CN112118143B - Traffic prediction model training method, traffic prediction method, device, equipment and medium - Google Patents
Traffic prediction model training method, traffic prediction method, device, equipment and medium Download PDFInfo
- Publication number
- CN112118143B CN112118143B CN202011293151.2A CN202011293151A CN112118143B CN 112118143 B CN112118143 B CN 112118143B CN 202011293151 A CN202011293151 A CN 202011293151A CN 112118143 B CN112118143 B CN 112118143B
- Authority
- CN
- China
- Prior art keywords
- flow
- sample
- traffic
- network
- prediction
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
- H04L41/14—Network analysis or design
- H04L41/147—Network analysis or design for predicting network behaviour
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
- H04L41/14—Network analysis or design
- H04L41/145—Network analysis or design involving simulating, designing, planning or modelling of a network
Abstract
The application provides a traffic prediction model, a training method, a prediction method, a device, equipment and a medium, wherein the method comprises the following steps: acquiring a flow data set, and intercepting the flow data set through sliding window operation to obtain sample data characteristics of flow of the size of a sliding window and a sample label corresponding to the sample data characteristics of the flow; the next flow value after the sample data characteristic of the sample label flow; acquiring sample attribute characteristics corresponding to the sample label; processing the sample data characteristics of the flow by using a gated cyclic neural network of the model, and inputting the output memory state and the sample attribute characteristics into a full-connection network of the model together for regression prediction; and updating the parameter weight through a back propagation algorithm according to the sample label and the prediction result to obtain a trained flow prediction model. Therefore, the prediction result is associated with the attributes of different flow rates, and the problems of low hysteresis and accuracy in flow rate peak prediction are solved.
Description
Technical Field
The present application relates to the field of network technologies, and in particular, to a traffic prediction model, a training method, a prediction method, an apparatus, a device, and a medium.
Background
With the popularization of networks, the scale of network traffic is constantly refreshed, and efficient and reasonable network resource allocation becomes important. On one hand, unreasonable network resource allocation may cause that part of network resources cannot be used normally due to exhaustion, and even cause network paralysis; and the other part of network resources are in an excess state, which seriously affects the internet surfing experience of the user. On the other hand, although the network reasonably allocates the network resources in the early stage, the network traffic is bursty, and the situation of resource shortage may occur when the network resources are originally sufficient. In order to solve the problem, an existing SDN (Software Defined Network ) controller can alleviate the problem to a certain extent by detecting a link condition and then scheduling. However, at present, the SDN controller is already congested when performing scheduling, so that the requirements of higher level and better quality of service cannot be met. For this reason, it is necessary to enable advanced prediction of traffic, so that the SDN controller can perform reasonable scheduling in advance.
Gated Recurrent neural networks, such as LSTM (Long Short-Term Memory) and GRU (Gated Recurrent Unit), are special Recurrent neural networks that can learn Long-Term dependencies in sequences and predict values of next sequences based on historical actual values. When the flow prediction is performed through the LSTM or GRU neural network, most time nodes can be well predicted, overall MAPE (Mean Absolute Percentage Error) and RMSE (Root Mean Square Error) are also small, but for the sample data characteristics of the flow with weak stationarity, the prediction of the flow peak value has serious problems of hysteresis and low accuracy. The concrete expression is as follows: at the t +1 moment when the flow peak actually occurs, the predicted flow is very similar to the flow at the t moment, so that when the flow peak occurs, the difference between the actually occurring flow and the predicted value is large, and the predicted value cannot be adopted as a judgment basis in an actual application scene. This greatly reduces the value of traffic prediction by LSTM or GRU.
Disclosure of Invention
An object of the embodiments of the present application is to provide a traffic prediction model, a training method, a prediction method, an apparatus, a device, and a medium, so as to improve accuracy of traffic prediction.
The embodiment of the application provides a method for training a flow prediction model, which comprises the following steps: acquiring a flow data set arranged according to an acquisition time sequence; intercepting the flow data set through sliding window operation to obtain sample data characteristics of flow of the size of the sliding window and a sample label corresponding to the sample data characteristics of the flow; the sample label is a next flow value corresponding to the sample data feature in the flow data set; acquiring sample attribute characteristics corresponding to the sample label; processing the sample data characteristics of the flow by using a gated cyclic neural network, and inputting the memory state output by the gated cyclic neural network and the sample attribute characteristics corresponding to the sample labels into a fully-connected network together for regression prediction; and updating the parameter weights of the gated cyclic neural network and the fully-connected network through a back propagation algorithm according to the sample label and the prediction result to obtain a trained flow prediction model.
According to the model obtained through the training of the implementation process, when flow prediction is carried out, the processing result of the gated cyclic neural network on the sample data characteristics of the previous flow and the sample attribute characteristics during prediction can be combined at the same time, so that on the basis of traditional prediction through the gated cyclic neural network, the sample attribute characteristics of richer flow are combined, prediction on the flow is not carried out according to the flow evolution rule of the historical moment, but prediction is carried out according to different attribute characteristics of different flows, the prediction result is associated with the attributes of different moments to be predicted, and the problems of low hysteresis and accuracy in prediction of a flow peak value are solved. The scheme of the embodiment of the application can guarantee the diversity of data characteristics, is favorable for automatically learning the potential law of the flow in practical application, can improve the accuracy of prediction, and solves the problem of hysteresis existing in flow prediction generally.
The obtaining of the sample attribute feature corresponding to the sample label includes: and converting the attribute of the sample label into a sample attribute characteristic corresponding to the sample label.
Further, the attributes of the sample label include at least one of:
the time of collection of the sample label;
the sample tags the events that occurred at or before the time of collection.
In the actual application process, the flow rate is often particularly high at certain time, or when or after certain events occur. The time, the event and the like can be used as the attributes of the flow, and the sample attribute features can be extracted from the attributes, so that when prediction is carried out, the time of the flow needing to be predicted can be combined, or an event which occurs at the time point corresponding to the flow needing to be predicted or an event which occurs before the time point needing to be predicted can be combined, and the problem of inaccurate prediction of the flow peak value can be solved.
Further, the converting the attribute of the sample label into the sample attribute feature corresponding to the sample label includes: and carrying out one-hot encoding on the attributes of the sample label to obtain the sample attribute characteristics corresponding to the sample label.
Further, the acquiring the traffic data sets arranged in the acquisition time sequence includes: aggregating the original network traffic collected according to the time sequence into network traffic with a preset time interval to generate a training set; and carrying out exception handling on the network traffic in the training set to obtain the traffic data set.
Further, performing exception handling on the network traffic in the training set, including: determining abnormal network traffic in the training set; and carrying out exception eliminating treatment on the abnormal network flow.
