Adaptive Resource Forecasting With Incremental Neural Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing consumption forecasting models struggle to adapt to changes in consumption habits, leading to inefficiencies in resource provisioning systems.
Innovation Solution
An adaptive system utilizing a neural network with incremental learning and self-associative components to generate resource estimates, incorporating a predictive part and a self-associative part to maintain past learning while adapting to new data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional forecasting models are used, then resource provisioning can be maintained, but the system cannot adapt to changes in consumption habits
Solution Approach 1:
The patent implements a dynamic neural network that continuously adapts to changing consumption patterns while maintaining forecast accuracy. The system uses online learning mechanisms where the neural network is trained incrementally on new data streams, allowing it to dynamically adjust its parameters and structure in response to changing consumption habits without losing previously learned patterns
Solution Approach 2:
The system incorporates feedback loops where actual consumption data is continuously compared with forecasted values, and the differences (errors) are used to update and refine the neural network model. This feedback mechanism enables the system to learn from past performance and continuously improve its adaptability while maintaining reliability through error correction
2Adaptability or versatility
If the system adapts quickly to new habits, then it loses memory of past habits
Solution Approach 1:
The neural network is pre-trained on historical consumption data before deployment, establishing a foundation of past habits. This preliminary action ensures that the system starts with knowledge of historical patterns, which then serves as a baseline that continues to influence predictions even as the system adapts to new consumption behaviors through continuous learning
Solution Approach 2:
The system employs a nested architecture where multiple neural network layers and multiple time-scale models are combined. Inner layers capture short-term adaptations to new habits, while outer layers maintain long-term memory of past habits. This nested structure allows simultaneous operation at different temporal scales, enabling quick adaptation without forgetting historical patterns
3Productivity
If resource production is optimized based on forecasts, then production costs are reduced, but resource availability may be compromised
Solution Approach 1:
The system implements partial production optimization by generating forecasts at multiple confidence levels and producing resources accordingly. Instead of fully optimizing production based on a single forecast, the system produces a base level of resources guaranteed to meet minimum demand, then optimizes additional production based on forecasted excess demand, achieving cost savings while maintaining adequate availability
Solution Approach 2:
The neural network forecast acts as an intermediary between production capabilities and consumption demands. It translates complex, variable consumption patterns into actionable production guidance that balances efficiency and availability. The forecast intermediary enables production systems to operate efficiently while maintaining reliability by providing probabilistic predictions that inform risk-based production decisions
Data Source
Figure 1A
Figure 1B
Figure 1C
AI summary
This description relates to a control system configured to supply and/or receive in a buffer tank, a quantity of a resource intended for or received from a target system (102), the control system comprising: - at least one sensor (106) configured to measure, periodically and in a first time interval, quantities of the resource supplied and/or received; - a processing device (110) configured to perform a processing operation on the measurements made by the at least one sensor, the processing operation corresponding to the generation of a first time series whose values are sums of the quantities measured at several time instants; the processing device further comprising a neural network (202) to generate at least one quantity estimate on the basis of the first time series, and to control an actuator (108).