Mobile network processing method, device and server based on traffic data

Abstract

The application provides a mobile network processing method and device based on traffic data and a server, relates to communication technology, and the method comprises the following steps: obtaining a data prediction instruction, wherein the data prediction instruction comprises a data identifier, and obtaining a traffic data sequence corresponding to the data identifier; performing traffic prediction on the traffic data sequence according to a preset traffic prediction model, and obtaining target traffic data in a target time slice; determining a network slice allocation strategy corresponding to the target traffic data according to the target traffic data in the target time slice, and outputting a mobile network according to the network slice allocation strategy. The method greatly relieves the load pressure of the current server, guarantees the stability of the mobile network, and solves the technical problem that the mobile network is unstable due to the low accuracy of the traffic prediction result.

Description

Technical Field

[0001] This application relates to communication technology, and more particularly to a mobile network processing method, apparatus, and server based on traffic data. Background Technology

[0002] Currently, in order to ensure the stability of 5G networks, multiple dedicated, virtualized, and isolated logical networks are typically constructed based on network slicing technology to meet the differentiated network capability requirements of different scenarios.

[0003] In existing technologies, when constructing an isolated network, the traffic data sequence of the base station is usually obtained, and the traffic is predicted by the Auto Regressive Moving Average (ARMA) model to complete the prediction of the future traffic of the base station and construct the isolated network.

[0004] However, in existing technologies, if the sampled traffic data sequence is highly volatile, the ARMA model will still be non-stationary after second-order differencing, resulting in errors in the stationarity processing. This leads to inaccurate traffic predictions by the ARMA model, making it impossible to provide users with a stable isolation network. Summary of the Invention

[0005] This application provides a mobile network processing method, apparatus, and server based on traffic data to solve the technical problem of mobile network instability caused by low accuracy of traffic prediction results.

[0006] In a first aspect, this application provides a mobile network processing method based on traffic data, including:

[0007] Obtain a data prediction instruction, wherein the data prediction instruction includes a data identifier, and obtain a traffic data sequence corresponding to the data identifier;

[0008] Traffic prediction is performed on the traffic data sequence according to a preset traffic prediction model to obtain target traffic data within the target time slice; wherein, the preset traffic prediction model is trained based on multiple historical time slices and historical traffic sequences within each historical time slice, and the target time slice includes at least one target moment.

[0009] According to the target traffic data in the target time slice, a network slice allocation strategy corresponding to the target traffic data is determined, and a mobile network is output according to the network slice allocation strategy; wherein the network slice allocation strategy is used to indicate the rule information of allocating virtual network elements and outputting a mobile network according to the allocated virtual network elements.

[0010] Further, the target traffic data in the target time slice is obtained by performing traffic prediction on the traffic data sequence according to a preset traffic prediction model, comprising:

[0011] Determining the target time slice for predicting traffic in the preset traffic prediction model;

[0012] According to the preset traffic prediction model, the target traffic data in the target time slice is obtained by performing traffic prediction on the traffic data sequence.

[0013] Further, the network slice allocation strategy corresponding to the target traffic data is determined according to the target traffic data in the target time slice, comprising:

[0014] If it is determined that the target traffic data in the target time slice is greater than or equal to a threshold value, the maximum traffic value that the current server can load in the target time slice and the remaining traffic value in the target traffic data except for the maximum traffic value are determined;

[0015] According to the maximum traffic value and the remaining traffic value, a network slice allocation strategy corresponding to the target traffic data is determined.

[0016] Further, the method further comprises:

[0017] According to a preset feature extraction function, the traffic data sequence is preprocessed to obtain a preprocessed traffic data sequence;

[0018] According to the preset traffic prediction model, the target traffic data in the target time slice is obtained by performing traffic prediction on the traffic data sequence, comprising:

[0019] According to the preset traffic prediction model, the target traffic data in the target time slice is obtained by performing traffic prediction on the preprocessed traffic data sequence.

[0020] Further, the traffic data sequence is preprocessed according to a preset feature extraction function to obtain a preprocessed traffic data sequence, comprising:

[0021] According to the preset feature extraction function, the traffic data sequence is smoothed;

[0022] determining a number of time slices of the flow data sequence; wherein each time slice includes a plurality of flow data;

[0023] determining a maximum value and / or a minimum value of the flow data in each time slice, and determining the maximum value and / or the minimum value as an extreme point in the corresponding time slice;

[0024] determining the extreme point and flow data respectively located at two ends of the flow data sequence as feature points;

[0025] determining a slope of a line between two adjacent feature points based on the two adjacent feature points;

[0026] if it is determined that the slope is greater than a threshold value, performing a correction process on the flow data between the two adjacent feature points, and obtaining a pre-processed flow data sequence according to the corrected flow data.

[0027] Further, the method further comprises:

[0028] obtaining a plurality of historical time slices and a historical flow sequence in each historical time slice;

[0029] for two adjacent historical time slices in the plurality of historical time slices, performing encoding training on a rolling long short-term memory network according to a flow data sequence of a previous historical time slice in the two adjacent historical time slices, to obtain encoding feature information; and performing flow prediction training on an unfolded long short-term memory network according to the encoding feature information, to obtain predicted flow data of a subsequent historical time slice in the two adjacent historical time slices;

[0030] generating the predicted flow data in the last historical time slice in the plurality of historical time slices, and generating a flow prediction model according to the rolling long short-term memory network and the unfolded long short-term memory network; wherein the flow prediction model is used to predict target flow data in a target time slice according to an obtained flow data sequence.

[0031] In a second aspect, the application provides a mobile network processing device based on flow data, comprising:

[0032] a first obtaining unit configured to obtain a data prediction instruction, wherein the data prediction instruction includes a data identifier, and obtain a flow data sequence corresponding to the data identifier;

[0033] a prediction unit configured to perform flow prediction on the flow data sequence according to a preset flow prediction model, to obtain target flow data in a target time slice; wherein the preset flow prediction model is trained based on a plurality of historical time slices and a historical flow sequence in each historical time slice, and the target time slice includes at least one target time point;

[0034] a first determining unit, configured to determine a network slice allocation strategy corresponding to the target traffic data according to the target traffic data in the target time slice;

[0035] an output unit, configured to output a mobile network according to the network slice allocation strategy; wherein the network slice allocation strategy is used to indicate rule information of allocating virtual network elements and outputting the mobile network according to the allocated virtual network elements.