Through the implementation process, abnormal data in the flow data set can be removed, so that the flow data used for training is more reliable, and the accuracy of prediction can be further improved.
Further, determining abnormal network traffic in the training set includes: and determining abnormal network flow in the training set through an isolated forest algorithm.
In the implementation process, the abnormal network flow is determined through the isolated forest algorithm, the occupation of a memory and a CPU is low, and the data preprocessing is facilitated.
The embodiment of the present application further provides a traffic prediction method, including: forming sample data characteristics of the traffic to be predicted according to time sequence of the network traffic within a preset time before the time to be predicted; acquiring attribute information corresponding to the flow to be predicted, and converting the attribute information into sample attribute characteristics; inputting the sample data characteristics of the flow to be predicted into a gated cyclic neural network of a flow prediction model; and inputting the memory state output by the gated cyclic neural network and the sample attribute characteristics into a fully-connected network of the flow prediction model together for regression prediction to obtain predicted flow.
Through the implementation process, the processing result of the sample data characteristics of the previous flow and the sample attribute characteristics corresponding to the time to be predicted can be simultaneously combined to predict, the prediction does not depend on the network flow in the preset duration before the time to be predicted, and also depends on the attribute information corresponding to the time to be predicted, so that the prediction result is associated with the attributes of different times to be predicted, and the problems of hysteresis and low accuracy in the prediction of the flow peak value are solved.
Further, the attribute information includes at least one of:
time information of the flow to be predicted;
events about to occur in the flow to be predicted;
the current most recent event that occurred.
In practical application, the flow rate is often particularly high at certain times, or is particularly high when or after certain events occur. Therefore, at least one of the time information of the flow to be predicted, the event which will occur in the flow to be predicted and the current latest event is used as attribute information to be input into the model for prediction, so that the problem of inaccurate prediction of the flow peak value can be solved by combining the time of the flow to be predicted, the event which occurs in the flow to be predicted or the event which occurs before the time to be predicted when the flow to be predicted is predicted.
Further, converting the attribute information into sample attribute features, including: and carrying out one-hot coding on the attribute information to obtain the sample attribute characteristics.
Further, after the network traffic within a preset time before the time to be predicted forms the sample data features of the traffic according to the time sequence, the sample data features of the traffic to be predicted are input to the front of the gated cyclic neural network of the traffic prediction model, and the method further comprises the following steps: determining abnormal network traffic in the sample data characteristic quantity of the traffic; and carrying out exception eliminating treatment on the abnormal network traffic to obtain the sample data characteristics of the processed traffic.
Through the implementation process, the abnormal data in the sample data characteristics of the flow can be removed, so that the flow data for prediction is more reliable, and the accuracy of prediction can be further improved.
The embodiment of the application also provides a flow prediction model, which is obtained by training any one of the above training methods of the flow prediction model.
The flow prediction model obtained by training can be used for predicting the sample data characteristic processing result and the sample attribute characteristic of the previous flow by combining the gated cyclic neural network, the prediction does not depend on the historical network flow, but depends on the attribute information, so that the prediction result is associated with the attribute of the flow to be predicted, and the problems of low hysteresis and accuracy in the prediction of the flow peak value are solved.
The embodiment of the present application further provides a training apparatus for a traffic prediction model, including: the system comprises an acquisition module, a feature processing module and a training module; the acquisition module is used for acquiring the flow data sets arranged according to the acquisition time sequence and acquiring the sample attribute characteristics corresponding to the sample labels; the characteristic processing module is used for intercepting the flow data set through sliding window operation to obtain the sample data characteristics of the flow of the size of the sliding window and sample labels corresponding to the sample data characteristics of the flow; the sample label is a next flow value corresponding to the sample data characteristic of the flow in the flow data set; and the training module is used for processing the sample data characteristics of the flow by using a gated cyclic neural network, inputting the memory state output by the gated cyclic neural network and the sample attribute characteristics corresponding to the sample labels into a fully-connected network together for regression prediction, and updating the parameter weights of the gated cyclic neural network and the fully-connected network through a back propagation algorithm according to the sample labels and the prediction result to obtain a trained flow prediction model.
An embodiment of the present application further provides a traffic prediction apparatus, including: a preparation module and a prediction module; the preparation module is used for forming the sample data characteristics of the traffic to be predicted according to the time sequence of the network traffic within the preset time before the time to be predicted; acquiring attribute information corresponding to the flow to be predicted, and converting the attribute information into sample attribute characteristics; the prediction module is used for inputting the sample data characteristics of the flow to be predicted into a gated cyclic neural network of a flow prediction model; and inputting the memory state output by the gated cyclic neural network and the sample attribute characteristics into a fully-connected network of the flow prediction model together for regression prediction to obtain predicted flow.
An embodiment of the present application further provides an electronic device, including: the method comprises the following steps: a processor, a memory, and a communication bus; the communication bus is used for realizing connection communication between the processor and the memory; the processor is configured to execute one or more programs stored in the memory to implement any of the above-described methods of training a flow prediction model or to implement any of the above-described methods of flow prediction.
Also provided in an embodiment of the present application is a readable storage medium storing one or more programs, where the one or more programs are executable by one or more processors to implement any of the above methods for training a flow prediction model or any of the above methods for flow prediction.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are required to be used in the embodiments of the present application will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present application and therefore should not be considered as limiting the scope, and that those skilled in the art can also obtain other related drawings based on the drawings without inventive efforts.
Fig. 1 is a schematic view of a flow prediction model provided in an embodiment of the present application;
fig. 2 is a schematic flowchart of a method for training a traffic prediction model according to an embodiment of the present disclosure;
fig. 3 is a schematic flow chart of a traffic prediction method according to an embodiment of the present application;
fig. 4 is a schematic structural diagram of a training apparatus for a flow prediction model according to an embodiment of the present application;
fig. 5 is a schematic structural diagram of a flow rate prediction apparatus according to an embodiment of the present disclosure;
fig. 6 is a schematic structural diagram of an electronic device according to an embodiment of the present application.
Detailed Description
The technical solutions in the embodiments of the present application will be described below with reference to the drawings in the embodiments of the present application.
The first embodiment is as follows:
at present, in order to realize the prediction of the flow, a gated recurrent neural network (such as LSTM, GRU, etc.) is often used as a prediction model, and a trained gated recurrent neural network is used to process the historical network flow before the time to be predicted, so as to obtain a predicted value.
However, in an actual application scenario, the use condition of the network flow often has many emergency situations, and the stationarity of the data change of the network flow is not strong, so that the problem of serious hysteresis and low accuracy exists for the prediction of the flow peak value by simply adopting a gated recurrent neural network for prediction. To this end, in the embodiment of the present application, a new flow prediction model is provided, which can be seen from fig. 1, and includes: gated recurrent neural networks and fully connected networks.
