Joseph Mathew

[jma1999]

Weekly Notebook Entry - Week 01 to 03

Overview:

• Week Span: Start of semester to [09-09-2024]

  Tasks for This Week:

• [x] Task-1: Gain overall familiarity with the mission and vision of VIP-SMUR.
• [x] Task-2: Understand objectives and research interests of the various sub-groups and decide which sub-group fits my personal research/career goals.
• [x] Task-3: Find learning material to get familiar with programming, data analytics, and ML - in order to be technically useful to the team and the tool we're trying to build.

  Challenges & Solutions:

• Main challenge was in understanding the ML-model that was built so far beyond the surface level of understanding - this took some time, hardwork, and research - however, it is an improving ongoing process.

General Takeaway, Skills & Knowledge Acquired:

• Enrolled in the edX course offered by NUS for Data Science - it is helping me to develop my data analytics and programming skillsets.
• Along with the aforementioned data analytics course - also taking a machine learning course which is helping me understand the different types of ML-models and how to interface with them - this is of particular use as it helps me think of how I can transfer different types of data from the architectural and building sciences components of our research to the ML-model in the right way.

Team Meetings:

• Meeting Highlights: Got to know the team members, and got to understand the diverse backgrounds we are all from.
• Peer Feedback: Peers are all helpful, hardworking, and competent.

Personal Reflection:

• Expanding my current skills and knowledge to data science and machine learning has a learning curve - but it feels very fulfilling to be able to gain relevant knowledge that fits into my overall goals as well as to be able to gain additional technical skills for contributing to the team.

Visual Documentation:

None for this week.

Additional Notes & Resources:

None for this week.

Plans for Next Week:

• [x] Next Week's Task-1: To understand the process of UBEM.
• [x] Next Week's Task-2: To find parameters of commercial buildings that are most likely to affect their energy consumption.
• [x] Next Week's Task-3: To look at different methods by which UBEM is executed and how different building parameters are inputted.

[kastnerp]

@jma1999 We need notebooks here asap!

[jma1999]

Weekly Notebook Entry - Week 04

Overview:

• Week Span: [09-09-2024] to [09-15-2024]

Tasks for This Week:

• [x] Task-1: To understand the process of UBEM.
• [x] Task-2: To look at different methods by which UBEM is executed and how different building parameters are inputted.

Tasks:

Task: [To understand different existing methods of BEM as well as UBEM] - (completed)

• Description: There are many existing ways in which building energy modeling is done, as well as, different methods by which building geometry and related data are processed to run a thermal simulation. I wanted to see how we can leverage these existing methods to help our use case. Also, to understand where we stand with respect to achieving our particular thermal simulation objectives.

• Link to Work: [Link] (Slides 7, 8, 9)

• Outcome & Reflection: We found two methods - shoebox & autozone - both seem viable.

Challenges & Solutions:

• The challenge was in figuring out how the randomly generated data that would represent different building parameters could be used to generate accurate energy usage data.
• Solution is to run those data into a Grasshopper based thermal simulation to get the usage outputs.

General Takeaway, Skills & Knowledge Acquired:

• Doing a literature review in an efficient way is a new skill acquired.

Team Meetings:

• Meeting Highlights: The discussions revolved around getting energy usage data from the existing randomly generated building data (and the data obtained from various studies).
• Peer Feedback: There were some things that required clarification regarding how the method used for running a single building's thermal simulation can now be scaled to an urban scale - as well as in identifying which parameters were to be given more importance - which will be the next set of tasks.

Personal Reflection:

• BEM is a very interesting process. The process just got more interesting now that we have an ML-model which needs to be trained with data that it can understand - which is very different from data that a Rhino/Grasshopper environment outputs - like surfaces and geometry.

Visual Documentation:

Plans for Next Week:

• We now know how to get thermal usage data for one building and want to figure out how we can scale this process for multiple buildings at an urban scale
• We also realised that out of the randomly generated data, some parameters may stay constant for commercial buildings, some may now affect it at all, and some may affect it a lot - so we need to find out what parameters are to be chosen with what level of importance assigned to them.

