AI Power Consumption Prediction for Unknown Home Devices
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems struggle to accurately predict power consumption in homes, especially for electronic devices with unknown usage patterns, necessitating improved methods to provide reliable energy consumption information.
Innovation Solution
An artificial intelligence apparatus utilizing a neural network model to estimate power consumption data for unknown devices by classifying mixed power data into individual data, incorporating energy and usage pattern information, and predicting consumption based on registered devices and external factors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If machine learning models are trained offline and deployed to edge devices for real-time power consumption prediction, then real-time prediction capability is improved, but the complexity of model deployment and maintenance worsens
Solution Approach 1:
The system segments the machine learning model into separate training and deployment phases. Training is performed offline on powerful servers, while the trained model is deployed to edge devices for real-time inference. This segmentation allows complex model training to be separated from real-time prediction execution, achieving both high prediction speed and manageable deployment complexity.
Solution Approach 2:
The system performs preliminary model training offline before deployment to edge devices. By pre-training the model with historical data and pre-processing in advance, the system prepares the model structure and parameters beforehand, enabling real-time predictions to be made quickly without performing complex training operations during runtime.
2Measurement precision
If deep learning models are used for power consumption prediction, then prediction accuracy is improved, but the computational resources required worsens
Solution Approach 1:
The system segments computational tasks between offline training phase and online inference phase. Resource-intensive deep learning model training is performed offline on powerful servers, while the deployed model on edge devices performs only lightweight inference operations. This segmentation achieves high prediction accuracy through sophisticated deep learning architectures while minimizing real-time computational resource consumption.
Solution Approach 2:
The system performs preliminary model training with complex deep learning architectures offline, preparing optimized model weights and structures in advance. During real-time operation, the pre-trained model requires minimal computational resources for inference, thus achieving high accuracy without excessive real-time computational resource consumption.
3Productivity
If machine learning models are deployed on edge devices, then real-time prediction is achieved, but the device resources required worsens
Solution Approach 1:
The system segments the model lifecycle into offline training and online deployment phases. Complex model training that requires substantial computational resources is performed offline, while the deployed model on edge devices performs efficient real-time predictions with minimal resource requirements. This segmentation enables real-time prediction efficiency while reducing the quantity of device resources needed at runtime.
Solution Approach 2:
The system creates an optimized copy of the trained model for deployment on edge devices. The full model training is performed once offline, and the resulting model weights and structure are copied to edge devices for efficient real-time inference. This copying approach avoids replicating the resource-intensive training process on limited-edge devices, achieving high prediction efficiency with minimal device resources.
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
The present disclosure relates to an artificial intelligence device and method capable of predicting an amount of power consumption occurring in a home, can obtain power consumption data for pre-registered electronic devices, check whether there is a specific electronic device for which the power consumption data has not been obtained, and if there is a specific electronic device for which the power consumption data has not been obtained, obtain mixed power data consumed in a current home, classify the obtained mixed power data into a plurality of individual power data, extracts individual power data matching the specific electronic device, estimate power consumption data for the specific electronic device based on the extracted individual power data, and predict an amount of power consumption in a home based on power consumption data for all pre-registered electronic devices.