Multi-path routing system based on flow prediction and method thereof

Abstract

The invention relates to the technical field of computers, in particular to a multi-path routing system and system based on traffic prediction, and the system comprises a traffic prediction module which is used for obtaining historical data of network traffic; predicting the network flow in a period of time in the future based on the historical data; the route selection module is in communication connection with the traffic prediction module and receives predicted traffic data sent by the traffic prediction module; a plurality of paths meeting the bandwidth requirement are selected for data transmission; the data transmission module is in communication connection with the route selection module and executes data transmission according to the multiple paths selected by the route selection module; the network monitoring module is in communication connection with the data transmission module and monitors data traffic in a network in real time; sending the real-time flow data to a routing selection module; the route adjustment module is in communication connection with the route selection module and the network monitoring module and is used for receiving the real-time flow data sent by the network monitoring module; and dynamically adjusting the path selected by the routing module based on the real-time flow data.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and more specifically to a multi-path routing selection system based on traffic prediction and a system thereof. Background Art

[0002] As the scale of networks continues to expand and become more complex, traditional single-path routing methods can no longer meet the needs of modern networks for efficiency, reliability, and flexibility. In recent years, multi-path routing technology has attracted widespread attention because of its ability to improve network resource utilization and enhance network robustness. However, existing multi-path routing methods still have many shortcomings.

[0003] First, most existing methods make routing decisions based only on the current network status and lack the ability to predict future network traffic. This results in routing selection often lagging behind network changes and being unable to effectively cope with burst traffic or periodic traffic fluctuations. Second, existing routing selection algorithms usually only consider a single performance indicator (such as the shortest path or minimum delay) and have difficulty balancing multiple performance objectives. Furthermore, many methods use global optimization strategies when adjusting routing, which has high computational complexity and is difficult to adapt to the real-time needs of large-scale networks.

[0004] In addition, existing technologies also face the following problems: insufficient traffic prediction accuracy, making it difficult to accurately grasp the dynamic changes in network traffic; lack of flexibility in routing selection, making it impossible to dynamically adjust strategies based on the needs of different applications; network monitoring granularity is coarse, making it difficult to detect and respond to local network anomalies in a timely manner; and the routing adjustment mechanism is too simple, which can easily cause network shocks due to frequent adjustments. Summary of the invention

[0005] In response to these problems, the present invention proposes a multi-path routing selection system and method based on traffic prediction. The system achieves accurate prediction of network traffic by introducing advanced machine learning algorithms; adopts multi-objective optimization methods to seek the best balance between bandwidth utilization, delay minimization and load balancing; and designs a distributed and adaptive routing adjustment mechanism that can quickly respond to network changes while maintaining overall stability.

[0006] The present invention provides a multi-path routing selection system based on traffic prediction, comprising:

[0007] Traffic prediction module, used for:

[0008] Get historical data on network traffic;

[0009] Based on the historical data, predict the network traffic in a future period of time;

[0010] A routing selection module, which is in communication with the traffic prediction module, is used to:

[0011] Receiving predicted traffic data sent by the traffic prediction module;

[0012] Based on the predicted traffic data, select multiple paths that meet bandwidth requirements for data transmission;

[0013] A data transmission module, in communication with the routing selection module, is used to:

[0014] performing data transmission according to the plurality of paths selected by the routing selection module;

[0015] A network monitoring module is communicatively connected with the data transmission module and is used to:

[0016] Real-time monitoring of data traffic in the network;

[0017] Sending real-time traffic data to the routing module;

[0018] A route adjustment module is connected in communication with the route selection module and the network monitoring module, and is used to:

[0019] Receiving real-time traffic data sent by the network monitoring module;

[0020] Based on the real-time traffic data, the path selected by the routing module is dynamically adjusted.

