Graph Signal Processing Energy Forecasting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing energy forecasting systems rely on assumptions of independent and identically distributed data, which may not be true for time series data, and deep learning models are computationally heavy, leading to inefficiencies and inaccuracies, especially in building load forecasting.
Innovation Solution
A Graph Signal Processing (GSP) based approach is used to construct a weighted adjacency matrix from energy consumption parameters, applying total variation minimization to generate a smooth signal for accurate forecasting, handling missing values and long-term predictions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep learning models are used for energy forecasting, then forecasting accuracy is improved, but computational complexity increases
Solution Approach 1:
The patent segments the energy forecasting problem into two parts: (1) constructing a graph structure from energy consumption data where nodes represent time points and edges represent temporal relationships, and (2) applying graph signal processing techniques to extract features and make predictions. This segmentation allows using simpler GSP methods instead of complex deep learning models while maintaining accuracy.
Solution Approach 2:
The patent replaces the mechanical deep learning model processing approach with graph signal processing methods. Instead of using neural networks to process sequential data, the patent transforms the time series into a graph structure and applies spectral graph theory, which is computationally more efficient while achieving comparable or better forecasting accuracy.
2Productivity
If traditional machine learning methods are used for energy forecasting, then computational efficiency is improved, but forecasting accuracy deteriorates
Solution Approach 1:
The patent transforms the traditional one-dimensional time series forecasting problem into a two-dimensional graph structure problem. By representing energy consumption data as a graph where nodes are time points and edges capture temporal dependencies, the method enables application of graph signal processing techniques that improve accuracy while maintaining computational efficiency.
Solution Approach 2:
The patent changes the fundamental parameters of the forecasting approach by using graph Laplacian eigenvalues and eigenvectors as features instead of traditional statistical features. This parameter transformation allows capturing complex temporal patterns more effectively while using efficient linear algebra operations rather than computationally intensive deep learning training.
3Reliability
If graph signal processing is used to construct weighted adjacency matrix from energy consumption parameters, then handling of missing values and long-term predictions is improved, but data processing complexity increases
Solution Approach 1:
The patent introduces a graph structure as an intermediary representation of the energy consumption data. The graph acts as a mediator that naturally handles missing values through its structure - missing data points correspond to isolated nodes or nodes with fewer connections, which can be handled through graph signal processing techniques like graph interpolation or imputation based on neighborhood information.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Energy consumption modelling requires to consider various factors affecting the energy consumption in buildings, to be able to effectively forecast future consumption. Even though some of the state of the art deep learning based approaches are able to address these requirements to some extent, they are computationally heavy. The disclosure herein generally relates to energy forecasting, and, more particularly, to a method and system for graph signal processing (GSP) based energy modelling and forecasting. The system monitors and collects information on energy consumption in a building and values of associated energy consumption parameters. This input data is further processed using GSP to generate a building energy consumption model, from which a smooth signal is obtained by applying total variation minimization. The system further performs forecasting using the smooth signal.