[0036] Further, the prediction unit comprises:

[0037] a first determining module, configured to determine a target time slice for predicting traffic in the preset traffic prediction model;

[0038] a second determining module, configured to perform traffic prediction on the traffic data sequence according to the preset traffic prediction model to obtain target traffic data in the target time slice.

[0039] Further, the first determining unit comprises:

[0040] a third determining module, configured to, if it is determined that the target traffic data in the target time slice is greater than or equal to a threshold value, determine a maximum traffic value that can be loaded by a current server in the target time slice and a residual traffic value outside the maximum traffic value in the target traffic data;

[0041] a fourth determining module, configured to determine a network slice allocation strategy corresponding to the target traffic data according to the maximum traffic value and the residual traffic value.

[0042] Further, the apparatus further comprises:

[0043] a preprocessing unit, configured to perform preprocessing on the traffic data sequence according to a preset feature extraction function to obtain a preprocessed traffic data sequence;

[0044] The prediction unit is specifically configured to:

[0045] perform traffic prediction on the preprocessed traffic data sequence according to a preset traffic prediction model to obtain target traffic data in a target time slice.

[0046] Further, the preprocessing unit comprises:

[0047] a smoothing module, configured to perform smoothing processing on the traffic data sequence according to a preset feature extraction function;

[0048] a fifth determining module, configured to determine a number of time slices of the traffic data sequence; wherein each time slice comprises a plurality of traffic data;

[0049] The sixth determining module is configured to determine a maximum value and / or a minimum value of the traffic data in each time slice, and determine the maximum value and / or the minimum value as an extreme point in the corresponding time slice;

[0050] The seventh determining module is configured to determine the extreme point and traffic data located at two ends of the traffic data sequence respectively as feature points.

[0051] The eighth determining module is configured to determine a slope of a line connecting two adjacent feature points based on the two adjacent feature points.

[0052] The correcting module is configured to, if it is determined that the slope is greater than a threshold value, perform a correction process on the traffic data between the two adjacent feature points, and obtain a preprocessed traffic data sequence according to the traffic data after the correction.

[0053] Further, the apparatus further comprises:

[0054] The second obtaining unit is configured to obtain a plurality of historical time slices and a historical traffic sequence in each historical time slice.

[0055] The training unit is configured to, for two adjacent historical time slices in the plurality of historical time slices, perform encoding training on a rolling long short-term memory network according to a traffic data sequence of a previous historical time slice in the two adjacent historical time slices, to obtain encoding feature information, and perform traffic prediction training on an unfolded long short-term memory network according to the encoding feature information, to obtain predicted traffic data of a subsequent historical time slice in the two adjacent historical time slices.

[0056] The generating unit is configured to generate the predicted traffic data in a last historical time slice in the plurality of historical time slices, and generate a traffic prediction model according to the rolling long short-term memory network and the unfolded long short-term memory network; wherein the traffic prediction model is configured to predict target traffic data in a target time slice according to an obtained traffic data sequence.

[0057] In a third aspect, a server is provided, which comprises a memory and a processor, and the memory stores a computer program capable of running on the processor, and the processor implements the method in the first aspect when executing the computer program.

[0058] In a fourth aspect, a computer readable storage medium is provided, which stores computer execution instructions, and the computer execution instructions are used to implement the method in the first aspect when executed by a processor.

[0059] In a fifth aspect, a computer program product is provided, which comprises a computer program, and the computer program is used to implement the method in the first aspect when executed by a processor.

[0060] The application provides a mobile network processing method and device based on traffic data and a server. A data prediction instruction is obtained, wherein the data prediction instruction comprises a data identifier, and a traffic data sequence corresponding to the data identifier is obtained. A preset traffic prediction model is used to predict the traffic data sequence to obtain target traffic data in a target time slice, wherein the preset traffic prediction model is trained based on a plurality of historical time slices and historical traffic sequences in each historical time slice, and the target time slice comprises at least one target moment. A network slice allocation strategy corresponding to the target traffic data is determined according to the target traffic data in the target time slice, and a mobile network is output according to the network slice allocation strategy, wherein the network slice allocation strategy is used to indicate rule information of allocating a virtual network element and outputting the mobile network according to the allocated virtual network element. In the scheme, the traffic data sequence corresponding to the data identifier is obtained, the traffic data sequence is input into the preset traffic prediction model, the traffic prediction model is used to predict the traffic data sequence, and the target traffic data in the target time slice can be obtained. The network slice allocation strategy corresponding to the target traffic data is determined, the virtual network element in the current server is moved according to the network slice allocation strategy, and the mobile network is output according to the virtual network element in the current server. Therefore, the target traffic data in the target time slice is accurately obtained by prediction, the virtual network element is allocated according to the network slice allocation strategy corresponding to the target traffic data, the mobile network can be output according to the mobile virtual network when the target time slice is reached, the load pressure of the current server is greatly relieved, the stability of the mobile network is ensured, and the technical problem that the mobile network is unstable due to low accuracy of the traffic prediction result is solved. BRIEF DESCRIPTION OF DRAWINGS

[0061] The accompanying drawings, which are incorporated herein and form part of the specification, illustrate embodiments consistent with the present disclosure and, together with the description, further serve to explain the principles of the present disclosure.

[0062] Figure 1 A flowchart of a mobile network processing method based on traffic data provided by an embodiment of the application;

[0063] Figure 2 A flowchart of another mobile network processing method based on traffic data provided by an embodiment of the application;

[0064] Figure 3 An architecture diagram of a UR-LSTMs provided by an embodiment of the application;

[0065] Figure 4 Another architecture diagram of a UR-LSTMs provided by an embodiment of the application;

[0066] Figure 5A network structure diagram of a mobile network processing method based on traffic data provided by an embodiment of the present application;

[0067] Figure 6 A scene diagram of a mobile network processing method based on traffic data provided by an embodiment of the present application;

[0068] Figure 7 A scene diagram of another mobile network processing method based on traffic data provided by an embodiment of the present application;

[0069] Figure 8 A flow diagram of still another mobile network processing method based on traffic data provided by an embodiment of the present application;

[0070] Figure 9 A structure diagram of a mobile network processing device based on traffic data provided by an embodiment of the present application;

[0071] Figure 10 A structure diagram of another mobile network processing device based on traffic data provided by an embodiment of the present application;

[0072] Figure 11 A structure diagram of a server provided by an embodiment of the present application.

[0073] The specific embodiments of the present disclosure have been shown through the above-described drawings, and will be described in more detail hereinafter. The drawings and the written description are not intended to limit the scope of the present disclosure in any way, but to explain the concept of the present disclosure to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0074] The exemplary embodiments will be described in detail herein below with reference to the accompanying drawings. In the following description, the same drawings refer to the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all the embodiments consistent with the present disclosure.

