Building Energy Forecast Training with Selective Data Replacement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Building managers face challenges in generating accurate energy forecasts due to the use of inaccurate data from external services, which can lead to incorrect predictions and inefficient energy management.
Innovation Solution
A building system that selectively uses data by identifying time-steps with inaccurate forecasts through metrics like CV-RMSE and CUSUM values, replacing these with predetermined values to ensure only accurate data is input into prediction models, thereby improving forecast accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If external services are used to provide forecast data, then the building system can obtain forecast data for energy prediction, but the accuracy of the forecast data deteriorates due to inaccurate data from external services
Solution Approach 1:
The patent introduces an intermediary validation layer between external forecast services and the prediction model. This layer computes accuracy metrics (CV-RMSE, CUSUM) to assess forecast quality and selectively filters or replaces inaccurate forecasts with alternative data sources or placeholder values, thereby mediating between data availability and data accuracy requirements
Solution Approach 2:
The system dynamically changes the parameter of forecast data quality by computing accuracy metrics (CV-RMSE, CUSUM) for different forecast time-steps and data types. Based on these parameter assessments, the system adjusts which forecasts are used, replaced, or flagged as inaccurate, transforming static forecast data into dynamically quality-adjusted input
2Ease of operation
If all forecast data from external services is used without differentiation, then the data processing is simple, but the energy forecast accuracy deteriorates due to inclusion of inaccurate data
Solution Approach 1:
The patent segments the forecast data processing into distinct phases: initial forecast retrieval, accuracy metric computation (CV-RMSE, CUSUM), quality assessment, and selective usage or replacement. This segmentation transforms a simple but inaccurate process into a multi-stage process that maintains simplicity where possible while adding precision where needed
Solution Approach 2:
The system implements feedback loops where historical forecast accuracy is continuously measured using CV-RMSE and CUSUM metrics. This feedback information is used to adjust future forecast usage decisions, creating a self-correcting system that improves energy forecast accuracy based on observed performance patterns
3Productivity
If inaccurate forecast data is used as input, then the prediction model can operate continuously, but the energy prediction accuracy deteriorates leading to inefficient energy management
Solution Approach 1:
The patent applies preliminary action by computing accuracy metrics and assessing forecast quality before the prediction model uses the data. Inaccurate forecasts are identified and replaced with placeholder values or alternative sources in advance, ensuring the prediction model receives pre-validated input without interrupting operational continuity
Solution Approach 2:
The system discards inaccurate forecast data identified through CV-RMSE and CUSUM metric analysis and recovers by using alternative data sources or placeholder values. This selective discarding and recovering maintains model operation while preventing inaccurate data from degrading prediction quality
Data Source
AI summary
A building system for training a prediction model with augmented training data. The building system comprising one or more memory devices configured to store instructions thereon that, when executed by one or more processors, cause the one or more processors to obtain a first training data set comprising data values associated with a data point of the building system and with a plurality of time-steps and energy values associated with consumption of the building system at each of the plurality of time-steps; generate an augmented training data set comprising a second training data set, the second training data set comprising the energy values and the data values of the first training data set but with a data value replaced with a predetermined value at a time-step of the plurality of time-steps; and generate a prediction model by training the prediction model.