[0021] Preferably, the traffic prediction module comprises:

[0022] A data preprocessing unit, used for filtering, smoothing, denoising, normalizing and differential processing the historical data;

[0023] A model training unit, used for training a machine learning model based on the preprocessed historical data;

[0024] The traffic prediction unit is used to use the trained machine learning model to predict the network traffic in the future.

[0025] Preferably, the machine learning model includes at least one of a linear regression model, a recurrent neural network model and a convolutional neural network model.

[0026] Preferably, the routing selection module comprises:

[0027] A topology construction unit is used to construct a network topology graph G(N,Q,E), where N is a node set, Q is an edge attribute set, and E is an edge set;

[0028] A path calculation unit, configured to calculate a plurality of paths that meet bandwidth requirements based on the network topology diagram and the predicted traffic data;

[0029] The routing table updating unit is used to update the routing table according to the calculated multiple paths.

[0030] Preferably, the path calculation unit uses an improved Dijkstra algorithm to calculate multiple paths.

[0031] Preferably, the network monitoring module comprises:

[0032] Traffic collection unit, used to collect data traffic in the network in real time;

[0033] A threshold judgment unit, used to judge whether the real-time collected data flow exceeds a preset threshold;

[0034] The trigger unit is used to trigger the routing adjustment module to perform path adjustment when the real-time collected data flow exceeds a preset threshold.

[0035] Preferably, the routing adjustment module includes:

[0036] The edge set maintenance unit is used to maintain a network link set Q that needs special attention;

[0037] The average flow calculation unit is used to calculate the average value of the current network flow;

[0038] The adjustment strategy generating unit is used to generate a routing adjustment strategy based on the edge set Q and the average traffic.

[0039] Preferably, the adjustment strategy generating unit generates the routing adjustment strategy through the following steps:

[0040] When the real-time flow value is greater than the threshold Δs, the average value of the new time series value is calculated;

[0041] Select any edge e in the current edge set and determine whether the average flow value v of e is greater than v*+Δc;

[0042] If v is greater than v*+Δc, add the edge e to the edge set Q;

[0043] If v is not greater than v*+Δc, e is deleted from the edge set Q.

[0044] Preferably, the system is deployed on a cloud platform node cluster, and each node obtains the network topology structure, historical data of network traffic and the shortest path between any two nodes on the network topology map through a cloud platform interface.

[0045] A multi-path routing selection method based on traffic prediction comprises the following steps:

[0046] Get historical data on network traffic;

[0047] Based on the historical data, predict the network traffic in a future period of time;

[0048] Based on the predicted network traffic, multiple paths that meet bandwidth requirements are selected for data transmission;

[0049] performing data transmission according to the selected multiple paths;

[0050] Real-time monitoring of data traffic in the network;

[0051] The selected path is dynamically adjusted based on the real-time monitored data traffic.

[0052] The beneficial effects of the present invention are mainly reflected in the following aspects:

[0053] At the macro level, the system significantly improves the overall performance and resource utilization of the network. Through accurate traffic prediction and intelligent routing selection, the system can more effectively allocate network resources, reduce congestion, and improve throughput. At the same time, the multi-path transmission strategy enhances the fault tolerance and reliability of the network, ensuring continuous data transmission even in the event of partial link failure.

[0054] From a microscopic perspective, the present invention has achieved innovation and breakthroughs in multiple key links. In terms of traffic prediction, an integrated learning method is used to integrate the advantages of multiple prediction models, greatly improving the accuracy and robustness of the prediction. In the routing selection algorithm, the improved Dijkstra algorithm and NSGA-II multi-objective optimization algorithm are introduced, which not only improves the computational efficiency, but also provides multiple Pareto optimal solutions for decision-making, increasing the flexibility of selection.

[0055] The various modules of the present invention exhibit good synergy. For example, the traffic prediction module provides forward-looking data support for the routing selection module, so that routing decisions can respond to potential network changes in advance. The cooperation of the network monitoring module and the routing adjustment module realizes real-time perception and rapid response to the network status, forming a closed-loop adaptive system.

