Modeling, Control And Optimization For Indoor Environment And Energy Consumption Prediction

Nowadays, energy crisis and environmental pollution are common challenges that ev-ery country in the world should face. With the rapid development of economy, energy con-sumption in China is considerable. Building energy consumption have accounted for more than a quarter of the total social energy consumption and the ratio is still in increases year by year. Against this background, the building energy-saving technologies have received more and more attention. From the control point of view, building energy system can be seen as multivariate nonlinear complex system. To achieve the goal of building energy efficiency many problems are involved, such as optimal control of building environment, building energy prediction and management, etc.. Based on the control subjects, this dis-sertation focuses on the areas of building environment control and energy management, and the main contributions are concluded as follows.●For energy saving and indoor environment optimization purpose, Heating Venti-lation Air Conditioning (HVAC) systems need indoor temperature field’s dynamic feed-back information. Usually, this kind of data is provided by Computational Fluid Dynamics (CFD) models. However, the complexity and time-consuming of CFD methods can’t sat-isfy the real-time requirement. This study proposes a new indoor thermal environment modeling method by introducing Proper Orthogonal Decomposition (POD) model reduc-tion technique, which can help satisfy the real-time and precision requirements simultane-ously. The POD model reduction falls in the category of projection methods, which can transfer the infinite dimensional nonlinear systems into low order linear ones with the co-operation of discretization techniques. The concrete implemention includes several steps. First, construct indoor thermal environment by CFD dynamic simulation, and in this pro-cess, sample temperature fields by snapshotting method; Second, apply the Finite Volume Method (FVM) for the discretization of the energy equation in time and space domains, and construct its state-space expression. Then, apply POD model reduction on the ob- tained thermal environment, and project the energy equation onto the subspace optimally in energy sense, which obtains reduced linear system less than ten order. A simulation experiment of a two dimensional room is applied to investigate the performances of the proposed method. Results demonstrate the approaching abilities of the surrogate model compared with the CFD-based simulations, and the order is reduced to six, which prove the method’s effectiveness.●Based on the obtained POD reduced model, a temperature control strategy is pro-posed. The main feature is that, a "offline-online" strategy is used to construct the reduced temperature field, which can help increase the control precision by real-time feedback ther-mal information. The initial state of the reduced-order model is estimated by a temperature sensor combined with a Kalman filter. To investigate the control method’s performance, both the single neuron adaptive PID controller and the model predictive control strategy are used. Simulation results show that, with the constant velocity field hypothesis, the pro-posed temperature control strategy can precisely control tiny zones’temperature of the room, and thus has the potential of thermal comfort improvement and energy saving.●Aiming at the "Perfectly mixed" assumption of indoor air in most environmental optimization strategies, and taking advantage of POD approaches’fast and high-resolution modeling features, this study develops an optimization scheme, which can improve the thermal comfort, Indoor Air Quality (IAQ) and energy efficiency in a balanced way. In this optimization scheme, all environmental parameters, such as temperature, airflow, CO2emission and Predicted Mean Vote (PMV) distributions are obtained by CFD simulations. Then, POD method is used to reconstruct the reduced parameter-spaces. Genetic algo-rithm (GA) is chosen as the optimization approach. The control variables include supply air temperature and velocity of the displacement ventilation system. The optimization ob-jective includes several environmental indexes, such as system energy consumption, indoor thermal comfort, IAQ, vertical temperature difference, etc., In each iteration of GA, multi-dimensional interpolation within the obtained parameter subspaces are used to fast obtain the system response according to each candidate control variable pair, which can guaran- tee the real-time performance of the proposed optimization scheme. A simulation based example demonstrates the method’s effectiveness.●As a typical data driven modeling method, Neural Networks (NNs) have been widely used in building energy forecasting area in the past20years. This study combines the features of adaptive network-based fuzzy inference system (ANFIS) and GA, and proposes a new prediction method, called GA-ANFIS. In this method, ANFIS adaptively tunes the T-S type fuzzy system’s premise parameters and consequent parameters by data training. GA is used to optimize the rule-base’s parameters in ANFIS. The hierarchical structure of ANFIS helps solve the curse-of-dimensionality problem. A prediction test is applied using building energy data provided by ASHRAE (American Society of Heating Refrigerating and Air conditioning Engineer) web site. Results show that, the GA-ANFIS method has the same modeling time scale compared with NNs, while the prediction precision can improve up to20%.●The GA-ANFIS method is applied to forecast the building electric energy consump-tion of Yuquan Library and Marco Polo Holiday Hotel in Hangzhou. Energy data are pro-vided by Zhejiang Supcon Software Co., Ltd., and meteorological data are obtained from Meteorological Bureau of Zhejiang Province. The forecasting results illustrate the effec-tiveness of the proposed method.