Traffic prediction method, device and server

Abstract

The application provides a traffic prediction method and device and a server, relates to network information engineering technology, and comprises the following steps: obtaining a data acquisition instruction and acquiring a traffic data sequence according to the data acquisition instruction. A preset convolutional neural network model is used for feature extraction processing of the traffic data sequence to obtain a first feature vector; a preset long short-term memory network model is used for feature extraction processing of the traffic data sequence to obtain a second feature vector. The first feature vector and the second feature vector are fused to obtain a fusion vector. Regression prediction is performed on the traffic data sequence to obtain a regression residual. A preset radial basis network model is used for traffic prediction on the fusion vector and the regression residual to determine target traffic information at a target moment, so that traffic prediction is completed. The method solves the technical problem that it is difficult to determine traffic data when traffic bursts and traffic imbalance occur.

Description

Technical Field

[0001] This application relates to network information engineering technology, and in particular to a traffic prediction method, apparatus and server. Background Technology

[0002] Currently, with the booming development of the 5G industry and the gradual implementation of 5G commercialization, user demand and business scale are expanding rapidly. The diversified development of 5G cannot be separated from the stable support of network resources.

[0003] In existing technologies, to provide users with stable network resources, network slices are typically determined in real-time based on the user's current traffic data, and network resources are then provided to the user according to these slices. Because current traffic data changes rapidly, this necessitates frequent determination of the user's current traffic data and subsequent provision of network resources, leading to longer service latency for the user. There is a need to reduce the frequency of real-time determination of the user's current traffic data to lower service latency.

[0004] Therefore, there is an urgent need for a method that can accurately determine traffic data during traffic bursts and traffic imbalances. Summary of the Invention

[0005] This application provides a traffic prediction method, apparatus, and server to solve the technical problem of difficulty in determining traffic data during traffic bursts and traffic imbalances.

[0006] In a first aspect, this application provides a traffic prediction method, including:

[0007] Obtain a data acquisition instruction, and acquire a traffic data sequence according to the data acquisition instruction;

[0008] The traffic data sequence is processed by a pre-defined convolutional neural network model to extract features and obtain a first feature vector; and the traffic data sequence is processed by a pre-defined long short-term memory network model to extract features and obtain a second feature vector; wherein, both the convolutional neural network model and the long short-term memory network model are trained based on historical traffic sequences;

[0009] The first feature vector and the second feature vector are fused to obtain a fused vector.

[0010] Regression prediction is performed on the traffic data sequence to obtain the regression residuals;

[0011] Traffic flow prediction is performed on the fusion vector and the regression residual according to the preset radial basis function network model to determine the target traffic flow information at the target time, thereby completing the traffic flow prediction; wherein, the preset radial basis function network model is trained based on historical traffic flow sequences, and the radial basis function network model includes the target time for predicting the target traffic flow information.

[0012] Further, the step of predicting the flow rate of the fusion vector and the regression residual based on a preset radial basis function network model to determine the target flow rate information at the target time includes:

[0013] Based on the target time in the preset radial basis function network model, the flow prediction is performed on the fusion vector and the regression residual according to the radial basis function network model to determine the target flow information at the target time; wherein, the target time includes at least one time.

[0014] Further, regression prediction is performed on the traffic data sequence to obtain regression residuals, including:

[0015] Regression prediction is performed on the traffic data sequence to obtain preliminary prediction information, and the actual value corresponding to the preliminary prediction information is determined.

[0016] Based on the preliminary prediction information and the actual value, the regression residuals are obtained.

[0017] Furthermore, the method also includes:

[0018] Obtain multiple historical traffic sequences;

[0019] The historical traffic sequence is processed by feature extraction according to a preset convolutional neural network model to obtain a first historical vector; the historical traffic sequence is processed by feature extraction according to a preset long short-term memory network model to obtain a second historical vector.

[0020] The first historical vector and the second historical vector are fused to obtain a historical vector;

[0021] Regression prediction is performed on the historical traffic sequence to obtain historical prediction information, and historical residuals are obtained based on the historical prediction information and the historical values ​​corresponding to the historical prediction information.

[0022] Based on the historical residuals and the historical vectors, the radial basis function network model is trained for traffic prediction to determine the predicted traffic information at historical moments.

[0023] Based on the historical forecast information and the forecast traffic information, the forecast residual is determined;

[0024] If the predicted residual is determined to meet a threshold, then the radial basis function network model is determined to have converged.

