AI-Powered Traffic Forecasting Program for Cacti-Based MPLS Uplinks

Abstract

The invention relates to a program used in the telecommunications and computer networks field, particularly in MPLS-based data communication networks, for uplink traffic monitoring, analysis, and forward capacity planning processes. It enables the analysis of network traffic data obtained from the Cacti monitoring platform and the generation of future traffic forecasts.

Description

1 TARIFF AI-Powered Traffic Forecasting Program for Cacti-Based MPLS Uplinks Technical Area 5 The invention has implications for telecommunications and computer networks, particularly for MPLS-based data communication. in uplink traffic monitoring, analysis and forward capacity planning processes in their networks By analyzing network traffic data obtained through the Cacti monitoring platform, It is related to a program that enables the generation of future traffic forecasts. State of the Art Currently, in telecommunications network management, uplink traffic monitoring and capacity are crucial. Planning processes are generally based on static reporting and human expert visual analysis. Cacti-based graphical analysis is commonly used, although classical time series analysis is also used, albeit rarely. is used. 15 Cacti collects data from network devices via the SNMP protocol and generates time series graphs. It generates these graphs. Network operators manually evaluate traffic trends based on these graphs. Although Cacti graphs visually present historical traffic values, they do not automatically advance forward traffic. No forecasting or planning support is available. SARIMA (Seasonal Autoregressive Integrated Moving Average) analyzes the past 20 years of time series. Holt Linear is a statistical method used to model trends and seasonality. The trend is forward-looking with the assumption of a linear trend for irregular or short time series. It is a method that can generate predictions. These models make forward-looking predictions based on past data. Although they can generate predictions, they have the following shortcomings: Lack of Appropriate Model Selection Based on Dataset Characteristics: 25 obtained from Cacti graphs MPLS uplink traffic data typically has a short time series. Models like SARIMA It attempts to capture patterns such as seasonality and trends, but the number of observations is small. Parameter estimation is not reliable. Simple models like Holt Linear only estimate linear parameters. It can capture increases / decreases, but it cannot adapt to sudden fluctuations or trend changes. The predictive reliability of current models on short-time series data is low, 30 It fails to adequately capture seasonal behavior and trend patterns. 2 Absence of Multiple Modeling and Selection Mechanisms: Only one model (e.g., only) When using SARIMA (or simply Holt Linear), a single profile suitable for all different uplink profiles is provided. A consistent method cannot be provided. For example: While certain uplinks exhibit a dominant trend, others show fluctuations. or erratic behaviors may be observed. A single model approach is suitable for different traffic characteristics. It cannot adapt and therefore causes incorrect projections on some uplinks. 5 Shortcomings of Adaptive Estimation with Learning Models: Current statistical models are classical and... It is fixed; their ability to spontaneously learn relationships within historical data is limited. For example... neural network-based methods (such as LSTM) can handle complex time dependencies within the data. Dynamically learning models are not included in current approaches. Complex and Automatic learning of short time series relationships is not supported, which means that going forward to 10 This leads to a loss of accuracy in forecasting and capacity planning. Lack of Automated Preprocessing and Data Cleaning Mechanisms: Traditional The approaches rely on manual processes working with raw data. The sum of multiple uplinks. Automatic calculation, identification of aggregations such as "Total uplinks", missing or out-of-control links. It does not include technical operations such as automatic extraction of data points. Analysis of raw network data 15 There is a preprocessing mechanism that enables automatic conversion to a usable format. It is not. Currently, network traffic trends can be determined based on a visual examination of Cacti charts. forecasting, reliable results due to the natural nature of short time series. Classical statistical models alone are not adaptable to different traffic profiles. It cannot provide and learning models are not included in the current technical approach, therefore traffic Reliable forward-looking predictions of trends cannot be made. These shortcomings, An automated, adaptive, and multi-model approach to traffic forecasting and analysis. This reveals the need for the system. Purpose of the Invention The main purpose of the invention is to monitor MPLS uplink traffic obtained via the Cacti monitoring platform. Automated data processing and forecasting adapted to short time series. The aim is to generate future traffic predictions using its architecture. Another purpose of the invention is to collect and analyze uplink traffic data on a node-by-node basis. converting traffic values ​​into time series and in structures with multiple uplinks It automatically combines the data. This eliminates the need for manual data preparation. is being removed. 3 Another purpose of the invention is to periodically receive uplink traffic data originating from Cacti over time. convert to series format and automatically determine the length and structure of the dataset. The goal is to evaluate the situation. In cases of insufficient data, an appropriate forecasting approach is selected. The generation of unstable model outputs is prevented; trends suitable for short time series are observed. In addition to forecasting methods based on historical traffic values, temporal 5 together artificial neural network-based prediction methods that can learn dependencies The goal is to use it. This allows for more stable and efficient uplink lines with different traffic characteristics. Reliable predictions are obtained; forward-looking forecasting is performed in a single step. Instead of performing each prediction, use each prediction result as input for the next prediction step. The goal is to create a step-by-step forecasting chain using this method, with a six-month forecast horizon of 10. This enables more realistic modeling of traffic trends throughout the region; the accuracy of predictions, the error between predicted values ​​and historical actual data It involves automatically evaluating using metrics. This includes the average and absolute errors. Metrics such as percentage, average, and absolute error are calculated by the system on a per-uplink basis. It is reported that it is not limited to a single uplink, but includes multiple MPLS uplink lines and node 15 Uplinks with different traffic profiles are designed to handle the same structure simultaneously. The aim is to enable comparable and scalable analysis of the lines. Figures that will help understand the invention. Figure 1 shows the general architecture of the system that is the subject of the invention. 20 Explanation of Part References 101. MPLS uplink traffic data obtained via the Cacti monitoring system. 102. Data collection module that enables the automatic acquisition of traffic data. 103. Preprocessing that enables time series generation and the merging of multiple uplink data. 25 module 104. Data sufficiency check module that evaluates the length and structure of the dataset. 105. Forecasting module that performs forecasting operations suitable for short time series. 106. Performance evaluation module that measures prediction accuracy using error metrics. 107. Output module 30 that presents prediction results graphically and numerically. 108. A system that intervenes when the data set is insufficient, alerts the user, and manages the process. terminating error / warning module 35 4 Detailed Description of the Invention This detailed description of the preferred configurations of the invention provides a better understanding of the subject matter. This is intended to facilitate understanding and will not impose any limiting effects. The invention is based on MPLS uplink traffic data (101) obtained via the Cacti monitoring system. It starts with MPLS uplink traffic data obtained through this Cacti monitoring system. (101) traffic measurements obtained, data enabling automatic acquisition of traffic data It is automatically collected at certain time intervals via the collection module (102). The raw traffic data obtained is used to create time series and combine data from multiple uplinks. The preprocessing module (103) converts the missing time series format into the time series format. or erroneous measurements are filtered out and a suitable data structure is created for forecasting. 10 The generated time series data includes data that evaluates the length and structure of the data set. The competency is analyzed by the control module (104) so ​​that predictions can be made. It is assessed whether the required minimum data length and continuity are met. In the event that the dataset is insufficient, the system that activates in cases of insufficient datasets. The error / warning module (108) is activated, alerting the user and terminating the process. It informs the user and outlines the process in order to prevent erroneous prediction results. ends it. If data sufficiency is ensured, the data can be used for forecasting suitable for short time series. The operations are transferred to the forecasting module (105) which performs the operations, and within this module 20 data on MPLS uplink traffic values ​​using time series forecasting algorithms Forward-looking predictions are being made. This forecasting process is carried out over a six-month period. This is being done gradually. The generated prediction results are presented in a performance database that measures prediction accuracy using error metrics. The accuracy of the estimation process is analyzed within the evaluation module (106) and Its stability is being measured. 25 In the final stage, the traffic forecasts and performance metrics obtained are used to support the forecast results. The output module (107) presents the graphical and numerical output to the user, This allows for the monitoring of MPLS uplink capacity usage trends for the future. is provided.

