SUSTLABS OHM ASSISTANT USER DOCUMENTATION
D001 SustLabs Ohm Assistant Consumer Product Note
Author: Rajan Mathew, Client Success Leader, SustLabs
Version 1.0 2022 0323
Note: While best efforts will be made to keep this documentation updated, please note that some text or illustration may differ with app or platform or type of hardware or updates.
SustLabs presents Ohm Assistant, the electricity activity tracker designed to help residential electricity users become energy wiser and ultimately make homes smarter. This is an IoT product for residential (and small office) electricity users to track overall and appliance level electricity consumption without deploying any sensors. Get ready to “sense the heart-beat of your home”.
Ohm Assistant is aimed to be a power monitoring solution which provides information regarding electricity consumption of data. Multiple positive outcomes include accurate forecasting of electricity bill, preventive care and servicing of heavy appliances and maintaining a record of all the appliances, with the ultimate objectives of savings & sustainable living.
Basis the information available, consumers can act to save power by identifying high consumption appliances, electricity leakages, usage patterns etc. to implement measures and bring savings to power consumption.
Real Time Live Tracking
Ohm Assistant is a real-time electricity activity tracker for the home. On the App Home Screen, LIVE implies that the data is being collected on real time basis.
Overview of Technology
Primarily SustLabs analyses & provides the overall electricity consumption of the house. Big Data is compiled and dissected through Machine Learning (ML) through in-built proprietary also helps in real time record of electricity consumption at the home. The electricity consumption information is displayed via a web or app interface.
Ohm Assistant seeks to inform users about the activity & performance of heavy appliances. Using real-time energy monitoring ML-BOT (Machine Learning Bot), electricity consumption activity is analysed right up to the appliance level. The machine learning processes work through monitoring & analysis at the global level and eliminates the need to deploy sensors at the appliance level.
Machine Learning Bot (ML-BOT) Working & Training
The ML Bot detects the overall consumption for categories of heavy appliances in the home. Categories include Refrigerator, Air Conditioner, Geyser, Washing Machine, Heating Devices (Iron, Kettle, Microwave, OTG, Toaster, Air Fryer, Grinder, Hair Dryer), Water Pump Motors etc. The ML-BOT requires sufficient training to enable it to differentiate between similar assets.
The ML-BOT collects data samples & transmits the same every 4-10s. The same is added to the raw database. This raw data is read every hour and aggregated using tools. This aggregated data is displayed on the Ohm Assistant Smartphone application for users to understand their energy consumption patterns in a simple language and format.
The Appliance Activity works in near real time to allow for analysis to deliver high quality analytical accuracy. Based on ML system intelligence with no human intervention, the Bot displays only selected appliance activity incidents, generated automatically, for selected incidents for which the system requires user validation, and here not all detections incidents.
Each specific Appliance Activity detection incident would be displayed up to 2 hours after the event. The aggregation on hourly basis is focussed on improving analytical accuracy, and if the interval of aggregation were to be reduced, the accuracy would be lower. The timestamp when the appliance was switched on is displayed here.
ML processes involve disaggregation of total energy consumption of a household into individual appliances, using smart energy meters and big data analytics. One technique used in this domain is termed non-intrusive appliance load monitoring (NIALM). However, there are unsolved problems with respect to its practicality and effectiveness at high sampling rates.
Appliance Detection Technology
Ohm uses adaptive machine learning models, trained on internally generated datasets down sampled at a 3-second resolution. Individual appliances including geysers, air-conditioners, refrigerators, ovens, fans, tube lights, bulbs, washing machines, irons, kettles etc. from multiple households intensively studied, thus creating an extensive database with features.
Various machine learning algorithms are applied on the available dataset, and using edge detection techniques, identification is done whether an event has taken place or not. Feature Extraction is carried out subsequently, where each positive event is matched with its equivalent negative event in a particular window.
Each event pair is then labelled by comparing its signatures with the existing database of appliance signatures. In case a match is not found for any particular pair, the user is requested to label that pair depending upon the time of its occurrence and the model as well as database are updated with new values.
Appliance Activity Feedback from Users
On a day-to-day basis, users must pay special attention to each Appliance Activity event that is displayed and provide required feedback for each activity. This is essential to ensure proper training of the Bot in the Machine Learning experience, to improve the quality of detection. Therefore, this list must be visited by the user and cleared through the feedback mechanism.
Click on the particular event under Appliance Activity and kindly provide proper feedback. If the specific incident is a proper prediction, click on YES, ITS FINE to confirm that this incident is a proper detection. If the event does not seem to match expected appliance usage, click WRONG DETECTION.