[0025] Secondly, this application provides a flow prediction device, comprising:

[0026] The first acquisition unit is used to acquire data acquisition instructions and acquire traffic data sequences according to the data acquisition instructions;

[0027] The second acquisition unit is used to perform feature extraction processing on the traffic data sequence through a preset convolutional neural network model to obtain a first feature vector; and to perform feature extraction processing on the traffic data sequence through a preset long short-term memory network model to obtain a second feature vector; wherein, both the convolutional neural network model and the long short-term memory network model are trained based on historical traffic sequences;

[0028] The first fusion unit is used to fuse the first feature vector and the second feature vector to obtain a fusion vector;

[0029] The third acquisition unit is used to perform regression prediction on the traffic data sequence and obtain the regression residual;

[0030] The prediction unit is used to predict the flow rate of the fusion vector and the regression residual based on a preset radial basis function network model, and to determine the target flow rate information at the target time to complete the flow rate prediction; wherein, the preset radial basis function network model is trained based on historical flow rate sequences, and the radial basis function network model includes the target time for predicting the target flow rate information.

[0031] Furthermore, the prediction unit is specifically used for:

[0032] Based on the target time in the preset radial basis function network model, the flow prediction is performed on the fusion vector and the regression residual according to the radial basis function network model to determine the target flow information at the target time; wherein, the target time includes at least one time.

[0033] Furthermore, the third acquisition unit includes:

[0034] The determination module is used to perform regression prediction on the traffic data sequence, obtain preliminary prediction information, and determine the actual value corresponding to the preliminary prediction information;

[0035] The acquisition module is used to obtain the regression residuals based on the preliminary prediction information and the actual values.

[0036] Furthermore, the device also includes:

[0037] The fourth acquisition unit is used to acquire multiple historical traffic sequences;

[0038] The fifth acquisition unit is used to perform feature extraction processing on the historical traffic sequence according to a preset convolutional neural network model to obtain a first historical vector; and to perform feature extraction processing on the historical traffic sequence through a preset long short-term memory network model to obtain a second historical vector.

[0039] The sixth acquisition unit is used to perform fusion processing on the first historical vector and the second historical vector to obtain a historical vector;

[0040] The seventh acquisition unit is used to perform regression prediction on the historical traffic sequence, acquire historical prediction information, and acquire historical residuals based on the historical prediction information and the historical values ​​corresponding to the historical prediction information.

[0041] The first determining unit is used to train the radial basis network model for traffic prediction based on the historical residuals and the historical vectors, and to determine the predicted traffic information at historical moments.

[0042] The second determining unit is used to determine the prediction residual based on the historical prediction information and the predicted flow information;

[0043] The third determining unit is used to determine that the radial basis network model has converged if the predicted residual satisfies a threshold.

[0044] Thirdly, this application provides a server, including a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the computer program to implement the method described in the first aspect.

[0045] Fourthly, this application provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the method described in the first aspect.

[0046] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the method described in the first aspect.

[0047] This application provides a traffic prediction method, apparatus, and server that acquires data acquisition instructions and acquires traffic data sequences according to these instructions. The traffic data sequences are processed by feature extraction using a pre-defined convolutional neural network model to obtain a first feature vector; and by feature extraction using a pre-defined long short-term memory network model to obtain a second feature vector. Both the convolutional neural network model and the long short-term memory network model are trained based on historical traffic sequences. The first and second feature vectors are fused to obtain a fused vector. Regression prediction is performed on the traffic data sequences to obtain regression residuals. Traffic prediction is then performed using a pre-defined radial basis function (RBF) network model on the fused vector and regression residuals to determine the target traffic information at a target time, thus completing the traffic prediction. The pre-defined RBF network model is trained based on historical traffic sequences and includes a target time for predicting the target traffic information. In this scheme, traffic data sequences are acquired according to data acquisition instructions, and the traffic data sequences are processed by feature extraction using a pre-defined convolutional neural network model to obtain a first feature vector and by feature extraction using a pre-defined long short-term memory network model to obtain a second feature vector. Then, the first and second feature vectors are fused to obtain a fused vector. Regression prediction is then performed on the traffic data sequence to obtain the regression residuals. Finally, both the fused vector and the regression residuals are used as input parameters for a pre-defined radial basis function (RBF) model. Based on the RBF model, traffic prediction is performed using the fused vector and regression residuals to determine the target traffic information at the target time, thus completing the traffic prediction. Therefore, by leveraging convolutional neural network and long short-term memory (LSTM) models from deep learning to quickly extract feature vectors from traffic data sequences, and then fusing the RBF model to predict the traffic data sequence and determine the target traffic information at the target time, the accuracy of the target traffic information at the target time is greatly improved. This enables rapid traffic fitting and solves the technical problem of difficulty in determining traffic data during traffic bursts and imbalances. Attached Figure Description

[0048] The accompanying drawings, which are incorporated in and form a part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure.