[0075] Currently, when talking about the 5th Generation Mobile Communication Technology (5G), many people only think that 5G is just a "fast" word, but this is only one of the advantages of 5G. In fact, 5G has three core application scenarios: Enhanced Mobile Broadband (eMBB) requires greater network throughput, peak rate and low latency; massive Machine Type of Communication (mMTC) has a huge number of connections to support massive Internet of Things devices; and ultra-reliable & low-latency communication (uRLLC) focuses on the high reliability and ultra-low latency of the network. Different application scenarios have different requirements for the quality of service of the network, which requires us to create a dedicated network for different application scenarios. However, the network structure has been throughout the world, and it is not realistic to create a dedicated network for different application scenarios under the current network size, and the huge cost is also unimaginable. In this background, network slicing technology emerges as the times require. It builds multiple dedicated, virtualized and mutually isolated logical networks without changing the physical infrastructure of the network to meet the differentiated requirements of different scenarios for network capabilities. The congestion, overload and configuration adjustment of any network slice do not affect the congestion, overload and configuration adjustment of other network slices. Therefore, how to efficiently, securely and stably perform network slicing has become a research topic.

[0076] Therefore, in the use of 5G, in order to ensure the stability of the 5G network, multiple dedicated, virtualized and mutually isolated logical networks are usually constructed according to the network slicing technology to meet the differentiated requirements of different scenarios for network capabilities.

[0077] In an example, when constructing an isolated network, a traffic data sequence of a base station is usually obtained, and a traffic prediction is performed on the traffic data sequence by using a moving average model (MA), an auto regressive model (AR), an auto regressive moving average model (ARMA), or the like, to complete prediction of future traffic of the base station and construct the isolated network. However, in the prior art, the ARMA model combines the advantages of the MA model and the AR model, and corrects the problem of insufficient stability by using a difference method, but still has the following shortcomings: if the sampled traffic sequence is a sequence with large fluctuations, the sequence is still unstable after second-order difference, which causes an error in stationary processing and leads to inaccurate prediction; the model training is time-consuming and is not suitable for a large-scale batch real-time prediction scenario such as network traffic. The ARIMA model based on the ARMA model and the Box-Jenkins method is still required to be stable for time series prediction and analysis, and can only capture the linear relationship between the front and rear data of the time series, but cannot capture the deep relationship such as the nonlinear relationship of the time series, so that a stable isolated network cannot be provided for a user.

[0078] The application provides a mobile network processing method and device based on traffic data and a server, and aims to solve the technical problems of the prior art.

[0079] The technical solutions of the application and how the technical solutions solve the above technical problems will be described in detail in the specific embodiments below. The specific embodiments below can be combined with each other, and the same or similar concepts or processes can not be described again in some embodiments. The embodiments of the application will be described below with reference to the drawings.

[0080] Figure 1 A flowchart of a mobile network processing method based on traffic data provided by an embodiment of the application is shown in FIG. 1, and the method comprises the following steps. Figure 1

[0081] In step 101, a data prediction instruction is obtained, wherein the data prediction instruction comprises a data identifier, and a traffic data sequence corresponding to the data identifier is obtained.

[0082] ​Exemplarily, the execution subject of the embodiment can be a server. First, a data prediction instruction needs to be acquired. Specifically, if a user clicks a data prediction button on a display screen corresponding to the server, a data prediction instruction is triggered to be generated, and the server acquires the data prediction instruction; or the server receives a data prediction instruction sent by another server, etc. The data identifier indicates the identity of the data, for example, the data identifier indicates a plurality of base stations. The server acquires the data prediction instruction, and based on the data identifier in the data prediction instruction, acquires the traffic data sequence of the plurality of base stations corresponding to the data identifier. The sampling interval of the traffic data sequence can be minutes, hours, etc.

[0083] Step 102, performing traffic prediction on the traffic data sequence according to a preset traffic prediction model to acquire target traffic data in a target time slice; wherein the preset traffic prediction model is trained based on a plurality of historical time slices and historical traffic sequences in each historical time slice, and the target time slice includes at least one target time.

[0084] Exemplarily, the preset traffic prediction model is trained based on a plurality of historical time slices and historical traffic sequences in each historical time slice. The traffic prediction model includes a target time slice number for predicting traffic. The traffic data sequence is taken as input data of the traffic prediction model. According to the traffic prediction model, the traffic data sequence is predicted to output target traffic data in a target time slice corresponding to the target time slice number, wherein the target time slice includes at least one target time.

[0085] Step 103, determining a network slice allocation strategy corresponding to the target traffic data according to the target traffic data in the target time slice, and outputting a mobile network according to the network slice allocation strategy; wherein the network slice allocation strategy is used to indicate rule information of allocating virtual network elements and outputting the mobile network according to the allocated virtual network elements.

[0086] Exemplarily, the network slice allocation strategy is used to indicate rule information of allocating virtual network elements and outputting the mobile network according to the allocated virtual network elements, for example, if the load of the current server is too large, part of the virtual network elements can be migrated into other servers. The server determines a network slice allocation strategy corresponding to the target traffic data according to the target traffic data in the target time slice, and according to the network slice allocation strategy, part of the virtual network elements in the current server are migrated into other servers, or part of the virtual network elements are migrated into the current server, and finally the mobile network is output according to the virtual network elements in the current server.

[0087] In the embodiment of the present application, a data prediction instruction is obtained, wherein the data prediction instruction comprises a data identifier, and a traffic data sequence corresponding to the data identifier is obtained. The traffic data sequence is subjected to traffic prediction according to a preset traffic prediction model to obtain target traffic data in a target time slice; wherein the preset traffic prediction model is obtained based on a plurality of historical time slices and historical traffic sequences in each historical time slice, and the target time slice comprises at least one target time. A network slice allocation strategy corresponding to the target traffic data is determined according to the target traffic data in the target time slice, and a mobile network is output according to the network slice allocation strategy; wherein the network slice allocation strategy is used to indicate rule information of allocating virtual network elements and outputting a mobile network according to the allocated virtual network elements. In the present scheme, the traffic data sequence corresponding to the data identifier is obtained, the traffic data sequence is input into the preset traffic prediction model, the traffic data sequence is subjected to traffic prediction according to the traffic prediction model, and the target traffic data in the target time slice can be obtained. The network slice allocation strategy corresponding to the target traffic data is determined, the virtual network elements in the current server are moved according to the network slice allocation strategy, and the mobile network is output according to the virtual network elements in the current server. Therefore, the target traffic data in the target time slice is obtained by prediction, and the virtual network elements are allocated according to the network slice allocation strategy corresponding to the target traffic data, so that the mobile network can be output according to the mobile virtual network when the target time slice is reached, which greatly relieves the load pressure of the current server, guarantees the stability of the mobile network, and solves the technical problem of unstable mobile network caused by low accuracy of traffic prediction results.

