AI- new jack of all trades

AI presents challenges, however.  There is often an expectation amongst technology evangelists that an AI can make better, more informed decisions than human users. Users of AI systems (such as utility engineers or customers) may lose trust in a system where decisions are opaque, particularly if the system makes decisions that do not reflect the user’s underlying values.

For AI to be successful, users must be given space to build an understanding of its operation, and the ability to influence its outcomes to reflect their underlying values.

Key industry challenges

At a macro level, the utility industry is under pressure to contain prices while ensuring reliability & adequacy. Managing reliability & supply adequacy is becoming increasingly complex given the proliferation of variable energy resources. The useful lifetime of critical assets is at risk owing to 2-way power flow and peak demand put on by ultra DCFC’s (DC Fast Charging stations).

AI has many potential use-cases in a contemporary electricity network. They can best be described by considering the different domains (4 key quadrants) in which it might function.

How to build pragmatically? The renewable energy boom results in a far greater number of connections to the distribution network. These new connections are primarily not synchronous generators, which are more complex to analyze and integrate. AI offers an opportunity to improve DER connection processes, making them faster with better technical outcomes.

Monitoring the networks: The environment for utilities is getting more complex and data-rich. It is no longer possible to analyze this data in traditional ways. AI promises to unlock the insight from this data allowing an understanding that has previously been unattainable. Exploratory data analysis can uncover hidden insights & patterns to avert any imminent risks by learning from the past.

Operating the grid: Operation of a network with large numbers of non-traditional dispersed generation presents challenges to operate the network safely and securely. It becomes harder to predict and manage the network with large numbers of dynamic elements that may respond differently to grid disturbances and wholesale market dynamics.  Similarly, customers increasingly expect more responsive digital integration from utilities. Coupled with the increasing role of customer-owned DER in maintaining the grid, how would you engage customers differently & meaningfully without AI?

Maintaining the Assets: Continuing pressures to reduce expenditure by networks necessitates a better understanding of asset health using diverse online and offline condition monitoring metrics. Low-cost sensors integrated with IoT technologies offer data acquisition and asset monitoring in real-time but it is important to identify which assets are worth investing for predictive analytics.

How AI can help?

AI can help in multiple ways to ensure better network planning, operations, resource adequacy and personalized engagement with consumers. Let’s take an example to understand how AI can help in optimizing the integrated grid planning, given the exponential rise in scattered energy resources. Most of the network utilities have very vast operating regions and serve most distant customers as part of their electricity delivery obligation. When few customers are connected over long radial feeders, the costs of deployment & maintenance are very high. Forward-looking utilities are planning to leverage AI, Machine & Deep learning models to find out the most optimum ways of delivery future grid solutions.

Globally, utilities are considering non-network (non-wire) alternatives to defer the expensive network augmentation where constraints can be reliably addressed. There are many similar examples, where utilities are planning to model new energy solutions as Microgrids, Energy Storage, EV’s, VPPs and DER’s to meet future energy requirements of customers in a very cost-effective way.

Below is a reference architecture that could be adopted by utilities to make informed decisions while contemplating future network planning.

It follows the simple methodology of predicting how much energy demand is expected in the span of 10-30 years mapped to the edge of the grid. Based on demand, multiple cases will run to model the future network and also identifying the best transition states in between. A variety of AI/ML learning models can be leveraged based on data availability, use-cases, and decision modeling. Optimization plays an important role which can be done via linear or non-linear algorithms. Reinforcement learning will play a crucial role by improvising on changing network & demand scenarios.

Role of AI in future networks/ DSO space

Distribution markets act to democratize access to energy markets, enabling customers to realize additional value from their assets from grid services or other similar markets. Energy markets are complex however, spanning technical and economic domains to optimally dispatch diverse resources. It is unlikely customers will want to manage participation in these complex markets on their own. Similarly, distribution markets will result in large numbers of market participants in remote parts of the network. This will be increasingly challenging to manage in traditional ways.

AI has an enabling role in future DSO/TSO space. It can act as the customer’s agent, participating in complex markets on their behalf. It can seamlessly and automatically orchestrate large numbers of market participants on behalf of the grid operator. It can identify, diagnose and automate response to network contingency events such as voltage & frequency drops. AI also plays an important role in improving the network utilization indices by extracting maximum value from the existing asset base and recommending surgical network augmentations.

Key success factors

• Trust - giving people the safe space to understand an AI and influence its behaviour to reflect their underlying values (e.g. a customer may instruct an AI to increase the level of backup provided by a battery at the expense of some financial return)
• Embed AI in business decision-making process- It is important to have industry translators who can bridge the gap between “data science & decision science”. Not every data scientist can tell how to make decisions for effective network planning, operations and maintaining assets. Utilities need “AI translators” to ensure the outcomes of AI are considered in decision making process.
• Data requirements & Conflation models- Before embarking on AI path, it is critically important to consider & evaluate the data requirements. If you do not have enough “data confidence” consider investing in the right resources to ensure “right data” is available in “sufficient volumes”. Having data itself is the first step but conflation models is next mandate which brings the multiple systems data together and ready to be ingested by algorithms.
• AI/ ML algorithm selection- Finally, we need to ensure the right algorithms are selected and customized to achieve the expected business outcomes. As we need right cutlery on a dining table, for a seamless experience, we’d need a library of suitable models which most of open source systems provide today. Different business problems such as predictions, classifications and optimizations require “Champion Models” to win the business outcomes. These models are generally combination of multiple models working synchronously and competing to retain their winning position.

Conclusion

AI is a critical tool in the transition to a future grid. It’s use cases cover all parts of the value chain from grid expansion through to maintenance, operations, and customer engagement.

AI requires some care in its application, however. Not only is it important to select the “right tool for the job”, but a successful implementation also needs to consider other important factors. This may include process, data, and people considerations. Autonomous energy grids are poised to become an integral component of future energy systems with right & responsible application of AI.