[0049] Figure 1 A flowchart illustrating a traffic prediction method provided in an embodiment of this application;

[0050] Figure 2 A flowchart illustrating another traffic prediction method provided in an embodiment of this application;

[0051] Figure 3 A flowchart illustrating another traffic prediction method provided in an embodiment of this application;

[0052] Figure 4 A flowchart illustrating another traffic prediction method provided in an embodiment of this application;

[0053] Figure 5 A flowchart illustrating another traffic prediction method provided in an embodiment of this application;

[0054] Figure 6 This is a schematic diagram of the structure of a flow prediction device provided in an embodiment of this application;

[0055] Figure 7 This is a schematic diagram of another flow prediction device provided in an embodiment of this application;

[0056] Figure 8 This is a schematic diagram of the structure of a server provided in an embodiment of this application.

[0057] The accompanying drawings have illustrated specific embodiments of this disclosure, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concepts of this disclosure to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0058] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure.

[0059] In one example, to provide users with stable network resources, network slices are typically determined in real-time based on the user's current traffic data, and network resources are then provided to the user accordingly. Because current traffic data changes rapidly, frequent determination of the user's current traffic data is necessary to provide network resources, leading to longer service latency. There is a need to reduce latency by eliminating the need for frequent real-time determination of the user's current traffic data. Therefore, there is an urgent need for a method that can accurately determine traffic data during traffic bursts and traffic imbalances.

[0060] This application provides a traffic prediction method, apparatus, and server, which aim to solve the above-mentioned technical problems in the prior art.

[0061] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0062] Figure 1 This is a flowchart illustrating a traffic prediction method provided in an embodiment of this application, as shown below. Figure 1 As shown, the method includes:

[0063] Step 101: Obtain the data acquisition instruction and acquire the traffic data sequence according to the data acquisition instruction.

[0064] For example, the execution entity in this embodiment can be a server. First, a data acquisition instruction needs to be obtained. This can be achieved by receiving data acquisition instructions transmitted from other devices or other servers; alternatively, the user can click a data acquisition button on the server's corresponding display screen to trigger the generation of a data acquisition instruction, which the server then acquires. When the server acquires the data acquisition instruction, it can obtain a traffic data sequence based on the instruction. The traffic data sequence can be traffic data from multiple time points obtained from multiple base stations.

[0065] Step 102: Perform feature extraction processing on the traffic data sequence using a preset convolutional neural network model to obtain the first feature vector; and perform feature extraction processing on the traffic data sequence using a preset long short-term memory network model to obtain the second feature vector; wherein, both the convolutional neural network model and the long short-term memory network model are trained based on historical traffic sequences.

[0066] For example, since both the Convolutional Neural Network (CNN) model and the Long Short-Term Memory (LSTM) network model are trained based on historical traffic sequences, the server can perform feature extraction processing on the traffic data sequence using the preset CNN model to obtain a first feature vector, and perform feature extraction processing on the traffic data sequence using the preset LSTM network model to obtain a second feature vector.

[0067] For example, the traffic data sequence is a matrix, with the number of rows being the number of time points t, n, and the number of columns being the number of base stations. The intersection of the row where t=1 is located and the column where the first base station is located represents the traffic data of the first base station at t=1. Similarly, the intersection of the row where t=n is located and the column where the first base station is located represents the traffic data of the first base station at t=n. The traffic data at each intersection includes 20 indicators such as user experience rate, connection density, end-to-end latency, mobility, traffic density, user peak rate, and energy efficiency.

[0068] The traffic data sequence is input into a convolutional neural network model and a long short-term memory network model, respectively. The last layer of the convolutional neural network model outputs a 1*n-dimensional first feature vector C.r :

[0069]

[0070] The Long Short-Term Memory (LSTM) network model outputs an n*m dimensional second eigenvector H. r :

[0071] Wherein, the first feature vector C r For spatial features, the second feature vector H r For time features, n is the sequence length, m is the output dimension of the LSTM hidden layer, and r is the label of the extracted feature value, with no other meaning. and These are all eigenvalues ​​in the eigenvector.

