ReducedDisp
openCEM capacity expansion calculations (i.e. new generators, storage, hybrid and interconnectors) need to be carried simultaneously with system dispatch decisions in order to evaluate the performance of the resulting operational capacity of the system. openCEM by default computes capacity expansion decisions with a reduced set of 168-hour hourly dispatch periods (i.e. 7 days at 1 hour intervals). Seven day periods allow for the modelling of the inter-day interaction between intermittent renewable generation and storage, electric vehicle charging patterns and is one of the modes in which demand cycles.
openCEM select dispatch periods based on an aggregate classification of demand data (clustering) and the explicit inclusion of a peak load period and a "dark doldrums period". The aim of this selection is to include a range of demand and resource conditions that represent the demand and resource conditions for a given year.
openCEM splits the operational demand data in each simulated year into 7-day/168-hour dispatch periods and concatenates the demand in NSW, QLD, SA, TAS and VIC into "individuals" of 840 "features". openCEM uses hierarchical clustering to group together individuals who are relatively similar to each other into clusters. The figure below shows an example of all the individuals in a cluster that accounts for 48% of demand weeks in 2020 (ISP 2018 Neutral data).
The number of clusters in openCEM is configurable, but by default is 6 to strike a compromise between diversity of representative dispatch periods and computational burden. For each cluster, the average of each feature is computed and the closest (euclidean distance) individual to the cluster average is selected as a representative dispatch period. In clusters with a single individual, the average and the selected individual are one and the same.
Only the starting date for each representative dispatch period is returned to openCEM for capacity expansion calculations, cluster individuals are not used in simulations.
Demand data classification typically identifies periods of high demand as single individual clusters, however it may not cover the period of greatest demand across the system in a single demand period. This may lead to an underestimation of capacity, storage and transmission needed to maintain reliability during an extended period of high demand.
openCEM implements a "peak week index" which quantifies the aggregated demand across all regions with a 7 day moving average. The figure below shows the computed index for sample data.
The moving average window is centered so that maxima along the index represent the middle of a weekly period of high demand. openCEM then selects a demand period that contains the peak week index maximum. The peak week period start date is added to the set of clustered demand periods so that week is considered explicitly in capacity expansion calculations.
The dark doldrums index computes a 7 day moving average of an aggregate of the ratio of wind and solar generation potential (as described by traces) to operational demand. The objective of the dark doldrums index is to identify extended periods where low solar and wind resource intersect with relatively high electricity demand. This measure helps allocate sufficient dispatchable generation and storage.
openCEM includes a weekly dispatch period where the minimum of the dark doldrums index occurs at the period's halfway point. The dark doldrums dispatch period start date is added to the set of clustered demand periods so it is considered explicitly in capacity expansion calculations.
openCEM employs the stochastic programming module in Pyomo PySP to set up a stochastic optimisation of capacity in each simulated year with a single stage and as many scenarios as there are representative periods (e.g. 6 demand clustered cases + 1 peak week + 1 dark doldrum week by default).
The stochastic program shares new generation, storage, hybrid and interconnector decision variables between scenarios, and finds a set of decisions that are feasible across all scenarios and constitute the least cost on a weighted sum of the objective in each scenario. Cluster scenarios have a weight proportional to the number of individuals in the cluster, whereas peak week and dark doldrum week only have the weight of a single week. The sum of all dispatch periods weights adds to 1.
Return to Documentation or the home page