[0088] Figure 2 Another flowchart of a mobile network processing method based on traffic data provided by the embodiment of the present application is shown in FIG. 2, which comprises the following steps. Figure 2

[0089] Step 201, a plurality of historical time slices and historical traffic sequences in each historical time slice are obtained.

[0090] For example, the server obtains a plurality of historical time slices and historical traffic sequences S=(s0, s1, s2…sn) in each historical time slice. t-1 Further, the historical traffic sequence can be preprocessed to obtain a preprocessed historical traffic sequence S pre-train The preprocessing process can refer to step 205.

[0091] ​Step 202: For two adjacent historical time slices among multiple historical time slices, the rolling long short-term memory network is encoded and trained based on the traffic data sequence of the previous historical time slice in the two adjacent historical time slices to obtain encoded feature information; and based on the encoded feature information, the unfolded long short-term memory network is trained for traffic prediction to obtain the predicted traffic data of the next historical time slice in the two adjacent historical time slices.

[0092] For example, the Rolling-Unrolling LSTMs (UR-LSTMs) architecture originates from encoder-decoder models for text processing, where the input and output are sequence-to-sequence pairs. Such text processing models consist of an encoder that processes the words of the input sentence and a decoder that generates the words of the output sentence. Both the encoder and decoder are typically implemented using LSTMs; the difference is that, instead of English sentences, the model processes traffic sequences. The decoder initializes itself internally using the encoder's internal representation and iterates based on this to predict future traffic values. To enable continuous traffic prediction, the decoder is appended to each encoding time slice. This allows the model to predict traffic in a continuous manner and continuously improve. The encoder is called a Rolling Long Short-Term Memory network (R-LSTM), and the decoder is called an Unrolling Long Short-Term Memory network (U-LSTM = Unrolling LSTMs, U-LSTM).

[0093] Figure 3 An architecture diagram of UR-LSTMs provided for embodiments of this application is shown below. Figure 3 As shown in the figure This is a feature extraction function for traffic data sequences. For better understanding, [the following is a more detailed explanation of the function]. Figure 3 The time slice in is set to T o =1,T a =3(s0 is T) o s e s e+1 and s t-1 The three are T a Furthermore, the flow data sequence S = (s0, s1, s2…s) has already been defined previously. t-1 Preprocessing: The traffic data sequence currently used for model training should be the preprocessed traffic data sequence S. pre-train .

[0094] Figure 4 Another UR-LSTMs architecture diagram provided for embodiments of this application, such as... Figure 4 As shown, in the first stage of encoding and training R-LSTM, given a time slice t, the preprocessed traffic data sequence S from the previous historical time slice is used.pre-train input R-LSTM, output encoded feature information f s，t The R-LSTM is trained by encoding, and the encoded incoming traffic information is updated cyclically. The following formula (1) is executed:

[0095]

[0096] In formula (1), denotes the R-LSTM related to the preprocessed traffic data sequence S pre-train , which can be uniquely determined by learning parameters . And and are the hidden and cell states calculated in the preprocessed traffic data sequence S pre-train at time slice T=t, and t-1 in formula (1) is also a parameter, and are the hidden and cell states calculated at time slice T=t-1. The initial hidden and cell states of the R-LSTM are initialized with zero, and based on the hidden layer parameters of the zero initialization, the new hidden layer parameters can be obtained by encoding training formula (1), and the initial hidden and cell states are zero initialized as follows:

[0097]

[0098] And in the second phase of traffic prediction training of the U-LSTM, for each time slice t, the U-LSTM is used to predict the future traffic value. The U-LSTM itself is initialized with the hidden and cell states of the R-LSTM in the current time slice T=t:

[0099]

[0100] The encoded feature information f s，t is input into the U-LSTM, the U-LSTM is trained by traffic prediction, and the predicted traffic data of the latter historical time slice in the adjacent two historical time slices is obtained, and the current observed encoded feature information f s，t is iterated, and the number of iterations nt is equal to the number of time slices required when the time point of the predicted traffic occurs: T o +T a +t-1. Similar to the R-LSTM, when the U-LSTM generates the jth iteration, the formula (2) for calculating the hidden and cell states is as follows:

[0101]

[0102] In formula (2), This represents the preprocessed flow data sequence S. pre-train The relevant U-LSTM network can be learned from the parameters. The only certainty. And and In the preprocessed flow data sequence S pre-train The hidden and cell states are calculated in the j-th iteration.

[0103] For example, during the prediction process, based on the encoded feature information of the R-LSTM output and the hidden layer parameters, the calculation formula for the hidden and cell states of the U-LSTM is as follows:

[0104]

[0105] Where, input f s，t+j-1 This indicates that features are sampled from the future time slice t+j-1.

[0106] It is worth noting that the input f of U-LSTM s，t It does not depend on j, as it is fixed during the "unfolding" process. The main reason for "unfolding" the U-LSTM a different number of times on each time slice is to encourage the model to produce different predictions at different expected time scales.

[0107] Finally, the U-LSTM predicted flow rate is obtained by linearly transforming the hidden and cell states to obtain the predicted flow rate data p. t The linear transformation is performed according to the following formula (3):

[0108]

[0109] in, It is based on The hidden state obtained from the final iteration is... This represents the last hidden state of the final LSTM layer in the iteration. The hidden state in the last historical time slice out of multiple historical time slices; and All of these are learnable parameters.

[0110] In this step, for the adjacent two historical time slices in the plurality of historical time slices, the preprocessed traffic data sequence of the previous historical time slice in the adjacent two historical time slices is input into the R-LSTM, the R-LSTM is encoded and trained to obtain the encoded feature information, and the specific encoding and training process is referred to the first stage as indicated above. And the encoded feature information is input into the U-LSTM, the U-LSTM is trained for traffic prediction, and the predicted traffic data p of the subsequent historical time slice in the adjacent two historical time slices is obtained t , and the specific prediction training process is referred to the second stage as indicated above and the final linear transformation stage. According to the traffic data sequence in the plurality of historical time slices, the R-LSTM is encoded and trained, and the U-LSTM is trained for traffic prediction, until the last historical time slice is trained.

