AI Case Study
Verv plans to predict renewable energy generation and usage for its p2p marketplace using deep learning
Verv is to begin trialling p2p energy trading on blockchain platform Ocean Protocol. Consumers with excess renewable energy, such as solar, will be able to sell it to others on the blockchain. Verv claims its added benefit lies in its ability to predict future energy consumption via machine learning and large accumulated energy data set.
From NullTX: "Verv is a blockchain-based system for people to make and save money by trading energy with others in their area, using machine learning AI to automate optimization of the entire process for everyone." Verv implements AI by "Using signal processing and machine learning techniques, it disaggregates energy readings taken from your home’s main lines to determine how much energy each appliance in your house is using. From this information it can predict energy consumption and allows homeowners with solar panels and home batteries to trade energy with other neighbors using the blockchain-based Verv Energy Trading Platform facilitated by VLUX tokens".
From a Verv whitepaper: "Verv expects this to result in the efficient utilisation of electrical infrastructure, and therefore provide an assurance that the generated kWhs will be traded at the right time for the best economic return. To achieve this, the VHH has leveraged key developments in machine learning, particularly deep learning. In order to decide whether energy should be transacted, the Verv AI is being developed to take into consideration:
* Current battery state of charge
* Maximum battery capacity
* Forecast household demand for electricity
* Forecast generation of electricity from solar panels
* Any additional input parameters set by users (e.g. minimum kWh charge of battery at all times, or maximum trade amount)"
Verv received funds from the UK government to simulate their p2p trading market across the UK in 2017 and is intended to be trialling it in areas in London (Q1 2018). "Verv developed the blockchain-based solution, named Verv 2.0, because homeowners with renewable energy sources only use 30% of the energy they generate. While battery storage technology improves this to around 60%, there is still excess energy which is being underutilised (Eurobat 2016’s report). Verv’s P2P energy trading solution aims to tackle this problem by utilising the excess energy to improve access to affordable low carbon electricity for all households in the UK." (What Investment)
From Verv's whitepaper: "Advances in machine learning and AI have over the past few years transformed many industries, improving predictive capabilities and providing better models of human behaviour. For example, the development of complex machine learning algorithms has enabled retailers to predict customer activity, understand trends in consumer purchasing behaviour, and adapt advertisements and content for customers’ individual preferences. Verv believes these advances can be applied to the energy sector, using algorithms to disaggregate electricity data to appliance use information, and providing better predictions of energy use behaviour. Under this lens, Verv believes machine learning will become a critical enabler for energy trading and the balancing of supply and demand of electricity."
Due to the high frequency data sampling, What Investment reports that "consumers will get the best possible electricity prices, in the same way the foreign exchange market works".
According to Verv's whitepaper: "[Verv Home Hub's] existing presence in households means that Verv has an extensive dataset of detailed appliance-level energy usage, enabling it to train powerful models of consumption behaviour... Verv has designed the system to process these data feeds with neural networks, which are being trained to identify interconnected patterns so that Verv’s forecasts for generation, consumption, and battery activity can be continuously improved... Verv expects the AI algorithms should be able to learn not only customers’ electricity consumption behaviour, but also their patterns of engagement with energy trading. Rather than requiring ongoing input and decisions from the user, it is anticipated that the algorithms will learn how the user interacts with the platform to provide a customised experience that ensures long-term, sustainable customer engagement."
High-frequency sampled electricity data. According to Verv on Medium:
* "Granular detail on household electrical activity, both on i) appliances and ii) microgeneration/storage assets (capacity,
performance, state of charge)
* External factors influencing electricity generation/consumption, specifically weather forecast data, geolocational data, satellite data on cloud coverage and opacity."