Grey box modelling and advanced control scheme for building heating systems

This dissertation is aimed at generating new knowledge on Recurrent Neuro-Fuzzy Inference Systems (RenFIS) and to explore its application in building automation. Inferential sensing is an attractive approach for modeling the behavior of dynamic processes. Inferential sensor based control strategies are applied to optimize the control of residential heating systems and demonstrate significant energy saving and comfort improvement. Despite the rapidly decreasing cost and improving accuracy of most temperature sensors, it is normally impractical to use a lot of sensors to measure the average air temperature because the wiring and instrumentation can be very expensive to install and maintain. To design a reliable inferential sensor, of fundamental importance is to build a simple and robust dynamic model of the system to be controlled. This dissertation presents the development of an innovative algorithm that is suitable for the robust black-box model. The algorithm is derived from ANFIS (Adaptive Neuro-Fuzzy Inference System) and is referred to as RenFIS. Like all other modelling techniques, RenFIS performance is sensitive to the training data. In this study, RenFIS is used to model two different heating systems, hot water heating system and forced warm-air heating system. The training data is collected under different operational conditions. RenFIS gives better performance if trained with the data set representing overall qualities of the whole universe of the experimental data. The robustness analysis is conducted by introducing simulated noise to the training data. Results show that RenFIS is less sensitive than ANFIS to the quality of training data. The RenFIS based inferential sensor is then applied to design and inferential control algorithm that can improve the operation of residential heating systems. In current practice, the control of heating systems is based on measurement of the air temperature at one point within the building. The inferential control strategy developed through this study allows the control to be based on an estimate of the overall thermal performance. This is achieved through estimating the average room temperature using a RenFIS based inferential sensor and incorporating the estimate with conventional control technology. The performance of this control technology has been investigated through simulation study.

Grey box modelling and advanced control scheme for building heating systems

This dissertation is aimed at generating new knowledge on Recurrent Neuro-Fuzzy Inference Systems (RenFIS) and to explore its application in building automation. Inferential sensing is an attractive approach for modeling the behavior of dynamic processes. Inferential sensor based control strategies are applied to optimize the control of residential heating systems and demonstrate significant energy saving and comfort improvement. Despite the rapidly decreasing cost and improving accuracy of most temperature sensors, it is normally impractical to use a lot of sensors to measure the average air temperature because the wiring and instrumentation can be very expensive to install and maintain. To design a reliable inferential sensor, of fundamental importance is to build a simple and robust dynamic model of the system to be controlled. This dissertation presents the development of an innovative algorithm that is suitable for the robust black-box model. The algorithm is derived from ANFIS (Adaptive Neuro-Fuzzy Inference System) and is referred to as RenFIS. Like all other modelling techniques, RenFIS performance is sensitive to the training data. In this study, RenFIS is used to model two different heating systems, hot water heating system and forced warm-air heating system. The training data is collected under different operational conditions. RenFIS gives better performance if trained with the data set representing overall qualities of the whole universe of the experimental data. The robustness analysis is conducted by introducing simulated noise to the training data. Results show that RenFIS is less sensitive than ANFIS to the quality of training data. The RenFIS based inferential sensor is then applied to design and inferential control algorithm that can improve the operation of residential heating systems. In current practice, the control of heating systems is based on measurement of the air temperature at one point within the building. The inferential control strategy developed through this study allows the control to be based on an estimate of the overall thermal performance. This is achieved through estimating the average room temperature using a RenFIS based inferential sensor and incorporating the estimate with conventional control technology. The performance of this control technology has been investigated through simulation study.

Design of multi-loop control systems for distillation columns: review of past and recent mathematical tools

Design of a control structure in distillation columns involves selecting proper sets of manipulated and controlled variables (often including tray temperatures for inferential control of product compositions) and one-to-one pairing between the two sets. In this paper, various mathematical tools for achieving this goal are reviewed. First, traditional methods such as Singular Value Decomposition (SVD) and Relative Gain Array (RGA) that build upon a simplified steady-state or dynamic model of the column are explored. The role of optimization in systematizing the control design procedures is also investigated. Then, more recent inferential control techniques that rely on statistical methods such as Principal Component Regression (PCR), Partial Least Squares Regression (PLSR), and other machine learning techniques such as Artificial Neural Networks (ANN) and Support Vector Machine Regression (SVMR) are discussed extensively. The discussions include newer distillation technologies with complex configurations such as dividing-wall columns. Finally, the use of process simulators in aiding the control structure design of distillation columns is surveyed.

Nonlinear Model-Based Inferential Control of Moisture Content of Spray Dried Coconut Milk

The moisture content of a powder is a parameter crucial to be controlled in order to produce stable products with a long shelf life. Inferential control is the best solution to control the moisture content due to difficulty in measuring this variable online. In this study, fundamental and empirical approaches were used in designing the nonlinear model-based inferential control of moisture content of coconut milk powder that was produced from co-current spray dryer. A one-dimensional model with integration of reaction engineering approach (REA) model was used to represent the dynamic of the spray drying process. The empirical approach, i.e., nonlinear autoregressive with exogenous input (NARX) and neural network, was used to allow fast and accurate prediction of output response in inferential control. Minimal offset (<0.0003 kg/kg) of the responses at various set points indicate high accuracy of the neural network estimator. The nonlinear model-based inferential control was able to provide stable control response at wider process operating conditions and acceptable disturbance rejection. Nevertheless, the performance of the controller depends on the tuning rules used.

Fractional-Order-Based Inferential Control

This chapter presents the application of fractional differential operator in modelling and control of a three-tank interacting level process. In cases where the usage of sensors for the measurement of primary variable, which is the level of third tank in present case, is physically or economically not feasible, the measurement of secondary variable (i.e., second tank level) is used to determine the level of third tank for control purpose, known as inferential control scheme. The process is modeled and linearized around the operating points, resulting in third order plant, which is approximated to lower order integer and non-integer model. Both conventional integer order PI (IO-PI) & PID (IO-PID) and fractional order PI (FO-PI) & (FO-PID) controllers are implemented for this inferential control. Extensive simulation studies performed using MATLAB validate the supremacy of non-integer order model and controller over integer order model and controller. Genetic algorithm (GA) is being applied for both, firstly for reduced order model approximation and secondly for controller tuning.