[jma1999]

Weekly Notebook Entry - Week 05

Overview:

• Week Span: [09-16-2024] to [09-22-2024]

Tasks for This Week:

• [x] Task-1: To look at how our thermal simulation method for one building can be scaled for mutliple buildings.
• [x] Task-2: To find parameters that apply specifically to commercial buildings in terms of energy consumption.

Tasks:

Task: [Methods of UBEM & parameters for commercial buildings] - (completed)

• Description: Looked at 2 models of UBEM - one by MIT - called UMI, and the other by Berkeley - called CityBES. For the parameters - identified what inputs were given into these models.

• Link to Work: https://docs.google.com/presentation/d/1JNc6lz630-T52qAKNN2nWqR39b52dmzBbukfHky4EV4/edit#slide=id.g3077c1a0d6b_0_4 (Slides 23 & 24)

• Outcome & Reflection: Both UMI and CityBES achieve our objective - but we need to now figure out how to add additional parameters that can make our model unique.

Challenges & Solutions:

• The challenge was to run thermal simulation for multiple buildings in a fast way.
• Solution is to either use the tools found - like UMI / CityBES - or to find a dataset which contains the required data.

General Takeaway, Skills & Knowledge Acquired:

• Used UMI - it requires Rhino 8.

Team Meetings:

• Meeting Highlights: The discussions revolved around running thermal simulation for multiple buildings.
• Peer Feedback: During our discussions, we decided to move ahead with the ComStock data parameters for the majority part as it included heating and cooling energy consumption data, but decided to add in some parameters from our end that would be unique to our model (need to identify those).

Personal Reflection:

• We had discussions regarding using shading and footprint as unique parameters for our model - this would be an interesting process as we will need to figure out how this can be inputted into the model.

Visual Documentation:

Plans for Next Week:

• We were doing literature reviews and looking into existing tools for the most part so far - now it is time to dive into this and actually run our own simulations with the data that we have, for the city we are looking at for the moment. This will give us further issues to tackle - and a deeper understanding of the pipeline we will need to build.

[jma1999]

Weekly Notebook Entry - Week 06

Overview:

• Week Span: [09-23-2024] to [09-29-2024]

Tasks for This Week:

• [x] Task-1: Rebuild an urban-scale model for energy simulation (Tech Square as a case).

Tasks:

Task: [Test energy model for Tech Square as example] - (completed)

• Description: We wanted to see whether the parameters we've collected so far and methods we talked about could be applied to run an actual thermal simulation. This included modeling massing geometry for a section of Tech Square, choosing a building to run thermal simulation, identifying parameters that could be input into the ClimateStudio template that was used to run the simulation, and extracting contextual geometry that would be used for shading (as shading is one of the parameters we want to use as a unique feature to our model.)

• Link to Work: https://docs.google.com/presentation/d/1JNc6lz630-T52qAKNN2nWqR39b52dmzBbukfHky4EV4/edit#slide=id.g3077c1a0d6b_0_651 (Slides 29 to 35)

• Outcome & Reflection: It is clear that simulation for individual buildings are not an issue. However, this process can be made quicker by extracting geometry data from the datasets we are referring to. Furthermore, we can now look into ways of encoding footprint & contextual shading data into a format that the ML-model can understand.

Challenges & Solutions:

• The challenge was to run thermal simulation using the building data we are currently using, furthermore, to look at a portion of Atlanta in particular for this case.
• ClimateStudio works fine for the time-being - but as we move up in adding more parameters, and running more buildings, I suspect we will not be able to use existing templates for much longer.

General Takeaway, Skills & Knowledge Acquired:

• ClimateStudio - using the existing template to run thermal simulation using building data we currently have.

Team Meetings:

• Meeting Highlights: The discussions revolved around running an energy model for a building in Tech Square so that we can start zoning in to new issues and workflows.
• Peer Feedback: We want to be able to extract geometry directly from the datasets we are referring to so that we won't have to model multiple buildings individually.