[0056] In addition, the present invention solves a number of technical problems. For example, by introducing edge set Q and multi-level threshold mechanism, the contradiction between routing stability and sensitivity is effectively balanced. The use of microservice architecture and container technology solves the problems of system deployment and scalability. These innovations not only bring performance improvements, but also enhance the maintainability and scalability of the system.

[0057] In general, the present invention achieves superposition and complementarity of effects through the organic combination of multiple innovative points. It not only promotes the development of multi-path routing technology in theory, but also provides a comprehensive and efficient solution for traffic management and resource optimization in large-scale, dynamic and complex network environments in practical applications. This is of great significance for improving the performance and reliability of modern network infrastructure and supporting emerging network-intensive applications (such as large-scale IoT, edge computing, etc.). BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 It is a system logic block diagram of the present invention. DETAILED DESCRIPTION

[0059] Please refer to Figure 1 The present invention provides a multi-path routing selection system and method based on traffic prediction. The system includes a traffic prediction module 1, a routing selection module 2, a data transmission module 3, a network monitoring module 4 and a routing adjustment module 5. These modules work together to realize intelligent routing selection and dynamic adjustment based on traffic prediction, thereby improving network performance and resource utilization.

[0060] First, the traffic prediction module 1 is used to obtain historical data of network traffic and predict network traffic in the future based on these historical data. Specifically, the traffic prediction module 1 uses advanced machine learning algorithms to analyze and model historical traffic data. Preferably, the present invention uses a time series prediction model, such as an autoregressive integrated moving average model (ARIMA) or a long short-term memory network (LSTM), to capture the time dependency and periodicity characteristics of network traffic. For example, the mathematical expression of the ARIMA model can be expressed as:

[0061] φ(B)(1-B) d X t =θ(B)ε t

[0062] Among them, X t is the flow value at time t, B is the lag operator, φ(B) and \θ(B) are the autoregressive and moving average polynomials respectively, d is the difference order, ε t is white noise. By fitting the model to historical data, we can get the predicted value of future traffic.

[0063] The routing module 2 is connected to the traffic prediction module 1 for receiving the predicted traffic data sent by the traffic prediction module 1. Based on the predicted data, the routing module 2 selects multiple paths that meet the bandwidth requirements for data transmission. In one embodiment of the present invention, the routing module 2 uses an improved Dijkstra algorithm to calculate multiple paths. The algorithm not only considers the distance of the path, but also incorporates the predicted traffic as a weight factor into the calculation. Specifically, the path weight can be expressed as:

[0064] w ij =α·d ij +β·f ij

[0065] Among them, w ij is the path weight from node i to j, d ij is the physical distance, f ij To predict traffic, α and β are trade-off factors. By adjusting the values ​​of α and β, a balance can be achieved between distance and traffic prediction. For example, when the network load is light, α = 0.7 and β = 0.3 can be set to consider physical distance more; when the network load is heavy, α = 0.3 and β = 0.7 can be set to consider traffic prediction more.

[0066] The data transmission module 3 is in communication connection with the routing selection module 2, and is used to perform data transmission according to the multiple paths selected by the routing selection module 2. The present invention adopts a multi-path parallel transmission strategy to divide the data stream into multiple sub-streams and transmit them simultaneously through different paths. This method not only improves the transmission efficiency, but also enhances the fault tolerance of the network. Preferably, the data segmentation adopts a dynamic weighted polling algorithm to dynamically adjust the allocation ratio according to the real-time bandwidth and delay of each path.

[0067] The network monitoring module 4 is connected to the data transmission module 3 for real-time monitoring of data traffic in the network and sending real-time traffic data to the routing module 2. The present invention adopts a distributed monitoring architecture, deploys monitoring probes at key nodes of the network, and collects key indicators such as traffic, delay, and packet loss rate. In order to reduce monitoring overhead, an adaptive sampling technology is used to dynamically adjust the sampling frequency according to the network status. For example, when the network status is stable, the sampling interval can be set to 5 minutes; when a sudden change in traffic is detected, the sampling interval can be shortened to 1 minute to more accurately capture network changes.

