*2.1. Modelling*

Low-load diesel application was simulated using Homer Pro software, developed by the U.S. national renewable energy agency to assist in the selection and sizing of power generation technologies. The model accepts the daily, seasonal and yearly profiles for resource and load, allowing the user to model diesel and renewable generation via control of generation dispatch order, reserve requirements, generator curtailment and efficiencies [22]. Homer Pro was selected as the appropriate software environment given the prevalence of this format within industry, providing for reduced barriers to utility review and consideration. For both the King Island and Moloka'i case studies measured resource and load data was used to define a 12-month simulation of hourly generation dispatch. Each case study was configured to represent the as-build system configuration, with observed system performance used to validate the model. Model configuration included generation dispatch and reserve definition as implemented for each case study. Model validation consisted of review across both modelled and observed annualised diesel generation run hours, renewable penetration and fuel consumption. For a known solar irradiance profile, the model develops an hourly solar resource estimate [23]. Solar PV power output is then calculated using Equation (1).

$$P\_{p\upsilon} = f\_{p\upsilon} \mathcal{Y}\_{p\upsilon} \frac{I\_t}{I\_s} \tag{1}$$

where *fpv* is the derating factor, *Ypv* is the rated capacity of solar PV (kW), *It* is the solar irradiance, and *Is* is one kW per square meter. The derating factor can be used to approximate reduced efficiency as may be experienced in specific configurations or environment conditions. For a known wind resource, the model uses a user-defined wind turbine power curve, the relationship between wind speed and power, to calculate wind generation. To facilitate this, an hourly wind resource profile can be estimated from the average wind speed, Weibull shape factor, autocorrelation factor and diurnal pattern, however, for King Island measured hub height hourly wind resource data was used. Postproduction losses applied to both wind and solar generation include electrical losses and unit availability. When the available renewable generation is insufficient to meet the system load, the model may schedule battery or diesel generation according to the maximum and minimum unit loadings. At all other times the system reserve requirements defined the battery and diesel response. The diesel generator fuel curve is assumed to be linear according to Equation (2), and is used to calculate fuel consumption.

$$F = F\_{\rm int} Y\_{\rm \\_gen} + F\_1 P\_{\rm \\_gen} \tag{2}$$

where *Fint* is the fuel curve intercept coefficient, *F1* is the fuel curve slope, *Ygen* is the unit rated capacity (kW) and *Pgen* is the generator output (kW). The units of F depend on the manufacturers preferred measurement units for fuel, typically either litres or gallons per hour. In regard to the battery performance the model uses the nominal voltage, capacity curve, allowable charge range, roundtrip efficiency and cycle life to simulate any battery contribution. The capacity curve details the discharge capacity of the battery versus the discharge current, and is supplied by the manufacturer. The maximum rate of charge or discharge is specified by the kinetic battery model [24]. A detailed overview of the simulation theory is provided in [25].

For each case study the utilities reported annual fuel consumption, dispatch scheme, generator run hours and renewable penetration were used to validate the model configuration ahead of low-load diesel simulation. For Moloka'I, the generation performance and dispatch model was first established using 2009 generation and load data, representative of the system prior to residential solar PV uptake. To this model, annual reported renewable investment was added iteratively to validate the performance under increasing solar PV penetrations. The approach yielded annual diesel fuel consumption estimates within 3% of the observed performance once measured data was corrected in consideration for the high fuel consumption rate of generator 7 (this unit was removed from the model post-calibration, given its inefficient operation and non-standard performance). For King Island much the same methodology and accuracy were employed/observed with the higher annual resource variability (wind compared to solar PV) addressed via validation of the model against operational data using the 5-year moving average (1999–2019). The models were largely insensitive to economic assumptions given the fuel consumption rates were directly compared across simulations to quantify system performance. Irrespective of this, actual incurred fuel costs were adopted to match real inflation. Low-load diesel application was assessed via revision of the diesel low-load limit from 30% to 10%. The efficiency of existing diesel assets under low-load operation was established in prior studies [26], remaining predominantly linear in relationship.

#### *2.2. Low-Load Diesel*

Low-load diesel application affords diesel generation improved range and flexibility, permitting system acceptance of additional renewable generation via a reduction in the diesel engine load limit. The low-load limit is set within the primary engine controller on a case by case basis. No hardware of software replacement is required, resulting in a low complexity, low cost and accessible approach applicable to all diesel generators [27,28]. Poor combustion and cylinder condition are responsible for historical restrictions surrounding low-load operation, however, a number of manufactures now warrant low-load applications, reflecting the increased awareness and viability of the practice [29]. Due to poor low-load efficiencies, fuel consumption per kWh of diesel generation increase at low-load, however, given both the increased renewable capture and the low volume of kWh's produced by diesel generation at low-load, the practice has a net positive reduction in fuel usage, as confirmed via system

simulation. Net fuel reductions result, despite reduced e fficiency, given the acceptance of greater instantaneous renewable penetration. This occurs given the engines ability to further reduce load as renewable generation increases. Conventionally, once an engine hits its 30% load limit any additional renewable generation is spilt from the system. Under low-load application this additional generation is accepted via further diesel load reduction. The major operational concern in reducing an engine's low-load limit is the risk of reverse power acceptance given the reduced ability of the diesel assets to regulate upward renewable variability. This is commonly managed via the inclusion of a dump load to dissipate excess generation as heat.

