Adaptive Machine Learning Model Selection for Chiller Fault Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Chiller faults in HVAC systems are difficult to predict due to various influencing factors, leading to unplanned shutdowns and associated losses, and existing machine learning and deep learning models require careful training with high-quality historic data to make accurate predictions.
Innovation Solution
A chiller fault prediction system that generates single and cluster prediction models using machine learning and deep learning, allowing for the labeling of models as accurately or inaccurately predicting based on performance, and re-evaluates models to retain or switch states, trains models with parameters, and assigns new chillers to clusters for fault prediction, using techniques like Multivariate Gaussian Modeling and Long Short Term Memory with autoencoders.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple machine learning and deep learning models are trained with different hyper-parameters to predict chiller faults, then prediction accuracy is improved, but model selection complexity and computational resources increase
Solution Approach 1:
The system dynamically selects the most appropriate machine learning or deep learning model based on the specific chiller dataset characteristics and prediction requirements. Instead of using a fixed model, the system adapts model selection to match data patterns, improving prediction accuracy while managing complexity through automated selection criteria.
Solution Approach 2:
The system trains multiple models with different hyper-parameters (learning rates, batch sizes, network architectures) to optimize prediction accuracy. By systematically varying model parameters and selecting the best-performing configuration, the system achieves high accuracy without manually managing all model variations, as the selection process is automated based on performance metrics.
2Measurement precision
If historic chiller data is used to train prediction models, then model accuracy is improved, but data quality requirements and preprocessing complexity increase
Solution Approach 1:
The system performs comprehensive data preprocessing, cleaning, and validation before model training by establishing data quality standards and processing pipelines in advance. This preliminary action ensures that only high-quality, properly formatted historic chiller data is used for training, improving model accuracy while automating the preprocessing complexity so it does not burden end users.
3Reliability
If cluster-based modeling is implemented for chillers with similar characteristics, then prediction reliability is improved for new chillers, but system complexity and computational requirements increase
Solution Approach 1:
The system segments the chiller population into distinct clusters based on similar operational characteristics, failure patterns, and performance metrics. By grouping chillers with analogous behaviors, the system can apply specialized prediction models to each cluster, improving reliability for new chillers by matching them to appropriate historical patterns while managing complexity through automated clustering algorithms.
4Measurement precision
If continuous model re-evaluation and retraining is performed, then prediction accuracy is maintained over time, but computational resources and processing time increase
Solution Approach 1:
The system implements periodic re-evaluation and retraining of prediction models at scheduled intervals rather than continuously. This periodic action maintains prediction accuracy over time by updating models with new data while avoiding the excessive computational burden of continuous retraining, thus balancing accuracy maintenance with reasonable processing time and resource utilization.
Data Source
AI summary
A model management system for building equipment includes one or more memory devices configured to store instructions that, when executed on one or more processors, cause the one or more processors to determine whether fault data exists in equipment data used to generate a plurality of shutdown prediction models for the building equipment, generate a first performance evaluation value for each of the plurality of shutdown prediction models using a first evaluation technique in response to a determination that the fault data exists in the equipment data, generate a second performance evaluation value for each of the plurality of shutdown prediction models using a second evaluation technique in response to a determination that the fault data does not exist in the equipment data, and select one of the plurality of shutdown prediction models based on the first performance evaluation value and the second performance evaluation value.