Wherein, the gated cyclic neural network inputs a sequence formed by network traffic. The sequence is processed by a gated recurrent neural network, and a memory state is output.
In the embodiment of the application, the output end of the gated cyclic neural network is connected with the input end of the fully connected network, and the memory state output by the gated cyclic neural network and the sample attribute characteristics of the time required to be predicted are input into the fully connected network together for regression prediction, so that the predicted flow is obtained.
It should be noted that, in the embodiment of the present application, the gated recurrent neural network may be implemented by using an LSTM network, a GRU network, or the like.
It should be noted that, for the foregoing traffic prediction model, training needs to be performed first, and based on the trained traffic prediction model, accurate prediction of network traffic can be achieved.
To this end, an embodiment of the present application further provides a method for training a flow prediction model, please refer to fig. 2, where the method includes:
s201: a flow data set arranged in an acquisition time order is acquired.
It should be understood that in an actual application scenario, because the stationarity of the network traffic is not strong, there are many abnormal values that are not suitable for training in the original network traffic obtained by sampling.
Therefore, in the embodiment of the application, network traffic in a network can be sampled according to a preset sampling interval, and then original network traffic obtained by sampling is aggregated into network traffic of a preset time interval, and then a traffic data set is obtained by arranging according to a time sequence. Therefore, the influence of the abnormal value on the training sample is reduced by means of flow aggregation.
For example, the original network traffic may be sampled every 10 seconds, and then an average value of all the collected original network traffic in every five minutes is used as the aggregated network traffic of the five minutes, and then the traffic data sets are obtained by arranging in time sequence.
However, it should be understood that, although the influence of the abnormal value on the training sample can be reduced to some extent by means of traffic aggregation, for a case that there is more abnormal original network traffic within an aggregation time period, the network traffic obtained by aggregation still belongs to the abnormal value.
Therefore, in the embodiment of the present application, the network traffic at the preset time interval obtained by aggregation may be used as a training set, and the network traffic in the training set is subjected to exception handling, so as to obtain a traffic data set.
In order to perform exception handling on the network traffic in the training set, in the embodiment of the present application, the abnormal network traffic in the training set may be determined by an isolated forest algorithm, wavelet denoising, 3sigma smoothing, and other algorithms, and then the abnormal network traffic is subjected to exception handling, so that the data reliability in the traffic data set is higher on the premise of ensuring the data consistency in the traffic data set.
For example, the algorithm formula S (x, n) = 2 may be used–(E(h(x)))/c(n)And obtaining abnormal scores corresponding to the network flows in the training set, and judging whether the abnormal scores are larger than a preset score threshold value or not, so that the network flows of which the abnormal scores are larger than the preset score threshold value are determined as abnormal network flows.
In the above formula, E (h (X)) is the expected path length of the sample X in a batch of isolated trees, and c (n) is the average value of the path lengths of the samples when the number of samples is n. The samples are network traffic in the training set, and n is the total number of the network traffic in the training set.
After the abnormal network traffic is determined, in the embodiment of the application, the determined abnormal network traffic value can be replaced by the mean value between neighbors, or the normal network traffic value is determined again by adopting a lagrange interpolation method, so that the abnormal network traffic is eliminated.
For example, when it is determined that the network traffic at time t is abnormal, the average of the network traffic at time t-1 and the network traffic at time t +1 may be used as the network traffic at time t.
In this embodiment, in order to facilitate model processing, normalization processing may be performed on each network traffic in the traffic data set, and each network traffic in the traffic data set may be normalized to a value between [0, 1 ].
Illustratively, the formula x = (x-x) may be usedmin) /( xmax - xmin) And obtaining the normalized value of each network flow in the flow data set. Wherein x ismaxIs the maximum value, x, of the network traffic in the traffic data setminThe value is the minimum value of the network flow in the flow data set, x is the value of the current calculated network flow, and x is the value of the calculated network flow after normalization.
S202: and intercepting the flow data set through sliding window operation to obtain the sample data characteristics of the flow of the size of the sliding window and the sample labels corresponding to the sample data characteristics of the flow.
In the embodiment of the present application, the sample label is the next flow value after the sample data feature of the flow in the flow data set.
For example, assuming that data in the traffic dataset is (a 1, a2, A3, a4, a5, A6 …) and the sliding window size is 3, a4 is a sample label of the sample data characteristics (a 1, a2, A3) of the traffic, a5 is a sample label of the sample data characteristics (a 2, A3, a 4) of the traffic, and A6 is a sample label of the sample data characteristics (A3, a4, a 5) of the traffic.
S203: and acquiring sample attribute characteristics corresponding to the sample labels.
In the embodiment of the application, the attribute of the sample label can be converted into the sample attribute feature corresponding to the sample label.
It should be understood that in practical applications, network traffic is often particularly large or small at some time, for example, for a scenario where an office intranet is managed, it may be that network traffic is particularly large around 9 am every day, and network traffic is particularly small after 6 pm. I.e. there will be a certain correlation between network traffic and time. In addition, network traffic is often particularly large or small when certain special events occur, such as the network of a certain fan association, which may be particularly large when events such as live or rebroadcast of a game occur. I.e., there will be some correlation between network traffic and events.
To this end, in an embodiment of the present application, the attribute of the specimen label may include at least one of a collection time of the specimen label, and an event that occurs at or before the collection of the specimen label. Which attribute or attributes need to be used to convert the sample attribute features can be determined according to the needs of the actual scene.
The acquisition time in the embodiment of the present application may be specified to a certain minute, a certain hour, a certain day, even a certain week, etc., and the specified granularity may be set by an engineer according to actual needs.
Events in embodiments of the present application may include external events that occur at or before the time the sample label is collected, such as product releases, sporting events, and the like.
It should be understood that the attribute of the sample label may be, besides the collection time of the sample label and the event occurring before or during collection of the sample label, the remaining content related to the size of the network traffic, and is not limited in the embodiment of the present application.
In order to enable the traffic prediction model to identify the attribute of the sample label, in the embodiment of the present application, the attribute of the sample label may be converted into a sample attribute feature corresponding to the sample label through a single hot coding or the like.
It should be noted that the one-hot encoding is only an optional implementation manner in the embodiment of the present application, and besides, the conversion of the attribute of the sample label into the sample attribute feature corresponding to the sample label may be implemented by using other encoding manners, as long as the converted data can be identified by the traffic prediction model and the feature of the attribute of the sample label can be retained, which does not limit the processing manner in the embodiment of the present application.
S204: and processing the sample data characteristics of the flow by using a gated cyclic neural network of the flow prediction model, and inputting the memory state output by the gated cyclic neural network and the sample attribute characteristics corresponding to the sample labels into a full-connection network of the flow prediction model together for regression prediction.
