HTM-MPC Building Control for Multi-Step Anomaly Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing Building Automation Systems (BAS) face challenges in effectively monitoring and controlling complex building mechanical and electrical systems, as they are limited to finite domain controllers that can only forecast single time steps and provide one-dimensional system behavior indications, lacking the ability to handle multi-dimensional characterization and anomaly detection.
Innovation Solution
The integration of Hierarchical Temporal Memory (HTM) networks with Model Predictive Control (MPC) extends the finite MPC domain to an infinite domain controller, enabling forecasting multiple time steps ahead and creating multi-dimensional characterization curves, allowing for more comprehensive system state indication and anomaly detection by using HTM networks with spatial and temporal poolers to process input data and generate predictions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If traditional finite domain controllers are used in BAS, then the system structure is simple and easy to implement, but the forecasting capability is limited to single time steps and provides only one-dimensional system behavior indications
Solution Approach 1:
The patent transitions from finite domain controllers to infinite domain HTM networks, enabling the system to process multi-dimensional characterization data instead of single-dimensional inputs. This dimensional expansion allows the controller to handle complex multi-variable system behaviors and provide comprehensive forecasting across multiple time steps and dimensions simultaneously.
Solution Approach 2:
The patent implements dynamic forecasting by using HTM networks that can predict system behavior across multiple future time steps rather than single-step forecasting. The temporal memory component continuously updates predictions based on evolving system states, enabling adaptive forecasting that adjusts to changing system conditions dynamically.
2Measurement precision
If HTM networks with spatial and temporal poolers are integrated with MPC, then multi-dimensional characterization and anomaly detection capabilities are enhanced, but the device complexity and computational requirements increase
Solution Approach 1:
The patent segments the control system into distinct functional modules: MPC module for predictive control, HTM network for pattern recognition, spatial pooler for feature extraction, and temporal pooler for temporal pattern analysis. This modular segmentation allows each component to specialize in specific tasks, improving anomaly detection precision while enabling independent optimization and maintenance of each module.
Solution Approach 2:
The patent introduces HTM networks as an intermediary layer between the MPC controller and the building systems. This intermediary processes and interprets multi-dimensional system data, identifying anomalies and patterns before presenting refined information to the MPC module, thereby enhancing detection accuracy while managing computational complexity through layered processing.
3Power
If single-step forecasting is used in traditional BAS controllers, then the computational load is low and response time is fast, but the system cannot provide comprehensive future behavior predictions
Solution Approach 1:
The patent implements preliminary action by forecasting multiple future time steps ahead using HTM networks before actual system events occur. The temporal memory component continuously predicts system behavior across multiple future states, allowing the MPC controller to proactively adjust control actions based on anticipated future conditions rather than reacting to past events alone.
Data Source
AI summary
An embodiment includes duplicating an input dataset being input to a model predictive control (MPC) module for input to a first Hierarchical Temporal Memory (HTM) network. The embodiment also includes generating system behavior data using the MPC module for characteristic data of the input dataset. The embodiment also includes generating first HTM prediction data from the input dataset and the system behavior data using the first HTM network, the first HTM prediction data comprising predictions for respective dimensions of the system. The embodiment also includes generating second HTM prediction data from the first HTM prediction data and system output data using a second HTM network, the second HTM prediction data comprising a distinction between the first HTM prediction and the system output data. Finally, the embodiment includes determining that the distinction of the second HTM prediction data indicates an anomaly and adjusting system input data based on the anomaly.