A cross-domain migration spatio-temporal integration network method

Abstract

The application belongs to the technical field of mobile traffic prediction, and particularly relates to a cross-domain transfer spatio-temporal integrated network method, comprising the following steps: S1, constructing an integrated learning model, using integrated learning to initially train a basic learner, and combining a spatio-temporal network to capture spatio-temporal features, which is used to overcome the instability and overfitting defects existing in a multi-dimensional spatio-temporal cross-domain environment, wherein the integrated learning model comprises a traffic data set structure, spectral clustering, a long short-term memory (LSTM) and a convolutional neural network (CNN). The application proposes a new transfer learning algorithm, which trains a mobile traffic prediction model from source domain data rich in data, and then migrates the model to a mobile edge network environment with sparse data, so as to improve the prediction accuracy of the model, and at the same time proposes a new transfer mapping framework, that is, learning a reflection function from experience data and optimizing the reflection function to predict when to migrate or how to transfer to obtain the best prediction performance.

Description

Technical Field

[0001] This invention belongs to the field of mobile traffic prediction technology, specifically relating to a cross-domain transfer spatiotemporal integrated network method. Background Technology

[0002] Mobile edge computing (MEC) has become a hot topic in next-generation computing architectures. This new computing paradigm shifts network computing resources and task data to the edge network to improve task computing performance and reduce network bandwidth requirements. With the continuous advancement of MEC technology in the internet industry, it has led to a surge in network traffic, with mobile traffic now accounting for 20% of total web application network traffic. This figure is particularly pronounced with the emergence of short-video mobile applications such as TikTok and Bilibili, where mobile devices now generate 86% of total mobile traffic.

[0003] One of the keys to achieving high-quality computing services for MEC is accurate prediction of wireless network traffic, as mobile network services such as offloading, spectrum access, and computing resource allocation require the ability to analyze and predict network traffic (which affects data rates). Although researchers and industry have conducted in-depth research on this topic, traffic prediction for existing mobile edge networks still faces many challenges: (1) Mobile devices have different service demands at different locations and times, making wireless traffic prediction difficult. (2) Wireless traffic prediction for mobile devices involves temporal and spatial characteristics, and the complex spatial and temporal dependencies make prediction even more difficult. (3) Due to varying levels of development of mobile edge network monitoring infrastructure, sensor data is relatively sparse in some areas, resulting in lower accuracy in wireless traffic prediction.

[0004] Therefore, to address the aforementioned technical problems, it is necessary to provide a cross-domain transfer spatiotemporal integration network method.

[0005] The information disclosed in this background section is intended only to enhance the understanding of the overall background of the invention and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to provide a cross-domain transfer spatiotemporal integrated network method that can solve the problems mentioned in the background art.

[0007] To achieve the above objectives, a specific embodiment of the present invention provides the following technical solution:

[0008] A cross-domain transfer spatiotemporal integration network method includes the following steps:

[0009] S1. Construct an ensemble learning model, use ensemble learning to initially train basic learners, and combine it with a spatiotemporal network to capture spatiotemporal features to overcome the instability and overfitting shortcomings in multidimensional spatiotemporal cross-domain environments. The ensemble learning model includes a traffic dataset structure, spectral clustering, long short-term memory (LSTM), and convolutional neural network (CNN).

[0010] S2. To address the diversity and similarity between the target domain and the source domain, data preprocessing is performed on different cross-domain data. By summarizing historical transfer experience, an experience-based mechanism is established, and transfer reflection functions between the source and target domains are designed. The feature differences between different target and source domains are learned and parameterized to optimize knowledge transfer and improve learning performance in potential spatial cross-domain environments.

[0011] In one or more embodiments of the present invention, in S1, for spatiotemporal relationships, the spatial location of a mobile device is defined as X×B, and there exist n mobile networks with different degrees of data sparsity, i.e., v∈{1,...,n}. Given u base stations, the wireless traffic spatiotemporal sequence of each network data point can be denoted as... Where T is the total time interval. Let be the traffic flow matrix at point t, defined as:

[0012]

[0013] in, It is used to measure the traffic capacity of mobile networks that is related to time and space, i.e.