Claims

REQUESTS 1. Starting with MPLS uplink traffic data obtained via Cacti monitoring system (101), Data collection module (102) that enables the automatic acquisition of traffic data, time A preprocessing module that enables data series creation and the merging of multiple uplink data. (103), data adequacy check module 5 which evaluates the length and structure of the data set. (104), forecasting that performs forecasting operations suitable for short time series module (105), performance evaluation which measures prediction accuracy with error metrics module (106), output module (107) which presents the prediction results graphically and numerically, It intervenes when the data set is insufficient, alerts the user, and guides the process. It is a program that includes a terminating error / warning module (108), and its feature is; 10 MPLS uplink traffic data obtained via Cacti monitoring system (101), traffic through the data collection module (102) which enables the automatic collection of data automatic receipt at specific time intervals Time series generation of raw traffic data and multiple uplink data. Pre-processing module (103) that enables the merging of time series format 15 transformation, elimination of missing or erroneous measurements, and suitable data for forecasting. the creation of its structure, data that evaluates the length and structure of the dataset of time series data generated. The competency is analyzed by the control module (104) so ​​that predictions can be made. whether the required minimum data length and continuity are ensured 20 evaluation, in cases where data sufficiency cannot be ensured, or when the data set is insufficient. Error / warning module (108) that activates, alerts the user and terminates the process preventing the production of erroneous model outputs by informing the user. and termination of the process, 25 If data sufficiency is ensured, the data can be adapted to short time series. transfer to the forecasting module (105) which performs the forecasting operations and this MPLS uplink traffic is predicted using time series forecasting algorithms within the module. the creation of forward-looking predictions regarding their values, Performance 30 measures the accuracy of the predictions generated using error metrics. The estimation process is analyzed within the evaluation module (106). measuring its accuracy and stability, The traffic forecasts and performance metrics obtained are presented graphically and in terms of prediction results. the output module (107) which presents the output to the user in digital form It involves carrying out its operations. 35

Images