Endogenous Approaches for Methodological Improvement

Time Resolution Enhancement:

Selection of high temporal resolution plays a crucial role in representing the dynamics of diurnal and seasonal electricity load curves in energy system models. It also allows a user to depict high granular RE availability factor by time slice and operational constraints. For this, higher temporal disaggregation of load and RE has been performed by considering twenty annual time slices in the long-term UK-MARKAL model. High resolution helps to improve load balancing and electricity dispatch simulations and analyze the role of demand and supply side storage options [20]. There are instances of adopting a high number of annual time slices with hourly diurnal resolution to capture hourly, daily and seasonal supply and demand dynamics [21, 106, 107]. Coarse temporal resolution leads to sub-optimal investment in future RE technologies as well as flexible capacity. Higher time resolution is, therefore, crucial to assess the real impact of a particular energy policy, especially in high RE-penetrated systems. But, high temporal resolution doesn’t necessarily mean that model can handle RE stochasticity; it is only the variability which is being captured. Deterministic representation of system uncertainty is sufficient in planning studies, as long as the time resolution of system operation and the model are similar [108].

Limited number of time slices can be compensated by improving other aspects like model formulation, and higher spatial resolution. Short-term system fluctuations have been captured by only 49 annual time slices (four seasons with three typical days, four diurnal

2.4 Addressing RE Intermittency in Energy system Models 29 and an additional super peak time slice) in LIMES-EU⁺model [109]. The model has a high spatial representation of the planning area (Some part of Europe and the Middle East/North Africa (MENA) regions), which has been divided into 20 geographical entities connected by 32 transmission corridors. Similarly, major modeling enhancement has been performed by considering reserve capacity, day-night and seasonal electricity storage, RE curtailment, industrial DSM, smart grids and endogenous transmission and distribution network with 78 yearly time slice in Belgium-TIMES model [110]. Twelve annual time slices have been utilized in EU-TIMES model with major mathematical improvement in TIMES model (The integrated MARKAL EFOM system) formulation by including a detailed representation of power flow in transmission lines [111, 112]. Also, different availability factors of RE technologies pertaining to each time slice is considered in this model.

In long-term energy system planning models, it is not feasible to consider consecutive time series of demand or RE generation. Selection of time resolution depends mainly on annual, seasonal and daily load curve, rather than RE time series fluctuations. Normally, there is no yardstick or guideline available to select optimum number of time slices for a given system, and it is often decided by examining available data, study focus, system in hand, and computational resources. Recently, some studies have utilized only a set of typical days within a year considering the historical RE and demand time series. This approach can decrease computational requirements of planning models by reducing temporal resolution but without compromising on the reliability of results. Different methods such as heuristics, random selection, optimization models, hierarchical clustering algorithm, and hybrid methods can be adopted for this purpose [113, 114]. Consideration of only six typical days,i.e. 48 selected annual time slices in LIMES-EU model is sufficient to capture temporal dynamics, which can only be obtained with much higher temporal resolution in standard approach [114].

Uncertainty in historical wind power and electricity price has been represented in a long-term model by two stage scenario tree constructed by joint probability distribution [115]. The focus in this approach is on simultaneous optimization of investment (first stage) as well as on operational decisions (second stage). Compared to a deterministic version of the model with peak reserve constraint, stochastic model reports less wind capacity and system cost for model years. Stochastic approach makes it possible to endogenously optimize reserve capacity for intermittent RE integration.

Spatial Resolution Enhancement:

Spatial resolution selection in energy system models is often driven by political or economic boundary, rather than RE resource variations. Various attempts have been made to improve

spatial resolution by dividing the planning area into many regions. Fifty load areas and 124 existing and new transmission corridors have been considered in SWITCH model for capacity expansion study in western North America [22]. Five model regions are considered within German transmission system control area to track spatial variation of RE resources, as well as the effect of transmission capacity expansion to meet the future load growth and wind energy integration [116]. Thirty two model regions have been considered in a TIMES based planning model of USA power sector namely, FACETS [23]. Despite high granularity of spatial assumptions in these attempts, power flow between model regions is often represented as a transportation problem,i.e. the physical aspects of electrical load flow are ignored. Power flow constraints like bus voltage and angle limit, and transmission line capacity limit are often not considered. Linearized DC power flow equations with simplified N-1 security constraints have been incorporated in long-term model generator TIMES for better representation of transmission grid [117]. This feature has been utilized in some studies for long-term system planning [111, 26].

Enhancing the Representation of Technical Constraints:

To simulate the effect of RE intermittency, high temporal and spatial resolution should be complemented by a detailed description of operational constraints in the mathematical formulations of models [25, 118]. Various approaches have tried to endogenously incorporate additional operational constraints to capture the effect of RE variability.

Residual load duration curve (RLDC)⁶captures the relation between RE variability and demand, and reveals the associated integration challenges [119]. In long-term models, it allows defining constraints on the minimum production level of thermal units. It also helps to determine the storage capacity, flexible generation, and RE curtailment for expected variations in residual load and RE generation [120]. Thus, it provides simultaneous optimization of investment planning and system operation. Traditional RLDC representation can be improved by piece-wise linearization into three parts; base load, intermediate, and peak including reserve margin. In REMIND-D model, representation of demand and RE variation using this approach leads to 35%, and 27% reduction in generation from variable RE plants in the base, and ambitious green house gas mitigation scenario respectively in 2050 compared to a model version without any description of variability [24]

For any power system, critical operational constraint are related to generating units as described in Section 2.1 and 2.2. There are recent attempts to include these constraints in long- term modeling frameworks like OSeMOSYS, TIMES, and eMix [25, 118, 121, 122]. The

6RLDC is the load duration curve minus total RE generation.

2.4 Addressing RE Intermittency in Energy system Models 31 motive of these approaches is to endogenously simulate short-term power system operation within long-term planning models.