Towards an Internet of Energy, Part II: Steps to building Autonomous Distribution Grids
Towards an Internet of Energy, Part II: Steps to building Autonomous Distribution Grids
The future of the grid will be a network of interconnected distribution networks. To achieve that, we need to map the network in real-time; a more complex task than it seems
As we discussed in Part I, the future of the grid will be a network of interconnected distribution networks, much like the Internet is a network of networks. But what are the steps towards achieving this vision? What are the barriers and why haven’t they been overcome?
A once passive grid now needs active management
The distribution grid is undergoing rapid change. As more and more electrification occurs through technologies like EVs and heat pumps, demand is going up. Meanwhile, distributed generation through rooftop solar and batteries is turning a grid designed for unidirectional power flow into a bi-directional, dynamic grid.
Despite these changes, the grid has not changed much in the 100+ years since it was first built. The grid was built for a world with unidirectional flow of power from centralised power plants and generators to users. There was very little to control and coordinate, and 99.99% of the grid is managed on human-to-human communication (e.g. linemen phoning in changes as they work on the power lines).
The grid is basically operating on a switchboard by switchboard operators. Most of the operator’s work is monitoring and coordinating via phone calls with field crews. These operators need to coordinate resources to:
Balance supply and demand on the grid
Ensure safety
Ensure reliability
All while operating on physical constraints measured in 1’s/10’s/100’s of cycles at 50Hz (i.e. <0.1s at times), with real-world consequences and implications for safety if things go wrong.
The current system operation at the control center is relatively simple yet completely dependent on the operator's capability of understanding the situation based on his/her experience. An operator sees the information in the form of data/alerts in different systems, comprehends the current situation based on past experience and decides, based on assumptions and experience, how to handle the situation. The monitoring, analysis and control in the current system is manually driven by the system operators.
This is a far cry from the autonomous systems of the internet but since the grid had fairly stable flows of power from centralised generators to users, and demand followed a fairly predictable pattern with limited load growth in recent decades, this was a relatively easy job. Until recently. As the distribution grid gets more complex and interconnected, it will get harder and harder to manage the distribution grid.
Defining connectivity: The importance of knowing network topology
The reason the internet is able to operate autonomously is because independent networks are able to communicate via standardised protocols. Any new network or device can connect and communicate via TCP/IP - the address system of the internet. As we discussed in Part I, all routers know the connectivity of all other routers and are able to route data accordingly. The topology of the internet is known.
In the distribution grid, this topology is unknown. Connectivity may be estimated, or even defined, but the distribution grid is constantly changing as linemen open and close connections to complete repairs, power is re-routed to avoid overloading substations and the daily operations of the grid occur. Most estimates are inaccurate and out-of-date, and less than 1% of topology changes make it into distribution grid planning and control models. There is no understanding of the real-time connectivity of the as-operated grid. No-one knows how nodes in the distribution grid are electrically coupled.
The distribution networks have historically operated on a passive basis. They were built under the assumption that power flows unidirectionally from substations to customers. The control centers have not had to perform control actions and therefore there has been limited need for real-time visibility into the network. As a result, they have very little in the way of advanced monitoring and control techniques, and there is minimal automation. Manual processes and human-to-human communication limit the flow and reliability of network information. Distribution equipment is built to handle estimated worst-case conditions and the system somewhat runs on its own.
Now, operators are having to actively manage the grid and they are flying blind. Nobody knows what is connected to what on the grid, in real-time. No-one knows how the grid is configured every second as different sub-stations and switches open and close. You can’t see the as-operated state of the distribution grid. In summary, operators don’t know:
Which meters are connected to which feeders
Which distribution lines are open and closed
How this is changing in real-time
As a result, operators can’t see:
Loops in the network that could cause issues
Alternative network paths and configurations as they balance power and prevent equipment from exceeding capacity
Changes that are happening as Linemen open and close distribution feeders, as none of the information gets communicated back to the control center
Meanwhile, transmission system operators know nothing about the distribution nodes and observe only their aggregate effect at defined load sinks. They can’t see how changes to the distribution grid will impact the transmission system.
In the distribution grid, connectivity identification is a latent & error prone process that’s reliant on human communication. There are thousands of nodes and physical junctures on one distribution feeder, and operators have minimal knowledge on the status of any of them. Despite some attempts to solve this, fewer than 1% of nodes on a given distribution feeder have switches with status data available in real-time, and low voltage networks (the stuff closer to the customer) and the associated customer connections may be entirely undefined. Where network topology is defined, parameters may change as maintenance crew repair lines, or the grid expands and upgrades to host new load, but updates don’t make it back into the network model.
Currently network models are updated and calibrated infrequently, yet changes to network connectivity occur frequently and remain unreported. At the same time connected loads are becoming more uncertain as more DERs and renewables come online. Without a communication system that supports it, a proper map of network topology requires humans to maintain which inevitably leads to out-of-date data and errors.
Fundamentally, grid operators and planners are working on a network map where they have:
Static topology data for a highly dynamic and constantly changing network
Undefined network topology for an interconnected network
Unknown connectivity & capacity of nodes for an expanding network
As DERs and distributed generation continue to input into the distribution grid and impact the nature, size and direction of power flows, the amount of data and system visibility required for an operator to respond to unplanned events and operate the grid is becoming significant.