The Android platform where new developments release first, features a pathbreaking enhancement to the Appliance Activity feedback, that offers the user a superior & sharper feedback mechanism, with the option to TAG YOUR APPLIANCE, and map the specific detection incident to specific appliances, complements Train Your Bot.
Note that Appliance Activity Feedback is an irreversible process, therefore feedback must be accurate and given only when the user is confident about the event. In the latest product iteration (Android), if unsure about the detection, an option to provide this feedback (Unsure) is also available.
Further, if there is any major discrepancy noticed in the detection, such as incorrect devices or missed detections, users can raise a support ticket, and the SustLabs team will further analyse the same.
Sharpening Analysis by Train Your Bot
Directly train the Ohm Assistant Bot to recognize the specific appliances, to vastly improve the detection accuracy with this feature. Follow the Appliance Training Guidelines, leaving only essential appliances (light, fan, Wi-Fi) on if in use.
To avoid mistraining, switch off all heavy appliances (AC, Geyser, Washing Machines, Heating Devices etc.), especially those appliances to be trained. Complete the stability check and then proceed to train the specific appliances following the on-screen instructions.
Please note that the prerequisite for stability is ensuring that no heavy appliances are running, else the same affects stability and interferes in the training process. There are three levels on the Train Your Bot Process to be completed for each specific device.
Diagnosis & Health (Currently available only for Air Conditioners & Geysers)
The Diagnosis functionality is designed to evaluate appliance health, and further help get information regarding possible energy leakage sources in the home through the specific devices. Ensure that all heavy devices are turned off, to minimise interference and create the stable setting for optimal diagnosis tests. This is a manual process where appliance health status is reported as Healthy, Unhealthy, Critical & Unknown.
Importance of Defining ALL Devices Under Vyas:
VYAS is developed as a recording tool for all home appliances, as a “Ledger”, maintaining the list and details of the devices associated with the bot. wherein users can maintain details like make, brand, purchase date, warranty information, service date, AMC as well as star rating of the appliance.
Ensure that each and every heavy appliance is distinctly defined in Vyas, with proper categorization, to ensure that the specific appliance is included in the detection and analysis processes, else appliances not defined will not be properly detected, and this would reflect under Others.
The importance of accurate appliance information comes into play for advanced diagnosis, where the analysis is based on the Appliance Model, and the working of the device during the diagnostic test is compared to the BEE database, which has threshold values defined. Using this, we can provide feedback as to whether the device is operating in healthy, unhealthy or critical mode.
Once appliances are defined by the user, then only these appliances & categories will be included in the detection & analysis. (Any false device will also falsely be displayed in the detection). Thus, proper updating of Vyas is important to identify & help distinguish the number of identical appliances.
Wi-Fi Constant & Strong Connection Requirement:
The Ohm Assistant requires Wi-Fi 24x7 for the data to flow in seamlessly. Any disruption or break in connection will cause detections to be missed, since data recording will be incomplete, and incomplete data is often useless.
Please ensure that the Wi-Fi connection is strong, since there are occasions where data collection points are weak because of several competing devices at the Wi-Fi router APN, which will affect the quality of recording of data points.
IMPORTANT NOTE ON PRODUCT LIMITATIONS
Note that this is an evolving technology, and thus certain limitations exist as follows. The same are specified under the Terms & Conditions which are binding on the usage of the Ohm Assistant product as follows.
- The detection that the bot does are the predictions and not the actual consumption of the appliances.
- As of now, we can detect only heavy appliances in the house i.e., Air conditioners, geysers, washing machines, fridges, and heating devices (comprising iron, kettle, oven, microwave etc.).
- We operate at an average detection accuracy of 85%. The accuracy may affect depending on the size of the house.
- Appliance segregation is provided at an overall category level and not on individual device level or phase wise.
- For example, we would be able to predict the consumption of air conditioners all together and not how much each air-conditioner is consuming.
- Depending on the size of house and number of appliances, some appliances may not get classified correctly.
- We do not explicitly mention leakage or theft causing appliances.
- Train your bot may or may not improve the appliance detection accuracy.
- Diagnosis is currently available only for Air Conditioners & Geysers.
- Wi-Fi availability affects the appliance detection accuracy.
- Some of the features in the App are under development or testing and thus will be made available for use when ready.
- In continuation of the previous point, as of date, the Android App is the most mature, whereas the IOS App and WebApp are still evolving. This means that while core features are available in all apps, the enhancements and optimised features are available in the Android App. We are working to bring parity to all apps and will keep our stakeholders updated as this happens.