[0014] In one or more embodiments of the present invention, due to The value at t+1 depends only on the traffic of neighboring mobile networks and is related to distant mobile networks. Therefore, the autocorrelation coefficient in the spatiotemporal domain can be defined as:

[0015]

[0016] in, It is the average value that changes over time;

[0017] To address the issue of uneven distribution of mobile network traffic, the Pearson correlation coefficient is used to measure spatial correlation, i.e.:

[0018]

[0019] Where C(.) is the covariance operator, s is the standard deviation, the spatiotemporal data is regarded as a spatial point on a time series, a spectral clustering strategy is used for clustering, and then d is used ijExpressing the Euclidean distance between two points, we obtain:

[0020]

[0021] Where, ω ij It is data point x i x j The weights between them.

[0022] In one or more embodiments of the present invention, a graph consisting of all datasets is constructed by calculating the weights between data points. Then, the subgraph obtained by connecting the edges of the sum of the weights of the graph is made as small as possible, and the graph with the sum of the weights is made as large as possible. The measure of the graph is:

[0023]

[0024] According to KNN, the adjacency matrix M is obtained as follows:

[0025]

[0026] In one or more embodiments of the present invention, three data input types are defined based on the spatiotemporal characteristics of mobile network traffic flows: the first type is input data as {D t-1 ,…,D t-m The first type is data stream sequence before the target time interval, m∈N}; the second type is time-related data, m data; the third type is cross-domain dataset.

[0027] In one or more embodiments of the present invention, a deep network is established based on three data input types. The deep network includes constructing a spatiotemporal model, embedding m-data, and constructing cross-domain data, wherein the m-data is temporal metadata.

[0028] In one or more embodiments of the present invention, in S2, let E v =((S) v ,T v ),l v ) represents the transfer learning memory pool, where, For the source domain, For the target domain, It is the characteristic matrix, and it exists. The transferred knowledge is parameterized as W v The performance improvement rate for each transfer learning memory pool is [percentage missing]. The learning performance on the test dataset after transferring Wv from the source domain is: The learning performance on the test dataset after not transmitting Wv from the source domain is:

[0029] In one or more embodiments of the present invention, when a pair of new domains When this occurs, the mobile network proxy will optimize the data to be transmitted. Knowledge, among which, As input, the agent learning reflection function of the mobile network intelligent system is f(W) v ,S v ,T v ), reflection function f(W v ,S v ,T v ) as an approximation l v Used to guide how to transmit and improve performance improvement rates.

[0030] In one or more embodiments of the present invention, the reflection function f(W) is used. v ,S v ,T v From memory pool Approximate output l v , l v Updates are influenced by two factors: the difference between the target domain and the source domain, and the target...

[0031] The degree of domain recognition is based on which the difference between the target domain and the source domain is measured using the following formula:

[0032]

[0033] in, express The i-th sample in the model is Θ, which is the mapping from the latent space to the reproducing kernel Hilbert space. κ(.,.) is the kernel function, which affects the maximum mean dispersion distance, and thus the function f(W). v ,S v ,T v The value of ).

[0034] In one or more embodiments of the present invention, unlabeled samples are used to evaluate the recognition performance of the target domain. In traffic prediction of mobile networks, unlabeled discrimination criteria are considered:

[0035]

[0036] Where: τ υ Metrics used to measure the recognition performance of ensemble learning models in the target domain.

[0037]

[0038] It is a local scatter covariance matrix, and

[0039]

[0040] in, express and The r-nearest neighbor. By maximizing τ v , The first principle was guaranteed, and

[0041]

[0042] Compared with existing technologies, the present invention proposes a cross-domain transfer spatiotemporal integrated network method, which proposes a new transfer learning algorithm. This algorithm trains a mobile traffic prediction model from data-rich source domain data and then transfers the model to a mobile edge network environment with sparse data to improve the prediction accuracy of the model. At the same time, it proposes a new transfer mapping framework, which learns and optimizes the reflection function from empirical data to predict when or how to transfer to obtain the best prediction performance. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0044] Figure 1 Diagram of an ensemble learning architecture for predicting mobile wireless traffic in mobile edge networks;

[0045] Figure 2 A flowchart for spectrum clustering in mobile networks.

