In this study, the effectiveness of a model predictive control (MPC) strategy for an office building subject to occupancy disturbance and time-varying electricity pricing was investigated. An artificial neural network and metaheuristics algorithm were employed for prediction and optimization, respectively. Prior to the MPC implementation, different condition of prediction and control horizon was estimated. As a result, MPC managed the room temperature well and saved total operation cost.
In this research we proposed a method to analyze the energy pattern in university buildings using K-means clustering method. Energy consumption in Science, Non-science and office buildings of university is analyzed and their respective base energy, energy consumption due to human activities and air conditioning energy consumption is calculated. This method will help for the planning of energy conservation in buildings.
Energy saving is required for air conditioning systems. Optimal operation using conventional energy simulation has a problem that it takes time to create a model. Therefore, we will consider how to automatically determine the optimal operation by using Bayesian optimization. In this paper, Bayesian optimization and energy simulation are compared and evaluated. As a result, it is shown to be possible to achieve the same accuracy as energy simulation with a small amount of observation data.
In this study, we developed a long-term performance prediction method of the earth-to-air heat exchanger (EAHE), and also implemented reinforcement learning (DQN) to this method in order to construct a control law that achieves both energy savings and suppression of dew condensation.
Energy and water consumption data of the DECC database were analyzed. In non-certified office buildings, energy and water consumption per square meter of buildings built before 2011 were 18.2% and 35.4% larger than those built after 2012 respectively. In non-certified public buildings, energy and water consumption per square meter of buildings built before 2011 were 27.9% and 24.2% larger than those built after 2012 respectively.
We proposed a control method of heat source system with water heat storage tanks to minimize CO2 emissions named as CADR (carbon activated demand response). CADR considers dynamic coefficient of CO2 emissions and optimizes the chiller operation using model predictive control. Compared to the conventional method, the CADR reduced 30% of CO2 emissions.
In the paper, we describe pre-processing methods based on the actual power consumption data of a comprehensive university. Data obtained from buildings is huge and its measurement items are diverse. Those obtained from buildings are expected to be used in the advanced data science field in the future.