Comment: The Challenges of Grid Modelling in Japan

By Hugo Batten, Managing Director, APAC; Reece Jackson, Grid Lead, APAC; Rowan von Spreckelsen, Head of New Markets, APAC; Kento Yoshimura, Market Lead, Japan; James Ha, Research Lead, APAC; Zelin Chen, Project Leader, Japan; Dan Yamada, Analyst, Japan

After previously exploring an eclectic range of topics—the challenges of battery modelling, modelling non-equilibrium outcomes in energy systems with long-run equilibrium models, the Philippine power system—our APAC team are now diving into grid modelling, with a particular focus on Japan. We also recently released a podcast episode on this very topic where we additionally discussed Australia and the Philippines and our grid-modelling services.

We wanted to focus on grid modelling as academic literature spends a lot of time discussing market modelling, but relatively less time on grid modelling, neither the approaches nor the quality and reliability of outputs. By “grid modelling”, we generally mean power flow transmission modelling typically used to forecast grid losses, curtailment, or nodal prices, although, of course, there are a range of types of grid models. In addition, several markets are seeing rising levels of grid-related losses or curtailment, and many regions are discussing policies to sharpen locational signals. This is likely to make power flow transmission modelling (intelligently integrated with capacity expansion/dispatch modelling) increasingly important to renewable and storage investment cases.

With our team in Tokyo undertaking both market modelling and asset-specific grid modelling for over 2 years now, paired with our most recent subscriber report on Japanese grid modelling, we aim to provide a high-level summary of what grid modelling looks like for Japan, including the following topics:

1. A brief description of our approach to grid modelling
2. Why grid modelling matters in Japan
3. A Japanese grid modelling case study—offshore wind in Tohoku

1. A brief description of Aurora’s approach to grid modelling

a) Our approach to grid modelling

We build all our various models in-house. We think it is critical to own our supply chain—both the modelling code and input assumptions—so we can provide truly independent and rigorous analysis to our clients. We have a team of over 60–70 modellers who sit in 5 global modelling hubs in Oxford, Austin, Singapore, Gurugram, and Sao Paulo. Our modellers develop and maintain our code base across our modelling suite, giving us the flexibility to adapt our models to client and market needs and to expand into software offerings: Origin, Chronos, Amun, and Solaris.

We have 3 major types of model that we use most frequently. These models do interact but are separate with distinct approaches to solve different problems:

• Energy market models (Origin)
• Electricity network models (AER-EN)
• Asset dispatch models (particularly for flexible assets) (e.g., Chronos)

Fundamentally, these grid models are trying to solve where and how the grid might impact asset economics over time, as well as looking at the impact of transmission augmentation. These grid models approximate AC (alternating current) power flows in high-voltage transmission networks across a given region at 30-min granularity. Our grid models are a DC (direct current) approximation, which is industry standard for long-term network forecasting.

Critically, this grid modelling is integrated with our market models such that results are internally consistent between market outcomes and grid outcomes. We incorporate key data underpinning the connectivity, impedances, and flow limits of each network element and calculate power flows that are consistent with our market model scenarios (in terms of assumed future generation and retirements, storage, network augmentations, etc).

Considering slight variations by markets, the key outputs from this fundamental power flow solve are: key points of thermal congestion within the network (e.g., quantifying thermal grid curtailment, line utilisation and/or headroom), grid losses, and asset specific curtailment outcomes.

We then regularly stress-test and back-cast the results of our grid/power flow models against metrics provided by system operators, like AEMO in Australia, to assess reliability and accuracy.

We also then design scenarios (across market and grid inputs) that specifically test grid outcomes—for example, assuming all future transmission augmentation is delayed or, the placement of renewables and storage across the grid over time. Increasingly, clients also ask us to run isolated asset-specific sensitivity analysis—for example, a renewable plant builds in close proximity, a particular transmission line is delayed, a smelter closes early, or a change to detailed interconnector design (e.g., recently Tohoku-Hokkaido undersea cables).

b) Some common pitfalls we see in grid modelling

We see a range of fairly common issues with grid modelling globally. Some of the bigger issues include:

●    Methodological issues

One form of methodological issue is using historical regressions instead of full network flow modelling. Regression based on historical grid outcomes is probably not a terrible approach to forecasting grid outcomes over the short-term, assuming there are no major changes to nearby generation or transmission additions/augmentation.

However, over the long term, and assuming there are changes to either generation locations or grid build-out, the regression approach is self-evidently unlikely to capture non-linear shifts in outcomes. This approach is still fairly widely used in US, but our understanding is that some of the forecasts have proven to be materially off-the-mark in more renewable-intensive systems like CAISO and ERCOT.

Another methodological issue we see is inconsistent market and network modelling.

It is difficult to fully integrate a 30-minute dispatch/capacity expansion model with a 30-minute power flow transmission model. Having said that, it is possible to create feedback loops between the 2 modelling suites such that input assumptions and outputs are consistent.