[0046] Figure 3 This is a performance comparison chart of the proposed algorithm and the baseline algorithm on the MAE metric.

[0047] Figure 4 This is a performance comparison chart of the proposed algorithm and the baseline algorithm on the RMSE metric.

[0048] Figure 5 This is a performance comparison chart of the proposed algorithm and the baseline algorithm on the R2 metric. Detailed Implementation

[0049] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this invention.

[0050] like Figures 1 to 2 As shown, a cross-domain transfer spatiotemporal integration network method according to an embodiment of the present invention includes the following steps:

[0051] S1. Construct an ensemble learning model, use ensemble learning to initially train basic learners, and combine it with a spatiotemporal network to capture spatiotemporal features to overcome the instability and overfitting in multidimensional spatiotemporal cross-domain environments. The ensemble learning model includes traffic dataset structure, spectral clustering, long short-term memory (LSTM), and convolutional neural network (CNN).

[0052] S2. To address the diversity and similarity between the target domain and the source domain, data preprocessing is performed on different cross-domain data. By summarizing historical transfer experience, an experience-based mechanism is established, and transfer reflection functions between the source and target domains are designed. The feature differences between different target and source domains are learned and parameterized to optimize knowledge transfer and improve learning performance in potential spatial cross-domain environments.

[0053] In S1, for spatiotemporal relationships, the spatial location of a mobile device is defined as X×B, and there exist n mobile networks with different degrees of data sparsity, i.e., v∈{1,...,n}. Given u base stations, the spatiotemporal sequence of wireless traffic for each network data point can be denoted as... Where T is the total time interval. Let be the traffic flow matrix at point t, defined as:

[0054]

[0055] in, It is used to measure the traffic capacity of mobile networks that is related to time and space, i.e.

[0056] Since the spatiotemporal domain analysis applies to all similar mobile network traffic flows, for simplicity, the data type of the traffic flow is omitted. At the same time, due to The value at t+1 depends only on the traffic of neighboring mobile networks and is related to distant mobile networks. Therefore, the autocorrelation coefficient in the spatiotemporal domain can be defined as:

[0057]

[0058] in, It is the average value that changes over time;

[0059] To address the issue of uneven distribution of mobile network traffic, the Pearson correlation coefficient is used to measure spatial correlation, i.e.:

[0060]

[0061] Where C(.) is the covariance operator, s is the standard deviation, the spatiotemporal data is regarded as a spatial point on a time series, a spectral clustering strategy is used for clustering, and then d is used ij Expressing the Euclidean distance between two points, we obtain:

[0062]

[0063] Where, ω ij It is data point x i x j The weights between them.

[0064] By calculating the weights between data points, a graph consisting of all datasets is constructed. Then, the subgraph obtained by connecting the edges of the sum of the weights of the graph is made as small as possible, and the graph with the sum of the weights is made as large as possible. The measure of the graph is:

[0065]

[0066] According to KNN, the adjacency matrix M is obtained as follows:

[0067]

[0068] The flowchart of spectrum clustering for mobile networks is as follows: Figure 2 As shown, based on the spatiotemporal characteristics of mobile network traffic flow, three data input types are defined: the first type is input data as {D}. t-1 ,…,D t-m The first type is data stream sequence before the target time interval, m∈N}; the second type is time-related data, m data, such as 5 hours of Friday, February 10; the third type is cross-domain dataset.

[0069] Based on three data input types, a deep network is built. The deep network includes building a spatiotemporal model, embedding m-data, and building cross-domain data, where m-data is temporal metadata.

[0070] Among them, data D t The input format is similar to video stream data, with M data frames. To process this type of input, a convolutional neural network (CNN) is introduced. While CNNs have strong spatial data modeling capabilities and can effectively extract features from local regions, they cannot capture temporal features. To capture temporal features, we fuse LSTM and CNN, designing a multi-layer CNN-LSTM spatiotemporal network.