[0068] The route adjustment module 5 is connected to the route selection module 2 and the network monitoring module 4 in communication, and is used to receive the real-time traffic data sent by the network monitoring module 4, and dynamically adjust the path selected by the route selection module 2 based on these data. The present invention adopts a threshold-based trigger mechanism and a smooth adjustment strategy to avoid the impact of frequent route changes on network stability. Specifically, when the deviation between the real-time traffic and the predicted traffic exceeds a preset threshold (for example, 20%), the route adjustment process is triggered. During the adjustment, the exponentially weighted moving average method is used to calculate the new route weight, and the formula is as follows:

[0069] w new =γ·w old +(1-γ)·w current

[0070] Among them, w new is the new routing weight, w old is the old routing weight, w current is the currently calculated weight, and γ is the smoothing factor. By setting the γ value appropriately (e.g. 0.8), a balance can be achieved between fast response and stability.

[0071] The traffic prediction module 1 of the present invention also includes a data preprocessing unit, a model training unit and a traffic prediction unit. The data preprocessing unit filters, smoothes, de-noises, normalizes and differentially processes the historical data to improve the data quality. For example, median filtering is used to remove outliers, wavelet transform is used for denoising, and the data is scaled to the [0,1] interval by maximum and minimum value normalization. The model training unit trains the machine learning model based on the preprocessed historical data. The present invention preferably uses a cross-validation method to evaluate the model performance to avoid overfitting. Specifically, a 5-fold cross-validation can be used to divide the data set into 5 parts, and 4 parts are used in turn as training sets and 1 part as a validation set. The traffic prediction unit uses the trained machine learning model to predict the network traffic in the future. The selection of the prediction time window requires a balance between prediction accuracy and practicality. The present invention recommends using a prediction window of 15 minutes to 1 hour, which can capture short-term traffic changes without accumulating excessive errors due to long-term predictions.

[0072] In a preferred embodiment of the present invention, the machine learning model includes at least one of a linear regression model, a recurrent neural network model, and a convolutional neural network model. The linear regression model is suitable for capturing simple linear trends, has a fast calculation speed, and is suitable as a baseline model. The recurrent neural network model, especially the long short-term memory network (LSTM), can effectively capture the long-term dependencies of time series and is suitable for processing network traffic data with periodicity and long-term trends. The convolutional neural network model is good at extracting local features and multi-scale patterns, and is suitable for processing traffic data with complex spatiotemporal correlations. The present invention adopts the idea of ​​ensemble learning to weightedly fuse the prediction results of these models to obtain more accurate and robust prediction results. The weights can be dynamically adjusted based on the performance on the validation set, for example, using an additive weight optimization algorithm:

[0073]

[0074] in, is the weight of the i-th model at the t-th iteration, y is the true value, is the predicted value of the i-th model, and η is the learning rate (which can be set to 0.01). In this way, the model with good prediction performance will obtain a higher weight, thereby improving the overall prediction accuracy. The routing selection module 2 of the present invention includes a topology construction unit 21, a path calculation unit 22 and a routing table update unit 23. These units work together to achieve intelligent routing selection based on network topology and predicted traffic.

[0075] The topology construction unit 21 is used to construct a network topology graph G (N, Q, E), where N is a node set, Q is an edge attribute set, and E is an edge set. In a preferred embodiment of the present invention, the topology construction unit 21 adopts a distributed topology discovery algorithm to dynamically update the topology structure by exchanging link status information among network nodes. This method can timely reflect network changes and improve the accuracy of routing selection. Specifically, each node periodically broadcasts its own link status information to neighboring nodes, including connected neighboring nodes, link bandwidth, delay and other attributes. Nodes that receive the broadcast information will update local topology information and further propagate it. In order to reduce broadcast storms, the present invention adopts an update mechanism based on serial numbers, and forwards only when the received information is newer than the locally stored information.

