Data Storage and Data Mining of Building Monitoring Data with Contexts

In order to improve the energy efficiency of buildings, many sensors have been being installed in buildings for monitoring temperature, humidity, electricity consumption, etc. In the Building Energy Management Systems (BEMS), it seems that visualization of the monitoring data is the main objective and that managers still have to find problems from the visualized data with drawings of buildings as contexts. As the number of sensors and amount of data increase, it will be difficult to find problems. Although, data mining would be an effective means to automate this process, two problems exist. The first is despite the necessity of contextual data of the surrounding environment of sensors to discover meaningful knowledge, it is difficult to define the data model to store monitoring data and contexts beforehand and it is very difficult to change the data model afterward. The second problem is that simple data mining techniques are not effective for time-series and massive monitoring data. Thus, in this research, first, a flexible and variable data modeling technique, named A&A method, was proposed to incorporate various and vague contextual data into the sensor data model. Next, Data Mining for Sensor data of buildings using Contexts (DMSC) method was developed to support building managers to discover useful knowledge from a large amount of sensor data for better building energy management. To check the feasibility of the proposed methods, we stored actual monitoring data in our laboratory at Osaka University with the developed data model and executed data mining, using DMSC method.