In the embodiment of the present application, the fully-connected network is used for performing regression operation. It may output a predicted value for the next instant of sample data characteristics for traffic input into the gated recurrent neural network. The predicted value is influenced by the development rule of the sample data characteristics of the flow and the sample attribute characteristics at the prediction moment.
S205: and updating the parameter weights of the gated cyclic neural network and the fully-connected network through a back propagation algorithm according to the sample label and the prediction result to obtain a trained flow prediction model.
It should be understood that the prediction result output by the traffic prediction model is the predicted network traffic value at the next time of the sample data characteristic of the traffic. And the sample label of the sample data characteristic of the flow is the actual network flow value at the next moment corresponding to the sample data characteristic of the flow. Therefore, whether the model is trained completely or not can be determined by means of loss function calculation and the like based on the sample label and the prediction result. And when the training is not finished, updating the parameter weights of the gated cyclic neural network and the fully-connected network through a back propagation algorithm, and continuously iterating to obtain a trained flow prediction model.
It should be noted that, in the embodiment of the present application, before or during training, the determining and optimizing of the hyper-parameters of the flow prediction model may be implemented by a method such as a grid exhaustive search algorithm, a bayesian optimization algorithm, a stochastic search algorithm, and the like.
In the embodiment of the application, the hyper-parameters of the traffic prediction model include the size of a sliding window, the learning rate, the number of neurons of a gated recurrent neural network and a fully connected network, and the like.
If the hyper-parameter is optimized in the training process, the optimized optimal hyper-parameter combination can be obtained through the algorithm, and the optimal hyper-parameter combination is utilized to update the parameter weights of the gated cyclic neural network and the fully-connected network by using a back propagation algorithm.
After the trained flow prediction model is obtained, the flow prediction model can be used for flow prediction. As shown in fig. 3, fig. 3 is a flow prediction method provided in an embodiment of the present application, and the method includes:
s301: and forming the sample data characteristics of the traffic to be predicted according to the time sequence of the network traffic within the preset time before the time to be predicted.
In the embodiment of the application, after the latest network traffic is acquired each time, the sample data characteristics of the traffic to be predicted are automatically formed with the previously acquired traffic.
Similarly, in order to improve the prediction accuracy and prevent prediction distortion caused by an abnormal value in the acquired original network traffic, in the embodiment of the present application, the network traffic in the network may be acquired according to a preset acquisition interval, and then the acquired original network traffic is aggregated into the network traffic of a preset time interval, and then the latest network traffic is placed behind the previous network traffic to form sample data characteristics of the traffic to be predicted, which are arranged according to the time sequence. Therefore, the influence of the abnormal value on the sample data characteristics of the flow to be predicted is reduced in a flow aggregation mode.
However, it should be understood that, although the influence of the abnormal value on the sample data characteristics of the traffic to be predicted can be reduced to some extent by means of traffic aggregation, for a case that there is more abnormal original network traffic within an aggregation time period, the network traffic obtained by aggregation still belongs to the abnormal value.
Therefore, in the embodiment of the application, abnormal network traffic in the sample data features of the traffic to be predicted may be detected (for example, the abnormal network traffic is checked by using an isolated forest algorithm or the like), and then the detected abnormal network traffic is subjected to exception elimination processing, so as to obtain the sample data features of the processed traffic to be predicted.
In the embodiment of the application, the determined abnormal network traffic value can be replaced by the mean value between neighbors, or the normal network traffic value can be determined again by adopting a lagrange interpolation method, so that the abnormal network traffic can be processed.
S302: and acquiring attribute information corresponding to the flow to be predicted, and converting the attribute information into sample attribute characteristics.
In this embodiment of the present application, the attribute information corresponding to the traffic to be predicted includes at least one of the following:
time information of the traffic to be predicted (i.e. time of the time to be predicted), an event to be generated at the time corresponding to the traffic to be predicted, and a current latest event.
It should be understood that the attribute information corresponding to the traffic to be predicted should be consistent with the attribute selected when the traffic prediction model is trained, so as to ensure that the traffic prediction model has corresponding processing capability.
For example, only the acquisition time of the sample label is used when the traffic prediction model is trained, and then the attribute information corresponding to the traffic to be predicted should only include the time information of the traffic to be predicted. For another example, only the event that occurs when the sample label is collected is used in training the traffic prediction model, and then the attribute information corresponding to the traffic to be predicted should only include the event that will occur at the time corresponding to the traffic to be predicted. For another example, only the event that occurs before the sample label is acquired is used in training the traffic prediction model, and then the attribute information corresponding to the traffic to be predicted should only include the current latest event.
Similarly, if the acquisition time of the sample label and the event occurred during the acquisition of the sample label are used in the training of the traffic prediction model, the attribute information corresponding to the traffic to be predicted should include the time information of the traffic to be predicted and the event that will occur at the moment corresponding to the traffic to be predicted, and cannot include the current latest event. The rest of the cases are similar and will not be described again.
S303: and inputting the sample data characteristics of the flow to be predicted into a gated cyclic neural network of the flow prediction model.
The gated cyclic neural network performs forward operation on the characteristics of the sample data of the flow to be predicted, so that a memory state is output.
Taking LSTM network as an example, the forward algorithm is:
updating the forgetting door:
f(t)＝σ(wf*[h(t-1)，xt]+bf) Updating an input gate:
i(t)＝σ(wi*[h(t-1)，xt]+bi)
l(t)＝tanh(wc*[h(t-1)，xt]+bc)
updating an output gate:
o(t)＝σ(wo*[h(t-1)，xt]+bo)
and (3) memorizing the state:
long memory: c. C(t) = f(t)*c(t-1) +i(t)* l(t)
Short memory: h is(t) = o(t) *tanh(c(t))
Wherein the meaning of each parameter is as follows:
xtis an input feature of time step t, h(t-1)Short memory state for time step t-1, c(t-1)Long memory state, w, for time step t-1f、bfWeight parameter for forgetting gate, wi、bi、wc、bcTo input the weight parameter of the gate, wo、boTo update the weight parameters of the door.
Taking the GRU network as an example, the forward algorithm is as follows:
and (4) updating the door: r is(t) =σ(Wr * [h(t-1)， xt])
Resetting a gate: z is a radical of(t) =σ(Wz * [h(t-1)， xt])
And (3) memorizing the state:
h¢(t)= tanh(W * [r(t) * h(t-1)， xt])
h(t) = (1- z(t))*h(t-1) + z(t) * h¢(t)
wherein the meaning of each parameter is as follows:
xtis an input feature of time step t, h(t-1)For the memory state of time step t-1, WrTo update the weight parameter of the door, WzThe weight parameters of the gates are reset.
