Building ML Model Scheduling for Fast BMS Fault Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current building management systems face challenges in efficiently analyzing large datasets related to building equipment operation, requiring substantial time and computing power, and often necessitate expensive human analyst training, which can be time-consuming and costly, especially when dealing with limited analyst resources.
Innovation Solution
A building management system that utilizes machine learning techniques to generate predictive models based on operating data sets, allowing for automated analysis and control responses, including fault predictions and preventative measures, thereby reducing the burden on human analysts and optimizing resource usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If machine learning systems are used to analyze building equipment data sets, then analysis automation and productivity are improved, but the time and computing power required for analysis increase
Solution Approach 1:
The system pre-generates predictive models at scheduled intervals (e.g., daily, weekly, or monthly) and stores them for later use. When a prediction request is received, the pre-generated model is retrieved and executed immediately, avoiding the time-consuming process of generating a new model at the moment of prediction. This preliminary action resolves the contradiction by separating model generation (done in advance) from model execution (done quickly when needed).
Solution Approach 2:
The system segments the machine learning process into distinct phases: model generation phase (where computational heavy lifting occurs) and prediction phase (where quick inferences are made). By storing intermediate model artifacts and retrieving them for prediction requests, the system divides the workload temporally, allowing complex model training to occur during off-peak times while providing fast predictions during operational hours.
2Reliability
If human analysts are trained to analyze building equipment data, then analysis quality and reliability are improved, but training cost and time consumption increase
Solution Approach 1:
The system uses pre-generated predictive models as reusable artifacts that can be copied and stored for multiple prediction requests. Instead of retraining analysts (or regenerating models) for each analysis task, the same trained model can be copied and applied repeatedly to different data sets, maintaining consistent analysis quality while eliminating repeated training time and cost.
3Loss of energy
If limited analyst resources are used to parse large data sets, then operational cost is reduced, but analysis completeness and productivity decrease
Solution Approach 1:
The system enables self-service prediction by storing generated models and allowing automatic retrieval and execution without requiring analyst intervention for each prediction task. The pre-generated models serve themselves by being automatically deployed to answer prediction requests, freeing analyst resources while maintaining high productivity through automated model execution.
Data Source
AI summary
A building management system including one or more memory devices and one or more processors, the one or more memory devices configured to store instructions to be executed on the one or more processors, the processors configured to generate a first predictive model using a machine learning technique and an operating data set based on one or more operating parameters associated with at least one of a plurality of BMS subsystems. The processors are further configured to store the first predictive model at a first time interval, to receive a prediction request from a user input at a second time interval following the first time interval, and to retrieve the first predictive model. The one or more processors are further configured execute the retrieved predictive model to generate a first prediction in response to the prediction request; and initiate an automated control response based on the first prediction.