Our modelling suite ensures market modelling scenarios are aligned with network model scenarios so there are consistent dispatch profiles across the fleet; an alignment of price/revenue outcomes with grid curtailment results; and consistent economic and grid curtailment forecasts.

Often, grid modelling analysis is based on a few short-term network ‘snapshots’ (i.e. single intervals representing demand low, demand high, summer day, winter night, etc.). These modelling exercises are useful for understanding and stress testing connection points, but there is significant value in a wider integration of network and market modelling. Considering the market-network interaction across days, weeks and years allows a more representative long-term view on things such as grid curtailment to support with investment decisions.

●    Input issues

Not modelling the entire transmission network is one input issue we have observed.

Often due to data issues, we only see modelling of the high voltage (backbone) rather than the entire transmission network. Depending on the market, local lines often make up >80% of all transmission lines. Local lines can have a material impact on power flows and need to be incorporated into the inputs.

There is a bit of misunderstanding that it is acceptable to only model the high voltage backbone network “because those are the important lines”. By way of a metaphor, this is similar to trying to model traffic on major highways while ignoring every residential road: the same number of cars with less roads will inevitably make the traffic seem worse. It’s clear that considering the entire interconnected system of lower-voltage local lines while modelling is critical for modelling accuracy.

Additionally, we have observed issues in failing to capture future changes to the grid.

One of the few areas where we do not typically come to a fully independent view on input assumptions is on grid augmentation. We will typically use authoritative reports like AEMO’s Integrated System Plan (Australia), OCCTO’s Master Plan (Japan), or NGCP’s National Transmission Development Plan (Philippines) to provide a view on future grid augmentations and expansion and build that into our market and grid scenarios. Project Finance banks, in our experience, typically do want to see scenarios that are consistent with the grid operators future view and sensitivities around delays to that transmission build out.

Implementing announced transmission plans is critical as it creates new flow pathways, unlocks capacity for renewable build, and improves network robustness.

●    Output issues

Designing scenarios and sensitivities that test asset-specific grid risks is an output issue we have observed.

We rarely see dedicated grid scenarios and sensitivities that are explicitly designed to create a meaningful distribution of outcomes for individual assets.

Aurora models create both:

1. Sensitivities that explore the impact of specific events that impact power flows and individual asset outcomes, for example, a new asset with a similar production profile building nearby or delays to a specific transmission asset nearby; and
2. Scenarios that lead to materially better/worse grid outcomes for a region’s assets, for example, delays to all new build transmission or an overbuild of renewables in a given price zone.

Similarly, we have also noticed issues regarding calibration with existing grid curtailment/losses given complexity of grid.

Even with very accurate grid snapshots and transparent data from grid operators, extensive quality assurance/quality control is required to ensure that the start of forecasts are consistent with recent historical outcomes across the whole grid.

We often see forecasts that are not anchored to recent grid outcomes, which undermines confidence in the accuracy of forecasts, particularly for the Project Finance community.

2. Why grid modelling matters in Japan

a) Shifts in generation/storage supply

Like almost all developed countries, Japan’s power market is going through a profound shift. Firstly, among other companies, we forecast unabated coal and gas generation to decline sharply and to be replaced by a mix of renewable generation, decarbonised thermal, storage and other flexible assets. While there is some variation between regions, this broad outline is consistent across the 9 regions.

Within this shift in generation, unique to Japan is the proportion of decarbonised thermal generation that supports the country’s pathway to Net Zero. Due to limited land availability for renewables and political uncertainty about nuclear energy, the Japanese Government now offers favourable subsidies to existing unabated gas/coal plants to retrofit their turbines to operate with CCS or on a co-firing basis with hydrogen or ammonia. This, coupled with additional hydrogen / ammonia production subsidies, sees Japan’s unabated thermal generation fleet increasingly retrofitting to operate on a decarbonised basis.

In addition to this fundamental overhaul of the coal and gas generation fleet, market design changes with the introduction of more granular trading of balancing/ancillary markets is also driving a huge growth in the build-out of battery storage capacity. This storage capacity rapidly replaces existing oil/gas peaking plants as the predominant source of flexible supply over the long term. It is worth noting that while there are material differences between credible forecasts, we are broadly aligned with OCCTO and METI on long-term generation mixes in the medium-term (2030) and the long-term (2050).

b) Growth in the grid

To facilitate the transition to renewables and flexible assets, OCCTO has mapped out a detailed scenario for future transmission augmentations within their Master Plan. We incorporate both the large inter-regional augmentations laid out in this Master Plan, as well as the intra-regional network upgrades/additions laid out for each of the nine mainland regions in the individual transmission operators’ Ten-Year Plans. This enables us to forecast how the current network topography is planned to be adapted over time, capturing the impact on regional prices, economic curtailment, as well as asset-specific grid curtailment.