S304: and inputting the memory state output by the gated cyclic neural network and the sample attribute characteristics into a full-connection network of the flow prediction model together for regression prediction to obtain the predicted flow.
According to the scheme provided by the embodiment of the application, the processing result of the sample data characteristics of the previous flow and the sample attribute characteristics can be simultaneously predicted by combining the gated cyclic neural network, the prediction does not depend on the historical network flow, but also depends on the attribute information, so that the prediction result is associated with the attribute to be predicted, and the problems of hysteresis and low accuracy in the prediction of the flow peak value are solved.
Meanwhile, in the scheme of the embodiment of the application, the diversity of data characteristics can be ensured, the potential law of the flow in practical application can be automatically learned, and the accuracy of prediction can be improved.
Example two:
in this embodiment, on the basis of the first embodiment, the LSTM network and the GRU network are taken as examples respectively to further illustrate the scheme of the present application.
In one example, the gated recurrent neural network is an LSTM network, and includes the following steps:
step 301), aggregating the sampled original network traffic into network traffic at preset intervals to obtain a training set.
And 302), detecting abnormal network traffic existing in the training set through an 'isolated forest' algorithm.
Using the algorithmic formula S (x, n) = 2–(E(h(x)))/c(n)And obtaining abnormal scores corresponding to the network flows in the training set, and judging whether the abnormal scores are larger than a preset score threshold value or not, so that the network flows of which the abnormal scores are larger than the preset score threshold value are determined as abnormal network flows.
In the embodiment of the present application, when the preset score threshold may be 0.5.
After the abnormal network traffic is detected, the detected abnormal network traffic is replaced by the mean value between neighbors. For example, when abnormal network traffic is detected at the time t, the network traffic value at the time t-1 and the network traffic value at the time t +1 are averaged to replace the network traffic value at the time t.
Step 303), normalizing the training set processed in step 302) to [0, 1] by using a formula x = (x-xmin)/(xmax-xmin), wherein xmax is the maximum value of the network traffic in the traffic data set, xmin is the minimum value of the network traffic in the traffic data set, x is the value of the currently calculated network traffic, and x is the value of the calculated network traffic after normalization.
Step 304) obtaining sample data characteristics and sample labels of the traffic through sliding window operation.
And carrying out one-hot coding on the acquisition time of the sample label to obtain a time sample attribute characteristic, and carrying out one-hot coding on an external event to obtain an event sample attribute characteristic.
Step 305), constructing a traffic prediction model by using the LSTM network.
The whole flow prediction model is composed of an input layer, an LSTM network and a full-connection network. The input of the input layer is the sample data characteristic of the flow, and the LSTM network forward algorithm is as follows:
updating the forgetting door:
f(t)＝σ(wf*[h(t-1)，xt]+bf) Updating an input gate:
i(t)＝σ(wi*[h(t-1)，xt]+bi)
l(t)＝tanh(wc*[h(t-1)，xt]+bc)
updating an output gate:
o(t)＝σ(wo*[h(t-1)，xt]+bo)
and (3) memorizing the state:
long memory: c. C(t) = f(t)*c(t-1) +i(t)* l(t)
Short memory: h is(t) = o(t) *tanh(c(t))
Step 306), the memory state, the time sample attribute characteristic and the event sample attribute characteristic output in the step 305) are spliced to be used as the input of the full-connection network. This example selects [ c ](t)，h(t)Time sample attribute feature, event sample attribute feature]As an input to the full connection.
And 307), obtaining the hyper-parameters of the flow prediction model training by using a grid exhaustive search algorithm, wherein the hyper-parameters comprise the size of a sliding window, the learning rate, the number of neurons of an LSTM network and a fully connected network and the like, and training the parameter weight by using a back propagation algorithm by using the optimal hyper-parameter combination.
Illustratively, the sliding window size may be 10, the learning rate may be 0.001, the number of nerves of the LSTM network may be 32, and the number of layers of the fully connected network may be 2.
And 308), carrying out flow prediction on the weight of the flow prediction model obtained in the step 307).
In example two, the gated recurrent neural network is a GRU network, and includes the following steps:
step 401), aggregating the sampled original network traffic into network traffic at preset intervals to obtain a training set.
And step 402), detecting abnormal network traffic existing in the training set through an 'isolated forest' algorithm.
Using the algorithmic formula S (x, n) = 2–(E(h(x)))/c(n)And obtaining abnormal scores corresponding to the network flows in the training set, and judging whether the abnormal scores are larger than a preset score threshold value or not, so that the network flows of which the abnormal scores are larger than the preset score threshold value are determined as abnormal network flows.
After the abnormal network traffic is detected, the detected abnormal network traffic is replaced by the mean value between neighbors. For example, when abnormal network traffic is detected at the time t, the network traffic value at the time t-1 and the network traffic value at the time t +1 are averaged to replace the network traffic value at the time t.
Step 403), normalizing the training set processed in step 402) to [0, 1] by using a formula x = (x-xmin)/(xmax-xmin), wherein xmax is the maximum value of the network traffic in the traffic data set, xmin is the minimum value of the network traffic in the traffic data set, x is the value of the currently calculated network traffic, and x is the value of the calculated network traffic after normalization.
Step 404) obtaining sample data characteristics and sample labels of the traffic through sliding window operation.
And carrying out one-hot coding on the acquisition time of the sample label to obtain a time sample attribute characteristic, and carrying out one-hot coding on an external event to obtain an event sample attribute characteristic.
Step 405), a traffic prediction model is built by using the GRU network.
The whole flow prediction model is composed of an input layer, a GRU network and a full-connection network. The input of the input layer is the sample data characteristic of the flow, and the GRU network forward algorithm is as follows:
and (4) updating the door: r is(t) =σ(Wr * [h(t-1)， xt])
Resetting a gate: z is a radical of(t) =σ(Wz * [h(t-1)， xt])
And (3) memorizing the state:
h¢(t) = tanh(W * [r(t) * h(t-1)， xt])
h(t) = (1- z(t))*h(t-1) + z(t) * h¢(t)
step 406), the memory state, the time sample attribute characteristic and the event sample attribute characteristic output in the step 405) are spliced to be used as the input of the full-connection network. This example selects [ h ](t)Time sample attribute feature, event sample attribute feature]As an input to the full connection.
And 407), obtaining the hyper-parameters of model training by using a grid exhaustive search algorithm, wherein the hyper-parameters comprise the size of a sliding window, the learning rate, the number of neurons of an LSTM network and a fully-connected network and the like, and training the parameter weight by using a back propagation algorithm by using the optimal hyper-parameter combination.
