer, a notebook computer, a desktop computer, or the like; the server end can be configured as a stand-alone physical server, can be configured as a server cluster or a distributed system formed by multiple physical servers, can also be configured as a cloud server providing basic cloud computing services such as cloud service, cloud database, cloud computing, cloud function, cloud storage, network service, cloud communication, middleware service, domain name service, security service, CDN, and big data and artificial intelligence platform; and the software can be an application that implements the network prediction method, but is not limited to the above forms.

[0079] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld devices or portable devices, tablet devices, multi-processor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, and the like. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, and the like that perform specific tasks or implement specific abstract data types. The present application can also be practiced in a distributed computing environment, in which tasks are performed by remote processing devices connected by a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media, including storage devices.

[0080] It should be noted that in each specific embodiment of the present application, when relevant processing needs to be performed on data related to the identity or characteristics of the user, such as user information, user behavior data, user history data, and user location information, the user's permission or consent will be obtained first, and the collection, use, and processing of such data will comply with relevant laws, regulations, and standards. In addition, when the embodiments of the present application need to obtain sensitive personal information of the user, the separate permission or separate consent of the user will be obtained through a pop-up window or a jump to a confirmation page, and after obtaining the separate permission or separate consent of the user, the necessary user-related data for enabling the embodiments of the present application to normally operate will be obtained.

[0081] To this end, with reference to Figure 1 The embodiments of the present application provide a network prediction method, which is applied to a central controller. The controller can be a server, can be an electronic device, or can be a mobile terminal, and is not specifically limited here. The method comprises the following steps S110 to S120:

[0082] In step S110, user terminal data, base station data and terrain data in a target area are acquired; the user terminal data includes user terminal moving time and user terminal latitude and longitude; the base station data includes base station load and base station latitude and longitude;

[0083] In step S120, the user terminal data, the base station data and the terrain data are input into a network prediction model to obtain a network prediction result output by the network prediction model;

[0084] The prediction process of the network prediction model includes:

[0085] The base station data is subjected to interpolation calculation to obtain first base station data;

[0086] According to the first base station data, the user terminal data is subjected to time alignment and spatial fusion to obtain first user terminal data;

[0087] The user terminal is matched with the base station based on the user terminal moving time and the user terminal latitude and longitude to obtain a matching result;

[0088] According to the matching result, the first user terminal data, the first base station data and the terrain data are fused to obtain corresponding fused data;

[0089] The network prediction result of each base station is calculated according to the fused data.

[0090] In this step, first, user terminal data, base station data and terrain data in a target area are acquired; the target area can be any scene in which network prediction is needed, and the scene includes user terminals and base stations in the network in the scene, wherein the user terminals can move and change at different time points and transfer to different base stations; by acquiring different dimension data in the target area, comprehensive information basis is provided for subsequent prediction.

[0091] Specifically, the user terminal data refers to a data set including user terminal moving time and user terminal latitude and longitude, which can be acquired through real-time location information and time stamp reported by the user terminal, and is used to reflect user moving track and network access time distribution; the base station data refers to a data set including base station load and base station latitude and longitude, which can be acquired through traffic statistics and geographic position information collected by a base station controller, and is used to represent base station resource state and spatial distribution; the terrain data refers to terrain elevation, building distribution and ground cover type data of the target area, which can be acquired through a geographic information system or remote sensing surveying and mapping technology, and is used to analyze signal propagation loss and environmental interference factors.

[0092] Further, the acquired user terminal data, base station data and terrain data are input into a network prediction model for network prediction to obtain a prediction result corresponding to the target area.

[0093] The prediction process of the network prediction model specifically includes: firstly, interpolating the base station data to fill in the gaps of the base station data in the time sequence, ensuring the continuity of the data, thereby obtaining first base station data.

[0094] Specifically, the interpolation calculation refers to the continuity processing of the base station load data at discrete time points, preferably using a cubic spline interpolation algorithm to achieve, so as to eliminate the time sequence breakage problem caused by the difference in base station data sampling frequency.

[0095] Further, according to the first base station data, the user end data is time-aligned and space-fused to obtain first user end data, thereby solving the inconsistency of user end data and base station data in time and space dimensions.

[0096] Specifically, the time alignment refers to synchronizing the time stamp of the user end data with the interpolation time node of the base station data, preferably through linear interpolation or nearest neighbor matching method to eliminate the time dimension deviation of user behavior and base station state; the space fusion refers to the spatial correlation between the user end latitude and longitude and the base station coverage area, preferably through the Thiessen polygon division or the geographic fence matching algorithm to establish the spatial attribution relationship between the user and the base station.

[0097] Further, based on the user end moving time and the user end latitude and longitude, the user end is matched with the base station to obtain a matching result, so as to establish a dynamic association relationship between the user and the base station, and then according to the matching result, the first user end data, the first base station data and the terrain data are fused to obtain corresponding fusion data, realizing the integration of user behavior, base station state and geographical environment information through the fusion process.

[0098] Specifically, the dynamic matching refers to dynamically associating the service base station according to the user moving time and position, preferably through the spatio-temporal proximity calculation or trajectory prediction model to capture the base station switching behavior of the user at different time periods; the fusion data refers to the multi-dimensional data set integrating the user end moving characteristics, the base station load state and the terrain environment, preferably generated through tensor fusion or feature splicing method, thereby constructing a comprehensive prediction input containing user distribution, resource consumption and environmental attenuation.

[0099] Further, according to the fusion data, the network prediction result of each base station is calculated, finally realizing the generation of accurate network state prediction for each base station by using the comprehensive data set, and through the cooperative processing of multi-dimensional data and the dynamic matching mechanism, the adaptive prediction of complex network environment is realized, thereby improving the accuracy and real-time performance of network resource allocation, which helps to optimize the load balancing strategy between base stations, reduce the risk of network congestion and improve the consistency of user experience.

[0100] In some embodiments, the training process of the network prediction model in step S120 includes steps S210 to S280 as follows:

[0101] Step S210, preset an initial network prediction model;

[0102] Step S220, obtain training user terminal data, training base station data and training terrain data; the training user terminal data includes training user terminal moving time and training user terminal latitude and longitude;

[0103] Step S230, perform interpolation calculation on the training base station data to obtain first training base station data;

[0104] Step S240, according to the first training base station data, perform time alignment and spatial fusion on the training user terminal data to obtain first training user terminal data;

[0105] Step S250, based on the training user terminal moving time and the training user terminal latitude and longitude, match the training user terminal with the training base station to obtain a training matching result;

[0106] Step S260, according to the training matching result, fuse the first training user terminal data, the first training base station data and the training terrain data to obtain corresponding training fusion data;

[0107] Step S270, extract a training simplified feature matrix from the training fusion data;

[0108] Step S280, optimize the initial network prediction model according to the training fusion data and the training simplified feature matrix to obtain the network prediction model.

