Advanced HVAC Control and Artificial Intelligence Controllers
Indoor air quality control systems within a house typically use simple control algorithms to control the indoor environment. A more intelligent, efficient, and aware design can be implemented using more advanced algorithms and techniques. In order to allow for testing and implementation of new HVAC control logics, a flexible and extensible controlling program was designed to allow for the creation and introduction of advanced HVAC control systems into the house. In particular, the base system was designed with some basic artificial intelligence systems to improve efficiency and adaptive capabilities.
The use of AI (artificial intelligence) can potentially lead to efficient, adaptive, and better HVAC control. The primary example of AI suitable for HVAC control is the neural network. A neural network works through the use of a model of an individual neuron. The model is fairly straightforward – the neuron has a number of inputs, each of which carries an individual weighting. The weighted inputs are summed, and the result is modified by a transfer function, typically 1/(1-exp(-x)). This result is then taken as the output of the neuron. A neural network takes these modeled neurons and creates an interconnected network. This network is then “trained” by feeding in inputs, cascading the results through the network, and then determining the error and adjusting the network using the error. The resulting system is capable of adapting to non-linear, multivariate functions with high levels of noise and unknown functions.
Another important feature for the HVAC controller is the ability to easily and completely alter the manner in which the HVAC system operates its controls. The solution for implementing this system is produced in several parts – decentralized control, plug-in architecture, and rule-based agents. Decentralized control is the process of having local parts of the controller make independent decisions, informing the global controller, and then receiving final judgment on the system’s operation. This allows for control strategies to be tested at different levels of the HVAC system, and allows for new parts to be added or removed without reducing the systems ability to operate.
The plug-in architecture is key to allowing new and varied strategies to be tested within the system. A standard protocol for libraries to control various parts of the system was written, allowing for the function of the network to be altered or added to by simply introducing additional files to the system. This removes the need to completely re-write the base system for every alteration desired. This also allows different instances of the same controller program to operate under completely different logic, for side-by-side comparison on the same test bed system.
Rules-based agents provide functionality in permitting multiple control schemes. Traditional residential HVAC control only controls the temperature within the residential environment. Minor alterations in the temperature, massive changes in the temperature, and ventilation are all controlled by the same single algorithm, and only on the basis of temperature. Rules-based agents at the local level allow for control schemes to be used for different situations and desires within the HVAC system, attaining different goals under different conditions. This produces a more fine-grained and modular approach to the control systems.
Advantages of the system are that it is efficient, adaptive, and flexible HVAC control with the capabilities of testing new and innovative control schemes with minimal coding. Also is designed to allow multi-zone capabilities and be able to grow and adapt with new conditions and research within the house. Coded in C# for simplicity and ease of use and extension.
A disadvantage of the system is that it is not an embedded solution for efficient operation. Relatively expensive in computing power due to use of higher level coding languages.
The use of the One Wire sensor network for controlling the HVAC system locally, such as setting set points and gathering real-time useful data. Integration with the final HVAC system, and evaluation of control schemes for global controller. Consideration of new areas of interest in control goals for the HVAC system.