leon-thomm / nus-cs3237-iot-buoys Goto Github PK

• Seek advice from prof about drilling into the plastic container - is it advisable or should try alternative means (EG hot rod for incisions)
• Potential for waterproof temperature sensors (hot-gluing the sensors directly may have damaged them)

Given that we are working with time series data, we can use LSTM neural networks.

See this group that did weather forecasting with time series data for inspiration:
https://kendrik-wah.github.io/cs3237_weather_nowcasting/frontend/html/index.html

@leon-thomm

1. narrow down the use case - what state/s exactly do we want to predict, and how?
2. notice the additional accelerometer in the set, it is much lighter than the phone
3. it's ok to pull in additional data sources to feed the models, such as OpenWeatherMap API
4. architecture seems good
5. finish POC in two weeks from now

To include some type of realtime component, we thought of implementing a wake up mechanism, where a WeMOS sensing some significant change in data can wake up others to pull everyone in an awake mode, causing them to record with higher frequency.

power management

The above implies at least two different modes of operation for the WeMOS. Unfortunately, it seems like putting the WeMOS into an actual sleep mode is not very useful, as it will probably break the WiFi connection, and reconnecting to WiFi is 1. unreliable and can 2. strongly vary in delay. Therefore, for now, we keep the WeMOS awake and just use simple delays.

resources:

• a useful guide
• ESP8266 low power manual

wake up

We will try to simply use a dedicated MQTT channel to which every WeMOS subscribes and publishes to.

@Tino3141

Add SSH public keys here for access to hosts.

104.248.98.70 (iot.ckteo.com)

Data collection endpoint, stores data within postgresql & mongodb

SSH into respective accounts (ch, ck, yh, tino, leon; all accounts have sudo rights already) at port 3237.

• MQTT Server
• HTTP Endpoint