[0072] Step 103: Perform a fusion process on the first feature vector and the second feature vector to obtain a fused vector.

[0073] For example, Figure 2 A flowchart illustrating another traffic prediction method provided in this application embodiment is shown below. Figure 2 As shown, the server performs a fusion process on the first and second feature vectors to obtain a fused vector 1*m. The fusion process can be implemented using the following formula:

[0074]

[0075]

[0076]

[0077] Among them, W a b is an m*n dimensional weight matrix; a These are bias terms, all learned during training; It is C r The transpose of the matrix, The 'c' in the equation has no practical meaning. Let α be the hidden layer output vector at different time points. i The weights are those corresponding to the hidden layer output vectors. The first feature vector of the CNN and the second feature vector of the LSTM are fused into a set of weights using formula (1), and the weights are normalized to 0-1 using the softmax activation function in formula (2). Finally, the hidden layer output vectors of the LSTM at different time points are fused using formula (3). With the corresponding weight α i Multiply and sum them to obtain the final feature representation of the traffic data sequence, i.e., f. map .

[0078] Step 104: Perform regression prediction on the traffic data sequence and obtain the regression residuals.

[0079] For example, the server can perform regression prediction on the traffic data sequence using a preset regression model to obtain a 1*n-dimensional preliminary prediction information and determine the actual value corresponding to the preliminary prediction information. The regression residual can be obtained by subtracting the preliminary prediction information from the actual value.

[0080] Step 105: Based on the preset radial basis function network model, perform flow prediction on the fusion vector and regression residual to determine the target flow information at the target time, thereby completing the flow prediction; wherein, the preset radial basis function network model is trained based on historical flow sequences, and the radial basis function network model includes the target time used to predict the target flow information.

[0081] For example, the preset radial basis function (RBF) network model is trained based on historical traffic sequences. The RBF network model includes a target time for predicting target traffic information. Therefore, the server combines the regression residuals obtained from the regression prediction with a 1*m dimensional fusion vector extracted through dual channels of CNN and LSTM as input to the RBF network model. This allows the output of target traffic information at the target time, thereby completing traffic prediction and facilitating the determination of network slicing strategies based on the target traffic information.

[0082] In this embodiment, a data acquisition instruction is obtained, and a traffic data sequence is acquired according to the instruction. A first feature vector is obtained by performing feature extraction on the traffic data sequence using a preset convolutional neural network model; a second feature vector is obtained by performing feature extraction on the traffic data sequence using a preset long short-term memory network model; both the convolutional neural network model and the long short-term memory network model are trained based on historical traffic sequences. The first and second feature vectors are fused to obtain a fused vector. Regression prediction is performed on the traffic data sequence to obtain regression residuals. Traffic prediction is performed on the fused vector and regression residuals according to a preset radial basis function network model to determine the target traffic information at a target time, thus completing the traffic prediction; the preset radial basis function network model is trained based on historical traffic sequences, and includes a target time for predicting the target traffic information. In this scheme, a traffic data sequence is acquired according to the data acquisition instruction, a first feature vector is obtained by performing feature extraction on the traffic data sequence using a preset convolutional neural network model, and a second feature vector is obtained by performing feature extraction on the traffic data sequence using a preset long short-term memory network model. Then, the first and second feature vectors are fused to obtain a fused vector. Regression prediction is performed on the traffic data sequence to obtain regression residuals. Finally, both the fused vector and the regression residuals are used as input parameters for a pre-defined radial basis function (RBF) model. Based on the RBF model, traffic prediction is performed on the fused vector and regression residuals to determine the target traffic information at the target time, thus completing the traffic prediction. Therefore, by leveraging convolutional neural network and long short-term memory (LSTM) models from deep learning to quickly extract feature vectors from traffic data sequences, and then fusing the RBF model to predict the traffic data sequence and determine the target traffic information at the target time, the accuracy of the target traffic information at the target time is greatly improved. This enables rapid traffic fitting and solves the technical problem of difficulty in determining traffic data during traffic bursts and imbalances.

[0083] Figure 3 A flowchart illustrating another traffic prediction method provided in this application embodiment is shown below. Figure 3 As shown, the method includes:

[0084] Step 201: Obtain multiple historical traffic sequences.

[0085] For example, the server can obtain multiple historical traffic sequences.

[0086] Step 202: Perform feature extraction processing on the historical traffic sequence according to the preset convolutional neural network model to obtain the first historical vector; and perform feature extraction processing on the historical traffic sequence through the preset long short-term memory network model to obtain the second historical vector.

