Method and controller for controlling a chiller plant for a building and chiller plant
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Centralized air conditioning systems in non-residential buildings consume significant energy, and existing methods for reducing power consumption are often implementation-specific and dependent on static assumptions, lacking flexibility and adaptability to varying environmental and operational conditions.
Innovation Solution
A method using machine learning models to predict cooling load demand and optimize chiller plant control signals based on environmental and operational data, allowing for flexible operation and adaptation to different chiller plant implementations and environmental situations, while minimizing power consumption and optimizing performance parameters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If static rule-based analytics or manual adjustment methods are used for chiller plant control, then implementation simplicity is maintained, but adaptability to varying environmental and operational conditions deteriorates
Solution Approach 1:
The patent applies dynamics by transitioning from static rule-based control to dynamic machine learning models that continuously adapt to varying environmental and operational conditions. The ML models learn optimal control strategies from historical data and adjust control signals in real-time based on changing conditions, making the control system flexible and adaptive without requiring complex manual reconfiguration.
Solution Approach 2:
The control system performs self-service through automated machine learning models that independently analyze historical operational data, environmental data, and control signals to generate optimized control strategies. The system self-adjusts to varying conditions without requiring manual intervention or expert knowledge, reducing operational complexity while improving adaptability.
2Adaptability or versatility
If machine learning models are implemented for predicting cooling load demand and optimizing control signals, then adaptability to different conditions is improved, but computational complexity and data processing requirements increase
Solution Approach 1:
The patent applies preliminary action by training machine learning models offline using historical operational data before deployment. The models pre-learn optimal control strategies and relationships between environmental conditions, cooling loads, and control signals. During actual operation, the pre-trained models rapidly generate control signals without requiring complex real-time computations, reducing operational computational complexity while maintaining high adaptability.
3Loss of energy
If existing control methods are used, then system simplicity is maintained, but power consumption of the chiller plant increases
Solution Approach 1:
The patent applies feedback by using historical operational data including actual control signals and resulting cooling power consumption to train machine learning models. The models learn from past performance and continuously optimize control strategies to minimize power consumption. The system incorporates feedback loops where predicted cooling loads and optimal control signals are continuously refined based on actual measured outcomes, enabling automated energy optimization without excessive automation complexity.
Data Source
Figure 1
Figure 2
Figure 3~4
AI summary
According to the invention, environmental data (ED) of an environment of the building (BD) and cooling load demand data (CLD) are received as first training data, which are used for training a first machine learning model (NN1) to predict a cooling load demand from environmental data. Furthermore, control signals (CS) for the chiller plant (CP) and cooling power data (CPD) resulting from applying the control signals (CS) to the chiller plant (CP) are received as second training data. The second training data are used for training a second machine learning model (NN2) to predict a cooling power from control signals. Moreover, actual environmental data (ED) are received, from which a cooling load demand (CLDP) is predicted by the trained first machine learning model (NN1). Furthermore, candidate control signals (CCS) for the chiller plant (CP) are generated, and from the candidate control signals (CCS) a resulting cooling power (CPP) is predicted by the trained second machine learning model (NN2). From the candidate control signals (CCS), applicable control signals (ACS) are selected for which a predicted cooling power (CPP) fulfills the predicted cooling load demand (CLDP).