[0076] The path calculation unit 22 calculates multiple paths that meet the bandwidth requirements based on the network topology diagram and the predicted traffic data. The system of the present invention adopts an improved multi-constraint path algorithm, taking into account multiple constraints such as bandwidth, delay, and packet loss rate. Preferably, the path calculation adopts a heuristic search strategy to balance the calculation efficiency and path quality. For example, a variant of the A* algorithm can be used, in which the heuristic function h(n) is defined as:

[0077]

[0078] Where d(n,t) is the Euclidean distance from node n to target node t, b(n) is the available bandwidth of node n, l(n) is the current load of node n, and d max 、b max and b max are the maximum distance, bandwidth, and load, respectively. α, β, and γ are trade-off factors that can be adjusted dynamically according to network characteristics. For example, in a high-load situation, the value of γ can be increased to give more consideration to load balancing.

[0079] The routing table updating unit 23 updates the routing table according to the calculated multiple paths. The present invention adopts an incremental update strategy and only modifies the changed routing entries to reduce the update overhead. Preferably, the routing table adopts a hierarchical structure to divide the network into multiple areas, each area adopts detailed routing, and the areas adopt aggregate routing. This method can significantly reduce the size of the routing table and improve the search efficiency. For example, CIDR (Classless Inter-Domain Routing) technology can be used for address aggregation, and multiple continuous IP address segments can be represented by one routing entry.

[0080] In one embodiment of the present invention, the path calculation unit 22 uses an improved Dijkstra algorithm to calculate multiple paths. The algorithm not only considers a single shortest path, but also finds multiple candidate paths. In specific implementation, the algorithm maintains a set of candidate paths instead of a single shortest path in the traditional Dijkstra algorithm. In each iteration, not only the shortest path is updated, but also the suboptimal path that meets specific conditions is updated. For example, an expansion factor k can be defined to retain all paths whose distance does not exceed k times the shortest path. The pseudo code of the algorithm is as follows:

[0081]

[0082]

[0083] This improved Dijkstra algorithm can provide multiple alternative paths while ensuring the shortest path, increasing the flexibility and robustness of routing.

[0084] The network monitoring module 4 of the present invention comprises a flow collection unit 41, a threshold judgment unit 42 and a trigger unit 43. These units work together to achieve real-time monitoring and abnormality detection of network flow.

[0085] The flow collection unit 41 is used to collect data flow in the network in real time. The present invention adopts a distributed flow collection architecture and deploys flow collection probes at key nodes of the network. Preferably, a sampling-based flow measurement technology, such as sFlow or NetFlow, is adopted to reduce measurement overhead. For example, a sampling rate of 1 / 1000 can be set, that is, 1 packet is randomly selected for analysis for every 1000 packets. In order to improve the accuracy of the collection, the present invention also adopts an adaptive sampling strategy to dynamically adjust the sampling rate according to the flow change. Specifically, the exponentially weighted moving average (EWMA) method can be used to calculate the flow change rate:

[0086] EWMA t =α·FlowRate t +(1-α)·EWMA t-1

[0087] Among them, FlowRate t is the flow rate at time t, and α is the smoothing factor (which can be set to 0.3). When the EWMA value exceeds the preset threshold, the sampling rate is increased; when the EWMA value is lower than another threshold, the sampling rate is reduced.

[0088] The threshold judgment unit 42 is used to judge whether the data flow collected in real time exceeds a preset threshold. The present invention adopts a multi-level threshold mechanism to set a warning threshold and an emergency threshold. Preferably, these thresholds are not fixed, but dynamically adjusted according to historical data. For example, the moving percentile method can be used to calculate the threshold:

[0089] Threshold=Q3+k·IQR

[0090] Among them, Q3 is the 75% quantile of the flow data, IQR is the interquartile range, and k is the coefficient (the warning threshold can be set to k=1.5, and the emergency threshold can be set to k=3). This method can adapt to the long-term change trend of flow and reduce false alarms.

