IMPRINT

Project Title: Data-driven modeling, analytics, and optimization techniques to manage building thermal demand.

Principal Investigators: Prof. Krithi Ramamritham (Professor, Dept. of CSE, IITB) and Prof. Milind V. Rane (Professor, Dept. of Mechanical Engineering, IITB)

Funding Agency: IMPRINT (IMPACTING RESEARCH INNOVATION AND TECHNOLOGY) – Ministry of HRD, Govt. of India

Overview:

Rapidly growing energy demand has lead to poor quality of service (black-outs, brown-outs, and load-shedding), depletion of resources and impact on the environment [15]. We address these concerns by developing solutions for:

  • reducing consumption by adopting occupancy and need-based appliance usage and
  • flattening peaks by applying timely demand-response strategies including scheduling and resource allocation.

However, managing energy demands cannot be left to ad-hoc approaches that depend on human actions: Information and Communication Technologies have a key role to play through:

  • occupancy and needs-based appliance usage, and
  • timely demand-response strategies, including appropriate scheduling and resource allocation.

We exploit learnings from our prior work and extend its scope through further elaboration and experimentation. One highlight of our solution approach is that it makes minimal use of physical sensors by employing additional (soft sensing) resources, introducing the principle of observability.

In the project, we aim to build an integrated system that systematically uses different approaches to achieve the aim of reducing energy consumption using a minimal number of sensors. For achieving this goal, we define the fundamental blocks of our SMART energy management for buildings:

  • Sense Meaningfully, Analyse and Respond Timely (SMART) Cycle: After careful examination of how sensors work in an application, we deduced that all the sensors follow the SMART cycle, where the sensor has to
  • Sense Meaningfully the parameter it is designed to sense,
  • Analyze the sensed value (might even perform computation) to help in decision making and finally
  • Respond Timely with a decision or an output.

 

Research Outcomes:

  • Facet-Sensor Relationship Graph: By analyzing the input and output of different sensors, we claim that the output of one sensor used to observe a particular facet can be used to predict the values of other facets being sensed, i.e., predict the output of other sensors. For example, temperature and occupancy can be provided as inputs to an inference engine (like a Neural Network) which predicts power consumption.  So rather than using a smart meter at every place, we can use temperature sensors and counting doors (or RFID data) to predict power consumption at any given space within a building thereby using only two sensors to observe three facets. Hence, facet-sensor relationship graph helps us observe more facets using a minimal number of sensors.
  • Aggregation to combine and propagate the data: With a large number of sensors, it becomes important to use the data generated by them, in a meaningful way. This can be achieved by combining the data from various sensors using Aggregation Modules. For example, to determine the power consumption of a floor, the data from each sensor should be combined by taking their sum. Also, data from all the sensors should be synchronized to avoid inconsistencies in data.
  • Frameworks which help in reducing the number of physical sensors: Every building can be viewed as a tree. The flow of information in a tree representation of a building can happen either top down or bottom up, depending on various scenarios. A disaggregation framework is used when a top-down approach is followed. In a given building, we can use the disaggregation framework, with some inherent inaccuracy, without installing smart meters in each room. Thus the framework reduces the number of physical sensors when a top-down approach is present. A prediction framework is helpful during a bottom-up approach. We can use a prediction framework to predict the power consumption of the floor without actually deploying smart meters. Thus, employing these frameworks helps reduce the number of hard sensors required.
  • Using building state as Real-Time Database: We view the building as a cyber-physical system with the presence of a sensor network. Since this generates real-time data values, we use a real-time database to store these real-time values. These values mainly help us to determine the state of the building and access to real-time and archival data for performing analytics or inference. It is necessary that the data satisfy properties like logical consistency, temporal consistency, and correctness requirements. Satisfying these properties helps to estimate the state of the building by using real-time databases.

 

Unique and Innovative Aspects of the Project:

The novelty of our approach to smart energy management lies in judiciously combining:

  1. physics based and data-driven models,
  2. off-line planning with timely dynamic decision making, and
  3. minimal physical sensor infrastructure with sophisticated soft sensors (virtual sensors that can replace physical sensors) based on our notion of observability.

Our solutions will be showcased in campus buildings, and subsequently packaged and deployed in our partner’s’ premises