Personal Reflection:

• Now we will be looking into making pipelines that solve specific problems in our overall workflow - the current workflow to target is extracting geometry data from IDF/OSM files.

Visual Documentation:

Plans for Next Week:

• To build a pipeline for extracting geometry data from IDF/OSM files that the dataset we are referring to gives us.
• To also figure out ways to encode information like shading and footprint into a way that can be inputed into the ML-model - this will also make our tool unique compared to existing tools.

[jma1999]

Weekly Notebook Entry - Week 07

Overview:

• Week Span: [09-30-2024] to [10-06-2024]

Tasks for This Week:

• [x] Task-1: Extract geometry information from IDF/OSM file and merge them to existing energy demand dataset.
• [ ] Task-2: Develop algorithm to recognize and classify footprint shape & contextual shading. (ongoing)
• [ ] Task-3: Study about how to encode such non-numeric variable to regression model. (ongoing)

Tasks:

Task: [Extract geometry information from IDF/OSM file and merge them to existing energy demand dataset] - (completed)

• Description: In order to extract geometry data from IDF files - the process was straightforward - get the IDF file, then convert it into the correct version that Ladybug can read, then it just needs to be plugged in and the required geometry data can be extracted. However, for OSM files, the process was a little trickier - the OSM file isn't read directly by Ladybug (it is able to read IDF and gbXML formats without issues) - so we opened our OSM file in OpenStudio - then exported it into an IDF and tried running it in GH but it was giving weird version errors. So we tried exporting it in the other format - gbXML - this time it gave a clearer error regarding a thermal resistance value for a component which seemed to be below it's threshold - so the solution was a straight-forward fix - you can either edit it directly by opening the xml file in notepad / you can edit the value in OpenStudio - we tried both - both worked!

• Link to Work: https://docs.google.com/presentation/d/1YwYqGYVsC_ukeOOAJawuhVWrvwQyaXqNCMKWhPEq4w8/edit#slide=id.g3064c5bb2b6_0_27 (Slides 17 to 23)

• Outcome & Reflection: There is now a pipeline to convert IDF & OSM files into geometry. However, this needs to be reviewed. I suspect that some aspects of this workflow (such as editing thermal resistance values & the error with IDF file versions) are not efficient enough to make this scalable. Also, there is still a need to figure out how to make this workflow run for multiple files simultaneously, if required.

Challenges & Solutions:

• Most of the challenges that arose were regarding differing versions of IDF files/radiance/openstudio. These weren't difficult to fix and arose more likely due to personal inexperience with these plugins. The other error regarding low thermal resistance value for a copper pipe seems to be an unique issue to the fle we teste, and such unique issues might arise for other files as well.
• The solution to the issue regarding differing versions is to get the right verions. The solution to the latter error was to edit it in notepad - or to edit it in the OpenStudio UI.

General Takeaway, Skills & Knowledge Acquired:

• Much more comfortable now manipulating the ClimateStudio plugins, fixing errors on my own by using community forums/videos.
• Familiarised myself with the OpenStudio interface as well as editing portions of information of an xml file in a text editor.
• Furthermore, from the extra courses I've been doing on ML and data science - I've been getting familiar with machine learning models - so far covered KNN, Logistic Regression, and Naive-Bayes - planning to cover the particular models we're using over the next week.

Team Meetings:

• Meeting Highlights: The discussions revolved around extracting geometry data from the datasets we're referring to, as well as finding ways to encode unique building features we've identified for the ML model.
• Peer Feedback: The weeks goals were met, this has shown a few new problems that can be tackled - mainly regarding making the next pipeline for contextual shading and building footprint values.

Personal Reflection:

• Compared to previous weeks - now it feels like we're more in track to getting values that can train our model to be unique - mainly because we are looking into contextual shading and building footprint parameters. Furthermore, I have been getting more comfortable with ML terms from the courses that were shared with me. It is helping me to think about the data I'm trying to extract in ways I wouldn't have when I wasn't as familiar with ML.

Visual Documentation:

Plans for Next Week:

• New pipelines to make - for contextual shading and building footprint values.