[0091] The trigger unit 43 is used to trigger the routing adjustment module 5 to perform path adjustment when the real-time collected data flow exceeds the preset threshold. The present invention adopts a rule-based trigger mechanism, combining the duration and degree of flow exceeding the threshold to determine whether to trigger adjustment. For example, the following rules can be defined:

[0092] 1. If the traffic exceeds the warning threshold for 5 minutes, a slight adjustment is triggered;

[0093] 2. If the traffic exceeds the emergency threshold, emergency adjustment is triggered immediately;

[0094] 3. If the warning level exceeds the threshold more than three times within 30 minutes, a moderate adjustment is triggered.

[0095] This multi-level trigger mechanism can adopt corresponding adjustment strategies according to the severity of network conditions, which can not only respond to emergencies in a timely manner, but also avoid frequent adjustments caused by short-term fluctuations.

[0096] Through the above design, the network monitoring module 4 of the present invention can monitor the network traffic status in real time and accurately, provide a reliable basis for route adjustment, and thus improve the responsiveness and adaptability of the entire system. The route adjustment module 5 of the present invention includes an edge set maintenance unit 51, an average traffic calculation unit 52, and an adjustment strategy generation unit 53. These units work together to achieve dynamic adjustment and optimization of network routes.

[0097] The edge set maintenance unit 51 is used to maintain a network link set Q that needs to be focused on. In a preferred embodiment of the present invention, the set Q adopts a data structure of a priority queue and is dynamically sorted according to the importance and load of the link. Preferably, the importance of the link can be evaluated by its centrality measure in the network, for example, using the Betweenness Centrality indicator:

[0098]

[0099] Among them, σ st is the number of shortest paths from node s to node t, σ st (e) is the number of shortest paths from s to t that contain edge e. The higher the BC(e) value, the greater the importance of the link in the network.

[0100] The average flow calculation unit 52 is used to calculate the average value of the current network flow. The present invention uses a sliding window method to calculate the average flow to smooth short-term fluctuations and capture recent trends. Specifically, the exponentially weighted moving average (EWMA) method can be used:

[0101] EWMA t =α·F t +(1-α)·EWMA t-1

[0102] Among them, F t is the actual flow at time t, and α is the smoothing factor (which can be set to 0.2). This method can not only reflect the current flow status, but also take into account historical trends, avoiding frequent adjustments caused by instantaneous fluctuations.

[0103] The adjustment strategy generation unit 53 generates a routing adjustment strategy based on the edge set Q and the average traffic. The present invention adopts an adaptive adjustment algorithm to dynamically adjust the route according to the degree of change in the network status. Specifically, the adjustment strategy generation unit 53 generates a routing adjustment strategy through the following steps:

[0104] First, when the real-time flow value is greater than the threshold Δs, the average value of the new time series value is calculated. The selection of the threshold Δs requires a trade-off between sensitivity and stability. The present invention recommends setting Δs to 1.2 times the historical average flow. This can filter out small fluctuations and only respond to significant changes.

[0105] Secondly, select any edge e in the current edge set and determine whether the average flow value v of e is greater than v*+Δc. Where v* is the historical average flow of the edge, and Δc is the tolerance deviation. Preferably, Δc can be set to 15% of v*, which can detect obvious load increases while not being overly sensitive to normal fluctuations.

[0106] If v is greater than v*+Δc, the edge e is added to the edge set Q. This indicates that the load of the link has increased significantly, requiring special attention and possible adjustment. At the same time, the present invention also considers the historical performance of the link and introduces an aging mechanism. For example, a counter can be set for each link added to Q, which is reduced by 1 in each adjustment cycle and removed from Q when the counter returns to zero. This can avoid frequent adjustments caused by short-term fluctuations.

[0107] If v is not greater than v*+Δc, e is deleted from the edge set Q (if e was originally in Q). This indicates that the load of the link has returned to normal level and no special attention is needed.

