Automatic threshold selection of machine learning/deep learning model for anomaly detection of connected chillers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Chillers in HVAC systems often experience unexpected faults leading to unplanned shutdowns, causing efficiency losses and damage to other equipment, due to various influencing factors, making it challenging to predict chiller failures effectively.
Innovation Solution
A chiller threshold management system using machine learning and deep learning models, trained with historic data, to analyze chiller performance data and detect anomalies, allowing for the prediction of faults and potential shutdowns by setting and adjusting thresholds based on receiver operating characteristics and probability density functions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If machine learning and deep learning models are used to predict chiller faults, then the ability to detect anomalies and predict failures is improved, but the complexity of the system increases
Solution Approach 1:
The patent introduces an intermediary threshold selection system that bridges the gap between complex machine learning models and practical fault detection. The system automatically determines optimal thresholds using receiver operating characteristic (ROC) curves and probability density functions, serving as a mediator that simplifies the deployment of complex models without requiring manual threshold tuning or deep expertise in machine learning parameters.
Solution Approach 2:
The system implements self-service by automatically selecting thresholds based on the characteristics of the input data and model predictions. The threshold selection mechanism autonomously analyzes the distribution of prediction scores, identifies optimal separation points using ROC curves and probability density functions, and applies these thresholds without human intervention, allowing the system to adapt to different chiller configurations and operating conditions independently.
2Ease of operation
If automatic threshold selection is implemented, then the ease of operation is improved, but the computational requirements and processing time increase
Solution Approach 1:
The system performs preliminary action by pre-calculating and storing optimal thresholds during system initialization or model training phases. By determining thresholds in advance using ROC curves and probability density function analysis, the system eliminates the need for real-time threshold computation during operational monitoring, thereby reducing processing time and computational load during actual fault detection operations.
3Measurement precision
If multiple evaluation techniques are used for threshold selection, then the measurement precision is improved, but the device complexity increases
Solution Approach 1:
The system applies dynamics by adaptively selecting and combining different evaluation techniques based on the specific characteristics of the data and model being used. Rather than rigidly applying a single method or always using all methods, the system dynamically adjusts its approach, choosing the most appropriate evaluation technique (ROC analysis, probability density functions, or other methods) based on the situation, thereby balancing precision requirements with system complexity.
Data Source
AI summary
A chiller threshold management system for a building, including one or more memory devices and one or more processors. The one or more memory devices are configured to store instructions to be executed on the one or more processors. The one or more processors are configured to determine whether chiller fault data exists in chiller data used to generate a plurality of chiller prediction models. The one or more processors are further configured to generate a first threshold evaluation value for each of the plurality of chiller prediction models using a first evaluation technique in response to a determination that chiller fault data exists in the chiller data, and generate a second threshold evaluation value for each of the chiller prediction models using a second evaluation technique in response to a determination that chiller fault data does not exist in the chiller data. The one or more processors are configured to select a first threshold for each of the plurality of chiller prediction models based on the first threshold evaluation values in response to the determination that chiller fault data exists in the chiller data, and select a second threshold for each of the plurality of chiller prediction models based on the second threshold evaluation values in response to the determination that chiller fault data does not exist in the chiller data.