Flexibility Aggregation via Adaptive Machine Learning Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for aggregating electrical energy flexibilities at distribution network levels face challenges in quickly estimating and predicting flexibilities, especially for manually controlled loads, due to the localized and intermittent nature of energy production, leading to unreliable forecasts and high computational costs.
Innovation Solution
A method involving real-time measurement of power curves during flexibility activation, followed by a posteriori calculation of actual flexibility achieved, with automatic adaptation of individual flexibility models through learning algorithms, enabling robust and rapid aggregation across various site sizes and time periods.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If deterministic forecasting models are used for manually controlled loads, then the implementation is simple, but the forecast reliability is poor due to unpredictable human behavior
Solution Approach 1:
The system implements a feedback loop where actual flexibility responses from manual loads are measured and used to iteratively improve the statistical model. The model is retrained periodically with new data, allowing it to learn from past human responses and progressively improve forecast accuracy while maintaining the simplicity of statistical approaches.
2Measurement precision
If statistical flexibility forecasts are performed before each activation, then the forecast accuracy improves, but the computational cost becomes prohibitive for small aggregates
Solution Approach 1:
The system performs preliminary actions by pre-processing and storing historical flexibility data in structured formats during periods when computations are less demanding. Statistical models are pre-trained on historical data and stored for rapid deployment, allowing fast flexibility estimates during critical activation periods without repeating full computational processes.
Solution Approach 2:
The system dynamically adjusts the level of computational effort based on aggregate size and time constraints. For small aggregates where speed is critical, it uses pre-computed models and simplified calculations. For larger aggregates where accuracy is more important and computational resources are available, it performs more comprehensive statistical analyses.
3Reliability
If flexibility aggregation is performed at distribution network level with numerous small sites, then the local balance is improved, but the computational complexity increases due to the large number of sites
Solution Approach 1:
The system segments the aggregation process into modular components: data collection from individual sites, statistical modeling at the site level, aggregation of site-level models, and flexibility estimation. This segmentation allows each component to be optimized independently and enables parallel processing, reducing overall computational complexity while maintaining accuracy for numerous small sites.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The present invention relates in particular to a method for aggregating electrical energy consumption flexibilities from a plurality of sites, which is implemented by computer means and which comprises the following steps: a) sending a request for consumption flexibilities by an electricity network operator (ENSO) to an aggregator (AG); b) aggregation by said aggregator (AG) of the individual flexibilities of the available sites, these individual flexibilities being made up of as many individual flexibility models as there are available sites; c) activation of said available flexibilities, characterized in that it comprises the following steps: d) measurement of the actual power curves of said sites during the activation of step c); e) calculation a posteriori, that is to say after the end of the activation of step c), of the actual flexibility achieved for each of said sites;d) at each iteration of said steps b) to e), adaptation of said individual flexibility models by machine learning.;