Weekend Read: Artificial opportunities

SolarEdge’s product vice president, Ido Ginodi, explained how AI is being used to optimize energy management systems and how it handles fundamentally tough optimization problems and forecasting in a way that traditional control algorithms cannot – even in the home. Israel-based SolarEdge is well known in the PV industry and as complexity emerges between energy generation and storage, EV charging, data, and forecasting, Ginodi showed considerable enthusiasm for how his company is using AI’s advantages. “The lines between good solid algorithm approaches and AI are blurry,” Ginodi said. “But after spending a few years researching AI in academic settings, a lot of what people are doing, including us, in this field is truly AI-driven and it promotes our ability to offer state of the art energy optimization.” Ginodi explained that AI is not only required when an application grows in size from a single dwelling, with just one EV charger, to multi-dwelling buildings and commercial and industrial sites with multiple, possibly hundreds, of chargers. “I actually want to argue something a bit different: In the residential use case, AI is extremely important,” said Ginodi. “The problem of energy management is fundamentally a tough optimization problem. We started our journey with the concept of power optimization, optimizing the amount of juice that can be squeezed out of solar arrays. Now we are taking it a few steps ahead, optimizing a whole-site performance, which is an order of magnitude more complex.” The SolarEdge executive explained that an energy management system can optimize metrics for the end customer’s benefit. It does so while orchestrating elements such as PV generation, battery dispatch, EV charging, and load orchestration. Systems can also optimize heating, ventilation, and air conditioning integration for pre-heating and cooling, while accommodating dynamic tariffs and market participation, and even preparations for outages, by using data to make decisions. “It ends up having multiple degrees of flexibility,” said Ginodi. “It’s a lot and it’s fascinating, and in some places AI-driven solutions may generate results that are significantly better than what a naive algorithmic approach could have achieved. But we go further. We develop predictive models based on machine-learning regression techniques for consumption, production, import and export tariffs, and one for grid events. Once you have those four models, you can have classical algorithms make the decision on how you want to dispatch the different resources you have in a system.” For the end user, this translates to the management system either optimizing for profit, as is common, or optimizing for convenience or for decarbonization, per user preferences. Ginodi added that SolarEdge portfolio companies also work closely to incorporate AI capabilities into its offering. In particular, EV charging management company Wevo works to cost-effectively scale EV charging with predictive load management and capacity management. While static and dynamic load management technology is becoming more abundant in the industry, AI in the form of predictive modeling offers significant improvements to the concurrency factor – that is, the ability to fit more chargers under a given grid connection point. “Say an enterprise wants to offer electrified parking spots in its car park,” said Ginodi. “It’s extremely costly to offer 100 new spots at 11/22 kW each. That’s 1 MW or 2 MW of extra power required. A brute force approach would be to require the full power provisioned for the system but you don’t have to charge the vehicles together and you don’t even have to statically attach capacity to each charger. That’s dynamic load management. One step further, you can incorporate the predictions Wevo generate and build an optimal schedule for charging. The model assumes that cars will appear in a parking lot at a certain velocity and what will be the levels of local production and total consumption at each point in time. “With these predictions at hand, one can serve more vehicles and drivers. Up to 20 times more, compared to a naïve implementation.”