Distributed Power Prediction Using Cluster-Guided Reinforcement Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing AI models struggle to accurately predict power generation for newly installed distributed power plants due to differences in environmental conditions, requiring frequent retraining which is time-consuming and resource-intensive.
Innovation Solution
A method using a neural network model that determines whether to perform additional reinforcement learning based on environmental variables collected from a new power plant, by creating reference clusters from existing plants and comparing them to new clusters formed from the new plant's data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If AI models are trained on existing data to predict power generation, then prediction accuracy is improved for existing plants, but the model cannot accurately predict power generation for newly installed distributed power plants with different environmental conditions
Solution Approach 1:
The patent segments the prediction problem by creating multiple specialized neural network models, each trained on data from a specific distributed power plant. Each model learns the unique environmental characteristics and power generation patterns of its corresponding plant, enabling accurate predictions for that specific plant while maintaining the ability to handle diverse environmental conditions across different plants.
Solution Approach 2:
The patent changes the approach from using a single model with fixed parameters to creating multiple models with parameters optimized for specific environmental conditions. Each neural network model's parameters are adjusted and trained on data from particular environmental settings, allowing the system to adapt to varying environmental conditions across different power plant locations.
2Measurement precision
If the neural network model is retrained every time a new power plant is introduced, then prediction accuracy for the new plant is improved, but time and computational resources are consumed
Solution Approach 1:
The patent applies preliminary action by pre-training multiple neural network models during the setup phase, each on data from existing power plants. When a new power plant is introduced, the system can quickly select or combine pre-trained models rather than starting training from scratch, significantly reducing the time and computational resources required for new plant predictions.
Solution Approach 2:
The patent uses copying by creating multiple neural network model instances, each copied and trained on data from different power plants. These pre-trained model copies can be rapidly deployed or combined for new plant predictions, avoiding the need to retrain a single model from scratch for each new plant introduction.
3Reliability
If the neural network model is retrained frequently to accommodate new plants, then model performance is maintained, but computational resources and energy consumption increase
Solution Approach 1:
The patent performs preliminary training actions during the model development phase, creating a library of pre-trained neural network models. This preliminary action ensures model performance reliability is established in advance, eliminating the need for frequent retraining and thereby reducing ongoing computational energy consumption when new plants are introduced.
Solution Approach 2:
The patent creates multiple copies of neural network models trained on different datasets. Instead of retraining a single model frequently (which consumes energy), the system copies and combines pre-trained models, maintaining performance reliability while minimizing the energy required for model updates and new plant adaptations.
Data Source
AI summary
A power generation prediction method for distributed power plants using reinforcement learning according to an embodiment of the present disclosure is a method of predicting power generation of each of a plurality of distributedly installed power plants using a neural network model. The method includes: creating a plurality of reference clusters by clustering environmental variables accumulatively collected from each of the power plants; creating a new cluster by collecting new environmental variables from a new power plant and by clustering the new environmental variables; and additionally using the new environmental variables for reinforcement learning of the neural network model on the basis of similarity between the plurality of reference clusters and the new cluster.