[0087] For example, the server can perform feature extraction processing on the historical traffic sequence according to a preset convolutional neural network model to obtain a first historical vector, and perform feature extraction processing on the historical traffic sequence according to a preset long short-term memory network model to obtain a second historical vector, wherein the first historical vector is a spatial feature and the second historical vector is a temporal feature.

[0088] Step 203: Merge the first historical vector and the second historical vector to obtain the historical vector.

[0089] For example, this step can be referred to Figure 1 The fusion process of the first and second feature vectors in step 103 will not be described again.

[0090] Step 204: Perform regression prediction on the historical flow sequence to obtain historical prediction information, and obtain historical residuals based on the historical prediction information and the historical values ​​corresponding to the historical prediction information.

[0091] For example, the server can perform regression prediction on historical traffic sequences using a preset regression model to obtain historical prediction information, and then subtract the historical prediction information from the historical values ​​corresponding to the historical prediction information to obtain historical residuals.

[0092] Step 205: Based on historical residuals and historical vectors, train the radial basis function network model for traffic prediction to determine the predicted traffic information at historical moments.

[0093] For example, the server can combine the historical residuals obtained from regression prediction with the 1*m-dimensional historical vectors extracted through dual channels of CNN and LSTM, and use them as input to the radial basis function network model to determine the predicted flow information at historical moments through the training of the radial basis function network model.

[0094] Step 206: Determine the prediction residuals based on historical forecast information and predicted flow information.

[0095] For example, the server can subtract historical prediction information from predicted traffic information to obtain prediction residuals, and use these prediction residuals as corrections to the predicted traffic information.

[0096] Step 207: If the predicted residuals are determined to meet the threshold, then the radial basis function network model is determined to be converged.

[0097] For example, if the predicted residuals are determined to meet the threshold, the server determines that the radial basis function network model has converged; if the predicted residuals are determined not to meet the threshold, the server continues to train the radial basis function network model until the predicted residuals meet the threshold.

[0098] For example, let the historical vector be y and the historical residual be ε. iThe historical prediction information obtained from regression is y 回 The predicted residual is δ i The predicted traffic information is y 测 The regression analysis uses linear regression, and the regression process can be implemented using the following formula:

[0099] y 同 (i)=ky(i)+b-----(4)

[0100] ε i =y 同 (i)-y(i)-----(5)

[0101] y 测 (i)=y 同 (i)+δ i -----(6)

[0102] For example, Figure 4 A flowchart illustrating another traffic prediction method provided in this application embodiment is shown below. Figure 4 As shown, RBF is a three-layer neural network consisting of an input layer, hidden layers, and an output layer. The transformation from the input space to the hidden layer space is non-linear, while the transformation from the hidden layer space to the output layer space is linear. Compared to BP neural networks, RBF neural networks converge faster and are more versatile. The number of hidden layer units, i.e., the network structure, can be adaptively adjusted during the training phase according to the specific research problem. RBF is a highly efficient feedforward neural network with the characteristic of unique best approximation, effectively avoiding the problem of local optima.

[0103] After extracting historical vectors and performing regression prediction analysis on the original traffic data sequence, the trained traffic features and historical residuals can be used as inputs to the RBF to obtain predicted traffic information.

[0104] like Figure 4 It can be seen that the input layer has a dimension of 2n, the output layer has a dimension of n, and the hidden layer has a dimension h that is greater than or equal to m and less than or equal to 2n.

[0105] The activation function of a radial basis function neural network model can be expressed as:

[0106]

[0107] Where, x p It consists of the nth feature and the 1st to nth residual terms: 1≤p≤2n

[0108] c j Let h be the j-th center point, and h be the number of nodes in the hidden layer: 1 ≤ j ≤ h

[0109] m is the output prediction residual value;

[0110] i is the number of output nodes, 1≤i≤m

[0111] The output obtained from the radial basis function neural network model is:

[0112]

[0113] Finally, based on the above formula (6), the predicted flow information for historical moments can be calculated, that is, based on the predicted residual δ i As a correction to the predicted traffic flow information, the RBF fusion regression model is used to form a residual regression neural network to correct the error, which can effectively predict the residual and achieve traffic flow prediction.

[0114] Figure 5 A flowchart illustrating another traffic prediction method provided in this application embodiment is shown below. Figure 5 As shown, Figure 5 This refers to the training process of the radial basis function network model in steps 201-207 above.

[0115] Step 208: Obtain the data acquisition instruction and acquire the traffic data sequence according to the data acquisition instruction.

