ThesisPDF Available

Forecasting and optimization of ancillary services provision by renewable energy sources

Authors:

Abstract

As variable renewable energy plants penetrate significantly the electricity generation mix, they are expected to contribute to the supply of reserve power, albeit the high uncertainty levels on their production. A solution to reduce the uncertainty consists in aggregating renewable plants dispersed over several climates to obtain a smoother production profile and operate them within a Virtual Power Plant control system. In this thesis, a series of probabilistic forecasting models are proposed to assess the capacity of a variable renewable Virtual Power Plant to provide ancillary services with maximum reliability: these models are adapted decision-tree regression models, recurrent and convolutional neural networks, as well as distributions dedicated to extremely low quantiles. The combination of energy sources (Photovoltaics, Wind, Run-of-river Hydro) is considered in detail. Optimal strategies for the joint offer of energy and ancillary services by a variable renewable Virtual Power Plant are later defined, based on production forecasts and market uncertainties. Offer strategies explore several modelling options:dependence between renewable production and prices via a copula, controlled rate of reserve underfullfilment with a chance-constraint optimization, and finally offer of multiple ancillary services thanks to a Lagrangian formulation.
A preview of the PDF is not available
ResearchGate has not been able to resolve any citations for this publication.
Full-text available
Article
We propose a framework for evaluating the quality of solar irradiance probabilistic forecasts. The verification framework is based on visual diagnostic tools and a set of scoring rules mostly originating from the weather forecast verification community. Two types of probabilistic forecasts are used as a basis to illustrate the application of these verification approaches. The first one consists in ensemble forecasts commonly provided by national or international meteorological centres. The second one originates from statistical methods and produces a set of discrete quantile forecasts, the nominal proportions of which span the unit interval. These probabilistic forecasts are evaluated for two selected sites that experience very different climatic conditions. The first site is located in the continental US while the second one is situated on La Réunion Island. Although visual diagnostic tools can help identify deficiencies in generated forecasts, it is recommended that a set of numerical scores be used to assess the quality of probabilistic forecasts. In particular, the Continuous Ranked Probability Score (CRPS) seems to have all the features needed to evaluate a probabilistic forecasting system and, as such, may become a standard for verifying solar irradiance probabilistic forecasts and by extension probabilistic forecasts of solar power generation.
Full-text available
Thesis
Artificial Neural Networks are connectionist systems that perform a given task by learning on examples without having prior knowledge about the task. This is done by finding an optimal point estimate for the weights in every node. Generally, the network using point estimates as weights perform well with large datasets, but they fail to express uncertainty in regions with little or no data, leading to overconfident decisions. In this paper, Bayesian Convolutional Neural Network (BayesCNN) using Variational Inference is proposed, that introduces probability distribution over the weights. Furthermore, the proposed BayesCNN architecture is applied to tasks like Image Classification, Image Super-Resolution and Generative Adversarial Networks. The results are compared to point-estimates based architectures on MNIST, CIFAR-10 and CIFAR-100 datasets for the Image Classification task, on BSD300 dataset for Image Super-Resolution task and on CIFAR10 dataset again for Generative Adversarial Network task. BayesCNN is based on Bayes by Backprop which derives a variational approximation to the true posterior. We, therefore, introduce the idea of applying two convolutional operations, one for the mean and one for the variance. Our proposed method not only achieves performances equivalent to frequentist inference in identical architectures but also incorporate a measurement for uncertainties and regularisation. It further eliminates the use of dropout in the model. Moreover, we predict how certain the model prediction is based on the epistemic and aleatoric uncertainties and empirically show how the uncertainty can decrease, allowing the decisions made by the network to become more deterministic as the training accuracy increases. Finally, we propose ways to prune the Bayesian architecture and to make it more computational and time-effective.
Full-text available
Article
New methods are proposed for adjusting probabilistic forecasts to ensure coherence with the ag-gregation constraints inherent in temporal hierarchies. The different approaches nested within this framework include methods that exploit information at all levels of the hierarchy as well as a novel method based on cross-validation. The methods are evaluated using real data from two wind farms in Crete and electric load in Boston. For these applications, optimal decisions related to grid operations and bidding strategies are based on coherent probabilistic forecasts of energy power. Empirical evidence is also presented showing that probabilistic forecast reconciliation improves the accuracy of the probabilistic forecasts.
Full-text available
Article
