Building Energy Lag Optimization for Fine-Grained Weather Normalization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional weather normalization techniques fail to accurately estimate energy use as a function of temperature for periods less than 24 hours, leading to unreliable models when using fine-grained energy use data.
Innovation Solution
A system comprising participant stores, baseline data stores, a building lag optimizer, and a dispatch processor that shifts energy consumption data relative to outside temperature values to determine a building's energy lag, allowing for the generation of accurate machine learning model parameters and a dispatch schedule for demand response programs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional weather normalization techniques are used, then energy use can be estimated as a function of temperature, but the models are unreliable for periods less than 24 hours
Solution Approach 1:
The patent segments the energy consumption data into distinct components: base load consumption and weather-induced consumption. By separating these components and analyzing them independently, the system can accurately model short-term energy use patterns without the reliability issues that plague conventional aggregated approaches. This segmentation allows the system to focus on transient energy consumption periods separately from baseline consumption.
Solution Approach 2:
The patent introduces the dimension of energy lag (time delay) as a new parameter in the modeling approach. Instead of simply correlating energy consumption with current temperature, the system incorporates historical temperature data with varying lag times to capture the thermal inertia and delayed response of building systems. This dimensional addition enables reliable modeling at sub-24-hour time scales.
2Measurement precision
If fine-grained energy use data is used, then measurement precision is improved, but conventional techniques produce unreliable models
Solution Approach 1:
The patent introduces machine learning algorithms as intermediaries between the fine-grained energy data and the final predictive model. These algorithms process the high-resolution data, identify patterns in transient consumption periods, and generate robust model parameters that maintain reliability despite the detailed input granularity. The machine learning component acts as a mediator that transforms precise but complex data into reliable predictive relationships.
Solution Approach 2:
The system dynamically adjusts modeling parameters based on the characteristics of the fine-grained data being analyzed. It modifies the lag time windows, weighting factors, and model structure to optimize performance for high-resolution inputs. This parameter adaptability allows the system to fully leverage the precision of fine-grained measurements without succumbing to the reliability problems that conventional fixed-parameter techniques encounter.
3Loss of information
If energy consumption data is shifted by multiple lag values, then transient energy consumption periods are captured, but data processing complexity increases
Solution Approach 1:
Instead of analyzing all possible lag values indefinitely, the system applies a practical limit to the range of lag values examined. It focuses on the most significant lag periods that capture the essential transient energy consumption behavior while avoiding unnecessary computation with excessive or redundant lag values. This partial action approach maintains information completeness for critical transient periods while controlling processing complexity.
Solution Approach 2:
The patent replaces traditional statistical correlation methods with machine learning algorithms that are better suited for handling multiple lagged variables. These advanced algorithms efficiently process the shifted data across multiple lag values, automatically identifying the most influential lag periods and extracting meaningful patterns without the computational burden that would plague conventional statistical approaches.
Data Source
AI summary
A method for characterizing buildings, including retrieving a plurality of baseline energy use data sets for the buildings from a baseline data stores; generating energy use data sets for each of the buildings, each of the energy use data sets comprising energy consumption values along with corresponding time and outside temperature values, where the energy consumption values within each of the sets are shifted by one of a plurality of lag values relative to the corresponding time and outside temperature values, and where each of the plurality of lag values is different from other ones of the plurality of lag values; performing a machine learning model analysis on the each of the plurality of energy use data sets to yield corresponding machine learning model parameters and a corresponding residual; determining a least valued residual from all residuals, the least valued residual indicating a corresponding energy lag for the each of the buildings; and categorizing the buildings into types according to similar energy lags.