Building Energy Load Prediction With Kernel Smoothing Alerts
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for predicting building energy loads are complex, computationally expensive, and unsuitable for multiple buildings or those without detailed design data, often resulting in inaccurate or impractical energy monitoring and control.
Innovation Solution
A system and method using kernel smoothing with scaling factors determined by cross-validation error minimization to predict building energy loads, comparing predicted values with measured data to trigger alerts when thresholds are exceeded, adaptable for use across various buildings with minimal computational resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If detailed physical building models are used to calculate predicted building energy load values, then prediction accuracy is improved, but device complexity and computational cost increase significantly
Solution Approach 1:
The patent uses measured historical building energy load values and environmental variables to create a statistical copy of the building's energy behavior pattern, replacing the need for detailed physical models. This statistical model captures the relationship between environmental conditions and energy consumption without requiring complex building-specific parameters.
Solution Approach 2:
The patent replaces the mechanical/physical modeling approach with a statistical data-driven approach. Instead of using physical equations and building design parameters, the system uses measured historical data and statistical relationships to predict energy loads, significantly reducing computational complexity.
2Measurement precision
If detailed physical building models are used, then prediction accuracy for specific buildings is improved, but ease of manufacture and deployment across multiple buildings deteriorates
Solution Approach 1:
The patent creates a universal prediction system that can be deployed across multiple buildings of different types. The statistical model learns from historical data and adapts to each building's specific patterns without requiring building-specific physical models, making the system universally applicable while maintaining accuracy.
Solution Approach 2:
The system automatically learns building-specific energy patterns from measured historical data without requiring manual input of building design parameters or expert analysis. The model self-adjusts to each building's unique characteristics through data-driven learning, eliminating the need for manual model creation for each building.
3Device complexity
If polynomial functions are used to fit historical building energy load data, then model fitting is simplified, but prediction accuracy for variables outside training data range deteriorates
Solution Approach 1:
The patent uses a dynamic nearest-neighbor approach where the prediction model adapts based on the specific query conditions. Instead of a fixed polynomial model, the system dynamically selects and weights historical data points based on their similarity to current conditions, allowing accurate predictions both within and outside the original training range.
4Measurement precision
If neural network techniques are used to fit historical data, then prediction capability is improved, but computational resources and complexity increase
Solution Approach 1:
The patent uses a computationally efficient statistical model that requires minimal computational resources compared to neural networks. The nearest-neighbor approach with simple distance calculations and weighted averaging provides adequate prediction accuracy without the heavy computational burden of training and executing neural network models.
Data Source
AI summary
A system and method for predictive modeling of building energy consumption provides predicted building energy load values which are determined using kernel smoothing of historical building energy load values for a building using defined scaling factors for scaling predictor variables associated with building energy consumption. Predictor variables may include temperature, humidity, windspeed or direction, occupancy, time, day, date, and solar radiation. Scaling factor values may be defined by optimization training using historical building energy load values and measured predictor variable values for a building. Predicted and measured building energy load values are compared to determine if a preset difference threshold has been exceeded, in which case an alert signal or message is generated and transmitted to electronically and/or physically signal a user. The building energy monitoring system may be integrated with a building automation system, or may be operated as a separate system receiving building energy and predictor variable values.