This paper proposes a methodology for an efficient generation of correlated scenarios of Wind, Photovoltaics (PV) and small Hydro production considering the power system application at hand. The merits of scenarios obtained from a direct probabilistic forecast of the aggregated production are compared with those of scenarios arising from separate production forecasts for each energy source, the correlations of which are modeled in a later stage with a multivariate copula. It is found that scenarios generated from separate forecasts reproduce globally better the variability of a multi-source aggregated production. Aggregating renewable power plants can potentially mitigate their uncertainty and improve their reliability when they offer regulation services. In this context, the first application of scenarios consists in devising an optimal day-ahead reserve bid made by a Wind-PV-Hydro Virtual Power Plant (VPP). Scenarios are fed into a two-stage stochastic optimization model, with chance-constraints to minimize the probability of failing to deploy reserve in real-time. Results of a case study show that scenarios generated by separately forecasting the production of each energy source leads to a higher Conditional Value at Risk than scenarios from direct aggregated forecasting. The alternative forecasting methods can also significantly affect the scheduling of future power systems with high penetration of weather-dependent renewable plants. The generated scenarios have a second application here as the inputs of a two-stage stochastic unit commitment model. The case study demonstrates that the direct forecast of aggregated production can effectively reduce the system operational cost, mainly through better covering the extreme cases. The comprehensive application-based assessment of scenario generation methodologies in this paper informs the decision-makers on the optimal way to generate short-term scenarios of aggregated RES production according to their risk aversion and to the contribution of each source in the aggregation.
Full-text available
Article
With increasing renewable energy generation capacities connected to the power grid, a number of decision-making problems require some form of consistency in the forecasts that are being used as input. In everyday words, one expects that the sum of the power generation forecasts for a set of wind farms is equal to the forecast made directly for the power generation of that portfolio. This forecast reconciliation problem has attracted increased attention in the energy forecasting literature over the last few years. Here, we review the state of the art and its applicability to day-ahead forecasting of wind power generation, in the context of spatial reconciliation. After gathering some observations on the properties of the game-theoretical optimal projection reconciliation approach, we propose to readily rethink it in a distributed setup by using the Alternating Direction Method of Multipliers (ADMM). Three case studies are considered for illustrating the interest and performance of the approach, based on simulated data, the National Renewable Energy Labaratory (NREL) Wind Toolkit dataset, and a dataset for a number of geographically distributed wind farms in Sardinia, Italy.
Article
To decarbonize the heating sector, residential consumers may install heat pumps. Coupled with heating loads with high thermal inertia, these thermostatically controlled loads may provide a significant source of demand side flexibility. Since the capacity of residential consumers is typically insufficient to take part in the day-ahead electricity market (DAM), aggregators act as mediators that monetize the flexibility of these loads through demand response (DR). In this paper, we study the strategic interactions between an aggregator, its consumers and the DAM using a bilevel optimization framework. The aggregator-consumer interaction is captured either as a Stackelberg or a Nash Bargaining Game, leveraging chance-constrained programming to model limited controllability of residential DR loads. The aggregator takes strategic positions in the DAM, considering the uncertainty on the market outcome, represented as a stochastic Stackelberg Game. Results show that the DR provider-aggregator cooperation may yield significant monetary benefits. The aggregator cost-effectively manages the uncertainty on the DAM outcome and the limited controllability of its consumers. The presented methodology may be used to assess the value of DR in a deregulated power system or may be directly integrated in the daily routine of DR aggregators.
Article
Wind power plays a leading role in the development of renewable energy. However, the random nature of wind turbine power and its associated uncertainty create challenges in dispatching this power effectively in the power system, which can result in unnecessary curtailment of the wind turbine power. Improving the accuracy of wind turbine power forecasting is an effective measure for resolving such problems. This study uses a deep learning network to forecast the wind turbine power based on a long short-term memory network (LSTM) algorithm and uses the Gaussian mixture model (GMM) to analyze the error distribution characteristics of short-term wind turbine power forecasting. The LSTM algorithm is used to forecast the power and uncertainties for three wind turbines within a wind farm. According to numerical weather prediction (NWP) data and historical power data for three turbines, the forecasting accuracy of the turbine with the largest number of training samples is the best of the three. For one of the turbines, the LSTM, radial basis function (RBF), wavelet, deep belief network (DBN), back propagation neural networks (BPNN), and Elman neural network (ELMAN) have been used to forecast the wind turbine power. This study compares the results and demonstrates that LSTM can greatly improve the forecasting accuracy. Moreover, this study obtains different confidence intervals for the three units according to the GMM, mixture density neural network (MDN), and relevance vector machine (RVM) model results. The LSTM method is shown to have higher accuracy and faster convergence than the other methods. However, the GMM method has better performance and evaluation than other methods and thus has practical application value for wind turbine power dispatching.