[0116] For example, this step can be referred to Figure 1 Step 101 in the text will not be repeated here.

[0117] Step 209: Perform feature extraction processing on the traffic data sequence using a preset convolutional neural network model to obtain a first feature vector; perform feature extraction processing on the traffic data sequence using a preset long short-term memory network model to obtain a second feature vector; wherein, both the convolutional neural network model and the long short-term memory network model are trained based on historical traffic sequences.

[0118] For example, this step can be referred to Figure 1 Step 102 in the text will not be repeated here.

[0119] Step 210: Perform a fusion process on the first feature vector and the second feature vector to obtain a fused vector.

[0120] For example, this step can be referred to Figure 1 Step 103 in the text will not be repeated here.

[0121] Step 211: Perform regression prediction on the traffic data sequence to obtain preliminary prediction information and determine the actual value corresponding to the preliminary prediction information.

[0122] For example, the server can perform regression prediction on the traffic data sequence, obtain preliminary prediction information, and determine the actual value corresponding to the preliminary prediction information.

[0123] Step 212: Obtain the regression residuals based on the preliminary forecast information and the actual values.

[0124] For example, the server subtracts the initial prediction information from the actual value to obtain the regression residual.

[0125] Step 213: Based on the target time in the preset radial basis network model, perform flow prediction on the fusion vector and regression residual according to the radial basis network model to determine the target flow information at the target time, so as to complete the flow prediction; wherein, the target time includes at least one time.

[0126] For example, firstly, the server determines the number of target times in the preset radial basis function network (RBF) model, and inputs the fused vector and regression residuals into the RBF model. Traffic prediction is then performed through the RBF model, outputting the target traffic information for the target times, thus completing the traffic prediction. In this embodiment, multiple historical traffic sequences are acquired. Feature extraction processing is performed on the historical traffic sequences using a preset convolutional neural network model to obtain a first historical vector; feature extraction processing is also performed on the historical traffic sequences using a preset long short-term memory (LSTM) network model to obtain a second historical vector. The first and second historical vectors are fused to obtain a historical vector. Regression prediction is performed on the historical traffic sequences to obtain historical prediction information, and historical residuals are obtained based on the historical prediction information and the corresponding historical values. Based on the historical residuals and historical vectors, the RBF model is trained for traffic prediction to determine the predicted traffic information for historical times. Based on the historical prediction information and the predicted traffic information, the prediction residuals are determined. If the prediction residuals meet a threshold, the RBF model is considered converged. A data acquisition instruction is obtained, and a traffic data sequence is acquired according to the data acquisition instruction. The traffic data sequence is processed by a pre-defined convolutional neural network model to extract features, obtaining a first feature vector; and a pre-defined long short-term memory network model is used to extract features, obtaining a second feature vector. Both the convolutional neural network model and the long short-term memory network model are trained based on historical traffic sequences. The first and second feature vectors are fused to obtain a fused vector. Regression prediction is performed on the traffic data sequence to obtain preliminary prediction information, and the actual values ​​corresponding to the preliminary prediction information are determined. Based on the preliminary prediction information and the actual values, regression residuals are obtained. Based on the target time in the pre-defined radial basis function network model, traffic prediction is performed on the fused vector and regression residuals according to the radial basis function network model to determine the target traffic information at the target time, thus completing the traffic prediction; the target time includes at least one time. Therefore, by leveraging convolutional neural network and long short-term memory (LSTM) models from deep learning to quickly extract feature vectors from traffic data sequences, a radial basis function (RBF) model is integrated to predict the traffic data sequences and determine the target traffic information at the target time. This significantly improves the accuracy of the target traffic information at the target time, enabling rapid traffic fitting and solving the technical problem of difficulty in determining traffic data during traffic bursts and imbalances. Furthermore, traffic error prediction is achieved through RBF, a reverse learning method. RBF has a simple structure, strong nonlinear approximation capability, and can quickly fuse features and accurately predict traffic errors. It possesses global optimality and best approximation performance not found in other neural networks. Therefore, integrating RBF with a regression model to form a residual regression neural network to correct errors can effectively predict residuals and achieve traffic prediction.

[0127] Figure 6 This is a schematic diagram of the structure of a flow prediction device provided in an embodiment of this application, as shown below. Figure 6 As shown, the device includes:

[0128] The first acquisition unit 31 is used to acquire data instructions and acquire traffic data sequences according to the data acquisition instructions.