Step 408), carrying out flow prediction on the weight of the flow prediction model obtained in the step 407).
The isolated forest algorithm adopted by the embodiment is used for exception screening, occupies low memory and CPU, and is beneficial to data preprocessing. The flow prediction model brings time characteristics and event characteristics into the full-connection network, so that the diversity of data characteristics can be ensured, the potential rule of the algorithm automatic learning data is facilitated, the prediction accuracy is improved, and the problem of hysteresis existing in flow prediction generally is solved.
Example three:
based on the same inventive concept, the embodiment of the application also provides a training device of the flow prediction model and a flow prediction device. Referring to fig. 4 and 5, fig. 4 shows a training apparatus 100 of a traffic prediction model corresponding to a training method of a traffic prediction model according to the first embodiment, and fig. 5 shows a training apparatus 200 of a traffic prediction model corresponding to a training method of a traffic prediction model according to the first embodiment. It should be understood that the specific functions of the apparatuses 100 and 200 can be referred to the above description, and the detailed description is omitted here as appropriate to avoid redundancy. The devices 100 and 200 include at least one software functional module that can be stored in a memory in the form of software or firmware or solidified in an operating system of the devices 100 and 200. Specifically, the method comprises the following steps:
referring to fig. 4, the training apparatus 100 for a flow prediction model includes: an acquisition module 101, a feature processing module 102, and a training module 103. Wherein:
the obtaining module 101 is configured to obtain a flow data set arranged according to a collection time sequence, and obtain a sample attribute feature corresponding to the sample label.
The characteristic processing module 102 is configured to intercept the traffic data set through a sliding window operation, so as to obtain a sample data characteristic of traffic of the size of the sliding window and a sample label corresponding to the sample data characteristic of the traffic; the sample tag is a next flow value corresponding to the sample data feature in the flow data set.
The training module 103 is configured to process the sample data features of the traffic by using a gated cyclic neural network, and input the memory state output by the gated cyclic neural network and the sample attribute features corresponding to the sample labels together into a fully-connected network for regression prediction; and updating the parameter weights of the gated cyclic neural network and the fully-connected network through a back propagation algorithm according to the sample label and the prediction result to obtain a trained flow prediction model.
In this embodiment of the present application, the obtaining module 101 is specifically configured to convert the attribute of the sample label into a sample attribute feature corresponding to the sample label.
In an embodiment of the present application, the attribute of the sample label includes at least one of:
the time of collection of the sample label;
the sample tags the events that occurred at or before the time of collection.
In this embodiment of the present application, the obtaining module 101 is specifically configured to perform one-hot encoding on the attribute of the sample label to obtain a sample attribute feature corresponding to the sample label.
In this embodiment of the present application, the obtaining module 101 is specifically configured to aggregate original network traffic collected according to a time sequence into network traffic at a preset time interval, and generate a training set; and carrying out exception handling on the network traffic in the training set to obtain the traffic data set.
In this embodiment of the present application, the obtaining module 101 is specifically configured to determine abnormal network traffic in the training set; and carrying out exception eliminating treatment on the abnormal network flow.
In this embodiment of the present application, the obtaining module 101 is specifically configured to determine abnormal network traffic in the training set through an isolated forest algorithm.
Referring to fig. 5, the flow prediction apparatus 200 includes: a preparation module 201 and a prediction module 202. Wherein:
the preparation module 201 is configured to form sample data characteristics of traffic to be predicted according to a time sequence of network traffic within a preset time before a time to be predicted; and acquiring attribute information corresponding to the flow to be predicted, and converting the attribute information into sample attribute characteristics.
The prediction module 202 is configured to input the sample data characteristics of the flow to be predicted into a gated recurrent neural network of a flow prediction model; and inputting the memory state output by the gated cyclic neural network and the sample attribute characteristics into a full-connection network of a flow prediction model together for regression prediction to obtain predicted flow.
In an embodiment of the present application, the attribute information includes at least one of:
time information of the flow to be predicted;
events about to occur in the flow to be predicted;
the current most recent event that occurred.
In this embodiment of the application, the preparation module 201 is specifically configured to perform one-hot encoding on the attribute information to obtain the sample attribute feature.
In this embodiment of the application, the preparation module 201 is further configured to, after forming, according to a time sequence, sample data features of traffic to be predicted for network traffic within a preset duration before a time to be predicted, input the sample data features of the traffic to be predicted to a gated recurrent neural network of the traffic prediction model, and determine abnormal network traffic in the sample data features of the traffic to be predicted; and carrying out exception elimination treatment on the abnormal network traffic to obtain the processed sample data characteristics of the traffic to be predicted.
It should be understood that, for the sake of brevity, the contents described in some embodiments are not repeated in this embodiment.
Example four:
the embodiment provides an electronic device, which can be seen in fig. 6 and includes a processor 601, a memory 602 and a communication bus 603. Wherein:
the communication bus 603 is used for connection communication between the processor 601 and the memory 602.
The processor 601 is configured to execute one or more programs stored in the memory 602 to implement the method for training the flow prediction model and/or the method for flow prediction in the first embodiment or the second embodiment.
It is understood that the structure shown in fig. 6 is only an illustration, and the electronic device may further include more or less components than those shown in fig. 6, or have a different configuration from that shown in fig. 6, and is not limited in the embodiment of the present application.
It is understood that the electronic device shown in fig. 6 may be any device in a network, and may be, for example, an SDN controller.
The present embodiment also provides a readable storage medium, such as a floppy disk, an optical disk, a hard disk, a flash Memory, a usb (Secure Digital Memory Card), an MMC (Multimedia Card), etc., in which one or more programs for implementing the above steps are stored, and the one or more programs can be executed by one or more processors to implement the method for training the traffic prediction model and/or the method for traffic prediction in the first embodiment or the second embodiment. And will not be described in detail herein.
In the embodiments provided in the present application, it should be understood that the disclosed apparatus and method may be implemented in other ways. The above-described embodiments of the apparatus are merely illustrative, and for example, the division of the units is only one logical division, and there may be other divisions when actually implemented, and for example, a plurality of units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection of devices or units through some communication interfaces, and may be in an electrical, mechanical or other form.
In addition, units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.
Furthermore, the functional modules in the embodiments of the present application may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.
In this document, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions.
In this context, a plurality means two or more.
The above description is only an example of the present application and is not intended to limit the scope of the present application, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, improvement and the like made within the spirit and principle of the present application shall be included in the protection scope of the present application.
