Search

This work proposes the use of reinforcement learning (RL) techniques with deep-learning models to address the maximum power point tracking (MPPT) control problem of a photovoltaic (PV) array. We implemented the deep deterministic policy gradient (DDPG) method, the inverted gradient (IGDDPG) method and the delayed twins (TD3) method to solve the MPPT control problem. Several simulation experiments were performed in the OpenAI Gym platform aiming to evaluate the performance of the proposed control strategies, under different operating conditions in terms of temperature and solar irradiance. The obtained results show that the use of deep reinforcement learning (DRL) achieves a successful performance for the MPPT control problem with a fast response and a stable behavior. Moreover, the algorithms do not require any previous knowledge about the dynamic behavior of the photovoltaic array.

The use of multimodal control techniques for batch dynamic systems offers a promising alternative to successfully perform tasks of supervision and control in industrial processes. These techniques seek to integrate different control strategies with partial objectives or «behaviors» using Lebesgue automatons overlooking achieve certain operational objectives. These automatons can identify optimal sequences of modes, also called control programs, using system simulations. Multimodal control programs consist of a sequence of modes, each of which comprises a feedback control law (x) and relevant termination conditions (x,T). In this paper the use of a reinforcement learning algorithm is described via to find, by simulations, optimal control program aligned to maximize productivity in a buffer system composed of two tanks. Fil: de Paula, Mariano. Universidad Nacional del Centro de la Provincia de Buenos Aires. Centro de Investigaciones en Física e Ingeniería del Centro de la Provincia de Buenos Aires. - Consejo Nacional de Investigaciones Científicas y Técnicas. Centro Científico Tecnológico Conicet - Tandil. Centro de Investigaciones en Física e Ingeniería del Centro de la Provincia de Buenos Aires. - Provincia de Buenos Aires. Gobernación. Comisión de Investigaciones Científicas. Centro de Investigaciones en Física e Ingeniería del Centro de la Provincia de Buenos Aires; Argentina Fil: Sánchez Reinoso, Carlos Roberto. Consejo Nacional de Investigaciones Científicas y Técnicas; Argentina Fil: Avila, Luis Omar. Universidad Nacional de San Luis. Facultad de Ciencias Físico Matemáticas y Naturales. Departamento de Informática. Laboratorio Investigación y Desarrollo en Inteligencia Computacional; Argentina. Consejo Nacional de Investigaciones Científicas y Técnicas; Argentina

[ES] El control de sistemas complejos puede ser realizado descomponiendo la tarea de control en una secuencia de modos de control, o simplemente modos. Cada modo implementa una ley de retroalimentación hasta que se activa una condición de terminación, en respuesta a la ocurrencia de un evento exógeno/endógeno que indica que la ejecución del modo debe finalizar. En este trabajo se presenta una propuesta novedosa para encontrar una política de conmutación óptima para resolver el problema de control optimizando alguna medida de costo/beneficio. Una política óptima implementa un programa de control multimodal óptimo, el cual consiste en un encadenamiento de modos de control. La propuesta realizada incluye el desarrollo y formulación de un algoritmo basado en la idea de la programación dinámica integrando procesos Gaussianos y aprendizaje Bayesiano activo. Mediante el enfoque propuesto es posible realizar un uso eficiente de los datos para mejorar la exploración de las soluciones sobre espacios de estados continuos. Un caso de estudio representativo es abordado para demostrar el desempeño del algoritmo propuesto., [EN] The control of complex systems can be done decomposing the control task into a sequence of control modes, or modes for short. Each mode implements a parameterized feedback law until a termination condition is activated in response to the occurrence of an exogenous/endogenous event, which indicates that the execution mode must end. This paper presents a novel approach to find an optimal switching policy to solve a control problem by optimizing some measure of cost/benefit. An optimal policy implements an optimal multimodal control program, consisting in a sequence of control modes. The proposal includes the development of an algorithm based on the idea of dynamic programming integrating Gaussian processes and Bayesian active learning. In addition, an efficient use of the data to improve the exploration of the continuous state spaces solutions can be achieved through this approach. A representative case study is discussed and analyzed to demonstrate the performance of the proposed algorithm.

The objective of this chapter is to optimize the photovoltaic power plant considering the effects of variable shading on time and weather. For that purpose, an optimization scheme based on the simulator from Sanchez Reinoso, Milone, and Buitrago (2013) and on evolutionary computation techniques is proposed. Regarding the latter, the representation used and the proposed initialization mechanism are explained. Afterwards, the proposed algorithms that allow carrying out crossover and mutation operations for the problem are detailed. In addition, the designed fitness function is presented. Lastly, experiments are conducted with the proposed optimization methodology and the results obtained are discussed.