No abstract

No abstract

No abstract

No abstract

The deployment of microgrids could be fostered by control systems that do not require very complex modelling, calibration, prediction and/or optimisation processes. This paper explores the application of Reinforcement Learning (RL) techniques for the operation of a microgrid. The implemented Deep Q-Network (DQN) can learn an optimal policy for the operation of the elements of an isolated microgrid, based on the interaction agent-environment when particular operation actions are taken in the microgrid components. In order to facilitate the scaling-up of this solution, the algorithm relies exclusively on historical data from past events, and therefore it does not require forecasts of the demand or the renewable generation. The objective is to minimise the cost of operating the microgrid, including the penalty of non-served power. This paper analyses the effect of considering different definitions for the state of the system by expanding the set of variables that define it. The obtained results are very satisfactory as it can be concluded by their comparison with the perfect-information optimal operation computed with a traditional optimisation model, and with a Naive model.

In this article, a parametric model for health condition monitoring of wind turbines is developed. The study is based on the assumption that a wind turbine’s health condition can be modeled through three features: rotor speed, gearbox temperature and generator winding temperature. At first, three neural network models are created to simulate normal behavior of each feature. Deviation signals are then defined and calculated as accumulated time-series of differences between neural network predictions and actual measurements. These cumulative signals carry health condition–related information. Next, through nonlinear regression technique, the signals are used to produce individual models for considered features, which mathematically have the form of proportional hazard models. Finally, they are combined to construct an overall parametric health condition model which partially represents health condition of the wind turbine. In addition, a dynamic threshold for the model is developed to facilitate and add more insight in performance monitoring aspect. The health condition monitoring of wind turbine model has capability of evaluating real-time and overall health condition of a wind turbine which can also be used with regard to maintenance in electricity generation in electric power systems. The model also has flexibility to overcome current challenges such as scalability and adaptability. The model is verified in illustrating changes in real-time and overall health condition with respect to considered anomalies by testing through actual and artificial data.

This paper proposes an approach for stress condition monitoring and maintenance assessment in wind turbines (WTs) through large amounts of collected data from the supervisory control and data acquisition (SCADA) system. The objectives of the proposed approach are to provide a stress condition model for health monitoring, to assess the WT's maintenance strategies, and to provide recommendations on current maintenance schemes for future operations of the wind farm. At first, several statistical techniques, namely principal component analysis, Pearson, Spearman and Kendall correlations, mutual information, regressional ReliefF and decision trees are used and compared to assess the data for dimensionality reduction and parameter selection. Next, a normal behavior model is constructed by an artificial neural network which performs condition monitoring analysis. Then, a model based on the mathematical form of a proportional hazards model is developed where it represents the stress condition of the WT. Finally, those two models are jointly employed in order to analyze the overall performance of the WT over the study period. Several cases are analyzed with five-year SCADA data and maintenance information is utilized to develop and validate the proposed approach.

No abstract