[0071] In addition, to preserve historical information, each unit in the CNN-LSTM network was designed to have its own storage unit. Use input gates Forgot the door and output gate The structure is accessed and modified. When new traffic flow data is input into the CNN-LSTM spatiotemporal network, if... Once activated, the data it carries will be stored. In the middle. If Once activated, the attribute information of that unit can be forgotten or discarded. Finally, Output the hidden state h based on the cell state. t The main computational methods of CNN-LSTM spatiotemporal networks are as follows:

[0072]

[0073] Where B(·) is the layer weight, b(·) is the bias, * is the convolution operation, σ is the activation function, and ⊙ is the Hadamard product. and h t In the spatiotemporal domain of CNN-LSTM, the network is a three-dimensional tensor. Let M represent the number of feature maps, and the output is...

[0074] In the time-related data, for example, the t-th time point is 15:00:00 on December 10, 2021. We extracted four types of metadata: hour (15), day (12), month (10), and year (2021), and generated a feature vector m. The feature vector m was input into a CNN-LSTM network, and its dimension changed from 4 to M×X×B. Let fom represent the output of m-data embedding:

[0075]

[0076] in, and These are the weights and biases of layer I. After repeating the operation, the new output is as follows:

[0077]

[0078] To model the impact of different domains on traffic flow and learn the data characteristics of cross-domain datasets, a two-layer CNN network can be used. In this network model, datasets from different domains are concatenated to generate a tensor Θ. c Then, after Θ c Perform a nonlinear transformation operation to obtain Θ c Initial characteristics:

[0079]

[0080] Among them, B c⊕ represents the concatenation operation. The function f(·) executes in the order of BN, ReLU, and convolution. Using a density mesh, the three outputs are fused through concatenation and written as...

[0081]

[0082] Among them, f o This is the initial feature map. (The last part, "f") o As input to density, the network architecture has I layers, each containing a function f(·), where the execution order of f(·) is Batch Normalization (BN), ReLU, and convolution. To capture the spatiotemporal features of traffic flow, a connectivity-rejecting pattern structure can be designed. Therefore, layer I accepts feature maps from all previous network layers.

[0083]

[0084] in, And the final output is The final prediction is

[0085]

[0086] Finally, the objective function is obtained:

[0087]

[0088] Among them, W θ These are the parameters of the neural network.

[0089] In S2, let E v =((S) v ,T v ),l v ) represents the transfer learning memory pool, where, For the source domain, For the target domain, It is the characteristic matrix.

[0090] Due to the sparsity of the data, there exists The transferred knowledge can be parameterized as W v , This represents the performance improvement rate for each transfer learning memory pool. This represents the learning performance on the test dataset after transferring Wv from the source domain. This represents the learning performance on the test dataset after not transmitting Wv from the source domain.

[0091] When a new pair of domains When this occurs, the mobile network proxy will optimize the data to be transmitted. Knowledge, among which, As input, the agent learning reflection function of the mobile network intelligent system is f(W) v ,S v ,T v ), reflection function f(W v ,S v ,T v ) as an approximation l v Used to guide how to transmit and improve performance improvement rates.

[0092] Using the reflection function f(W) v ,S v ,T v From memory pool Approximate output l v , l v The update is influenced by two factors: the difference between the target domain and the source domain, and the degree of identification of the target domain. Based on this, the difference between the target domain and the source domain is measured by the following formula:

[0093]

[0094] in, express The i-th sample in the model is Θ, which is the mapping from the latent space to the reproducing kernel Hilbert space. k(.,.) is the kernel function, which affects the maximum mean dispersion distance, and thus the function f(W). v ,S v ,T v The value of ). Therefore, the function f(W) v ,S v ,T v The learning process of ) is essentially to optimize the kernel function κ(.,.), therefore, the maximum average dispersion can characterize the lv of all target and source domains. Based on this, κ(.,.) is parameterized as

[0095]

[0096] Where, N κ Let δ be the number of PSD kernels and δ be the learning coefficient. Then we have:

[0097]

[0098]

[0099] in, and From the k-th kernel κ κ Calculation. The kernel function we use is as follows:

[0100] κ κ (x,y)=exp(-||xy||2 / ω k )

[0101] Where ω k This is the varying bandwidth. Since the maximum mean dispersion is difficult to measure the differences between different domains, we need to measure the variance between cross-domain data pairs. The difference between the target domain and the source domain can be written as...

