Machine learning apparatus, air conditioning system, and machine learning method
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Air conditioning systems face high workloads in optimizing heat quantity transfer due to varying operation conditions and loads, requiring extensive data collection and model building for each device combination, which is inefficient and labor-intensive.
Innovation Solution
A machine learning apparatus that learns to optimize the transfer of heat quantity by obtaining state variables from both heat-providing and heat-using devices, using a reward-based learning system to adjust temperature and flowrate, thereby reducing energy consumption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If simulation-based optimization is used to optimize heat quantity transfer, then energy consumption is reduced, but extensive data collection and model building for each device combination is required, resulting in high workload
Solution Approach 1:
The patent applies universality by creating a unified machine learning model that can handle multiple device combinations and operation conditions simultaneously. Instead of building separate simulation models for each device pair, a single neural network model is trained to predict optimal thermal medium parameters across various scenarios, reducing the overall complexity and workload while maintaining energy optimization benefits
Solution Approach 2:
The patent replaces the traditional mechanical simulation-based optimization system with a machine learning-based system. The complex simulation processes that require extensive data collection and model building for each device combination are substituted with a trained neural network that can quickly predict optimal parameters, significantly reducing the workload while achieving similar or better energy optimization results
2Measurement precision
If separate models are built for each device combination to optimize heat transfer, then accurate optimization is achieved, but the work load increases significantly
Solution Approach 1:
The patent merges multiple separate optimization models into a single unified machine learning model. Instead of building and maintaining separate simulation models for each device combination (heat-providing device, thermal transfer apparatus, and heat-using device), the patent combines all these scenarios into one neural network model that learns from diverse training data and generalizes across different configurations, reducing model building time while maintaining accuracy
Solution Approach 2:
The patent applies preliminary action by pre-training the machine learning model with comprehensive data covering various device combinations and operation conditions. This preliminary training phase allows the model to learn optimal parameters across different scenarios in advance, so that during actual operation, quick predictions can be made without re-building models for each new device combination, thus reducing time while maintaining accuracy
Data Source
AI summary
A machine learning apparatus for optimizing transfer of heat quantity is provided. A machine learning apparatus for learning at least one of a temperature and a flowrate at which a thermal transfer apparatus transfers a thermal medium in an air conditioning system including a device on a heat-providing side, a device on a heat-using side, and the thermal transfer apparatus configured to transfer the thermal medium from the device on the heat-providing side to the device on the heat-using side, the machine learning apparatus including: a state variable obtaining unit configured to obtain state variables including an operation condition of the device on the heat-providing side, an operation condition of the device on the heat-using side, and a value correlated with a heat quantity required by the device on the heat-using side; a learning unit configured to perform learning by associating the state variables with the at least one of the temperature and the flowrate; and a reward calculating unit configured to calculate a reward, based on a total value of a power consumption of the device on the heat-providing side, a power consumption of the device on the heat-using side, and a power consumption of the thermal transfer apparatus, wherein the learning unit performs learning by using the reward.