[0108] Based on the above judgment results, the adjustment strategy generation unit 53 generates a specific routing adjustment strategy. Preferably, the present invention adopts a gradual adjustment method to avoid the impact of drastic changes on the network. For example, the following formula can be used to calculate the new link weight:

[0109] w new =w old +β·(vv * ) / v *

[0110] Among them, w new and w old are the weights of the new and old links respectively, and β is the adjustment coefficient (which can be set to 0.1). This method can gradually adjust the routing according to the degree of load change, which can not only respond to changes in network conditions in a timely manner, but also maintain the overall stability of the network.

[0111] The system of the present invention is deployed on a cloud platform node group, and each node obtains the network topology structure, historical data of network traffic, and the shortest path between any two nodes on the network topology map through the cloud platform interface. This distributed deployment method has good scalability and fault tolerance, and can adapt to the needs of large-scale complex networks.

[0112] In a preferred embodiment of the present invention, the system adopts a microservice architecture to encapsulate each functional module as an independent service. This design makes the system more flexible, easy to expand and maintain. For example, the traffic prediction module 1 can be used as an independent prediction service for other modules to call. At the same time, the present invention also uses container technology (such as Docker) to deploy these microservices to improve resource utilization and deployment efficiency.

[0113] In order to ensure the high availability of the system, the present invention deploys key components in a master-slave replication manner. For example, the routing module 2 can deploy replicas on multiple nodes, use a consistent hashing algorithm to distribute the load, and use the Raft protocol to ensure the state consistency between the replicas. In this way, even if a node fails, the system can continue to operate normally.

[0114] The present invention also provides a multi-path routing selection method based on traffic prediction, comprising the following steps:

[0115] First, obtain historical data of network traffic. This method uses multi-source data fusion technology to comprehensively utilize network device logs, traffic collector data, and application layer statistical information. Preferably, use the ETL (Extract, Transform, Load) process to clean and integrate the original data to remove outliers and redundant information.

[0116] Secondly, based on historical data, predict network traffic in the future. This method adopts the idea of ​​ensemble learning and combines the advantages of multiple prediction models. Specifically, the ARIMA model can be used to capture linear trends, the LSTM network can capture long-term dependencies, and the Prophet model can handle periodic patterns. Then, the prediction results of each model are fused by weighted averaging, and the weights can be dynamically adjusted according to the performance of each model on the validation set.

[0117] Next, based on the predicted network traffic, multiple paths that meet the bandwidth requirements are selected for data transmission. This method uses a multi-objective optimization algorithm to balance multiple performance indicators, such as bandwidth utilization, end-to-end delay, and load balancing. Preferably, the NSGA-II (Non-dominated Sorting Genetic Algorithm II) algorithm is used to solve the Pareto optimal solution set and provide multiple candidate solutions for decision making.

[0118] Then, data transmission is performed according to the selected multiple paths. This method adopts an adaptive load balancing strategy to dynamically adjust the data allocation ratio of each path according to the real-time network status. Specifically, a weighted round robin algorithm can be used, and the weight is proportional to the available bandwidth of the path.

[0119] At the same time, the data traffic in the network is monitored in real time. The method uses a distributed traffic monitoring system and deploys monitoring probes at key nodes of the network. Preferably, an anomaly detection algorithm based on machine learning, such as Isolation Forest, is used to identify abnormal traffic patterns.

[0120] Finally, based on the real-time monitored data traffic, the selected path is dynamically adjusted. This method adopts the idea of ​​reinforcement learning and models the route adjustment problem as a Markov decision process (MDP). Specifically, the Deep Q-Network (DQN) algorithm can be used to optimize the adjustment strategy through continuous interaction with the environment. The state space includes information such as the current network topology and link load, the action space is the possible route adjustment operations, and the reward function is designed based on network performance indicators (such as throughput and latency).

