NCSI: Sequential Monte Carlo (SMC) Methods for Data Assimilation

Status	Completed
Mentor Name	Sanish Rai
Mentor's XSEDE Affiliation	Undergraduate Faculty
Mentor Has Been in XSEDE Community	1-2 years
Project Title	Sequential Monte Carlo (SMC) Methods for Data Assimilation
Summary	The student will continue to work on implementing sequential monte carlo (SMC) methods using XSEDE resources for data assimilation purposes. SMC methods will be implemented for simulating real-time applications (building occupancy estimation). The student will work to parallelize the SMC methods and produce some significant results.
Job Description	The student has already started working in the research. In Fall 2018, the student worked on the literature of the SMC methods, agent-based and graph methods and also started working on coding them. Currently, the student is working on a basic particle filter (SMC) and combining with the simulation model. As a next step, the student will start working on parallel computing and try to parallelize the SMC methods for data assimilation in building simulation. Student will research on existing methods of parallel SMC and will select the best method applicable to our model. From the model sensor data will be collected for occupancy behaviors for various time periods. Data analysis will also be performed on the collected data to study the behaviors of occupants. Parallel Sequential Monte Carlo methods algorithm will be implemented to estimate the behavior in real time. Existing algorithm will be modified to make the process efficient and experiments will be conducted. Results of parallel SMC method in building simulation model will be compared with sequential methods and the results will be analyzed. Other machine learning algorithms will also be explored for real-time application.
Computational Resources	As the next step, XSEDE resources will be used for research and implementation of parallel SMC algorithm. Simulation process of occupancy behavior is computationally expensive. Due to the high number of agents and building structure, as the number of occupants increases the process becomes slow. As such computational resources of XSEDE will be utilized to increase the efficiency of the simulation. Parallel processing for agent modeling will be researched. Data assimilation is the process of using real-time (sensor) data with sequential monte carlo method to make dynamic estimation. Due to complex states of sequential monte carlo methods, when the occupancy is increased data assimilation becomes slow and an efficient estimation cannot be achieved. The research aims to look into use of XSEDE resources to increase the computational efficiency of the overall real time estimation process. We also plan to use XSEDE for sequential implementation and compare the results.
Contribution to Community
Position Type	Apprentice
Training Plan	The student has been trained for sequential SMC methods and use of existing agent-based and graph-based simulation model for building occupancy estimation. The student will be trained for parallel algorithms and how to parallelize SMC methods. The student will follow tutorials from online to learn the basics of the parallel algorithm. Any training materials and resources from XSEDE will be helpful to the student to achieve the research goal.
Student Prerequisites/Conditions/Qualifications
Duration	Semester
Start Date	01/07/2019
End Date	04/26/2019