[0129] The second acquisition unit 32 is used to perform feature extraction processing on the traffic data sequence through a preset convolutional neural network model to obtain a first feature vector; and to perform feature extraction processing on the traffic data sequence through a preset long short-term memory network model to obtain a second feature vector; wherein, both the convolutional neural network model and the long short-term memory network model are trained based on historical traffic sequences.

[0130] The first fusion unit 33 is used to fuse the first feature vector and the second feature vector to obtain a fusion vector.

[0131] The third acquisition unit 34 is used to perform regression prediction on the traffic data sequence and obtain the regression residual.

[0132] The prediction unit 35 is used to predict the flow rate of the fusion vector and regression residual based on the preset radial basis function network model, and to determine the target flow rate information at the target time to complete the flow rate prediction. The preset radial basis function network model is trained based on the historical flow rate sequence and includes the target time for predicting the target flow rate information.

[0133] The apparatus in this embodiment can execute the technical solutions in the above method. Its specific implementation process and technical principles are the same, and will not be repeated here.

[0134] Figure 7 This is a schematic diagram of another flow prediction device provided in an embodiment of this application. Figure 6 Based on the illustrated embodiments, as Figure 7 As shown, prediction unit 35 is specifically used for:

[0135] Based on the target time in the preset radial basis function network model, the flow prediction is performed on the fusion vector and regression residual according to the radial basis function network model to determine the target flow information at the target time; wherein, the target time includes at least one time.

[0136] In one example, the third acquisition unit 34 includes:

[0137] The determination module 341 is used to perform regression prediction on the traffic data sequence, obtain preliminary prediction information, and determine the actual value corresponding to the preliminary prediction information.

[0138] The acquisition module 342 is used to obtain the regression residuals based on the preliminary prediction information and the actual values.

[0139] In one example, the device also includes:

[0140] The fourth acquisition unit 41 is used to acquire multiple historical traffic sequences.

[0141] The fifth acquisition unit 42 is used to perform feature extraction processing on the historical traffic sequence according to the preset convolutional neural network model to obtain the first historical vector; and to perform feature extraction processing on the historical traffic sequence through the preset long short-term memory network model to obtain the second historical vector.

[0142] The sixth acquisition unit 43 is used to fuse the first historical vector and the second historical vector to obtain the historical vector.

[0143] The seventh acquisition unit 44 is used to perform regression prediction on the historical flow sequence, acquire historical prediction information, and acquire historical residuals based on the historical prediction information and the historical values ​​corresponding to the historical prediction information.

[0144] The first determining unit 45 is used to train the radial basis network model for traffic prediction based on historical residuals and historical vectors, and to determine the predicted traffic information at historical moments.

[0145] The second determining unit 46 is used to determine the prediction residual based on historical prediction information and prediction flow information.

[0146] The third determining unit 47 is used to determine the convergence of the radial basis network model if the predicted residuals satisfy the threshold.

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

[0148] Figure 8 This application provides a schematic diagram of the structure of a server, as shown in the embodiment of the present application. Figure 8 As shown, the server includes: a memory 51 and a processor 52.

[0149] The memory 51 stores a computer program that can run on the processor 52.

[0150] Processor 52 is configured to perform the methods provided in the embodiments described above.

[0151] The server also includes a receiver 53 and a transmitter 54. The receiver 53 is used to receive instructions and data sent by external devices, and the transmitter 54 is used to send instructions and data to external devices.

[0152] This application also provides a non-transitory computer-readable storage medium, which, when the instructions in the storage medium are executed by the processor of a server, enables the server to perform the methods provided in the above embodiments.

[0153] This application also provides a computer program product, which includes: a computer program stored in a readable storage medium, at least one processor of the server can read the computer program from the readable storage medium, and the at least one processor executes the computer program to cause the server to perform the solution provided in any of the above embodiments.