[0111] Step 203, generating predicted traffic data in the last historical time slice in the plurality of historical time slices, and generating a traffic prediction model according to the rolling long short-term memory network and the unfolded long short-term memory network; wherein the traffic prediction model is used to predict the target traffic data in the target time slice according to the obtained traffic data sequence.

[0112] Illustratively, the server generates predicted traffic data in the last historical time slice in the plurality of historical time slices, and generates a traffic prediction model according to the rolling long short-term memory network and the unfolded long short-term memory network, and then the trained traffic prediction model can predict the obtained traffic data sequence and obtain the target traffic data in the target time slice.

[0113] Step 204, obtaining a data prediction instruction, wherein the data prediction instruction includes a data identifier, and obtaining a traffic data sequence corresponding to the data identifier.

[0114] Illustratively, this step can refer to step 101 in Figure 1 , and will not be repeated here.

[0115] Step 205, preprocessing the traffic data sequence according to a preset feature extraction function to obtain a preprocessed traffic data sequence.

[0116] In one example, step 205 includes: smoothing the traffic data sequence according to a preset feature extraction function; determining the number of time slices of the traffic data sequence; wherein each time slice includes a plurality of traffic data; determining the maximum value and / or the minimum value of the traffic data in each time slice, and determining the maximum value and / or the minimum value as extreme points in the corresponding time slice; determining the extreme points and the traffic data respectively located at both ends of the traffic data sequence as feature points; determining the slope of the line connecting two adjacent feature points based on the two adjacent feature points; if it is determined that the slope is greater than a threshold value, performing correction processing on the traffic data between the two adjacent feature points, and obtaining the preprocessed traffic data sequence according to the corrected traffic data.

[0117] Exemplarily, Figure 5 A network structure diagram of a mobile network processing method based on traffic data provided by an embodiment of the present application is shown in Figure 5 In order for the U-LSTM and the R-LSTM to focus more on processing respective subtasks, the SCP (Sequence Completion Pre-training) technology needs to be used to pretrain the traffic sequence before the traffic sequence is input into the network for training. For the convenience of understanding, the time slice T is set to 1, and T is set to 3. The specific preprocessing process is as follows: o a

[0118] Figure 5 is a feature extraction function of the traffic data sequence, and specifically as follows:

[0119] 1. According to a preset feature extraction function, the sampled traffic data sequence is smoothed;

[0120] 2. The x-axis is divided in units of time slice length (min), and the y-axis represents the traffic data size. In a plurality of time slices (referred to as a time domain), the points at which the maximum value or / and the minimum value of y is obtained in the time domain are found, and the points at which the maximum value or / and the minimum value are set as extreme points of the time domain;

[0121] 3. The end points at both ends of the traffic data sequence and the extreme points obtained in each time domain are collectively set as feature points;

[0122] 4. According to the obtained feature points, the time data sequence is divided into a plurality of parts, and the adjacent two feature points are connected into a straight line. The greater the absolute value of the absolute value of the slope of the straight line, the greater the signal strength and the degree of change of the feature interval in the y-axis direction;

[0123] ​​​5、extracting a region with a variation greater than a threshold in the y-axis direction, and performing correction processing on the flow data between the two adjacent feature points, the correction processing including removing the extracted region, or replacing the flow data corresponding to the maximum extreme point in the extracted region with the flow data corresponding to the highest peak value of the region, to obtain encoded feature information, the encoded feature information being the preprocessed flow data sequence. Therefore, the preprocessing process is suitable for anomaly detection of the flow data sequence, and further obtains the preprocessed encoded feature information.

[0124] Step 206, determining a target time slice for predicting flow in the preset flow prediction model.

[0125] Exemplarily, the server determines a target time slice for predicting flow in the preset flow prediction model.

[0126] Step 207, performing flow prediction on the preprocessed flow data sequence according to the preset flow prediction model, to obtain target flow data in the target time slice corresponding to the target time slice number.

[0127] Exemplarily, the server performs flow prediction on the preprocessed flow data sequence according to the preset flow prediction model, to obtain target flow data in the target time slice.

[0128] For example, the flow data of the base station is first sampled, once per minute, and the flow data sequence is denoted as S=(s0, s1, s2…s t-1 As mentioned above, the UR-LSTMs need to process two sub-tasks, namely the "encoding" stage and the "prediction" stage, as shown below Figure 6 Figure 6 A scene diagram of a mobile network processing method based on flow data provided by an embodiment of the present application, Figure 6 including an encoding stage, a prediction stage, and an unknown stage.

[0129] It should be particularly noted that the "prediction" stage has already occurred, that is, the flow data in the "prediction" stage has already occurred, rather than not occurring and needing to be predicted. The so-called "prediction" stage means that after the "encoding" stage in the previous paragraph, the prediction stage starts at T=e+1, and the flow s t at time node T=t is predicted, and the predicted value of the flow is P e+1 . The flow can be predicted once every time node, until the last prediction P t of s t-1 ​, of course, the closer to the time T=t, the more accurate the prediction will be. As mentioned above, the prediction time, the architecture can be implemented in a multi-scale manner for how long in advance of the time node that needs to be predicted. That is, the prediction time can change, unlike many current prediction methods that specify that the time span from the start of prediction to the predicted time node is constant. Thus, the model can be more flexible in prediction. Of course, for the number of time slices T o = e in the "encoding" phase and the number of time slices T a = t-e-1 in the "prediction" phase, the parameters t and e need to be set in advance. Since the length of the encoding feature information (long-term memory) obtained after preprocessing the traffic data sequence will affect the training of the network, different parameter settings will affect the accuracy of the prediction.

[0130] The above is the process of predicting the traffic s t at the time node T=t, since the number of time slices in the "prediction" phase is not 1, s t+1 can be predicted from the second time node in the "prediction" phase, s t+2 can be predicted from the third time node in the "prediction" phase, and so on. Then at T=t-1, the predicted traffic values (s t , s t+1 , s t+2 …) in the future can be considered as much as possible, so as to make better network slicing decisions.

[0131] Step 208, if it is determined that the target traffic data in the target time slice is greater than or equal to the threshold value, the maximum traffic value that the current server can load in the target time slice and the remaining traffic value outside the maximum traffic value in the target traffic data are determined.

