Demand Forecast Adjustment via Error Removal
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Demand-sensing machine learning pipelines in production are difficult and costly to modify, making it challenging to incorporate timely information efficiently without resource-intensive changes.
Innovation Solution
A computing apparatus and method that receives historical data, cleans and generates features, trains a machine-learning model, forecasts data, collects real-time data, determines errors, and adjusts forecasts by removing errors, allowing for efficient updates without frequent retraining of the model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the machine learning pipeline is modified to incorporate timely information, then forecasting accuracy is improved, but implementation cost and resource consumption increase
Solution Approach 1:
The patent segments the machine learning pipeline into distinct modules: a trained model component that remains fixed, a real-time data collection component, an error determination component, and an adjustment component. This segmentation allows the system to incorporate timely information without requiring comprehensive pipeline modifications, thus improving forecasting accuracy while controlling implementation complexity
Solution Approach 2:
The patent performs preliminary actions by pre-training the machine learning model on historical data before deployment. This pre-trained model serves as a stable foundation that can be applied to new data without retraining, allowing timely information incorporation while avoiding the resource-intensive process of retraining the entire pipeline
2Measurement precision
If real-time data is incorporated frequently, then forecasting accuracy is improved, but computational resources and time consumption increase
Solution Approach 1:
The patent extracts only the essential adjustment mechanism from the full machine learning pipeline. Instead of retraining the entire model with new data, the system extracts real-time data, determines errors relative to the pre-trained model, and applies adjustments only where needed. This extraction approach improves forecasting accuracy while minimizing computational resource consumption
Solution Approach 2:
The patent applies partial action by performing error determination and adjustment only on specific forecast outputs rather than reprocessing the entire dataset. The system collects real-time data over a second time interval that is less than the forecast window, determining errors and forming adjusted forecasts for only the necessary portions, thus improving accuracy while reducing energy and computational resource usage
3Measurement precision
If the machine learning model is retrained with new data, then forecasting accuracy is improved, but operational costs and time increase
Solution Approach 1:
The patent performs preliminary model training on historical data before deployment, creating a pre-trained model that serves as a permanent reference. When new real-time data becomes available, the system does not retrain the model but instead uses the pre-trained model to determine errors and apply adjustments. This preliminary action eliminates the need for time-consuming retraining while maintaining improved forecasting accuracy
Solution Approach 2:
The patent creates a conceptual copy of the model's forecasting capability by determining errors between the pre-trained model's forecasts and actual real-time data. Instead of recreating the model through retraining, the system copies the essential forecasting function and adjusts it based on error analysis, significantly reducing the time required to incorporate new information
Data Source
AI summary
Disclosed are systems and methods that relate to demand forecasting based on machine learning of historical data, while adjusting forecasts based real-time data.