[0121] Through the above steps, the method of the present invention can adaptively select and adjust multiple transmission paths according to traffic prediction and real-time monitoring results, thereby improving the overall performance and resource utilization of the network. This method is particularly suitable for large-scale, dynamically changing network environments, such as data center networks, wide area networks, and 5G core networks.

[0122] It should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modification, replacement, and improvement made within the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A multi-path routing selection system based on traffic prediction, characterized in that: include: Traffic prediction module, used for: Get historical data on network traffic; Based on the historical data, predict the network traffic in a future period of time; A routing selection module, which is in communication with the traffic prediction module, is used to: Receiving predicted traffic data sent by the traffic prediction module; Based on the predicted traffic data, select multiple paths that meet bandwidth requirements for data transmission; A data transmission module, in communication with the routing selection module, is used to: performing data transmission according to the plurality of paths selected by the routing selection module; A network monitoring module is communicatively connected with the data transmission module and is used to: Real-time monitoring of data traffic in the network; Sending real-time traffic data to the routing module; A route adjustment module is connected in communication with the route selection module and the network monitoring module, and is used to: Receiving real-time traffic data sent by the network monitoring module; Based on the real-time traffic data, the path selected by the routing module is dynamically adjusted.

2. The system according to claim 1, characterized in that The traffic prediction module includes: A data preprocessing unit, used for filtering, smoothing, denoising, normalizing and differential processing the historical data; A model training unit, used for training a machine learning model based on the preprocessed historical data; The traffic prediction unit is used to use the trained machine learning model to predict the network traffic in the future.

3. The system according to claim 2, characterized in that The machine learning model includes at least one of a linear regression model, a recurrent neural network model and a convolutional neural network model.

4. The system according to claim 1, characterized in that The routing module comprises: A topology construction unit is used to construct a network topology graph G(N,Q,E), where N is a node set, Q is an edge attribute set, and E is an edge set; A path calculation unit, configured to calculate a plurality of paths that meet bandwidth requirements based on the network topology diagram and the predicted traffic data; The routing table updating unit is used to update the routing table according to the calculated multiple paths.

5. The system according to claim 4, characterized in that The path calculation unit calculates multiple paths using an improved Dijkstra algorithm.

6. The system according to claim 1, characterized in that The network monitoring module includes: Traffic collection unit, used to collect data traffic in the network in real time; A threshold judgment unit, used to judge whether the real-time collected data flow exceeds a preset threshold; The trigger unit is used to trigger the routing adjustment module to perform path adjustment when the real-time collected data flow exceeds a preset threshold.

7. The system according to claim 1, characterized in that The routing adjustment module includes: The edge set maintenance unit is used to maintain a network link set Q that needs special attention; The average flow calculation unit is used to calculate the average value of the current network flow; The adjustment strategy generating unit is used to generate a routing adjustment strategy based on the edge set Q and the average traffic.

8. The system according to claim 7, characterized in that The adjustment strategy generating unit generates a routing adjustment strategy by the following steps: When the real-time flow value is greater than the threshold Δs, the average value of the new time series value is calculated; Select any edge e in the current edge set and determine whether the average flow value v of e is greater than v*+Δc; If v is greater than v*+Δc, add the edge e to the edge set Q; If v is not greater than v*+Δc, e is deleted from the edge set Q.

9. The system according to claim 1, characterized in that The system is deployed on a cloud platform node cluster, and each node obtains the network topology structure, historical data of network traffic, and the shortest path between any two nodes on the network topology map through the cloud platform interface.

10. A multi-path routing selection method based on traffic prediction, characterized in that: The following steps are involved: Get historical data on network traffic; Based on the historical data, predict the network traffic in a future period of time; Based on the predicted network traffic, multiple paths that meet bandwidth requirements are selected for data transmission; performing data transmission according to the selected multiple paths; Real-time monitoring of data traffic in the network; The selected path is dynamically adjusted based on the real-time monitored data traffic.

Images