This lack of data and visibility causes all sorts of problems, including:
Long interconnection queues, and higher fees, as utilities struggling to estimate the impact on the grid
Safety issues for linemen as they find distribution switches that they thought were closed, open; and switches that they thought were opened, closed
Challenges in preventing outages, renewable curtailment, equipment overloading and managing congestion
Ineffective demand response programs as the utility doesn’t know which devices are connected to an overloading substation/line
Additionally, this is impacting the ability for utilities, regulators and states to reach their energy goals. For example, an accurate understanding of the state of the distribution grid has been identified as crucial for successful implementation of NY State’s Reforming the Energy Vision and High-Performing Grid initiatives. Many of the promised benefits of interactive grids and an Internet of Energy hinge on having accurate network maps.
Putting the communication and application layers on the same physical layer
Although this is a relatively new problem, there have been some attempts to define network connectivity through software and limited sensor equipment. However, these attempts haven’t worked. They either estimate values, and so have inaccuracy, or they have low latency and are slow to send the data back, meaning they will quickly be out-of-date. Most data collected by the grid is not at the temporal or spatial granularity needed to map the network.
So why have these solutions failed? We said that the data is inaccurate or out-of-date, but why is that? To explain, we need to take a short detour into network abstraction layers.
I like to think of networks as having three layers. You have the application layer, this is the thing being moved around the network; the communication layer, this is the protocol that defines how things are connected and routed; and the physical layer, this is the physical platform that underpins the network.
For example, for our road network, cars (the application layer) move along roads (the physical layer) coordinated by traffic signals and road signs (the communication layer). For the internet, data packets (applications) move along optical fibers (physical layer) coordinated by TCP/IP (communication layer).
So why have these existing solutions failed? Earlier we said that they are either inaccurate or out-of-date. This is because the communication layer is not on the same physical layer. The communication topology of the solutions doesn’t follow the physical topology of the electric grid. To use an analogy, the way we are defining network topology right now is like sending a letter to someone to see which cell tower their phone is connected to. You may get a response, and it may be accurate, but it will also be quickly out-of-date.
A truly connected network requires communication to occur on the same physical layer as the application; otherwise the application and communication get out of sync. You would be trying to call someone who was never connected to the network in the first place, or being directed to drive down a road that closed for repairs. For the grid, a communication protocol that maps network topology and connectivity needs to run on the physical wires that the electrons flow through.
This is where existing “Internet of Energy” theses fall short. In these theses, all the interconnected devices, such as solar panels, batteries, and smart thermostats, will be able to communicate and dynamically manage the grid. However, just because a device is connected to the internet and can communicate with other devices through an API, doesn’t mean you know how it is electrically coupled to the grid and other devices on it. The communication occurs over a different network that doesn’t map to the physical network the devices are connected to. Two devices may communicate to adjust their power outputs, or a central message may be sent to them, but you have no idea if they are electrically connected, or what else they are connected to, and therefore no idea how will it impact the grid or whether you can even achieve what you want to do in the first place. Due to this, attempts to dynamically manage the grid fall short.
Power-line communication to the rescue
So how do you get the communication layer for an autonomous grid onto the same wires that electrons move through? This is the key question that unlocks a true Internet of Energy. If you can map electrical connectivity at the same time as sending electrons, then the implications are huge.
Luckily, this has been done before; just not at the scale of an entire grid. If you’ve ever had to extend the WiFi network in your home, you may have come across Power-line Communication. Here data is sent between two wireless adapters that plug into your electric sockets in different parts of your house. These adapters are able to send data to each other through the wiring in your house by overlaying a high-frequency signal onto the 50Hz AC cycle.
Additionally, there is a brand of smart meter that sends data back to the utility via a protocol called TWACS (Two-Way Automatic Communication System), rather than the typical radio signals that smart meters send which don’t provide insights into electric coupling. TWACS uses a similar principle to modulate the electric current to send meter data and commands through the wires.
While there are a number of challenges with these technologies, it is theoretically possible to send unique IDs over the power line which, if received by another device, guarantees they are electrically coupled. In a similar way to how routing tables in routers are updated based on receiving signals from connected routers, grid-connected devices, smart meters and substations would be able to update their electrical coupling tables based on receiving signals from each other.
It will take a lot of infrastructure upgrades as each node will need to have the ability to send, receive and process the signals; but this can be done over time. An increase in DERs is one of the driving factors for distribution network modernisation, including the push for grid visibility and network topology. New devices coming online with DER rollouts, and smart meters and other grid infrastructure coming to the end of their life, means the infrastructure to support autonomous networks can be installed as this happens, with minimal disruption and stranded assets.
Once you’ve defined network connectivity, you can start to move towards an autonomous grid
As these devices are rolled out, we’ll fill in more and more of the real-time map of distribution grid connectivity. Not only will this solve the problems faced by network operators we discussed above, but you can start to move towards real-time autonomous control of the grid. Just like there is no ‘operator of the internet’, we could get to a point where there is no ‘operator of the grid’. You’ll still need network engineers and people on the ground to fix issues, but when it comes to balancing load and managing operations, this could become autonomous.
We can also start to unlock other use cases that have been discussed but never properly enacted. Things like: genuine P2P energy markets and renewable energy purchasing as you know if you’re electrically coupled to who you’re buying from; accurately allocating costs and avoided cost for infrastructure upgrades as you’ll know if you’re connected to the distribution feeder; real-time locational marginal pricing as you’ll know the path an electron will take and the cost associated with it. We may even reach a point where adding a new device, generation source or wire to the grid is as easy, and regulation-free, as connecting a new router to the internet.
Defining an electrical connectivity protocol and coupling communication topology with physical grid topology will radically change our grid and how it is operated. It will unlock a true Internet of Energy and have far-reaching implications. We’ll talk about some of the implications of this for utilities in Part III