Wireless Network Traffic Forecasting via Iterative Residual Decomposition

The patent applies segmentation by decomposing the traffic time series into distinct components: intrinsic component (systematic patterns including seasonality and trends) and random component (residual variations). This is achieved through iterative approximation where functionals are applied to extract intrinsic patterns, and residuals are calculated to isolate random components. The segmentation enables separate modeling and forecasting of each component, improving overall forecasting accuracy.

The patent introduces functionals as intermediary mathematical tools that bridge the raw traffic data and the separated components. These functionals act as mediators that systematically extract intrinsic patterns from the mixed traffic signal, enabling the separation process. The functionals include operations such as averaging, differencing, and seasonal adjustment that transform the original data into component-separated form.

The patent applies periodic action by explicitly modeling and extracting seasonal patterns from the traffic data. Functionals are designed to identify and separate periodic components (such as daily, weekly, or monthly cycles) from the irregular high-resolution data. By extracting these periodic patterns, the method transforms irregular data into a form where systematic variations are isolated, improving predictability of the remaining components.

The patent segments the high-resolution traffic data into intrinsic components (including periodic seasonal patterns) and random components. This segmentation allows the systematic periodic patterns to be separately analyzed and forecasted, while the random components are handled differently, thereby improving overall predictability despite the high time resolution of the input data.

The patent applies preliminary action by performing component separation and functional approximation before the forecasting step. Instead of directly sampling and filtering the raw traffic data, the method first applies functionals to extract intrinsic components and separate random components. This preliminary processing preserves more information in the original data structure, allowing subsequent forecasting to work with richer, less degraded data.

The patent substitutes the conventional mechanical sampling and filtering system with a functional-based mathematical transformation approach. Instead of physically sampling data at discrete intervals and applying filters, the method uses functionals to systematically decompose and analyze the continuous traffic signal, preserving information that would be lost in traditional sampling processes.