[0132] Exemplarily, the server compares the target traffic data in the target time slice with the preset threshold value, and if it is determined that the target traffic data in the target time slice is greater than or equal to the threshold value, the maximum traffic value that the current server can load in the target time slice and the remaining traffic value outside the maximum traffic value in the target traffic data are determined. If it is determined that the target traffic data in the target time slice is less than the threshold value, it is determined that the current server can load the target traffic data in the target time slice, and there is no need to compare with the threshold value.

[0133] Step 209, according to the maximum traffic value and the remaining traffic value, a network slice allocation strategy corresponding to the target traffic data is determined.

[0134] Exemplarily, the server determines the network slice allocation strategy corresponding to the target traffic data according to the maximum traffic value and the residual traffic value, that is, the number of virtual network elements that need to be reserved in the current server and the number of virtual network elements that need to be migrated out can be determined.

[0135] Specifically, NFV (Network Functions Virtualization) is a network function virtualization technology, which is based on large shared OTS (Off-The-Shelf) servers and explores the virtualization use of network entities through software-defined manner. The virtual machine (Virtual Machines) technology used in NFV is a kind of virtualization technology. Based on the network function virtualization technology, the complete decoupling of software and hardware is realized, so as to realize the intelligent migration in and out of the NFV network element. Therefore, the server can generate a network slice allocation strategy through the network slice technology of NFV, and output a mobile network according to the network slice allocation strategy.

[0136] The network with the NFV technology can integrate network device resources in a unified resource pool, and dynamically allocate the resource pool according to the abstracted network service and functional requirements. Figure 7 Another scene schematic diagram of the mobile network processing method based on traffic data provided by the embodiment of the present application is shown in FIG. 2. Figure 7 As shown in FIG. 2, Figure 7 When the server with the NFV technology detects that the load of the server 1 shared by the VNF 1 and the VNF 2 is too large, the server will migrate part of the virtual network elements (VNF 1 and VNF 2) in the server 1 to the server 2 (annotated arrows 1 and 2) according to the traffic perception level of the VNF 1 and the VNF 2, and migrate them to the server 2 (annotated arrows 3 and 4) that meets the traffic demand, so as to realize the intelligent allocation of network resources. This is the network slice decision-making process based on traffic perception, and the network slice allocation strategy is obtained. Figure 7

[0137] Step 210: outputting a mobile network according to the network slice allocation strategy.

[0138] Exemplarily, the server can output a mobile network according to the network slice allocation strategy.

[0139] ​In the embodiments of the present application, a plurality of historical time slices and a historical traffic sequence in each historical time slice are obtained. The encoding feature information is obtained by encoding and training the rolling long short-term memory network according to the traffic data sequence of the previous historical time slice in the adjacent two historical time slices in the plurality of historical time slices, and the prediction traffic data of the latter historical time slice in the adjacent two historical time slices is obtained by performing traffic prediction training on the unrolled long short-term memory network according to the encoding feature information. The prediction traffic data in the last historical time slice in the plurality of historical time slices is generated, and the traffic prediction model is generated according to the rolling long short-term memory network and the unrolled long short-term memory network; wherein the traffic prediction model is used to predict the target traffic data in the target time slice according to the obtained traffic data sequence. A data prediction instruction is obtained, wherein the data prediction instruction includes a data identifier, and a traffic data sequence corresponding to the data identifier is obtained. The traffic data sequence is preprocessed according to a preset feature extraction function to obtain a preprocessed traffic data sequence. The target time slice in the preset traffic prediction model for predicting traffic is determined. The preprocessed traffic data sequence is subjected to traffic prediction according to the preset traffic prediction model, and the target traffic data in the target time slice corresponding to the target time slice number is obtained. If it is determined that the target traffic data in the target time slice is greater than or equal to a threshold value, the maximum traffic value that the current server can load in the target time slice and the remaining traffic value outside the maximum traffic value in the target traffic data are determined. According to the maximum traffic value and the remaining traffic value, a network slice allocation strategy corresponding to the target traffic data is determined. The mobile network is output according to the network slice allocation strategy.

[0140] Therefore, by predicting the target traffic data in the accurate target time slice and allocating the virtual network elements according to the network slice allocation strategy corresponding to the target traffic data, the mobile network can be output according to the movement when the target time slice is reached, which greatly relieves the load pressure of the current server, guarantees the stability of the mobile network, and solves the technical problem that the mobile network is unstable due to the low accuracy of the traffic prediction result. Moreover, unlike many time series prediction methods, the UR-LSTMs in the present application can capture the nonlinear relationship between the data before and after the time series, and do not require the traffic data sequence to have strong stationarity, and can be trained on a large scale; and the traffic conditions that may occur in the future at the base station in the access network layer are considered, that is, the traffic at multiple time points in the future is considered, so as to comprehensively consider as the basis for network slicing. With continuous updating of the base station traffic state, the training set can be continuously updated, so as to iterate, and the model parameters are continuously optimized, and the accuracy of the predicted traffic will show better and better results, achieving the purpose of dynamic decision of network slicing; the present application also allows considering the prediction time at different scales, and many current prediction algorithms limit the prediction time to be constant, so that when special circumstances require early prediction, the task cannot be completed, but the UR-LSTMs can achieve this goal; the SCP technology is used for preprocessing the traffic data sequence, so that the two sub-networks in the RU-LSTMs can focus on the sub-tasks they need to complete.

[0141] Exemplarily, Figure 8 A flowchart of another traffic data-based mobile network processing method provided by an embodiment of the present application is shown in FIG. 6. Figure 8 As shown in FIG. 6, the first step is to sample the traffic data sequence, the second step is to preprocess the traffic data sequence by using the SCP technology, the third step is a feature extraction operation, to obtain the encoded feature information in the preprocessed traffic data sequence, the fourth step is to train according to the encoded feature information and obtain the UR-LSRMs, the fifth step is to predict the load of the base station traffic according to the UR-LSRMs, and the sixth step is to determine the network slice allocation strategy corresponding to the target traffic data according to the predicted target traffic data.

[0142] Figure 9 A structural diagram of a traffic data-based mobile network processing device provided by an embodiment of the present application is shown in FIG. 7. Figure 9 As shown in FIG. 7, the device comprises:

[0143] A first acquisition unit 31 is configured to acquire a data prediction instruction, wherein the data prediction instruction comprises a data identifier, and acquire a traffic data sequence corresponding to the data identifier.

[0144] The prediction unit 32 is configured to perform traffic prediction on the traffic data sequence according to a preset traffic prediction model to obtain target traffic data in a target time slice; the preset traffic prediction model is trained based on a plurality of historical time slices and historical traffic sequences in each historical time slice, and the target time slice includes at least one target moment.