Staying on top of these network developments is no small task, and our team in Tokyo spend a significant amount of time reviewing the various committee meeting drafts, proposals, and decisions on the specific pathways / voltage options that are under consultation at any given point in time. Different pathways and connection points for new lines (particularly the big, new interconnectors like those in Hokkaido and Kyushu) have a huge impact on how power flows across a particular region. This, in turn, drives a wide range in outcomes for asset-specific grid curtailment studies.

c) Evolving grid policy

In an effort to enable a higher rate of new asset connections, Japanese policy makers have updated regulations that will effectively re-allocate risks and costs amongst relevant parties in instances of grid-congestion. More specifically, they are removing the strict requirement to obtain firm connection titles, i.e. only approving those grid connections that will likely never see curtailment. Where previously large swathes of the Japanese grid were considered to have ‘no spare capacity’, the network is opening up to allow renewables developers to connect to a wider range of substations, with the understanding that if overloading occurs on relevant transmission lines, it will be their assets—and ultimately their balance sheets—that will bear the brunt. Without exhaustively running through the finer policy nuances, in summary, this shift towards non-firm connections provides some significant opportunities for developers while simultaneously increasing their exposure to grid-risk.

As shown above, the new policy will see that, from present day, all* new renewables developments connect to the grid under a non-firm title. Therefore, going forward, when transmission lines overload across the network, these non-firm assets will be subject to output control in order to manage and alleviate these constraints. This is true for network constraints that occur on both the backbone (highest voltage) and local (lower voltage) transmission assets. Crucially, this generation curtailment is set to be entirely uncompensated, placing the financial risk onto the asset owners.

This new paradigm of grid connection policy emphasises the importance of selecting a strong grid connection. By way of example, as seen in our modelling, near identical solar assets in the same price region, Kyushu, can experience drastically different outcomes due to grid dynamics. In a given year, one asset that is located on high-voltage infrastructure nearby residential load centres might face no curtailment while the other asset, that connects to a lower-voltage, sparse area of the grid, might lose more than 15% of its production potential.

The Japanese market has seen more sophisticated developers and utilities actively modelling grid outcomes using real power flow modelling, but there is still a fairly wide range of understanding of grid challenges. It is reminiscent of Australia 6–7 years ago before a string of assets got hit with low Marginal Loss Factors and saw very material revenue hits (largely on radial lines long distances from demand centres). Japan is likely to experience the same as non-firm connections are rolled out.

3) Japan grid modelling case study—offshore wind in Tohoku

With non-firm connection rules, offshore wind bidders are trying to identify connection points that minimise their exposure to grid risk. As an example in Tohoku, completed offshore wind zone auctions have been concentrated in the west. Upcoming auction sites with non-firm connections are focused in the north and east. Furthermore, beyond auction rounds 1–4, we anticipate significant additional offshore wind capacity to commission as part of yet-to-be-announced future rounds. This raises the natural question—if these future sites are expected to connect under a non-firm title, which points-of-injection will be most robust to the introduction of new capacity?

For various participants, we have undertaken detailed power flow modelling to determine the available technology-specific connection capacity for each substation.

After the commissioning of offshore wind site rounds 1–3, further spare/robust hosting capacity is located in three main areas. Accounting for planned future network augmentations, we have calculated the offshore wind hosting capacity for high-voltage costal substations in Tohoku in the year 2032:

1. Despite the commissioning of the auction site rounds 1, 2, and 3, our modelling indicates that there would still be significant hosting capacity in west Tohoku due to significant transmission augmentation in the early 2030s and thermal plant retirements.
2. However, further connection capacity is more limited in the northwest around the Sea of Aomori–North due other nearby offshore wind and export interconnector flows.
3. Connection points near to the Sakata and Mutsu Bay zones have significant spare capacity.

The amount of ‘spare’ capacity is notably very related to an asset’s tolerance for curtailment exposure. The above hosting capacity calculations assume a <6% curtailment tolerance. The following analysis changes this assumption and looks at the Tohoku grid from the perspective of an offshore wind farm with a near-zero curtailment risk appetite:

4. A zero-curtailment tolerance requires a prospective asset to have essentially zero exposure to grid constraints.
5. By adopting this more conservative criteria, several connection points within the north and southwest of Tohoku are deemed to have zero spare hosting capacity.
6. Potential connection locations within the northeast and southeast of the region are forecast to have significant grid hosting capacity, even with the stricter requirement.

This type of integrated market and grid analysis will be increasingly salient for offshore wind developers. As an example, our forecasts indicate a new 500 MW, non-firm offshore wind in the Kuji zone would be expected to experience between 1–5% total curtailment—a roughly even split between economic and grid curtailment driven by a combination of grid congestion, a 1.8 GW nuclear re-start, and long-term network augmentations.

Please feel free to reach out to any of the authors if you would like to discuss grid modelling in Japan, Australia, the Philippines, or South Korea.