A main challenge for integrating the intermittent photovoltaic (PV) power generation remains the accuracy of day-ahead forecasts and the establishment of robust performing methods. The purpose of this work is to address these technological challenges by evaluating the day-ahead PV production forecasting performance of different machine learning models under different supervised learning regimes and minimal input features. Specifically, the day-ahead forecasting capability of Bayesian neural network (BNN), support vector regression (SVR), and regression tree (RT) models was investigated by employing the same dataset for training and performance verification, thus enabling a valid comparison. The training regime analysis demonstrated that the performance of the investigated models was strongly dependent on the timeframe of the train set, training data sequence, and application of irradiance condition filters. Furthermore, accurate results were obtained utilizing only the measured power output and other calculated parameters for training. Consequently, useful information is provided for establishing a robust day-ahead forecasting methodology that utilizes calculated input parameters and an optimal supervised learning approach. Finally, the obtained results demonstrated that the optimally constructed BNN outperformed all other machine learning models achieving forecasting accuracies lower than 5%.

Accurate photovoltaic (PV) power forecasting is an essential tool that enables the efficient integration of solar electricity and energy trading. This work proposes a novel methodology for the implementation of a forecasting tool that provides aggregated PV production day‐ahead forecasts for the PV systems installed at a distribution level. Specifically, the tool is a cloud‐based platform, comprising of a data quality block, a weather forecasting model, a machine learning power prediction step and an up‐scaling aggregation stage. In this context and in the absence of a fully observable distribution system, the aggregated PV generation was estimated using a clustering approach and up‐scaling the measured generation of reference systems to the aggregated installed capacity. The results demonstrated that the implemented tool exhibited high forecasting accuracies, lower than 10% given by the mean absolute percentage error when applied to the distribution system of Cyprus. Furthermore, the comparative benchmarking of the up‐scaling techniques against the estimated aggregated PV generation demonstrated that the best‐performing approach was the hybrid model, which provided a normalised root mean square error of 10.29% and mean absolute percentage error of 9.11%. Finally, useful information is provided for establishing a robust day‐ahead forecasting methodology that is based on an optimal supervised learning and hybrid up‐scaling approach.

Nowadays, we have witnessed a significant increase of renewable sources across the existing power grids as a result of the undeniable need to mitigate the effects of environmental pollution and to assist the economic development of the countries. However, due to the lack of rotational inertia and their intermittent nature, renewables have increased the severity of the occurring disturbances and instability phenomena, which in turn, place a demand for more accurate modeling of the grid and the power system components. In this paper, we propose a novel network formulation for AC microgrids that allows the consideration of less conservative conditions for the network whilst facilitating the adoption of higher order inverter-based generation dynamics and the stability analysis and control. In particular, we formulate the network as a dynamical multi-input/multi-output system, which we show is both stable and passive, even when the network's dynamic and lossy nature is considered. Then we present the opportunities and the advantages offered by the proposed framework and provide useful guidelines for a more accurate modeling of power system components. Finally, we verify the proposed network formulation through a numerical application on the Four Machine Two Area Kundur test system and a dynamic simulation on typical Medium Voltage (MV) distribution lines.

The ongoing efforts by many countries worldwide to increase the share of renewable energy sources in power generation has changed the nature of existing power grids and introduced numerous stability-related issues. To address these problems, more extensive and detailed stability analysis studies are therefore needed. Driven by this need, in this article, we present a passivity-based framework for stability analysis and control design that allows more accurate modeling of both the network and the power system components while facilitating the derivation of completely decentralized stability results. In particular, the proposed approach relies on the formulation of the network as a dynamical multivariable system, which is shown to be passive, even when the network’s dynamic and lossy nature are taken into account. The application of decentralized passivity conditions on bus dynamics is then further exploited, together with the incorporation of more accurate dynamical models for the power system components, to guarantee the asymptotic stability of the interconnected system. Moreover, we discuss the opportunities provided by the proposed approach regarding the incorporation of higher order dynamics, the derivation of significant stability results in a completely decentralized manner and the design of effective distributed control mechanisms. These opportunities are finally verified through the design of a demand-side voltage droop controller and several dynamic simulations on two typical test systems.