[0145] The first determination unit 33 is configured to determine a network slice allocation strategy corresponding to the target traffic data according to the target traffic data in the target time slice.

[0146] The output unit 34 is configured to output a mobile network according to the network slice allocation strategy; the network slice allocation strategy is used to indicate rule information of allocating virtual network elements and outputting the mobile network according to the allocated virtual network elements.

[0147] The device of the embodiment can execute the technical solutions in the above method, and the specific implementation process and technical principles are the same, which will not be repeated here.

[0148] Figure 10 Another structure diagram of a mobile network processing device based on traffic data provided by the embodiment of the application is shown in Figure 9 Based on the embodiment shown in Figure 10 The prediction unit 32 includes:

[0149] The first determination module 321 is configured to determine a target time slice for predicting traffic in the preset traffic prediction model.

[0150] The second determination module 322 is configured to perform traffic prediction on the traffic data sequence according to the preset traffic prediction model to obtain target traffic data in the target time slice.

[0151] In one example, the first determination unit 33 includes:

[0152] The third determination module 331 is configured to, if it is determined that the target traffic data in the target time slice is greater than or equal to a threshold value, determine a maximum traffic value that the current server can load in the target time slice and a residual traffic value outside the maximum traffic value in the target traffic data.

[0153] The fourth determination module 332 is configured to determine a network slice allocation strategy corresponding to the target traffic data according to the maximum traffic value and the residual traffic value.

[0154] In one example, the device further includes:

[0155] The preprocessing unit 41 is configured to perform preprocessing on the traffic data sequence according to a preset feature extraction function to obtain a preprocessed traffic data sequence.

[0156] The prediction unit 32 is specifically configured to:

[0157] According to the preset traffic prediction model, the preprocessed traffic data sequence is predicted to obtain target traffic data in a target time slice.

[0158] In one example, the preprocessing unit 41 comprises:

[0159] The smoothing module 411 is configured to perform smoothing processing on the traffic data sequence according to a preset feature extraction function.

[0160] The fifth determination module 412 is configured to determine the number of time slices of the traffic data sequence, wherein each time slice comprises a plurality of traffic data.

[0161] The sixth determination module 413 is configured to determine the maximum value and / or the minimum value of the traffic data in each time slice, and determine the maximum value and / or the minimum value as extreme points in the corresponding time slice.

[0162] The seventh determination module 414 is configured to determine the extreme points and the traffic data respectively located at both ends of the traffic data sequence as feature points.

[0163] The eighth determination module 415 is configured to determine the slope of the line between two adjacent feature points based on the two adjacent feature points.

[0164] The correction module 416 is configured to, if it is determined that the slope is greater than a threshold value, perform correction processing on the traffic data between the two adjacent feature points, and obtain the preprocessed traffic data sequence according to the corrected traffic data.

[0165] In one example, the device further comprises:

[0166] The second acquisition unit 42 is configured to acquire a plurality of historical time slices and a historical traffic sequence in each historical time slice.

[0167] The training unit 43 is configured to encode feature information, encode and train the rolling long short-term memory network according to the traffic data sequence of a previous historical time slice in two adjacent historical time slices in the plurality of historical time slices, to obtain the encoded feature information, and perform traffic prediction training on the unfolded long short-term memory network according to the encoded feature information, to obtain predicted traffic data of a subsequent historical time slice in the two adjacent historical time slices.

[0168] The generation unit 44 is configured to generate the predicted traffic data in the last historical time slice in the plurality of historical time slices, and generate a traffic prediction model according to the rolling long short-term memory network and the unfolded long short-term memory network; wherein the traffic prediction model is configured to predict target traffic data in a target time slice according to the acquired traffic data sequence.

[0169] The device of the embodiment can execute the technical solutions in the above method, and the specific implementation process and technical principles are the same, which will not be repeated here.

[0170] Figure 11 A structural schematic diagram of a server provided by the embodiment of the application is shown in the figure, and the server comprises a memory 51 and a processor 52. Figure 11

[0171] The memory 51 stores a computer program that can run on the processor 52.

[0172] The processor 52 is configured to execute the method provided by the above embodiment.

[0173] The server further comprises a receiver 53 and a transmitter 54. The receiver 53 is configured to receive instructions and data sent by an external device, and the transmitter 54 is configured to send instructions and data to the external device.

[0174] The embodiment of the application further provides a non-transitory computer readable storage medium, when instructions in the storage medium are executed by a processor of a server, the server can execute the method provided by the above embodiment.

[0175] The embodiment of the application further provides a computer program product, the computer program product comprises a computer program, the computer program is stored in a readable storage medium, at least one processor of a server can read the computer program from the readable storage medium, and the at least one processor executes the computer program so that the server executes the scheme provided by any one of the above embodiments.

[0176] Other embodiments of the disclosure will be apparent to those skilled in the art from consideration of the specification and practice of the application disclosed herein. This application is intended to cover any variations, uses or adaptations of the disclosure that are deemed to fall within the general principles of the disclosure and include commonly known or customary practice in the art. The specification and examples are to be considered exemplary only, and the true scope and spirit of the disclosure are indicated by the following claims.

[0177] It should be understood that the present disclosure is not limited to the precise structures described and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present disclosure is limited only by the appended claims.​

Claims

1. A mobile network processing method based on traffic data, characterized in that, The method comprises: obtaining data prediction instructions, wherein the data prediction instructions comprise a data identifier, and obtaining a traffic data sequence corresponding to the data identifier; performing traffic prediction on the traffic data sequence according to a preset traffic prediction model to obtain target traffic data in a target time slice; wherein the preset traffic prediction model is trained based on a plurality of historical time slices and historical traffic sequences in each historical time slice, the traffic prediction model comprises a rolling long short-term memory network and an unrolled long short-term memory network, for adjacent two historical time slices in the plurality of historical time slices, the rolling long short-term memory network is trained according to a traffic data sequence of a previous historical time slice to obtain encoding feature information, and the unrolled long short-term memory network is trained according to the encoding feature information to obtain predicted traffic data of a next historical time slice, and the target time slice comprises at least one target time point; determining a network slice allocation strategy corresponding to the target traffic data according to the target traffic data in the target time slice, and outputting a mobile network according to the network slice allocation strategy; wherein the determination of the network slice allocation strategy comprises: if it is determined that the target traffic data in the target time slice is greater than or equal to a threshold value, determining a maximum traffic value that can be loaded by a current server in the target time slice and a residual traffic value outside the maximum traffic value in the target traffic data, and determining a network slice allocation strategy according to the maximum traffic value and the residual traffic value; the network slice allocation strategy is used to indicate the allocation of virtual network elements and the rule information of the output of the mobile network according to the allocated virtual network elements.