[0102]

[0103] To ensure consistency with the maximum average dispersion, rewrite in

[0104]

[0105] Furthermore, unlabeled samples are used to evaluate the recognition performance of the target domain. In traffic prediction of mobile networks, the unlabeled discrimination criterion is considered:

[0106]

[0107] Where: τ υ Metrics used to measure the recognition performance of ensemble learning models in the target domain.

[0108]

[0109] It is a local scatter covariance matrix, and

[0110]

[0111] in, express and The r-nearest neighbor. By maximizing τ v , This ensured the first principle was upheld, and

[0112]

[0113] Implement the second principle. Because of τ v Depends on the kernel function, according to τ v From τ v,k Obtained κ k , τ v Then it can be written as:

[0114]

[0115] Where, τ v =[τ e,1 ,…,τ v (N k Based on the above analysis, we can derive the following optimization problem.

[0116]

[0117] in,

[0118]

[0119] Based on the above formula, it can be seen that the loss function consists of five parts. Let be the Huber regression loss, b be the bias, L1 be the training error in the target domain, and G be the error in the target domain. t It is a feature extractor. L2 is used to constrain the feature distance between the target and source domains. L3 is used to ensure the discriminability between the target and source domains. L4 is used to prevent overloading. L0 can be optimized by adjusting the reflection function f(W). v ,S v ,T v ,δ * ,λ * ,μ * ,b * The calculation uses the latest eigenfactor matrix W to optimize the transition method, i.e., W Nv+1 f(W) should be made v ,S v ,T v ,δ * ,λ * ,μ * ,b * Maximize the value of ). Therefore, we have

[0120]

[0121] In the formula, φ2 is the complexity factor of W. Let W be the Frobinius norm of the matrix. The first and second terms of the equation are obtained from the following:

[0122]

[0123] Finally, the problem was optimized using the conjugate gradient method. Stochastic gradient descent (SGD) was used to solve L1-L4:

[0124]

[0125] have

[0126]

[0127] Among them, C 180 ⊙ is the inverted convolution kernel by 180°, and ⊙ is the element-wise product.

[0128]

[0129] have

[0130]

[0131] Experimental example:

[0132] Two real-world datasets were used, containing mobile network data related to telecommunications, weather, news, social networks, and electricity in Milan and the province of Trentino. Due to the datasets originating from different companies using different standards, their spatial distribution irregularities clustered within a grid containing 1000 city units. In optimizing traffic flow, data from multiple time periods and the previous week were used to estimate missing current traffic parameters. The simulation was performed using 8×NVIDIA TESLAV10032G GPUs on Linux. The hyperparameters used in the experiment are shown in the table below:

[0133]

[0134]

[0135] The root mean square error (RMSE), mean absolute error (MAE), and R-squared error (R²) were used as evaluation metrics to verify the convergence performance of the model.

[0136] To evaluate the superiority and accuracy of the algorithms, the following traffic prediction algorithms were selected as baselines: LSTM, Support Vector Regression (SVR), Temporal Graph Convolutional Network (T-GCN), Spatiotemporal Graph Convolutional Network (ST-GCN), and DenseNet.

[0137] This application employs an ensemble learning framework, using a clustering algorithm to classify mobile edge networks into different categories based on feature similarity between different mobile target domains, and then uses a CNN-LSTM hybrid model to fit the actual wireless traffic of the mobile edge networks. To address the varying data sparsity across different mobile network regions, a transfer learning algorithm based on experience memory pools and reflection functions is proposed to achieve cross-domain traffic prediction. Finally, we evaluate the predictive performance of the proposed TSTEN using two real-world datasets from Milan and Trentino. Figure 3 , Figure 4 and Figure 5 Analysis of three metrics—MAE, RMSE, and R2—shows that TSTEN outperforms existing baselines in cross-domain traffic prediction accuracy.