[0154] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0155] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A flow prediction method, characterized in that, include: Obtain a data acquisition instruction, and acquire a traffic data sequence according to the data acquisition instruction; The traffic data sequence is processed by a pre-defined convolutional neural network model to extract features and obtain a first feature vector; and the traffic data sequence is processed by a pre-defined long short-term memory network model to extract features and obtain a second feature vector; wherein, both the convolutional neural network model and the long short-term memory network model are trained based on historical traffic sequences; The first feature vector and the second feature vector are fused to obtain a fused vector. Regression prediction is performed on the traffic data sequence to obtain the regression residuals; Traffic flow prediction is performed on the fusion vector and the regression residual according to the preset radial basis function network model to determine the target traffic flow information at the target time, thereby completing the traffic flow prediction; wherein, the preset radial basis function network model is trained based on historical traffic flow sequences, and the radial basis function network model includes the target time for predicting the target traffic flow information; The method further includes: Obtain multiple historical traffic sequences; The historical traffic sequence is processed by feature extraction according to a preset convolutional neural network model to obtain a first historical vector; the historical traffic sequence is processed by feature extraction according to a preset long short-term memory network model to obtain a second historical vector. The first historical vector and the second historical vector are fused to obtain a historical vector; Regression prediction is performed on the historical traffic sequence to obtain historical prediction information, and historical residuals are obtained based on the historical prediction information and the historical values ​​corresponding to the historical prediction information. Based on the historical residuals and the historical vectors, the radial basis function network model is trained for traffic prediction to determine the predicted traffic information at historical moments. Based on the historical forecast information and the forecast traffic information, the forecast residual is determined; If the predicted residual is determined to meet a threshold, then the radial basis function network model is determined to have converged.

2. The method according to claim 1, characterized in that, The step of predicting traffic flow based on the fusion vector and the regression residual according to a preset radial basis function network model to determine the target traffic flow information at the target time includes: Based on the target time in the preset radial basis function network model, the flow prediction is performed on the fusion vector and the regression residual according to the radial basis function network model to determine the target flow information at the target time; wherein, the target time includes at least one time.

3. The method according to claim 1, characterized in that, Performing regression prediction on the traffic data sequence to obtain regression residuals includes: Regression prediction is performed on the traffic data sequence to obtain preliminary prediction information, and the actual value corresponding to the preliminary prediction information is determined. Based on the preliminary prediction information and the actual value, the regression residuals are obtained.

4. A flow prediction device, characterized in that, include: The first acquisition unit is used to acquire a data acquisition instruction and acquire a traffic data sequence according to the data acquisition instruction; The second acquisition unit is used to perform feature extraction processing on the traffic data sequence through a preset convolutional neural network model to obtain a first feature vector; and to perform feature extraction processing on the traffic data sequence through a preset long short-term memory network model to obtain a second feature vector; wherein, both the convolutional neural network model and the long short-term memory network model are trained based on historical traffic sequences; The first fusion unit is used to fuse the first feature vector and the second feature vector to obtain a fusion vector; The third acquisition unit is used to perform regression prediction on the traffic data sequence and obtain the regression residual; The prediction unit is used to predict the flow rate of the fusion vector and the regression residual based on a preset radial basis function network model, and to determine the target flow rate information at the target time to complete the flow rate prediction; wherein, the preset radial basis function network model is trained based on historical flow rate sequences, and the radial basis function network model includes the target time for predicting the target flow rate information; The device further includes: The fourth acquisition unit is used to acquire multiple historical traffic sequences; The fifth acquisition unit is used to perform feature extraction processing on the historical traffic sequence according to a preset convolutional neural network model to obtain a first historical vector; and to perform feature extraction processing on the historical traffic sequence through a preset long short-term memory network model to obtain a second historical vector. The sixth acquisition unit is used to perform fusion processing on the first historical vector and the second historical vector to obtain a historical vector; The seventh acquisition unit is used to perform regression prediction on the historical traffic sequence, acquire historical prediction information, and acquire historical residuals based on the historical prediction information and the historical values ​​corresponding to the historical prediction information. The first determining unit is used to train the radial basis network model for traffic prediction based on the historical residuals and the historical vectors, and to determine the predicted traffic information at historical moments. The second determining unit is used to determine the prediction residual based on the historical prediction information and the predicted flow information; The third determining unit is used to determine that the radial basis network model has converged if the predicted residual satisfies a threshold.

5. The apparatus according to claim 4, characterized in that, The prediction unit is specifically used for: Based on the target time in the preset radial basis function network model, the flow prediction is performed on the fusion vector and the regression residual according to the radial basis function network model to determine the target flow information at the target time; wherein, the target time includes at least one time.

6. The apparatus according to claim 4, characterized in that, The third acquisition unit includes: The determination module is used to perform regression prediction on the traffic data sequence, obtain preliminary prediction information, and determine the actual value corresponding to the preliminary prediction information; The acquisition module is used to obtain the regression residuals based on the preliminary prediction information and the actual values.

7. A server, characterized in that, The method includes a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the computer program to implement the method of any one of claims 1-3.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1-3.

9. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method of any one of claims 1-3.

Images