2. The method of claim 1, wherein, The method further comprises: determining a target time slice for predicting traffic in the preset traffic prediction model; performing traffic prediction on the traffic data sequence according to the preset traffic prediction model to obtain target traffic data in the target time slice.

3. The method of claim 1, wherein, The method further comprises: preprocessing the traffic data sequence according to a preset feature extraction function to obtain a preprocessed traffic data sequence; The method further comprises: performing traffic prediction on the preprocessed traffic data sequence according to the preset traffic prediction model to obtain target traffic data in a target time slice.

4. The method of claim 3, wherein, The method further comprises: performing smoothing processing on the traffic data sequence according to a preset feature extraction function; determining the number of time slices of the traffic data sequence; wherein each time slice comprises a plurality of traffic data; determining the maximum value and / or the minimum value of the traffic data in each time slice, and determining the maximum value and / or the minimum value as extreme points in the corresponding time slice; determining the extreme points and traffic data located at both ends of the traffic data sequence as feature points; Determine a slope of a line connecting two adjacent feature points based on the two adjacent feature points. If it is determined that the slope is greater than a threshold value, correct the flow data between the two adjacent feature points, and obtain a preprocessed flow data sequence based on the corrected flow data.

5. The method according to any one of claims 1 to 4, characterized in that, The method further includes: Obtain a plurality of historical time slices and a historical flow sequence in each historical time slice; For two adjacent historical time slices in the plurality of historical time slices, encode and train a rolling long short-term memory network based on a flow data sequence of a previous historical time slice in the two adjacent historical time slices to obtain encoding feature information, and predict and train an unrolled long short-term memory network based on the encoding feature information to obtain predicted flow data of a subsequent historical time slice in the two adjacent historical time slices; Generate predicted flow data in a last historical time slice in the plurality of historical time slices, and generate a flow prediction model based on the rolling long short-term memory network and the unrolled long short-term memory network; wherein the flow prediction model is used to predict target flow data in a target time slice based on an obtained flow data sequence.

6. A mobile network processing device based on traffic data, characterized by, Comprise: A first obtaining unit is configured to obtain a data prediction instruction, wherein the data prediction instruction comprises a data identifier, and obtain a flow data sequence corresponding to the data identifier; A prediction unit is configured to predict flow of the flow data sequence based on a preset flow prediction model to obtain target flow data in a target time slice; wherein the preset flow prediction model is trained based on a plurality of historical time slices and a historical flow sequence in each historical time slice, the flow prediction model comprises a rolling long short-term memory network and an unrolled long short-term memory network, for two adjacent historical time slices in the plurality of historical time slices, the rolling long short-term memory network is encoded and trained based on a flow data sequence of a previous historical time slice to obtain encoding feature information, and the unrolled long short-term memory network is predicted and trained based on the encoding feature information to obtain predicted flow data of a subsequent historical time slice, and the target time slice comprises at least one target time; A first determining unit is configured to determine a network slice allocation strategy corresponding to the target flow data based on the target flow data in the target time slice; The first determining unit comprises: A third determining module is configured to determine a maximum flow value that a current server can load in the target time slice and a residual flow value other than the maximum flow value in the target flow data if it is determined that the target flow data in the target time slice is greater than or equal to a threshold value; A fourth determining module is configured to determine a network slice allocation strategy corresponding to the target flow data based on the maximum flow value and the residual flow value; An output unit is configured to output a mobile network based on the network slice allocation strategy; wherein the network slice allocation strategy is used to indicate rules information of allocating a virtual network element and outputting the mobile network based on the allocated virtual network element.

7. The apparatus of claim 6, wherein, The prediction unit comprises: The first determining module is configured to determine a target time slice for predicting traffic in the preset traffic prediction model; The second determining module is configured to perform traffic prediction on the traffic data sequence according to the preset traffic prediction model to obtain target traffic data in the target time slice.

8. The apparatus of claim 6, wherein, The device further comprises: The preprocessing unit is configured to perform preprocessing on the traffic data sequence according to a preset feature extraction function to obtain a preprocessed traffic data sequence. The prediction unit is specifically configured to: perform traffic prediction on the preprocessed traffic data sequence according to the preset traffic prediction model to obtain target traffic data in the target time slice.

9. The apparatus of claim 8, wherein, The preprocessing unit comprises: The smoothing module is configured to perform smoothing processing on the traffic data sequence according to a preset feature extraction function. The fifth determining module is configured to determine a number of time slices of the traffic data sequence; each time slice includes a plurality of traffic data. The sixth determining module is configured to determine a maximum value and / or a minimum value of the traffic data in each time slice and determine the maximum value and / or the minimum value as an extreme point in the corresponding time slice. The seventh determining module is configured to determine the extreme point and traffic data respectively located at two ends of the traffic data sequence as feature points. The eighth determining module is configured to determine a slope of a line between two adjacent feature points based on the two adjacent feature points. The correction module is configured to, if the slope is greater than a threshold value, perform correction processing on traffic data between the two adjacent feature points, and obtain a preprocessed traffic data sequence according to the corrected traffic data.

10. The device of any one of claims 6-9, wherein, The device further comprises: The second obtaining unit is configured to obtain a plurality of historical time slices and a historical traffic sequence in each historical time slice. The training unit is configured to, for two adjacent historical time slices in the plurality of historical time slices, perform encoding training on a rolling long short-term memory network according to a traffic data sequence of a previous historical time slice in the two adjacent historical time slices to obtain encoding feature information, and perform traffic prediction training on an unfolded long short-term memory network according to the encoding feature information to obtain predicted traffic data of a subsequent historical time slice in the two adjacent historical time slices. The generation unit is configured to generate predicted traffic data in a last historical time slice in the plurality of historical time slices, and generate a traffic prediction model according to the rolling long short-term memory network and the unfolded long short-term memory network; the traffic prediction model is configured to predict target traffic data in a target time slice according to an obtained traffic data sequence.

11. A server, characterized by The computer readable storage medium stores computer execution instructions, and the computer execution instructions are executed by the processor to implement the method in any one of claims 1-5.

12. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer execution instructions, and the computer execution instructions are executed by the processor to implement the method in any one of claims 1-5.

13. A computer program product, characterised in that, The computer readable storage medium stores computer execution instructions, and the computer execution instructions are executed by the processor to implement the method in any one of claims 1-5.

Images