[0138] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0139] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A cross-domain transfer spatio-temporal integration network method, characterized in that, The method comprises the following steps: S1, constructing an ensemble learning model, using ensemble learning to initially train the base learner, and combining a space-time network to capture space-time features, to overcome the instability and overfitting shortcomings in the multi-dimensional space-time cross-domain environment, wherein the ensemble learning model comprises a traffic data set structure, spectral clustering, long short-term memory (LSTM) and a convolutional neural network (CNN); S2, for the diversity and similarity of the target domain and the source domain, data preprocessing is performed on different cross-domain data, an experience-based mechanism is established by summarizing historical transfer experience, a transfer reflection function of the source domain, the target domain and the target domain is designed, feature differences between different target domains and source domains are learned and parameterized, knowledge transfer is optimized, and learning performance in a potential space cross-domain environment is improved; In the S2, set E v = ((S v , T v ), l v ) represents a transfer learning memory pool, wherein, is a source domain, is a target domain, is a feature matrix, and there is The knowledge transferred is parameterized as W v The performance improvement rate of each transfer learning memory pool is The learning performance on the test data set after transferring Wv from the source domain is The learning performance on the test data set after not transferring Wv from the source domain is When a new pair of domains When this occurs, the mobile network proxy will optimize the data to be transmitted. Knowledge, among which, As input, the agent learning reflection function of the mobile network intelligent system is f(W) v ,S v ,T v ), reflection function f(W v ,S v ,T v ) as an approximation l v Used to guide how to transmit and improve performance improvement rates, using the reflection function f(W) v ,S v ,T v From memory pool Approximate output l v , l v The update is influenced by two factors: the difference between the target domain and the source domain, and the degree of recognition of the target domain. The difference between the target domain and the source domain is measured by the following formula: wherein, represents the i-th sample in, Θ is the mapping from the latent space to the reproducing kernel Hilbert space, κ(.,.) is the kernel function, the kernel function influences the maximum mean discrepancy distance and thus the value of the function f(W v , S v , T v ).

2. The method of claim 1, wherein, In the S1, for the space-time relationship, the spatial position of the mobile device is defined as XxB, and there are n mobile networks with different degrees of data sparsity, i.e. v∈{1,...,n}, given u base stations, the wireless traffic space-time sequence of each network data point can be recorded as θ v ={θ v,u,t |t=1,...,T},where T is the total time interval, θ v,u,t is the traffic flow matrix at t, which is defined as: wherein, is for measuring the traffic capacity of a mobile network related to time and space, 3. The method of claim 2, wherein, The autocorrelation coefficient of the space-time domain is defined as: wherein is the average value over time; The spatial correlation is measured by the Pearson correlation coefficient, that is: where C(.) is the covariance operator, s is the standard deviation, the spatio-temporal data is considered as a spatial point on a time series, spectral clustering strategy is used for clustering, and d ij represents the Euclidean distance between two points, where ω ij is the weight between data points x i , x j .

4. The method of claim 3, wherein, By calculating the weight between data points, a graph composed of all data sets is constructed, and then the sum of the weights of the subgraph obtained by connecting the edges of the graph is as small as possible and the sum of the weights of the graph is as large as possible, and the measure of the graph is: According to KNN, the adjacency matrix M is obtained as:

5. The method of claim 4, wherein, According to the space-time characteristics of mobile network traffic flow, three kinds of data input types are defined: the first is the input data {D t-1 ,…,D t-m ,m∈N}, representing the data stream sequence before the target time interval; the second is the time-related data, m data; and the third is the cross-domain data set.

6. The method of claim 5, wherein, Based on the three types of data input, a deep network is established, the deep network comprises constructing a space-time model, m data embedding and constructing cross-domain data, wherein the m data is time metadata.

7. The method of claim 5, wherein, Unlabeled samples are used to evaluate the recognition performance of the target domain, and in the traffic prediction of the mobile network, the unlabeled discriminant criterion is considered: where: τ υ an index for measuring the recognition performance of the integrated learning model on the target domain, is the local scatter covariance matrix, and wherein denotes with r-nearest neighbor of

Images