Forecasting Models Using Planning Calendar Time Hierarchies
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing data processing technologies face challenges in accurately predicting time series patterns, particularly when using planning calendars that differ from the Gregorian calendar, leading to inconsistencies in forecasting and resource management in business operations.
Innovation Solution
A computer-implemented method that identifies cycle patterns in data observations using a planning calendar with a time hierarchy, allowing for the generation of predictive models that align with specific business calendars, enabling more accurate forecasting and resource scheduling by mapping dates to relevant hierarchy levels such as weeks, periods, and quarters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a standard Gregorian calendar is used for time series analysis, then the forecasting model is simple to implement, but the forecasting accuracy deteriorates when business operations follow a different planning calendar
Solution Approach 1:
The patent transforms the time series data by changing the temporal parameters from standard calendar dates to planning calendar periods. Each timestamp is converted to represent a specific period in the planning calendar hierarchy (year, quarter, period, week, day), allowing the forecasting model to align with business operations while maintaining mathematical consistency for pattern recognition.
Solution Approach 2:
The patent introduces a hierarchical time dimension structure with multiple levels (year, quarter, period, week, day) that overlays the standard calendar system. This additional dimensional layer allows the model to capture business-specific temporal patterns without losing the chronological ordering, effectively adding a 'planning calendar dimension' to the time series analysis.
2Adaptability or versatility
If a planning calendar with custom periods and sub-periods is implemented, then the alignment with business operations is improved, but the data processing complexity increases
Solution Approach 1:
The patent segments the time series data according to the planning calendar hierarchy, dividing the continuous time stream into discrete periods and sub-periods. Each segment is tagged with its position in the hierarchy (e.g., period number, quarter number, year), enabling flexible querying and analysis at different granularities while simplifying the processing of recurring patterns within each segment.
Solution Approach 2:
The patent introduces an intermediary transformation layer that converts between standard calendar timestamps and planning calendar period identifiers. This intermediary module handles the complex mapping logic, acting as a buffer between the raw data and the forecasting model, thereby isolating the complexity from the core prediction algorithm and making the system more maintainable.
3Measurement precision
If historical data is transformed to match planning calendar periods, then the pattern recognition accuracy is improved, but the data transformation time increases
Solution Approach 1:
The patent performs preliminary transformation of historical data during the data ingestion phase, converting timestamps to planning calendar period identifiers upfront. By pre-computing the period mappings and storing them with the historical data, the system avoids repeated transformation operations during forecasting, significantly reducing the time cost when patterns need to be recognized and analyzed.
Solution Approach 2:
The patent creates a copied and transformed version of the time series data where timestamps are replaced with planning calendar period identifiers. This copied dataset maintains the same structural relationships but in a format optimized for pattern recognition, allowing the forecasting model to work with the transformed data without requiring real-time conversion during prediction operations.
Data Source
AI summary
The present disclosure relates to computer-implemented methods, software, and systems for identifying cycle patterns in data observations collected as time series based on a planning calendar with a time hierarchy. Time series can include data observations associated with a respective date. Model variables derived based on each date associated with each data observation can be determined. The model variables can map a date of a data observation to an occurrence of the date according to a time hierarchy of a planning calendar. The time hierarchy can include periods of sub-periods as a hierarchy level, where each period is defined to comprise a number of sub-periods based on a week-based pattern of the planning calendar. A predictive model is generated for a predicted variable identified at the data observations based on the model variables. The predictive model is executed to predict values for the predicted variable over a time horizon.