#### **3. Low-Load Diesel Modelling Results**

For high-penetration hybrid diesel systems such as King Island, low-load diesel permits the acceptance of additional renewable generation from any reserves of surplus generation. Simulation of the annual King Island system performance in this manner identified average diesel fuel savings of 6.3% per annum (on a year by year basis savings varied with renewable resource, above average wind generation would result in increased fuel savings, while lower wind generation would decrease observed fuel savings). It can be seen from the results that a reduction in engine low limit serves to lower the systems diesel reserve requirements, reducing the time engines spend operating at their low-load limit. For King Island the application delivers both improve renewable penetration, via a reduction in renewable spillage, and an associated reduction in the requirement for energy storage [30]. For high renewable penetration systems such as King Island, low-load diesel application does not impact the systems renewable hosting capacity, acknowledging the system already hosts a renewable capacity exceeding its maximum load. For high-penetration renewable energy systems such as King Island, low-load diesel is also observed to rationalise the requirement for energy storage In this case the annual average battery utilisation (amount of energy stored, MWh) is reduced by 50%, allowing for a smaller battery capacity and reduced capital cost.

For low-penetration renewable systems, such as Moloka'i, the role of low-load diesel application is less obvious. Primarily, because low-penetration systems rarely spill renewable generation. Simulation of the Moloka'i system identified a range of alternate low-load benefit, including improved hosting capacity and reduced diesel OPEX. Simulation of the pre-2016 (prior to BESS and dump load integration) Moloka'i hosting capacity determined that the limit of uncontrolled solar PV was 2.5 MW. Simulation inclusive of the BESS and dump load, representing the current Moloka'i system, determined the hosting capacity to be 3.0 MW. A low-load diesel scenario was then considered with the diesel generation permitted to run down to 10% loading. A 10% load limit is considered conservative and was selected considering both the age of the assets, and the experience of the neighbouring Kauai Island utility cooperative in obtaining permits to operate at this level. Simulating a low-load limit of 10%, the system hosting capacity was increased to between 3.5 MW and 3.8 MW. Allowing adoption of low-load operation across only the high-speed engines, units 1 and 2, the hosting capacity was increased to 3.5 MW, a 17% increase. For adoption of low-load operation across diesel engines 1, 2 and 9, the hosting capacity was increased to 3.8 MW, a 27% increase. The results define a role for low-load diesel application in near-term relief of hosting capacity constraint. As the Moloka'i system is functionally at its hosting capacity envelope, and does not generally spill generation (less than half a percent of generation is dissipated via the resistive dump load), no immediate increase in renewable penetration was observed under low-load diesel application. Despite this, fuel savings were delivered via low-load diesel operation, achieved via a modified low-load diesel dispatch scheme. The existing Moloka'i dispatch strategy prioritises large low-speed engines (large), generators 7–9, with small high speed (small) generators 1&2, deployed to address load variability. Band allocations 1 through 5 define the generation intensity, with band increase associated with greater diesel capacity, Table 5. Fuel usage correlates to band allocation, with lower bands consuming less fuel. A modified low-load diesel dispatch scheme, Table 6, shifts the balance of generation from band 3 to band 2, via substitution of large low-speed generation for small high-speed generation. The modified low-load diesel dispatch scheme delivers both a reduction in fuel use, and significantly, a reduced maintenance obligation. The reduced maintenance spend results in lower maintenance costs attributed to small inertia and high-speed engines [31].


**Table 5.** Moloka'i diesel utilisation for a 3 MW solar PV scenario (current dispatch scheme).

**Table 6.** Moloka'i diesel utilisation for a 3 MW solar PV scenario (low-load dispatch scheme).


In addition to improved hosting capacity, low-load diesel application within the current Moloka'i system results in a fuel reduction of 1%, and an 8% reduction in maintenance expenditure, Table 7. The combined annual OPEX reduction is \$209,469 p.a., representing 2.7% of total annual operational expenditure. Economic modelling was then extended to consider a future 6 MW solar PV capacity scenario, Table 8. While in violation of the system's current hosting capacity, this scenario is useful in exploring the storage requirements required to support additional solar PV deployment. The proposed low-load diesel methodology reduces this requirement by 43%, approximately halving the storage capacity required for grid security under future high-penetration scenarios.

**Table 7.** Performance of existing and proposed low-load dispatch scheme 3 MW solar PV.


**Table 8.** Performance of existing and proposed low-load dispatch scheme 6 MW solar PV.