Claims (12)
1. A method for training a flow prediction model is characterized by comprising the following steps:
acquiring a flow data set arranged according to an acquisition time sequence;
intercepting the flow data set through sliding window operation to obtain sample data characteristics of flow of the size of the sliding window and a sample label corresponding to the sample data characteristics of the flow; the sample label is a next flow value corresponding to the sample data feature in the flow data set;
acquiring sample attribute characteristics corresponding to the sample label;
processing the sample data characteristics of the flow by using a gated cyclic neural network, and inputting the memory state output by the gated cyclic neural network and the sample attribute characteristics corresponding to the sample labels into a fully-connected network together for regression prediction;
updating the parameter weights of the gated cyclic neural network and the fully-connected network through a back propagation algorithm according to the sample label and the prediction result to obtain a trained flow prediction model;
the obtaining of the sample attribute feature corresponding to the sample label includes:
converting the attribute of the sample label into a sample attribute characteristic corresponding to the sample label;
the attributes of the sample label include at least one of:
the time of collection of the sample label;
the sample tags the events that occurred at or before the time of collection.
2. The method of training a flow prediction model of claim 1, wherein said obtaining a flow data set arranged in time order of acquisition comprises:
aggregating the original network traffic collected according to the time sequence into network traffic with a preset time interval to generate a training set;
and carrying out exception handling on the network traffic in the training set to obtain the traffic data set.
3. The method for training a traffic prediction model according to claim 2, wherein the performing exception handling on the network traffic in the training set comprises:
determining abnormal network traffic in the training set;
and carrying out exception eliminating treatment on the abnormal network flow.
4. The method of training a traffic prediction model according to claim 3, wherein determining abnormal network traffic in the training set comprises:
and determining abnormal network flow in the training set through an isolated forest algorithm.
5. A method for traffic prediction, comprising:
forming sample data characteristics of the traffic to be predicted according to time sequence of the network traffic within a preset time before the time to be predicted;
acquiring attribute information corresponding to the flow to be predicted, and converting the attribute information into sample attribute characteristics;
inputting the sample data characteristics of the flow to be predicted into a gated cyclic neural network of a flow prediction model; the flow prediction model is obtained by training according to the training method of any one of claims 1 to 4;
and inputting the memory state output by the gated cyclic neural network and the sample attribute characteristics into a fully-connected network of the flow prediction model together for regression prediction to obtain predicted flow.
6. The traffic prediction method of claim 5, wherein the attribute information comprises at least one of:
time information of the flow to be predicted;
events about to occur in the flow to be predicted;
the current most recent event that occurred.
7. The traffic prediction method of claim 5, wherein converting the attribute information into sample attribute features comprises:
and carrying out one-hot coding on the attribute information to obtain the sample attribute characteristics.
8. The traffic prediction method according to any one of claims 5 to 7, wherein after the network traffic within a preset time period before the time to be predicted forms the sample data feature of the traffic to be predicted according to the time sequence, the sample data feature of the traffic to be predicted is input to the flow prediction model before the gated recurrent neural network, and the method further comprises:
determining abnormal network traffic in the sample data characteristics of the traffic to be predicted;
and carrying out exception elimination treatment on the abnormal network traffic to obtain the processed sample data characteristics of the traffic to be predicted.
9. An apparatus for training a flow prediction model, comprising: the system comprises an acquisition module, a feature processing module and a training module;
the acquisition module is used for acquiring the flow data sets arranged according to the acquisition time sequence and converting the attributes of the sample labels into sample attribute characteristics corresponding to the sample labels;
the characteristic processing module is used for intercepting the flow data set through sliding window operation to obtain the sample data characteristics of the flow of the size of the sliding window and sample labels corresponding to the sample data characteristics of the flow; the sample label is a next flow value corresponding to the sample data characteristic of the flow in the flow data set;
the training module is used for processing the sample data characteristics of the flow by using a gated cyclic neural network, inputting the memory state output by the gated cyclic neural network and the sample attribute characteristics corresponding to the sample labels into a fully-connected network together for regression prediction, and updating the parameter weights of the gated cyclic neural network and the fully-connected network through a back propagation algorithm according to the sample labels and the prediction result to obtain a trained flow prediction model;
the attributes of the sample label include at least one of:
the time of collection of the sample label;
the sample tags the events that occurred at or before the time of collection.
10. A flow prediction device, comprising: a preparation module and a prediction module;
the preparation module is used for forming the sample data characteristics of the traffic to be predicted according to the time sequence of the network traffic within the preset time before the time to be predicted; acquiring attribute information corresponding to the flow to be predicted, and converting the attribute information into sample attribute characteristics;
the prediction module is used for inputting the sample data characteristics of the flow to be predicted into a gated cyclic neural network of a flow prediction model; inputting the memory state output by the gated cyclic neural network and the sample attribute characteristics into a fully-connected network of the flow prediction model together for regression prediction to obtain predicted flow; the flow prediction model is obtained by training according to the training method of any one of claims 1 to 4.
11. An electronic device, comprising: a processor, a memory, and a communication bus;
the communication bus is used for realizing connection communication between the processor and the memory;
the processor is configured to execute one or more programs stored in the memory to implement the method for training a flow prediction model according to any one of claims 1-4 or to implement the method for flow prediction according to any one of claims 5-8.