[0109] In this step, preferably, a hybrid architecture combining convolutional neural network and long short-term memory network is used to construct the initial network prediction model, and then the training user terminal data, the training base station data and the training terrain data are obtained, wherein the training user terminal data includes the time stamp and corresponding latitude and longitude coordinates of the user moving track, the training base station data includes the load data and position information of the base station, and the training terrain data includes the terrain elevation and building distribution information of the target area.

[0110] Further, a cubic spline interpolation method is used to perform interpolation calculation on the training base station data, and interpolation is performed on the base station load data in the time dimension to fill in the missing data points to obtain the first training base station data. Then, according to the first training base station data after interpolation, the training user terminal data is time-aligned and spatially fused to obtain the first training user terminal data, preferably, the time alignment and spatial fusion are performed by combining the Thiessen polygon to divide the base station coverage area, so as to associate the user terminal moving track with the base station spatial distribution.

[0111] Specifically, the user data is resampled to the same time interval as the base station data by time alignment, and the user data is associated to the corresponding base station area by spatial fusion to construct the Thiessen polygon of the base station coverage.

[0112] Further, based on the training user terminal moving time and latitude and longitude, the training user terminal is matched with the training base station to obtain a training matching result, wherein the matching process is preferably a nearest neighbor algorithm, and each user data point is associated to the nearest base station. Furthermore, according to the training matching result, the first training user terminal data, the first training base station data and the training terrain data are fused to obtain corresponding training fusion data, so as to align the user, base station and terrain data in time and space dimensions through the fusion process, and form a unified data structure.

[0113] Further, preferably, a dimension reduction method such as principal component analysis is used to extract a training simplified feature matrix from the training fusion data to extract key features and reduce data redundancy. Furthermore, according to the training fusion data and the training simplified feature matrix, an initial network prediction model is optimized by using a back propagation algorithm to obtain a network prediction model. Specifically, the model parameters can be adjusted through multiple rounds of iterative training to improve the prediction accuracy. Finally, the user terminal data, the base station data and the terrain data are effectively fused and processed, and the data dimension mismatch problem is solved through interpolation calculation and space-time alignment, providing a reliable foundation for network prediction in complex scenarios.

[0114] In some embodiments, the training base station data is subjected to interpolation calculation in step S230 to obtain the first training base station data, including the following step S310:

[0115] Step S310, the training base station data is processed by cubic spline interpolation to obtain the first training base station data;

[0116] In this step, in the cubic spline interpolation process, the cubic polynomial in each time interval needs to satisfy the continuity of the function value, the first derivative and the second derivative at the end points of adjacent intervals. The spline interpolation coefficients are solved by solving the system of equations, thereby obtaining the first training base station data. Finally, the continuous processing of the base station load data is realized, the load mutation noise is eliminated, the second derivative continuity of the base station load change is maintained, and the physical rationality of the interpolation data is improved.

[0117] Wherein, the cubic spline interpolation process forms a smooth transition curve between adjacent time points by constructing a cubic polynomial function, and the spline interpolation coefficients can be determined by solving the system of equations with boundary conditions and continuity equations, for example, under the natural spline condition, the second derivative at the end points is set to zero.

[0118] Specifically, the calculation formula of the interpolation calculation includes:

[0119] L interp(t) = a(t - t i ) 3 +b(t - t i ) 2 +c(t - t i )+d(t i ≤t<t i+1 );

[0120] wherein, L interp (t) is an interpolation base station load at time t, t i is a time point of the i-th base station load, t i+1 is a time point of the i+1-th base station load, t i and t i+1 are adjacent time points, and a, b, c and d are preset spline interpolation coefficients.

[0121] In some embodiments, in step S240, the training user end data is time-aligned and spatially fused according to the first training base station data to obtain the first training user end data, including steps S410 to S430 as follows:

[0122] Step S410, aligning the training user end data to a preset space-time scale to generate space-time aligned data.

[0123] Step S420, constructing a corresponding Thiessen polygon according to the first training base station data.

[0124] Step S430, spatially connecting the space-time aligned data and the Thiessen polygon based on the first training base station data to obtain the first training user end data.

[0125] In this step, the training user end data is aligned to a preset space-time scale to generate space-time aligned data. For example, the user end data can be uniformly aligned to a time scale of 5-minute intervals. Then, a corresponding Thiessen polygon is constructed according to the first training base station data.

[0126] Specifically, taking each base station position as a center point, a vertical bisector is drawn to form a polygon network, and the geographic space is divided into coverage areas centered on each base station.

[0127] Further, based on the first training base station data, the space-time alignment data and the Thiessen polygon are spatially connected to obtain first training user end data. The spatial connection specifically maps the user end position data to the corresponding base station polygon unit through a spatial connection operation to establish a dynamic association relationship between the user end moving track and the base station coverage range, thereby using the spatial exclusivity feature of the Thiessen polygon to ensure that each user end position only belongs to the base station area with the optimal signal coverage, and further improve the spatial association accuracy of the user and the base station in the training data, providing higher quality input data for subsequent network prediction model training.

[0128] In some embodiments, in step S120, the first user end data, the first base station data and the terrain data are fused according to the matching result to obtain corresponding fusion data, including the following steps S510 to S520:

[0129] Step S510, based on the terrain data, the first user end data and the first base station data are data processed to obtain second user end data and second base station data; the data processing at least includes interpolation processing, local outlier factor abnormality processing and wavelet threshold denoising processing;

[0130] Step S520, the second user end data, the second base station data and the terrain data are fused to obtain fusion data.

[0131] In this step, based on the terrain data, the first user end data and the first base station data are data processed to obtain second user end data and second base station data, wherein the data processing includes interpolation processing, local outlier factor abnormality processing and wavelet threshold denoising processing.

[0132] Specifically, the interpolation processing preferably adopts the Kriging interpolation method to construct a semi-variogram function model based on the terrain data, and to perform spatial interpolation on the missing points in the user end moving track and the base station load data, thereby improving the interpolation accuracy; the local outlier factor abnormality processing preferably adopts a density-based local outlier detection algorithm to calculate the local reachable density of each data point, and to identify the abnormal position points in the user end data and the outliers in the base station load data to identify and eliminate the abnormalities; the wavelet threshold denoising processing preferably adopts a wavelet multi-resolution analysis method, which specifically sets a threshold for coefficient contraction on different scales to realize noise suppression by performing wavelet decomposition on the user end latitude and longitude sequence and the base station load time series data.

[0133] Further, a weighted average method is used to assign weights according to data reliability, and the second user end data, the second base station data and the terrain data processed by the above processing are fused to obtain fusion data with higher quality and stability after multi-stage data processing, thereby providing reliable input for subsequent network prediction calculation and improving the accuracy and reliability of the network prediction result.

[0134] In some embodiments, the training reduced feature matrix is extracted from the training fusion data in step S270, including steps S610 to S660 as follows:

[0135] Step S610, calculating the capacity demand feature weight according to the training fusion data;

[0136] Step S620, extracting the training initial feature matrix from the training fusion data;

[0137] Step S630, mapping the training initial feature matrix into a query matrix, a key matrix and a value matrix;

[0138] Step S640, calculating the signal quality feature weight according to the query matrix, the key matrix and the value matrix;

[0139] Step S650, calculating the fusion weight according to the capacity demand feature weight and the signal quality feature weight;

[0140] Step S660, performing feature screening on the training initial feature matrix according to the fusion weight to obtain the training reduced feature matrix.

[0141] In this step, preferably, a feature importance evaluation method based on information entropy is adopted to calculate the capacity demand feature weight according to the training fusion data, so as to calculate the contribution degree of each feature to the network capacity demand and obtain the corresponding weight value, and then the training initial feature matrix is extracted from the training fusion data by dimension reduction technology such as principal component analysis, so as to extract key features from high-dimensional training fusion data and construct the initial feature matrix.

[0142] Further, preferably, the training initial feature matrix is mapped into the query matrix, the key matrix and the value matrix by linear transformation or nonlinear activation function, so as to obtain three matrices of different dimensions, and then the signal quality feature weight is calculated according to the query matrix, the key matrix and the value matrix.

[0143] Specifically, preferably, the correlation score between features is calculated by matrix multiplication and normalization exponential operation as the signal quality feature weight by using attention mechanism.

[0144] Further, the capacity demand feature weight and the signal quality feature weight are integrated to obtain a final fusion weight by using a weighted average or other fusion strategy, the training initial feature matrix is filtered according to the fusion weight to obtain a training simplified feature matrix, the correlation between features is enhanced, the model can better capture the dynamic changes of the network state, and specifically, a weight threshold can be set, features higher than the threshold are retained, and features lower than the threshold are removed, so that a simplified feature matrix is obtained, and effective filtering of redundant features in the training fusion data and accurate extraction of key features are realized.

[0145] In some embodiments, the initial network prediction model is optimized according to the training fusion data and the training simplified feature matrix in step S280 to obtain the network prediction model, including the following steps S710 to S740:

[0146] Step S710, the initial network prediction model is optimized according to the training simplified feature matrix to obtain a first network prediction model;

[0147] Step S720, the training fusion data is input into the first network prediction model to obtain a first network prediction result output by the first network prediction model;

[0148] Step S730, a loss function is calculated according to the actual network value of the training fusion data and the first network prediction result to obtain a corresponding loss error;

[0149] Step S740, the first network prediction model is optimized according to the loss error to obtain the network prediction model.

[0150] In this step, the gradient descent algorithm is preferably used to optimize the initial network prediction model according to the training simplified feature matrix to update the first network prediction model, so that the model can better fit the key features in the training simplified feature matrix, the model's ability to capture key features is improved, and then the training fusion data is input into the first network prediction model to obtain a first network prediction result output by the first network prediction model.

[0151] Further, the actual network value of the training fusion data and the first network prediction result are used to calculate a loss function to obtain a corresponding loss error. The loss function can be a commonly used function such as mean square error or cross entropy, and the prediction error of the model is quantified by comparing the difference between the predicted value and the actual value.

[0152] Further, preferably, the back propagation algorithm is adopted to optimize the first network prediction model according to the loss error, so as to obtain the network prediction model. Specifically, the loss error is back propagated along the network structure, and the model parameters are adjusted accordingly, so as to continuously reduce the prediction error and improve the model performance, thereby realizing double optimization of the network prediction model, ensuring the adaptability of the model to complex scenes, and improving the generalization ability and prediction accuracy of the model in complex dynamic scenes.

[0153] In some embodiments, after the user terminal data, the base station data and the terrain data are input into the network prediction model in step S120 to obtain the network prediction result output by the network prediction model, the following steps S810 to S820 are further included:

[0154] Step S810: calculating a network optimization strategy of the target region according to the network prediction result of each base station;

[0155] Step S820: performing network optimization on the target region according to the network optimization strategy, to obtain an optimized target region network.

[0156] In this step, after obtaining the network prediction result of each base station, the network optimization strategy of the target region is calculated according to the network prediction result of each base station. Preferably, the prediction results of each base station are subjected to cluster analysis, and base stations with similar loads are grouped into a group. Then, based on the clustering result, a high-load region is identified, and the available resources of adjacent low-load base stations are calculated, so as to formulate a load balancing strategy, such as adjusting the downtilt angle of the high-load base station, and transferring part of the traffic to the low-load base station. In addition, the transmission power and spectrum resource allocation of the base station can also be dynamically adjusted according to the predicted traffic distribution.

[0157] Further, the target region is subjected to network optimization according to the calculated network optimization strategy. Preferably, parameter adjustment instructions can be sent to the base station through a remote control interface to realize automatic adjustment of the antenna downtilt angle. For the region where temporary base stations need to be added, small and modular mobile base stations can be deployed to quickly improve the local network capacity. At the same time, the network topology and routing strategy are dynamically adjusted by using software-defined network technology, and the data transmission path is optimized, so as to obtain the optimized target region network by executing the above optimization measures, so as to improve the overall network performance, enhance the response ability of the network to burst traffic, and effectively improve the network resource utilization efficiency and user service quality, so that the network can better cope with complex and changeable communication environment.

[0158] In some embodiments, dynamic prediction and optimization are applied in a 5G network, multi-modal data in the 5G network is obtained, and the multi-modal data is spatio-temporally aligned and fused to integrate the multi-modal data (including user trajectories, base station load, GIS terrain data, etc.), solve the spatio-temporal alignment and noise interference problem, and generate a spatio-temporally aligned and denoised multi-modal data set D clean , to provide high-quality input for subsequent modeling.

[0159] Step one, preprocessing of multi-modal data:

[0160] First, the original data of the multi-modal data is obtained, and then the original data is standardized and preliminarily processed for noise to generate high-quality input data D processed , The processed data set is used as the input for spatio-temporal alignment.

[0161] The original data includes user trajectories, base station load, and GIS terrain data; the user trajectory T user (1Hz) is data recording the latitude and longitude coordinates and timestamps of the user's moving path; the base station load L base (1min) includes data of base station frequency, height, utilization rate, bandwidth occupancy rate, and number of connected users; the GIS terrain data G map (static) contains data of geographical information such as building outlines, elevation, vegetation distribution, etc.

[0162] Specifically, the standardization operation includes normalization processing and Z-score standardization processing, wherein the normalization processing is preferably linear mapping of the user trajectory latitude and longitude to the range [0, 1], and the formula is:

[0163]

[0164] In the formula, Lon norm is the normalized longitude, Lat norm is the normalized latitude, lon raw is the original longitude, lat raw is the original latitude, lon min is the minimum longitude of the region, lon max is the maximum longitude of the region, lat min is the minimum latitude of the region, and lat max is the maximum latitude of the region.

[0165] The Z-score standardization is preferably standardized for the base station load indicators, and the formula is:

[0166]

[0167] In the formula, the standardized i-th load indicator, original load indicator value, μ L the historical mean of the load indicator, σ L the standard deviation of the load indicator.

[0168] Further, the processed dataset is spatio-temporally aligned, preferably by aligning D processed to a unified spatio-temporal scale to generate a spatio-temporally aligned dataset D aligned , wherein D processed is aligned to a unified spatio-temporal scale by performing time alignment and spatial alignment on the processed dataset D .

[0169] Specifically, first, the window length is set to a fixed value Δt = 10 ms, i.e., a time window of 10 milliseconds, and the sliding step is defined to be consistent with the window length (non-overlapping windows), i.e., the window slides forward by 10 ms. Then, the alignment operation is performed to obtain the time-aligned tensor , wherein

[0170] , wherein, for the user trajectory T user (1 Hz), the last valid point in each window is retained (if there is no data in the window, the last point of the previous window is inherited), for the base station load L base (1 min), 10 ms granularity continuous data is generated by cubic spline interpolation, and for the GIS terrain data G map (static), the static data is directly copied to all time windows, and there is no need for time dimension processing.

[0171] Specifically, the interpolation formula is:

[0172] L interp (t) = a(t - t i ) 3 + b(t - t i ) 2 + c(t - t i ) + d(t i ≤ t < t i+1 .

[0173] In the formula, L interp (t) is the interpolated load at time t, t i is the time point of the i-th base station load, t i+1 is the time point of the i+1-th base station load, t i , t i+1 are adjacent time points, a, b, c, and d are spline interpolation coefficients, and the spline interpolation coefficients are specifically determined according to the load values L(t i and L(t i+1 ) of adjacent time points t i and t i+1 .i ) and L(t i+1 ), and the first-order derivative continuity condition (L′(t i )=L′(t i+1 ))Solve for a, b, c, d.

[0174] Furthermore, based on the base station location data, QGIS is used to generate coverage area polygons (one Thiessen polygon for each base station), and then the tensors are aligned according to time. Perform spatial connection with Thiessen polygons to associate each user trajectory point with the base station ID. If the trajectory point happens to be located at the boundary of multiple Thiessen polygons, the optimal base station can be selected by randomly selecting an adjacent base station and combining the base station signal strength.

[0175] Furthermore, it is preferred that the wireless communication signal transmission model corrects terrain occlusion error, wherein the terrain occlusion error is mainly caused by obstacles such as buildings, vegetation, and terrain undulations, resulting in non-line-of-sight (NLoS) attenuation on the signal propagation path.

[0176] Specifically, the process of correcting terrain obstruction errors in the wireless communication signal transmission model is to first identify the type of obstructions (buildings, vegetation, terrain) between the user and the base station based on GIS terrain data (static), then look up the table to obtain the baseline attenuation value based on the physical properties of the obstructions (such as wall material and tree density), and then dynamically superimpose the attenuation values ​​of multiple obstructions to obtain the final terrain attenuation coefficient.

[0177] Furthermore, for time-aligned tensors Fix missing values, remove outliers and denoise to generate a complete dataset D clean ∈R T×S×M Among them, the preferred method for repairing missing values ​​is interpolation filling and associating with neighboring base stations, the preferred method for removing outliers is local outlier factor anomaly processing, and the preferred method for denoising is wavelet threshold denoising,

[0178] Specifically, interpolation filling is to perform linear interpolation on unmatched trajectory points based on the spatiotemporal correlation of neighboring base stations; neighboring base station association is to associate to the neighboring base station with the strongest signal if interpolation fails; local outlier factor screening is to detect outliers on the data after spatiotemporal alignment and eliminate samples that exceed the preset threshold.

[0179] Furthermore, integrating D clean ∈R T×S×M , generate a unified feature matrix D fused Specifically, by clean The user trajectory, base station load, terrain attenuation and other features in the time window are spliced ​​into a multidimensional tensor D merged ∈R T×S×M , then the multidimensional tensor D merged ∈RT×S×M Convert to deep learning input format and encode base station ID, add dimension K (base station number) to get the final fused dataset D fused ∈R T×S×(M+K) .

[0180] Further, the final fused dataset D fused ∈R T×S×(M+K) Aggregated into 15-minute interval coarse-grained windows (covering 24 hours → 96 time steps), and then for each 15-minute window (900 seconds), calculate the statistical values of the trajectory features (such as average latitude and longitude, moving speed standard deviation), and then calculate the mean and peak values of the 15-minute window for the interpolated 10ms load data (such as utilization, bandwidth occupancy), and for the GIS terrain data (static), directly copy to all time windows without aggregation.

[0181] Further, the features of each 15-minute window (user trajectory statistics, base station load statistics, terrain attenuation coefficient) are spliced into feature vectors X t ∈R 94 , and then arrange 96 consecutive windows in chronological order to generate multi-modal time series data X seq ∈R 96×94 .

[0182] Step two, dynamic feature selection and modeling:

[0183] First, a double-channel attention mechanism (DCA) is designed to evaluate feature importance from the perspectives of signal quality prediction and capacity demand prediction, respectively. Specifically, the original feature matrix F fused ∈R raw is extracted from D N×D , N is the number of samples, D=128 is the initial dimension, and includes base station transmit power, user distance, historical traffic, and other multi-modal features.

[0184] Further, initialize the learnable parameters W Q , W k , W V ∈R 128×64 , and then generate query, key, and value matrices through linear transformation: Q=F raw W Q , K=F raw W k , V=F raw W V .

[0185] Specifically, in channel 1 (signal quality channel), input base station transmit power, user distance, obstacle density, etc., calculate feature weight W1∈R D×1, to obtain the feature weight vector W1 e R 128 , the formula is:

[0186] Q=F raw W Q , K=F raw W k , v=F raw W V ;

[0187] Wherein: W1 is the weight vector of the signal quality channel, for, Q=F raw W Q , K=F raw W k , V=F raw W V , respectively, query, key, value matrix, W Q , W k , W v e R 128×64 is the learning matrix, and d=64 is the scaling factor.

[0188] In channel 2 (capacity demand channel): input historical traffic peak, user application type (video / VR proportion), time period (such as weekdays / holidays) and other data in base station load (1 min) data, adopt similar mechanism to generate weight W2 e R D ×1 , to obtain the feature weight vector W2 e R128 identifying the importance of the capacity demand related features, the formula is:

[0189] Q=F raw W Q , K=F raw W k , v=F raw W V ;

[0190] Wherein: W2 is the weight vector of the capacity demand channel, Q=F raw W Q , K=F raw W k , V=F raw W v , respectively, query, key, value matrix, W Q , W k , W v e R 128×64 is the learning matrix, and d=64 is the scaling factor.

[0191] Further, adjust the weight proportion a according to the optimization target:

[0192] W fused =αW1+(1-α)W2(α∈[0,1]);

[0193] Where W fused It is the dynamic fusion result of the weight vectors of the signal quality channel and the capacity demand channel. W1 is the feature weight vector generated by channel 1 (signal quality channel), W2 is the feature weight vector generated by channel 2 (capacity demand channel), and α is the dynamic weight coefficient.

[0194] Furthermore, for W fused Sort in descending order and filter the Top-K features (K=0.5D):

[0195] F selected =F raw ⊙TOPK(W fused ,K);

[0196] Where, F raw is the original feature matrix, dimension F raw ∈RN×D (N is the number of samples, D is the feature dimension), W fused is the fused feature weight vector with dimension D×1, K is the number of key features filtered from the original feature matrix, and ⊙ represents the weighted filtering operation.

[0197] Specifically, the top K feature indices with the highest weights are selected through the TOPK function, and then the original feature matrix F is obtained. raw Extract these K features and get the simplified feature matrix F selected ∈R N×K .

[0198] Step 3: LSTM-GCN spatiotemporal hybrid prediction model:

[0199] The spatiotemporal hybrid prediction model is constructed by LSTM and GCN, and then the spatiotemporal hybrid prediction model is used to predict the multimodal time series data X of the historical 24 hours. seq ∈R 96×94 Output future traffic hotspots and base station load predictions

[0200] Furthermore, based on the future traffic hotspots and base station load prediction values and historical 24-hour multimodal time series data X seq ∈R 96×94 The loss function is calculated based on the actual observation value of .

[0201] Specifically, calculate the mean square error (MSE):

[0202]

[0203] Where: LMSE represents the average squared error between model prediction and actual observation, N represents the number of samples, Y pred represents the predicted value of the traffic and load of each base station in the future period, Y true represents the real traffic and real load of each base station in the future period, which is used to compare with the model prediction value Y pred Comparison, evaluate the prediction accuracy and optimize the model.

[0204] Further, the topology consistency loss is calculated:

[0205]

[0206] In the formula, L topo represents the sum of squares of the prediction difference of adjacent base stations.

[0207] Further, the total loss is calculated:

[0208] L total = 0.7L MSE + 0.3L topo

[0209] In the formula, L total represents the total loss that the model needs to minimize in the training process, L MSE represents the average squared error between model prediction and actual observation, L topo represents the sum of squares of the prediction difference of adjacent base stations.

[0210] Step four, real-time optimization and closed-loop control:

[0211] Based on the prediction results, the fuzzy control algorithm is used to adjust the antenna tilt angle and power distribution, and at the same time trigger the load balancing strategy, to realize the millisecond-level dynamic resource allocation and closed-loop feedback, continuously optimize the network performance, output dynamic optimization parameters and feedback error signal E.

[0212] Specifically, first according to the predicted value Y pred , adjust the antenna tilt angle θ and the transmission power P:

[0213]

[0214] In the formula, θ new represents the antenna downtilt angle after dynamic adjustment, θ old represents the current antenna downtilt angle before adjustment, η represents the learning efficiency (default 0.01), represents the gradient of the loss function L MSE relative to the antenna tilt angle θ, P new represents the adjusted base station antenna transmission power, P old represents the current base station transmission power before adjustment, L represents the loss function MSE N represents the number of samples, the number of base stations participating in the calculation, represents the predicted value of the i-th base station, the future traffic or load prediction value output by the model, represents the actual observation value of the i-th base station, the real-time measured traffic or load value, represents the partial derivative of the predicted value with respect to the antenna downtilt angle, indicating the degree of influence of the change of the antenna downtilt angle on the predicted value, the partial derivative of the predicted value with respect to the transmit power, indicating the degree of influence of the change of the transmit power on the predicted value.

[0215] Further, the actual base station traffic and load T real are collected, and the error signal is calculated:

[0216]

[0217] In the formula, E represents the mean square error of the predicted value and the actual observation value, N represents the number of base stations participating in the error calculation or the number of data points, represents the actual traffic and load value of the i-th base station in the future period, represents the prediction result of the model for the future traffic and load value of the i-th base station.

[0218] Further, back propagation is performed to trigger data cleaning strategy update and model weight fine-tuning:

[0219]

[0220] W model represents the set of trainable parameters in the model that need to be optimized, and η represents the learning efficiency (default 0.01), represents the gradient of the error signal E with respect to the model parameter W model .

[0221] Finally, the optimized parameters {θ new , P new} and the error signal E are obtained for optimizing the regional network.

[0222] In some embodiments, indoor and outdoor mixed scene coverage optimization is achieved through network prediction. First, multi-modal data space-time alignment and fusion are performed:

[0223] Step 1, data input and preprocessing:

[0224] 1. Data source and collection:

[0225] Scenario Description: A large-scale commercial complex, including an indoor shopping center (50,000 ㎡, 3-story structure) and an outdoor plaza (20,000 ㎡), deploying a distributed antenna system (DAS) and surrounding 5G macro base stations.

[0226] 2. Data Modalities:

[0227] Indoor Data: User positions, real-time signal strengths, and user application types collected by the DAS;

[0228] Outdoor Data: Macro base station MR data and user trajectory data;

[0229] Environmental Data: Building BIM model and GIS terrain data.

[0230] Step Two: Spatio-Temporal Alignment and Denoising:

[0231] 1. Spatio-Temporal Hash Coding Alignment:

[0232] Align indoor UWB data and outdoor MR data to a 50ms time window to solve the indoor and outdoor clock synchronization problem.

[0233] Use the COST231-Hata modified model to compensate for indoor wall shielding loss.

[0234] 2. Noise Processing:

[0235] Apply wavelet decomposition to remove high-frequency noise from DAS signal strength data, and combine LOF anomaly detection to remove abnormal positioning points.

[0236] Repair signal loss caused by vegetation shielding of outdoor macro base stations.

[0237] Further, dynamic feature selection and modeling:

[0238] 1. Feature Selection and Optimization:

[0239] Input Features: Original features with 96 dimensions, including "indoor user density distribution", "wall penetration loss index", "elevator shaft signal attenuation", "outdoor macro base station load rate", etc.

[0240] Dual-channel attention mechanism (DCA) application:

[0241] Channel 1 (Signal Quality Channel): Focus on "wall material impact", "DAS antenna transmit power", and "user distance";

[0242] Channel 2 (Capacity Demand Channel): Focus on "peak period pedestrian density", "AR / VR application proportion", and "outdoor macro base station switching frequency";

[0243] Dynamic fusion: Through weight adjustment, key features are selected 32 dimensions, including "dynamic wall attenuation factor" and "indoor-outdoor user migration rate".

[0244] Output: Reduced feature matrix, dimensionality reduced by 66%.

[0245] Further, the hybrid model of LSTM-GCN is predicted:

[0246] 1. LSTM time series modeling:

[0247] Input historical 48-hour data, predict future 1-hour indoor-outdoor user distribution trend.

[0248] 2. GCN spatial modeling:

[0249] Topology graph construction: DAS antenna nodes, macro base stations, and elevator shafts as graph nodes, wall penetration loss and user migration path as edge attributes.

[0250] 3. Joint training and optimization:

[0251] Parameter settings: learning rate 0.0005, batch size 128, training time 6 hours.

[0252] Further, real-time optimization and closed-loop control:

[0253] 1. Edge computing decision:

[0254] Predict the indoor AR navigation hotspot area in the next 30 minutes, dynamically increase the DAS antenna power (+3dBm), and trigger the seamless handover strategy from macro base station to DAS (handover delay <50ms).

[0255] 2. Closed-loop feedback mechanism:

[0256] Collect real-time KPIs (such as indoor RSRP coverage rate >-90dBm, handover success rate), calculate error signal.

[0257] Reverse update feature selection weight, and trigger model fine-tuning (learning rate decay to 0.0001).

[0258] Further, verify the implementation effect:

[0259] 1. Performance index comparison:

[0260] Indicators Conventional scheme (not optimized) This embodiment Lifting range Signal blind area 3000㎡ 900㎡ 70％↓ User experience score (MOS) 3.2 4.8 50％↑ Total antenna power consumption 850W 697W 18％↓ Indoor and outdoor switching success rate 88％ 98％ 10％↑ Prediction error rate (traffic) 25％ 8％ 68％↓

[0261] 2. Scene adaptability verification:

[0262] High-density period (holiday): Through dynamic adjustment of DAS power, indoor throughput peak value is increased from 3.5Gbps to 5.1Gbps, and user complaint rate is decreased by 75%.

[0263] Complex structure scene (elevator shaft): the GCN model accurately models the signal attenuation path of the elevator shaft, and the blind area elimination effect is improved by 40% (compared with the baseline model without modeling the GCN).

[0264] As Figure 2 shown, some embodiments of the present application provide a network prediction system, which includes an acquisition module 210 and a prediction module 220, specifically:

[0265] The acquisition module 210 is configured to acquire user terminal data, base station data, and terrain data in a target area; the user terminal data includes user terminal moving time and user terminal latitude and longitude; the base station data includes base station load and base station latitude and longitude.

[0266] The prediction module 220 is configured to input the user terminal data, the base station data, and the terrain data into a network prediction model to obtain a network prediction result output by the network prediction model.

[0267] The prediction process of the network prediction model includes:

[0268] The base station data is subjected to interpolation calculation to obtain first base station data.

[0269] The user terminal data is subjected to time alignment and spatial fusion according to the first base station data to obtain first user terminal data.

[0270] The user terminal and the base station are matched based on the user terminal moving time and the user terminal latitude and longitude to obtain a matching result.

[0271] The first user terminal data, the first base station data, and the terrain data are fused according to the matching result to obtain corresponding fusion data.

[0272] The network prediction result of each base station is calculated according to the fusion data.

[0273] In some embodiments, the prediction module 220 can include a preset initial network prediction model.

[0274] In some embodiments, the prediction module 220 can include acquiring training user terminal data, training base station data, and training terrain data; the training user terminal data includes training user terminal moving time and training user terminal latitude and longitude.

[0275] In some embodiments, the prediction module 220 can include performing interpolation calculation on the training base station data to obtain first training base station data.

[0276] In some embodiments, the prediction module 220 can include performing time alignment and spatial fusion on the training user terminal data according to the first training base station data to obtain first training user terminal data.

[0277] In some embodiments, the prediction module 220 can comprise matching the training user terminals with the training base stations based on the training user terminal moving time and the training user terminal latitude and longitude, to obtain a training matching result.

[0278] In some embodiments, the prediction module 220 can comprise fusing the first training user terminal data, the first training base station data and the training terrain data according to the training matching result, to obtain corresponding training fusion data.

[0279] In some embodiments, the prediction module 220 can comprise extracting a training simplified feature matrix from the training fusion data.

[0280] In some embodiments, the prediction module 220 can comprise optimizing an initial network prediction model according to the training fusion data and the training simplified feature matrix, to obtain the network prediction model.

[0281] In some embodiments, the prediction module 220 can comprise processing the training base station data through cubic spline interpolation, to obtain the first training base station data.

[0282] In some embodiments, the prediction module 220 can comprise:

[0283] L interp (t) = a(t - t i ) 3 +b(t - t i ) 2 +c(t - t i )+d(t i ≤t<t i+1 );

[0284] wherein, L interp (t) is an interpolated base station load at time t, t i is a time point of the i-th base station load, t i+1 is a time point of the i+1-th base station load, t i and t i+1 are adjacent time points, and a, b, c and d are preset spline interpolation coefficients.

[0285] In some embodiments, the prediction module 220 can comprise aligning the training user terminal data to a preset space-time scale, to generate space-time aligned data.

[0286] In some embodiments, the prediction module 220 can comprise constructing a corresponding Thiessen polygon according to the first training base station data.

[0287] In some embodiments, the prediction module 220 can include: based on the first training base station data, spatially connecting the spatio-temporal alignment data and the Thiessen polygon to obtain first training user end data.

[0288] In some embodiments, the prediction module 220 can include: based on the terrain data, data processing the first user end data and the first base station data to obtain second user end data and second base station data; the data processing at least includes interpolation processing, local outlier factor anomaly processing and wavelet threshold denoising processing.

[0289] In some embodiments, the prediction module 220 can include: fusing the second user end data, the second base station data and the terrain data to obtain fusion data.

[0290] In some embodiments, the prediction module 220 can include: calculating a capacity demand feature weight according to the training fusion data.

[0291] In some embodiments, the prediction module 220 can include: extracting a training initial feature matrix from the training fusion data.

[0292] In some embodiments, the prediction module 220 can include: mapping the training initial feature matrix into a query matrix, a key matrix and a value matrix.

[0293] In some embodiments, the prediction module 220 can include: calculating a signal quality feature weight according to the query matrix, the key matrix and the value matrix.

[0294] In some embodiments, the prediction module 220 can include: calculating a fusion weight according to the capacity demand feature weight and the signal quality feature weight.

[0295] In some embodiments, the prediction module 220 can include: performing feature screening on the training initial feature matrix according to the fusion weight to obtain a training simplified feature matrix.

[0296] In some embodiments, the prediction module 220 can include: optimizing an initial network prediction model according to the training simplified feature matrix to obtain a first network prediction model.

[0297] In some embodiments, the prediction module 220 can include: inputting the training fusion data into the first network prediction model to obtain a first network prediction result output by the first network prediction model.

[0298] In some embodiments, the prediction module 220 can include: calculating a loss function according to an actual network value of the training fusion data and the first network prediction result to obtain a corresponding loss error.

[0299] In some embodiments, the prediction module 220 can comprise optimizing the first network prediction model according to the loss error to obtain the network prediction model.

[0300] In some embodiments, the prediction module 220 can comprise calculating a network optimization strategy of the target area according to the network prediction result of each base station.

[0301] In some embodiments, the prediction module 220 can comprise optimizing the network of the target area according to the network optimization strategy to obtain the optimized network of the target area.

[0302] It should be noted that the network prediction system provided by the present embodiment is based on the same inventive concept as the network prediction method described above, and therefore the related content of the network prediction method described above also applies to the network prediction system. Therefore, the details are not repeated here.

[0303] In order to, the system obtains user terminal data, base station data and terrain data in the target area; the user terminal data includes user terminal moving time and user terminal latitude and longitude; the base station data includes base station load and base station latitude and longitude; the user terminal data, the base station data and the terrain data are input into the network prediction model to obtain the network prediction result output by the network prediction model. In this way, the network change can be accurately predicted according to the network historical data, the real-time dynamic resource allocation can be realized according to the prediction result, and the network prediction accuracy, real-time performance and resource utilization rate can be significantly improved.

[0304] The present application also provides an electronic device, which comprises a memory and a processor, the memory stores a computer program, and the processor implements the network prediction method described above when executing the computer program.

[0305] As Figure 3 , Figure 3 The hardware structure schematic diagram of the electronic device provided by the present application is shown in the following figure:

[0306] At least one battery;

[0307] At least one memory;

[0308] At least one processor;

[0309] At least one program;

[0310] The program is stored in the memory, and the processor executes the at least one program to implement the network prediction method described above.

[0311] The electronic device can be any intelligent terminal including mobile phone, tablet computer, personal digital assistant (PDA), vehicle-mounted computer, etc.

[0312] The electronic device of the embodiment of the present application is described in detail below.

[0313] The processor 1600 can be implemented in a general-purpose central processing unit (CPU), a microprocessor, an application specific integrated circuit (ASIC), or one or more integrated circuits, and is configured to execute related programs to implement the technical solutions provided by the embodiments of the present application.

[0314] The memory 1700 can be implemented in a read only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 1700 can store an operating system and other application programs. When the technical solutions provided by the embodiments of the present application are implemented by software or firmware, the related program codes are stored in the memory 1700 and are called and executed by the processor 1600 to implement a network prediction method according to the embodiments of the present application.

[0315] The input / output interface 1800 is configured to implement information input and output.

[0316] The communication interface 1900 is configured to implement the communication interaction between the device and other devices. The communication can be realized by a wired manner (for example, a USB, a network cable, etc.) or a wireless manner (for example, a mobile network, WIFI, Bluetooth, etc.).

[0317] The bus 2000 is configured to transmit information between the components (for example, the processor 1600, the memory 1700, the input / output interface 1800, and the communication interface 1900) of the device.

[0318] The processor 1600, the memory 1700, the input / output interface 1800, and the communication interface 1900 are connected to each other by the bus 2000 to realize the communication connection between them in the device.

[0319] The embodiments of the present application further provide a storage medium, which is a computer readable storage medium and stores computer executable instructions for causing a computer to execute the network prediction method.

[0320] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. In addition, the memory can include a high-speed random access memory and can also include a non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state memory device. In some embodiments, the memory can optionally include a memory that is remotely disposed relative to the processor, and these remote memories can be connected to the processor through a network. Examples of the above network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0321] The embodiments described in the embodiments of the present disclosure are used to more clearly illustrate the technical solutions of the embodiments of the present disclosure, and do not constitute a limitation on the technical solutions provided by the embodiments of the present disclosure. Those skilled in the art can know that, as technology evolves and new application scenarios appear, the technical solutions provided by the embodiments of the present disclosure are also applicable to similar technical problems.

[0322] Those skilled in the art can understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present disclosure, and can include more or fewer steps than those shown in the figures, or combine certain steps or different steps.

[0323] The device embodiments described above are only schematic, and units described as separate components can or can not be physically separate, i.e., can be located in one place, or can be distributed on multiple network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the embodiments.

[0324] Those skilled in the art can understand that all or some of the steps in the above disclosed method, the functional modules / units in the system and the device can be implemented as software, firmware, hardware and their appropriate combinations.

[0325] The terms "first", "second", "third", "fourth" and the like used in the specification of the present application and the above-described drawings, if any, are used to distinguish similar objects, and do not necessarily have to describe a particular order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0326] It should be understood that, in the application, "at least one" refers to one or more, and "multiple" refers to two or more. "And / or" is used to describe the association relationship of the associated objects, which means that there can be three relationships, for example, "A and / or B" can represent three cases of only A, only B and A and B existing at the same time, wherein A and B can be singular or plural. The character " / " generally represents an "or" relationship between the associated objects before and after it. "At least one of the following" or similar expressions means any combination of these items, including any combination of single or multiple items. For example, at least one of a, b or c can represent a, b, c, "a and b", "a and c", "b and c", or "a and b and c", wherein a, b and c can be single or multiple.

[0327] In several embodiments provided in the application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only illustrative, for example, the division of units is only a logical function division, and actual implementation can have another division manner, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the displayed or discussed units can be indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0328] The units described as separate components can or can not be physically separated, and the components displayed as units can or can not be physical units, that is, they can be located in one place, or they can be distributed on multiple network units. According to actual needs, part or all of the units can be selected to achieve the purpose of the embodiment scheme.

[0329] In addition, the functional units in each embodiment of the application can be integrated into a processing unit, or each unit can exist physically, or two or more units can be integrated into one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0330] If the integrated unit is implemented in the form of a software function unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such understanding, the technical solutions of the present application essentially or say the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes multiple instructions for causing an electronic device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (Read-Only Memory, ROM), a random access memory (Random Access Memory, RAM), a magnetic disk or an optical disk, and various program storage media.

[0331] The above is a specific description of the preferred implementation of the embodiments of the present application, but the embodiments of the present application are not limited to the above-mentioned implementation. Those skilled in the art can make various equivalent modifications or replacements without departing from the spirit of the embodiments of the present application, and these equivalent modifications or replacements are all included in the scope defined by the claims of the embodiments of the present application.

[0332] The embodiments of the present application have been described in detail above in combination with the drawings, but the present application is not limited to the above-mentioned embodiments. Those skilled in the art can make various changes within the scope of knowledge possessed by those skilled in the art without departing from the purpose of the present application.

Claims

1. A network prediction method, characterized in that: The method comprises: Acquire user terminal data, base station data, and terrain data within the target area; the user terminal data includes the user terminal movement time and the user terminal longitude and latitude; the base station data includes the base station load and the base station longitude and latitude; Inputting the user terminal data, the base station data, and the terrain data into a network prediction model to obtain a network prediction result output by the network prediction model; The prediction process of the network prediction model includes: Performing interpolation calculation on the base station data to obtain first base station data; performing time alignment and spatial fusion on the user terminal data according to the first base station data to obtain first user terminal data; Matching the user terminal with the base station based on the user terminal movement time and the user terminal latitude and longitude to obtain a matching result; fusing the first user terminal data, the first base station data, and the terrain data according to the matching result to obtain corresponding fused data; A network prediction result of each base station is calculated based on the fused data.

2. The network prediction method according to claim 1, characterized in that The training process of the network prediction model includes: Preset the initial network prediction model; Acquire training user terminal data, training base station data, and training terrain data; the training user terminal data includes the training user terminal movement time and the training user terminal latitude and longitude; Performing interpolation calculation on the training base station data to obtain first training base station data; According to the first training base station data, the training user terminal data is time-aligned and spatially fused to obtain first training user terminal data; Matching the training user terminal with the training base station based on the movement time of the training user terminal and the longitude and latitude of the training user terminal to obtain a training matching result; According to the training matching result, fusing the first training user terminal data, the first training base station data, and the training terrain data to obtain corresponding training fusion data; extracting a training reduced feature matrix from the training fusion data; The initial network prediction model is optimized according to the training fusion data and the training simplified feature matrix to obtain the network prediction model.

3. The network prediction method according to claim 2, characterized in that: The interpolation calculation of the training base station data to obtain first training base station data includes: Processing the training base station data by cubic spline interpolation to obtain the first training base station data; The calculation formula of the interpolation calculation includes: L interp (t)=a(t-t i ) 3 +b(t-t i ) 2 +c(t-t i )+d(t i ≤t<t i+1 ); Among them, L interp (t) is the interpolated base station load at time t, t i is the time point of the i-th base station load, t i+1 is the time point of the i+1th base station load, t i and t i+1 are adjacent time points, a, b, c, and d are preset spline interpolation coefficients; The step of performing time alignment and spatial fusion on the training user terminal data according to the first training base station data to obtain first training user terminal data includes: Aligning the training user-end data to a preset spatiotemporal scale to generate spatiotemporal aligned data; Constructing corresponding Thiessen polygons according to the first training base station data; Based on the first training base station data, the spatiotemporal alignment data and the Thiessen polygons are spatially connected to obtain the first training user terminal data.

4. The network prediction method according to claim 1, wherein: The fusing the first user terminal data, the first base station data, and the terrain data according to the matching result to obtain corresponding fused data includes: Based on the terrain data, the first user terminal data and the first base station data are processed to obtain second user terminal data and second base station data; the data processing includes at least interpolation processing, local outlier factor anomaly processing, and wavelet threshold denoising processing; The second user terminal data, the second base station data and the terrain data are fused to obtain the fused data.

5. The network prediction method according to claim 2, characterized in that: The step of extracting a training reduced feature matrix from the training fusion data includes: Calculating capacity demand feature weights based on the training fusion data; Extracting a training initial feature matrix from the training fusion data; Mapping the training initial feature matrix into a query matrix, a key matrix, and a value matrix; Calculating signal quality feature weights based on the query matrix, the key matrix, and the value matrix; Calculating a fusion weight according to the capacity demand feature weight and the signal quality feature weight; According to the fusion weights, the training initial feature matrix is ​​subjected to feature screening to obtain the training simplified feature matrix.

6. The network prediction method according to claim 2, characterized in that: The optimizing the initial network prediction model according to the training fusion data and the training simplified feature matrix to obtain the network prediction model includes: Optimizing the initial network prediction model according to the training streamlined feature matrix to obtain a first network prediction model; Inputting the training fusion data into the first network prediction model to obtain a first network prediction result output by the first network prediction model; Calculating a loss function based on the actual network value of the training fusion data and the prediction result of the first network to obtain a corresponding loss error; The first network prediction model is optimized according to the loss error to obtain the network prediction model.

7. The network prediction method according to claim 1, characterized in that: After inputting the user terminal data, the base station data, and the terrain data into the network prediction model and obtaining the network prediction result output by the network prediction model, the method further includes: Calculating a network optimization strategy for the target area according to the network prediction result of each base station; The target area is network optimized according to the network optimization strategy to obtain an optimized target area network.

8. A network prediction system, characterized in that: The system comprises: An acquisition module is used to acquire user terminal data, base station data and terrain data within the target area; the user terminal data includes the user terminal movement time and the user terminal latitude and longitude; the base station data includes the base station load and the base station latitude and longitude; A prediction module, configured to input the user terminal data, the base station data, and the terrain data into a network prediction model to obtain a network prediction result output by the network prediction model; The prediction process of the network prediction model includes: Performing interpolation calculation on the base station data to obtain first base station data; performing time alignment and spatial fusion on the user terminal data according to the first base station data to obtain first user terminal data; Matching the user terminal with the base station based on the user terminal movement time and the user terminal latitude and longitude to obtain a matching result; fusing the first user terminal data, the first base station data, and the terrain data according to the matching result to obtain corresponding fused data; A network prediction result of each base station is calculated based on the fused data.

9. An electronic device, characterized in that: It includes at least one control processor and a memory for communicating with the at least one control processor; the memory stores instructions that can be executed by the at least one control processor, and the instructions are executed by the at least one control processor to enable the at least one control processor to execute a network prediction method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute the network prediction method according to any one of claims 1 to 7.

Images