In the years when I developed my PhD research (2017–2020), the penetration of distributed energy resources (DERs)—particularly rooftop solar—was still relatively modest. At the time, the role of DER in system-wide power flows was emerging but not yet central to distribution-network planning.

Today, that landscape has transformed. What was once an academic foresight problem is now a real and pressing operational reality.

A DER landscape transformed: from marginal to majority contributor

When the models in my thesis were developed, rooftop solar contributed only a small fraction of total electricity supply, and reverse power flows were still largely viewed as isolated edge-cases.

Fast-forward to today in the National Electricity Market (NEM):

Rooftop solar has already supplied up to 80 % of total electricity demand in South Australia during certain daytime periods.
This is not theoretical — it has been observed in real operational data (AEMO-reported and industry-verified events in 2024).

This single statistic captures the scale of change:

• DER is no longer “supplementary generation”.
• It is now a dominant force shaping voltage behaviour, power-flow direction, and equipment loading at local levels.
• The challenges predicted in the thesis — reverse power flows, voltage rise, phase imbalance risk — are now everyday technical issues.

In short, what was once a research-driven model anticipating future challenges has become a toolset urgently needed by planners today.

Why localised DER impact modelling matters

High DER penetration is changing how distribution networks behave:

Reverse power flows

When rooftop solar exceeds local consumption, power flows backwards towards the zone substation. Most low-voltage networks were never designed for sustained bi-directional flows.

Voltage rise / voltage excursions

DER-driven export can push voltages above statutory limits, especially at feeder extremities.
This was a core topic of the thesis, where I developed machine-learning-based voltage risk classifiers to predict:

• Over-voltage excursions
• Under-voltage events
• Voltage-collapse

Spatial clustering effects

DER is rarely evenly distributed; some feeders have 40–60 % solar uptake while others remain low. This creates highly localised pockets of system stress.

Traditional tools are no longer sufficient

Conventional load-flow simulation becomes computationally prohibitive when assessing:

• Millions of potential DER uptake scenarios
• Highly stochastic rooftop solar generation
• Local configuration differences across thousands of feeders

The thesis addressed exactly this challenge.

The model: Neural-network-based prediction of DER impacts

During the research, I developed a Neural Network regression model capable of predicting the magnitude and probability of reverse power flows and voltage deviations in distribution networks.

Key capabilities

• Localised modelling
  Trained at the node/feeder level using topological and electrical characteristics.
• High-resolution scenario analysis
  Able to evaluate thousands of DER-penetration futures.
• Predictive rather than reactive
  Estimates voltage deviations and RPF magnitude before they materialise.
• Computational efficiency
  Orders-of-magnitude faster than traditional load-flow calculations once trained.

These strengths mean the model is particularly suited for planning use-cases where many scenarios must be assessed rapidly.

Why this is so relevant today

With states like South Australia already seeing 80 % of demand met by rooftop solar, distribution networks are undergoing a real-time transition. High-penetration DER challenges must now be:

• Predicted, not simply observed
• Quantified, not just estimated
• Localised, not averaged across whole networks
• Prioritised, not treated as uniform risk

The model developed in the thesis enables precisely this.

For network planners, the value is clear

• Identify future hotspots before voltage issues emerge
• Rank feeders by severity and probability of RPF/voltage excursions
• Optimise reinforcement budgets by investing where the risk is highest
• Inform regulatory submissions with quantifiable risk metrics
• Enable smarter DER integration (e.g., dynamic operating envelopes, storage sizing, smart inverter settings)

This aligns directly with the planning and investment needs faced by Australian DNSPs today.

A practical example: feeder prioritisation tool

Using the model, a DNSP can:

1. Define DER uptake scenarios (e.g., rooftop PV doubling in five years).
2. Run neural-network predictions for each feeder/node.
3. Estimate the probability and magnitude of reverse flows and voltage rise.
4. Generate a feeder risk ranking.
5. Prioritise reinforcement, storage deployment, or inverter-limit updates.

The outcome: A targeted, cost-efficient reinforcement roadmap driven by data, not assumptions.

Conclusion

When developed in 2017–2020, the model anticipated an emerging technical challenge. Today, with states like South Australia experiencing rooftop solar contributions of up to 80 % of demand, the challenge is here — and accelerating.

The neural-network-based framework from the thesis offers a scalable, data-driven way to forecast, quantify and manage DER impacts in modern distribution networks.

At GG Advisory, we help network operators, regulators, and energy-transition organisations turn these analytical capabilities into practical planning tools.

If you’d like to explore how this model can support your network or project, feel free to reach out.