12. A readable storage medium storing one or more programs, the one or more programs being executable by one or more processors to implement the method for training a flow prediction model according to any one of claims 1-4, or to implement the method for flow prediction according to any one of claims 5-8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011293151.2A CN112118143B (en) | 2020-11-18 | 2020-11-18 | Traffic prediction model training method, traffic prediction method, device, equipment and medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011293151.2A CN112118143B (en) | 2020-11-18 | 2020-11-18 | Traffic prediction model training method, traffic prediction method, device, equipment and medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112118143A CN112118143A (en) | 2020-12-22 |
| CN112118143B true CN112118143B (en) | 2021-02-19 |
Family
ID=73794647
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011293151.2A Active CN112118143B (en) | 2020-11-18 | 2020-11-18 | Traffic prediction model training method, traffic prediction method, device, equipment and medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112118143B (en) |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113017831A (en) * | 2021-02-26 | 2021-06-25 | 上海鹰瞳医疗科技有限公司 | Method and equipment for predicting arch height after artificial lens implantation |
| CN113139643A (en) * | 2021-03-09 | 2021-07-20 | 卓望数码技术（深圳）有限公司 | Network card flow model construction method, flow prediction method, equipment and storage medium |
| CN113408609A (en) * | 2021-06-17 | 2021-09-17 | 武汉卓尔信息科技有限公司 | Network attack detection method and system |
| CN113746696A (en) * | 2021-08-02 | 2021-12-03 | 中移（杭州）信息技术有限公司 | Network flow prediction method, equipment, storage medium and device |
| CN113595798B (en) * | 2021-08-02 | 2023-06-30 | 湖北工业大学 | Network flow prediction method and system for improving lightning connection process optimization algorithm |
| CN114285728B (en) * | 2021-12-27 | 2024-02-02 | 中国电信股份有限公司 | Predictive model training method, traffic prediction device and storage medium |
| CN114403486B (en) * | 2022-02-17 | 2022-11-22 | 四川大学 | Intelligent control method of airflow type cut-tobacco drier based on local peak value coding circulation network |
| CN115442246B (en) * | 2022-08-31 | 2023-09-26 | 武汉烽火技术服务有限公司 | Traffic prediction method, device, equipment and storage medium of data plane network |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107230351A (en) * | 2017-07-18 | 2017-10-03 | 福州大学 | A kind of Short-time Traffic Flow Forecasting Methods based on deep learning |
| CN108051035A (en) * | 2017-10-24 | 2018-05-18 | 清华大学 | The pipe network model recognition methods of neural network model based on gating cycle unit |
| CN109034449A (en) * | 2018-06-14 | 2018-12-18 | 华南理工大学 | Short-term bus passenger flow prediction technique based on deep learning and passenger behavior mode |
| CN109754113A (en) * | 2018-11-29 | 2019-05-14 | 南京邮电大学 | Load forecasting method based on dynamic time warping Yu length time memory |
| CN110288157A (en) * | 2019-06-27 | 2019-09-27 | 电子科技大学 | A kind of Runoff Forecast method based on attention mechanism and LSTM |
| CN110633871A (en) * | 2019-09-25 | 2019-12-31 | 大连理工大学 | Regional traffic demand prediction method based on convolution long-term and short-term memory network |
| CN110782663A (en) * | 2019-09-30 | 2020-02-11 | 电子科技大学 | Road network traffic flow short-time prediction method combining time-space characteristics |
| CN110991724A (en) * | 2019-11-27 | 2020-04-10 | 合肥工业大学 | Method, system and storage medium for predicting scenic spot passenger flow |
| CN111371644A (en) * | 2020-02-28 | 2020-07-03 | 山东工商学院 | Multi-domain SDN network traffic situation prediction method and system based on GRU |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109816095B (en) * | 2019-01-14 | 2023-04-07 | 湖南大学 | Network flow prediction method based on improved gated cyclic neural network |
| CN111597961B (en) * | 2020-05-13 | 2023-04-25 | 中国科学院自动化研究所 | Intelligent driving-oriented moving target track prediction method, system and device |
| CN111738500B (en) * | 2020-06-11 | 2024-01-12 | 大连海事大学 | Navigation time prediction method and device based on deep learning |
| CN111861028A (en) * | 2020-07-29 | 2020-10-30 | 浙江工业大学 | Method for predicting crime number based on spatio-temporal data fusion |
-
2020
- 2020-11-18 CN CN202011293151.2A patent/CN112118143B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107230351A (en) * | 2017-07-18 | 2017-10-03 | 福州大学 | A kind of Short-time Traffic Flow Forecasting Methods based on deep learning |
| CN108051035A (en) * | 2017-10-24 | 2018-05-18 | 清华大学 | The pipe network model recognition methods of neural network model based on gating cycle unit |
| CN109034449A (en) * | 2018-06-14 | 2018-12-18 | 华南理工大学 | Short-term bus passenger flow prediction technique based on deep learning and passenger behavior mode |
| CN109754113A (en) * | 2018-11-29 | 2019-05-14 | 南京邮电大学 | Load forecasting method based on dynamic time warping Yu length time memory |
| CN110288157A (en) * | 2019-06-27 | 2019-09-27 | 电子科技大学 | A kind of Runoff Forecast method based on attention mechanism and LSTM |
| CN110633871A (en) * | 2019-09-25 | 2019-12-31 | 大连理工大学 | Regional traffic demand prediction method based on convolution long-term and short-term memory network |
| CN110782663A (en) * | 2019-09-30 | 2020-02-11 | 电子科技大学 | Road network traffic flow short-time prediction method combining time-space characteristics |
| CN110991724A (en) * | 2019-11-27 | 2020-04-10 | 合肥工业大学 | Method, system and storage medium for predicting scenic spot passenger flow |
| CN111371644A (en) * | 2020-02-28 | 2020-07-03 | 山东工商学院 | Multi-domain SDN network traffic situation prediction method and system based on GRU |
Non-Patent Citations (3)
| Title |
|---|
| 基于LSTM与传统神经网络的网络流量预测及应用;王海宁;《移动通信》;20190815;第38-40页 * |
| 基于多特性分析的短时交通流量预测研究;徐璐;《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》;20200215;全文 * |
| 王海宁.基于LSTM与传统神经网络的网络流量预测及应用.《移动通信》.2019, * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112118143A (en) | 2020-12-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112118143B (en) | Traffic prediction model training method, traffic prediction method, device, equipment and medium | |
| CN111176953B (en) | Abnormality detection and model training method, computer equipment and storage medium | |
| WO2021103823A1 (en) | Model update system, model update method, and related device | |
| CN116028315A (en) | Operation early warning method, device, medium and electronic equipment | |
| US11197175B2 (en) | Forcasting time series data | |
| CN111160959B (en) | User click conversion prediction method and device | |
| CN116489038A (en) | Network traffic prediction method, device, equipment and medium | |
| CN115294397A (en) | Classification task post-processing method, device, equipment and storage medium | |
| CN115794578A (en) | Data management method, device, equipment and medium for power system | |
| CN117041017B (en) | Intelligent operation and maintenance management method and system for data center | |
| CN112651534A (en) | Method, device and storage medium for predicting resource supply chain demand | |
| CN117131457B (en) | AI model-based electric power big data acquisition and processing method and system | |
| CN114625477A (en) | Service node capacity adjusting method, equipment and computer readable storage medium | |
| CN114548493A (en) | Method and system for predicting current overload of electric energy meter | |
| CN114090393A (en) | Method, device and equipment for determining alarm level | |
| CN113886454A (en) | Cloud resource prediction method based on LSTM-RBF | |
| CN116937645A (en) | Charging station cluster regulation potential evaluation method, device, equipment and medium | |
| CN115018212B (en) | Power generation water consumption prediction analysis method and system and cloud platform | |
| CN110717577A (en) | Time series prediction model construction method for noting regional information similarity | |
| Peng et al. | Stock price prediction based on recurrent neural network with long short-term memory units | |
| CN115562940A (en) | Load energy consumption monitoring method and device, medium and electronic equipment | |
| CN115459982A (en) | Power network false data injection attack detection method | |
| CN111210361B (en) | Power communication network routing planning method based on reliability prediction and particle swarm optimization | |
| CN115314402A (en) | Network element load monitoring method and device, storage medium and electronic equipment | |
| CN113011674A (en) | Photovoltaic power generation prediction method and device, electronic equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |