Cai
Williams
a,
Hannes
Michaels
b,
Andrew F.
Crossland
c,
Zongtai
Zhang
a,
Natasha
Shirshova
a,
Roderick C. I.
MacKenzie
a,
Hongjian
Sun
a,
Jeff
Kettle
d,
Marina
Freitag
b and
Christopher
Groves
*a
aDepartment of Engineering, Durham University, South Road, Durham, DH1 3LE, UK. E-mail: chris.groves@durham.ac.uk
bDepartment of Chemistry, Bedson Building, Newcastle University, Newcastle upon Tyne, NE1 7RU, UK
cDurham Energy Institute, Durham University, South Road, Durham, DH1 3LE, UK
dJames Watt School of Engineering, University of Glasgow, Glasgow, Scotland, UK
First published on 19th September 2023
Many scenarios for Net Zero anticipate substantial growth of Solar PV generation to satisfy 30% of our electricity needs. However, this scale of deployment introduces challenges as supply may not meet demand, thereby necessitating energy storage and demand-side management. Here we demonstrate a different, complementary approach to resolving this challenge in which Solar PV generation can be made intrinsically less variable than commercial PV. Proof-of-concept dye-sensetised PVs for which the power conversion efficiency increases as light intensity reduces are demonstrated. Modelling of the UK mainland energy network predicts that these devices are more effective at displacing high carbon generation from coal and gas than commercial PV. The capacity factor of these PV devices are controlled by their design, and capacity factors >60% greater than silicon are predicted based on experimental data. These data demonstrate a new approach to designing PV devices in which minimising variability in generation is the goal. This new design target can be realised in a range of emerging technologies, including Perovskite PV and Organic PV, and is predicted to be more effective at delivering carbon reductions for a given energy network than commercial PV.
Broader contextAchieving Net Zero requires substantial expansion of clean renewable energy generation as well as methods to account for variable generation which may overlap poorly with demand. Thus, it is expected that energy storage and management of demand will be needed alongside Solar PV which generates energy in proportion to the strength of sunlight. In this paper we demonstrate a new, complementary solar PV technology where generation can be made more efficient at lower light levels, thereby introducing flexibility in energy generation that is intrinsically absent in present Solar PV. Using this approach, we demonstrate that the new PV devices are more effective at displacing high-carbon generation from coal or gas at the national scale than present solar PV technology. This represents a new approach to solar PV, in which the goal is minimising variability in generation, and with it, the need for energy storage and demand side management. |
In the electrical power industry, variability of renewable generation is quantified by the capacity factor, which is the ratio of the actual energy output to the theoretical maximum energy output over a set period. Capacity factors for commercial silicon PVs vary with the local climate, for example, ranging from 10% in the UK15 to over 25% in California.16 Generally, the capacity factors offered by silicon PV do not compare favourably with other forms of energy generation, for example, the UK's National Statistics Agency estimated in 202115 that Offshore wind had a capacity factor of 38%, whilst Biomass had 64%, and Nuclear had 59% (though we note that this is a low value for Nuclear from a historical perspective). Maximising capacity factor through generator design is a recognised approach in the wind power sector17 to address the challenges of variable generation listed above.18 However, at present, the techniques to increase capacity factor in solar PV are limited and external to the solar cell itself, such as installation of east/west facing panels to extend generation into the morning and evening,19,20 or the use of tracking mounts.21,22
In this paper we demonstrate a class of PV devices that have intrinsically higher capacity factor than conventional PV technology. These devices, which we term High Capacity Factor Photovoltaics, or CFPV, have a power conversion efficiency (PCE) which increases as irradiance reduces. This behaviour boosts generation at times of low irradiance to both minimise variability in power output and maximise power harvested for a given capacity of grid connection. The benefits that CFPV devices offer for matching supply and demand are demonstrated using a model of the mainland UK energy network. In this context, it is shown that CFPV generation matches demand more effectively than an equivalent capacity of silicon PV, and that this leads to two key benefits that increase with capacity factor: (i) a reduced need for high carbon intensity generation from coal or gas, and (ii) increased income from wholesale markets. CFPV devices based on a dye-sensitised architecture are demonstrated, showing that the capacity factor can be controlled through rational design. Further, the improvement in capacity factor through device design is predicted to be larger than available with external methods, such as single- and dual-axis tracking. Introducing control of the generation profile of Solar PV in this way is expected to reduce the need for, and costs associated with, energy storage and demand-side management elsewhere in the wider energy system.
![]() | ||
Fig. 1 PCE irradiance characteristics of CFPV devices as reported by33 (line), and those devices reported here (symbols) with tTiO2 = 12 (pentagons), 18 (squares), 24 (triangles) and 48 μm (circles). |
To quantify how the behaviour shown in Fig. 1 may impact national generation profiles and consequent carbon emissions, a plant dispatch model of the mainland UK electrical network (inc. England, Wales, and Scotland) was developed to predict and compare carbon savings that result from adding varying amounts of CFPV or silicon PV. The mainland UK was selected as the location for our analysis due to the good availability of generation data, as well as having a mismatch between irradiance (which peaks at midday) and electricity demand (which peaks in the late afternoon) as is common for national and large-scale electricity networks including Brazil, California, Chile, India, Iran, Spain and South Africa to name a few.34–36 The UK receives lower irradiance than many countries in the world due to its northerly latitude, however, we stress that increasing the capacity factor of solar PV also has value in sunnier locations. For example, tracking improves capacity factor, and in 2021 more than 85% of US deployed utility scale solar PV capacity (i.e. solar farms of more than 5 MW) incorporated some aspect of tracking.37 These tracking systems were installed in high-irradiance states such as California, Arizona and Texas, as well as less sunny mid-western states, showing there is an economic advantage in increasing capacity factor in both low- and high-irradiance locations.
With the location selected, energy supply within the plant dispatch model was divided into differing dispatch classes (DCs), which represent technologies that are similar in terms of their ability to be curtailed and carbon intensity. These are: DC1, which includes nuclear; DC2 which includes hydrogeneration, curtailable utility scale PV and Wind; DC3 which represents energy storage; whilst DC4 is carbon-intensive gas and coal. The capacities of each DC are listed in ESI† Table S1. Time-resolved energy demand for the year 2016 was assumed to be the sum of dispatched energy for each half-hourly time period,11 and this demand was met within the model with energy from increasing dispatch classes within each time window, whilst integrating the associated carbon intensity. A flow chart of the dispatch process is shown in ESI† Fig. S1, and a table of the carbon intensity for varying forms of generation is shown in ESI† Table S1. The plant dispatch model was modified from that of Crossland et al.12 to incorporate PV generation with arbitrary PCE-irradiance characteristics. A detailed description of the calculations and the use of these data is included in ESI† Section S2, but is summarized briefly here. PV generation in DC2 was modelled at a half-hourly resolution using insolation and generation data from the photovoltaic geographical information system38 (PVGIS) and balancing mechanism reporting service39 (BMRS). The power generation capability of silicon PVs in the model is based on real historical data, thus we did not specify a PCE-irradiance characteristic. Conversely, for CFPV, we scaled the generation for silicon PV in each time window by an enhancement factor, E derived from the measured PCE irradiance characteristics for CFPV devices:
![]() | (1) |
We highlight that the data related to UK energy mix (e.g.Fig. 2) will be reported in terms of energy dispatched, i.e. where the energy was drawn from when it was used to satisfy demand, as opposed energy generated. In most cases, energy generated was dispatched immediately (ESI† Table S1 and Fig. 2a–c), in which case dispatch and generation are synonymous. However, excess generation can also be used to charge energy storage that can be used later, decoupling when energy is generated and when it is dispatched. As an illustrative example, excess solar generation which charges storage which is later dispatched to satisfy demand would be counted as coming from storage rather than solar. Counting energy in this way enabled us to examine how changes in generation alters the usage of energy storage. Fig. S1 of the ESI† describes the process of charging storage in more detail.
![]() | ||
Fig. 2 Annual generation profiles of the UK (a)–(c), change in generation (d)–(f) and carbon savings (g)–(i) as a function of additional silicon PV (a), (d), (g), and CFPV for tTiO2 = 48 μm devices shown in Fig. 1b (b), (e), (h) and CFPV device33 (c), (f) and (i). Change in generation (d), (e) and (f) figures show change in generation against 2016 historical data. Carbon savings (g), (h) and (i) figures show the source of carbon savings from reduction in Gas (Light Blue) and Coal (Dark Blue) emissions. |
We took as our reference the mainland UK electrical grid with associated generation and storage assets in the year 2016, which included 11970 MWP installed PV that was dominated by silicon, and added to this an additional capacity of silicon PVs (CSi) or CFPVs (CCF). The values of C reported here are expressed as a fraction of installed UK PV capacity in 2016 (11
970 MWP) which defined the power generating capability when irradiance was 1000 Wm−2. Thus, a simulation for CCF = 0.5 assumes an additional capacity of CFPV assets equal to 50% of the UK PV capacity in DC2 in 2016. We frame our discussion in terms of installed capacity as it is this quantity that significantly affects connections of renewable energy to the grid, including procedure, duration and cost of grid connections.40 In other words, the substations and wires where the solar farm connects to the wider network have a defined capacity, meaning that the solar farm must operate safely on the network without causing any other issues, such as reduced power quality and loss of network protection.40 Further detail as to how the network limits the scope of distributed generation in the UK context is available from the Energy Networks Association in both summary41 and detailed form.42
This effectiveness in decarbonization can be quantified by the annual Carbon Dioxide equivalent (CO2e) emissions calculated by the plant dispatch model. Fig. 2g and i show predicted CO2e emissions savings as generation capacity was increased for both silicon PV and CFPV respectively. As expected, CO2e savings increased with generation capacity for both PV technologies, however, we observed the rate at which this occurs is non-linear as it becomes progressively more difficult to satisfy demand that does not align to irradiance. These data revealed that the CFPV device is approximately twice as effective at decarbonizing the grid for an equivalent amount of silicon PV over the range of C examined. Hence grid-scale CFPV was predicted to be more successful in reducing Carbon emissions than the equivalent capacity of installed silicon PV. When one reflects on the scale of PV installation that has been forecast to achieve Net Zero at the national level,4–6 and the consequent PV market penetration this implies, we argue that these data make a strong case for CFPV to be part of the energy generation mix to address some of the inflexibility that silicon PV has in meeting ‘hard to decarbonize’ energy demand.
The reason for the greater impact of CFPV as compared to silicon PV was its superior yield at irradiances below 1000 Wm−2. Fig. 3 shows a histogram of the irradiance values used in the model corresponding to July. Diurnal variation and weather results in a mean irradiance during daylight hours of 259 Wm−2, for which the CFPV device reported by Bristow and Kettle33 had a PCE enhancement of factor E(259 Wm−2) = 2.94 as compared to standard testing conditions (Fig. 1). Thus, the CFPV device was providing a power output in real conditions that was closer to its rated capacity than the silicon equivalent.
Device | PCE at 1000 Wm−2 (%) | PCE (140 Wm−2)/PCE (1000 Wm−2) | Annual CO2e savings when C = 1 (Mt) | Predicted UK capacity factor (%) | Predicted wholesale revenue for 5 MWP farm sited in UK (k€) | Predicted module costs (€/WP) |
---|---|---|---|---|---|---|
Conventional Si PV | 20 | 0.9 | 4.70 | 8.89 | 194 | 0.245 |
CFPV, tTiO2 = 12 μm | 9.0 | 1.00 | 4.61 | 8.87 | 191 | 0.355 |
CFPV, tTiO2 = 18 μm | 8.1 | 1.06 | 5.00 | 9.36 | 201 | 0.395 |
CFPV, tTiO2 = 24 μm | 7.6 | 1.10 | 5.28 | 9.67 | 209 | 0.420 |
CFPV, tTiO2 = 48 μm | 4.3 | 1.90 | 9.00 | 14.50 | 315 | 0.740 |
SolarPrint CFPV reported in 33 | 0.8 | 5.20 | 17.00 | 27.40 | 611 | 4.000 |
The silicon solar farm was estimated to yield €194000 over the year, whilst an equal capacity installation of CFPVs with characteristics reported by Bristow and Kettle33 was estimated to yield €611
000, more than 3 times as much. This additional income is due to increased generation when irradiance is below 1000 Wm−2, which in a UK context, is a large proportion of the time (see e.g.Fig. 3). However, the greater income that the CFPV device provides should be viewed in the context of lower efficiencies under standard testing conditions (PCE = 0.8% under AM1.5), since this would be expected to lead to greater module costs (particularly deriving from the TCO44) and land costs for the same generating capacity. To quantify this anticipated trade-off, we used the approach described in ESI† Section S3 to estimate the cost of commercially produced dye-sensitised CFPVs of the type described by Bristow and Kettle33 at 28.54 €m−2, and in turn, predicted the module cost per unit capacity (€/WP) shown in Table 1. These module costs were predicted to increase with reducing PCE under standard testing conditions, with CFPV modules from ref. 33 costing 16 times more than the Silicon equivalent. Allied to this, land rent costs also increased as PCE under standard testing conditions reduced, as shown in ESI† Table S3, though land rent costs are predicted to be smaller than that of the modules themselves for a 5 MWP solar farm.
Hence, for CFPV modules with characteristics as reported by Bristow and Kettle,33 we have demonstrated an intrinsic increase in capacity factor and thus more efficient utilisation of a grid connection, but that the low PCE at AM1.5 in this case leads to substantial increases in module and land costs. We now move on to discuss the second aspect of our investigation, namely to ascertain whether the CFPV behaviour of devices can be designed. Were this to be the case, CFPV devices could be designed to balance the characteristic behaviours highlighted for the Bristow and Kettle33 device.
![]() | ||
Fig. 4 Schematic for present devices in which TiO2 thickness was varied over the range tTiO2 = 12, 18, 24, 48 μm. |
Fig. 1 shows PCE as a function of irradiance for CFPV devices with varying tTiO2. PCE was largely constant with irradiance for the device with small tTiO2 (= 12 μm), and thus has similar PCE-irradiance behaviour as commercial silicon,48 albeit with a lower efficiency. As tTiO2 increased beyond 12 μm, it was observed that PCE reduced. Importantly in the context of this paper, the reduction in PCE depended on irradiance, with PCE under high irradiance reducing more than under low irradiance. We ascribed this reduction to the photocurrent generation of dye-sensitised PVs being limited by diffusion mass transport,49 which in turn reduced performance to an extent that depends upon both tTiO2 and irradiance. Critically, we demonstrated that the PCE-irradiance behaviour for these devices was controlled by the PV layer stack.
Model predictions for the designed CPFV devices are summarised in Table 1. The devices with tTiO2 = 12 μm, 18 μm and 24 μm, for which there is limited increase in PCE as irradiance reduces (E(140 Wm−2) ≤ 1.1), show limited improvements in capacity factor, decarbonisation and revenue generation. These data are shown in more detail in ESI† Fig. S2 and S3.
These data contrast to that for the tTiO2 = 48 μm device, for which the PCE increased by a factor of E(140 Wm−2) = 1.9 (Table 1), which in turn translated to a more substantial modification of generation and associated benefits. Focusing on capacity factor in particular, the device with tTiO2 = 12 μm had a capacity factor of 8.87%, which compared to 14.5% for tTiO2 = 48 μm, more than 60% larger. This enhancement in PV capacity factor by device design is larger than for both single-axis (∼10–35%) and dual-axis (∼30–45%) tracking systems50,51 and does not require mechanical tracking systems. Looking in more detail at these data, Fig. 2b, e and h show predicted annual generation mix, change in energy dispatch, and CO2e savings respectively as capacity of tTiO2 = 48 μm CFPV devices is increased. It can be seen that even for the modest value of E(140 Wm−2) = 1.9 given by this device results in up to 60% improvement in carbon savings compared to the silicon equivalent. These data emphasise that the benefits associated with CFPV are due to an increase in PCE with reducing light intensity, and not the absolute value of PCE at low irradiance.
We now turn to examine the costs associated with the designed CFPV devices. As it is the PCE under 1000 Wm−2 that determines the rated power of a solar installation, an increase in PCE under these conditions will lead to a reduction in cost. Thus, the predicted module costs for the designed CFPV devices are all substantially lower than for the Bristow and Kettle33 CFPV device which has lower PCE. Focusing on the tTiO2 = 48 μm CFPV device, the predicted module cost is only 3 times higher than the silicon equivalent, contrasting with 16 times higher for the Bristow and Kettle33 CFPV device.
These data have thus shown that changing tTiO2 thickness offers control over both CFPV benefits and module cost. However, this was realised by controlling efficiency at AM1.5 (Fig. 1), such that there was a trade-off between costs (related to PCE at AM1.5) and enhanced capacity factor (related to increase in PCE with reducing light intensity). As such, changing tTiO2 thickness alone did not realise low costs and enhanced capacity factor simultaneously. That said, we highlight that cost is not the only consideration from a system perspective, as electricity supply operators must ensure reliable power supply from a diverse range of sources to meet demand when faced with varying generation from renewables. For example, Nuclear is an important part of the energy mix in the UK despite its high levelised cost of energy because of the reliable base load it provides.52 Thus, more expensive and less variable solar generation may still represent value from a system perspective.
Nonetheless, the costs of the designed CFPV devices motivates us to consider how one may achieve low costs and CFPV benefits simultaneously. In ESI† Section S4, we present data for Dye Cells engineered to give high capacity factor behaviour by another route, namely by changing electrolyte concentration. These data show that we again control the PCE vs. irradiance characteristics through device design, but again this is through a reduction in PCE at AM1.5, in a similar fashion to Fig. 1. Notwithstanding this, we expect that there is still opportunity for small improvement in costs associated with dye-sensitised CFPVs as the present record efficiency for a dye-sensitised PV device under AM1.5 conditions is 13.5%,53 which is higher than the 10% achieved for the tTiO2 = 12 μm device here.
We suggest that the larger opportunity may be in the creation of CFPV devices with Organic23–27 or Perovskite28–31 absorber layers, as higher PCEs in these devices can mitigate costs whilst still realising the benefits of increasing PCE as light intensity reduces. To demonstrate this opportunity, we repeated the above plant dispatch analysis using data for a quarternary Organic PV device reported by Nam et al.,23 and a MAPI Perovskite PV device from Du et al.,28 both of which display CFPV behaviour. Fig. 5 shows the predicted capacity factor for these devices as a function of enhancement factor at l = 140 Wm−2 alongside all other devices considered in this paper. It is observed that the predicted capacity factor varies approximately linearly with enhancement factor for all device types, showing the generality of the approach. We highlight that the enhancement factor is accessible in lab experiments, offering a straightforward first-order method to assess suitability of candidate PVs for CFPV applications. More detailed assessments of suitability would require measurement of the PCE-irradiance characteristic and analysis using the available code. Finally, while the capacity factors for the Perovskite and Organic PV device are modest compared to the Dye-Cell devices developed here, we recall that these devices were not designed for CFPV applications and so optimisation is likely to yield improvements.
![]() | ||
Fig. 5 Predicted UK capacity factor for PV devices considered in this study as a function of enhancement factor at 140 Wm−2. Purple closed squares: dye-cell CFPV with tTiO2 from 12 μm (light) to 48 μm (dark, purple) – Table 1; purple open triangles: dye-cell CFPV with CuM2+ from 0 M (light) to 0.06 M (dark) – ESI† Fig. S6; purple closed star – Bristow and Kettle dye-cell PV33; blue filled circle – Du et al. Perovskite PV;28 orange filled circle – Nam et al. organic PV.23 |
Footnote |
† Electronic supplementary information (ESI) available. See DOI: https://doi.org/10.1039/d3ee00633f |
This journal is © The Royal Society of Chemistry 2023 |