New electricity generation technologies are often assessed using simple metrics such as average return, break‐even energy prices or levelised cost of electricity. These simple metrics do not always capture the full economic value of a technology, particularly those that can react quickly and efficiently to changes in demand. In a wholesale spot market, opportunities exist to capture greater revenues, as currently achieved by peak power plants. This study demonstrates the use of two complementary methods to determine how storage should be operated to maximise revenue. First, the authors formulate and solve the problem as a linear program. The results indicate that there are distinct control modes. They then use Pontryagin's principle to confirm that the optimal operating strategy has three distinct control modes: (i) store all collected power, without generating, (ii) generate using collected power only and (iii) generate at maximum capacity using both collected and stored power. The mode that should be used at any instant depends only on the spot price of electricity relative to a pair of critical prices. These critical prices depend on the total energy that will be collected and the turnaround efficiency of the storage system.

Concentrating solar thermal power plants with thermal energy storage is a potential source of clean electrical power. This work has the same four optimal control modes established in a previous paper where Pontryagin's principle demonstrated that to maximise revenue from a plant with unlimited storage the optimal modes were to: (i) store all collected power, without generating; (ii) generate using collected power only; (iii) generate at maximum capacity using both collected and stored power; (iv) generate a maximum capacity, storing any surplus power. This paper presents an expanded analysis to cover a more realistic case where there is limited storage capacity, introducing 2 differences. First, it is sometimes necessary to discard power if the store is full. Second, the critical prices used to determine the optimal control can change whenever the store becomes full or empty. Again we use Pontragin's principle to derive necessary conditions for an optimal control and present an algorithm to find the sequence of critical prices required to construct it. The analysis gives more insight into the nature of the optimal control than mathematical programming or probabilistic search techniques, and the calculation is fast‐finding the optimal control for a yearlong operation in about 30 seconds.

The proliferation of non-scheduled generation from renewable electrical energy sources such concentrated solar power (CSP) presents a need for enabling scheduled generation by incorporating energy storage; either via directly coupled Thermal Energy Storage (TES) or Electrical Storage Systems (ESS) distributed within the electrical network or grid. The challenges for 100% renewable energy generation are: to minimise capitalisation cost and to maximise energy dispatch capacity. The aims of this review article are twofold: to review storage technologies and to survey the most appropriate optimisation techniques to determine optimal operation and size of storage of a system to operate in the Australian National Energy Market (NEM). Storage technologies are reviewed to establish indicative characterisations of energy density, conversion efficiency, charge/discharge rates and costings. A partitioning of optimisation techniques based on methods most appropriate for various time scales is performed: from “whole of year”, seasonal, monthly, weekly and daily averaging to those best suited matching the NEM bid timing of five minute dispatch bidding, averaged on the half hour as the trading settlement spot price. Finally, a selection of the most promising research directions and methods to determine the optimal operation and sizing of storage for renewables in the grid is presented.

This study compares parametric statistical time series models, such as autoregressive moving average (ARMA), with nonparametric artificial neural networks, specifically long short‐term memory (LSTM) models, for univariate forecasting. Two time series are analyzed separately: wind power output from the Clements Gap wind farm and the regional electricity price for South Australia. One‐step‐ahead forecast performance is evaluated using normalized mean bias error (NMBE), normalized mean absolute error (NMAE), and normalized root mean square error (NRMSE). Three LSTM models were examined: a manually tuned model, a structurally equivalent model implemented in a different library, and a model with automated hyperparameter tuning. While LSTM models achieved competitive performance, statistical models often performed equally well or better. For price forecasts, the manually tuned LSTM achieved the lowest NMBE (), while ARMA(3,1) had the best NMAE (0.0253) and AR(6) the best NRMSE (0.220). For wind forecasts, the manually tuned LSTM again performed best overall (NMBE: 0.00034, NMAE: 0.143, NRMSE: 0.225), while the equivalent library model performed worst. These results highlight the need to subject nonparametric LSTM models to more rigorous and systematic evaluation relative to their parametric statistical counterparts.