Horizon Networks’ uses AI to boost climate grid resilience

A New Zealand distribution utility is turning to artificial intelligence to sharpen how it plans for climate risk, as grid operators face growing pressure to align long-term investment with increasingly volatile weather patterns.

Rhizome has deployed its gridADAPT platform with Horizon Energy Distribution Limited (Horizon Networks), in a 12-month pilot aimed at strengthening distribution grid resilience while improving the efficiency of capital expenditure.

The project, approved under New Zealand’s Innovation and Non-Traditional Solutions Allowance, reflects a broader shift among utilities toward data-driven planning tools that can quantify climate risk and translate it into actionable investment decisions.

Rethinking grid upgrades

At a system level, the challenge is becoming harder to ignore. Infrastructure designed for historical weather conditions is increasingly exposed to more frequent and severe events, forcing utilities to rethink how they prioritise upgrades and allocate limited resources.

Rhizome says its gridADAPT platform is designed to address that gap by combining multiple layers of data — from network topology and asset condition to historical outage records and environmental variables — into a unified analytical framework.

Talking to Enlit World, Mishal Thadani, co-founder and chief executive of Rhizome, explained that the system builds a contextual understanding of how and why failures occur.

“gridADAPT first combines data related to the physical characteristics of the utility system, such as the layout of the network, assets such as poles, transformers, and spans of conductor, as well as all of the historical data related to power outages,” he said.

That dataset is then layered with what the company refers to as 'State of the World' inputs, including high-resolution weather data, vegetation and topography. Machine learning models use this combined dataset to identify patterns linking extreme weather events to infrastructure failure.

“AI models learn from the geographical and system contexts of when power outages occurred, specifically during extreme events where there typically a high volume of system failures, to then predict where the likely failures are going to be for the next events,” Thadani added.

Have you read?
Siemens expands AI data centre play with partnership trio
National Grid taps startup for connections intelligence as UK tackles ‘speculative’ requests

Forward-looking planning

Adding to this, said Thadani, the platform extends beyond retrospective analysis. By incorporating Global Climate Models, it aims to enable utilities to assess how risk profiles may evolve under different climate scenarios, providing a forward-looking basis for planning decisions.

For Horizon Networks, which serves approximately 25,500 homes and businesses in the Eastern Bay of Plenty, the deployment comes in response to a tangible increase in weather-related disruption.

Feng Wu, the utility’s General Manager of Networks, told me: “Over the last five years, Horizon’s electricity network has been subject to increasing weather related events that has caused significant damage.”

Their pilot with Rhizome is intended to move the utility from a reactive posture toward a more structured and predictive approach to resilience planning.

“Our decision to partner with Rhizome is to explore how we can use available data and learnings to assess extreme weather risk across the network and support long-term resilience planning,” Wu said.

At the asset level, gridADAPT models the likelihood of failure by combining physical characteristics — such as asset type, age, manufacturer, and condition — with hazard intensity. For example, the system can quantify how the probability of failure for a 50-year-old pole compares to a newer asset under high wind conditions.

This approach blends physical fragility modelling with empirical failure data, allowing the platform to refine its predictions based on observed outcomes.

Investment decisions

For Horizon, this granularity is expected to translate into more targeted investment decisions.

“The platform is designed to use network datasets and climate projections to assess which parts of the network may be more exposed or vulnerable to extreme weather events,” Wu said.

“This should allow us to assess and ascribe risk rating to various assets to provide a more structured view of risk and help us make informed investment decisions that will be grounded by sound evidence.”

That emphasis on evidence-based planning is central to the value proposition: gridADAPT evaluates how different interventions — from replacing components to deploying protective technologies — reduce risk relative to their cost.

According to Thadani, prioritisation is driven primarily by cost-benefit outcomes. “There are many factors that help guide the priority of targeting a specific asset investment, though of most importance is understanding which investment will yield the greatest cost-benefit ratio,” he said.

The platform allows planners to simulate different investment strategies and optimise for specific objectives, such as reliability improvements or risk reduction, while still allowing for manual adjustments based on operational considerations.

Beyond capital allocation, the deployment also points to a shift in how utilities integrate climate science into day-to-day decision-making.

Accessibility

One of the lessons from previous projects, Thadani noted, has been the importance of making complex modelling outputs accessible to engineers and planners who may not specialise in climate analytics.

“Expressing data through gridADAPT early on in the process makes everything easier to communicate to the planners and engineers who aren’t as familiar with climate science and fragility modelling,” he said, adding that earlier user engagement has reduced time-to-value by 80% in recent implementations.

For Horizon Networks, that usability will support with determining whether the platform can be embedded into routine planning processes rather than remaining a standalone analytical tool.

While the primary focus of the pilot is long-term planning, there are potential operational benefits as well. As confidence in the system grows, the same insights used to guide investment could support improved preparation for extreme weather events.

“In saying that, as our confidence improves there is no reason why we won’t be able use this to better prepare and manage our response to anticipated events,” Wu said.

The broader implication is that planning and operations may become more closely linked, with predictive analytics informing not only where to invest, but how to respond to emerging risks in near real time.

Regulatory backing and future integration

Regulatory backing for the trial also highlights a growing recognition that traditional planning frameworks may need to evolve. By approving the deployment under an innovation allowance, New Zealand’s regulator is effectively creating space for utilities to test non-traditional approaches to grid management.

For Rhizome, the Horizon project adds to a portfolio of international deployments across different climate contexts, including partnerships in the United Kingdom and the United States. Each geography presents distinct challenges, from wildfire risk to storm exposure, reinforcing the need for adaptable modelling frameworks.

“Utilities everywhere are facing climate risks their infrastructure was never designed for,” Thadani said. “Today’s partnership with Horizon Networks puts them one step ahead of those impacts, keeping the lights on for their customers.”

For Horizon Networks, the outcome of the pilot will determine whether AI-driven planning becomes a core component of its resilience strategy.

“If the trial proves successful, the insights could help Horizon target resilience investments more effectively,” Wu said. “This will not only support better-informed decisions but will also allow us to have more meaningful engagement with our community and stakeholders.”

Should the results meet expectations, the utility is already considering broader deployment.

“Once proven we would consider how the platform could be applied more broadly across the network as part of our resilience planning toolkit,” Wu added.