Decision support system for photovoltaic fault detection avoiding meteorological conditions

TL;DR

The paper addresses PV fault detection under variable meteorological conditions by introducing a meteorology-agnostic decision support framework that combines fuzzy sets, OWA aggregation, and a state machine. It defines daily performance metrics $\rho_i(d)$ and pairwise differences $\delta_{ik}(d)$, uses these to build membership functions $\mu_{ik}$, and then applies an OWA-based aggregation to produce per-facility scores $y_i$ that feed a linguistic alert and state transitions. The main contributions are the fuzzy-mechanism for describing facility behavior, the learning of performance classifications, and the automated, natural-language alerting coupled with a temporal state machine, demonstrated on real GEN data with generally high accuracy and no false positives. This approach offers scalable, meteorology-independent fault detection for PV plants and has practical impact for reducing energy losses and enabling automatic daily reporting across heterogeneous installations.

Abstract

A fundamental issue about installation of photovoltaic solar power stations is the optimization of the energy generation and the fault detection, for which different techniques and methodologies have already been developed considering meteorological conditions. This fact implies the use of unstable and difficult predictable variables which may give rise to a possible problem for the plausibility of the proposed techniques and methodologies in particular conditions. In this line, our goal is to provide a decision support system for photovoltaic fault detection avoiding meteorological conditions. This paper has developed a mathematical mechanism based on fuzzy sets in order to optimize the energy production in the photovoltaic facilities, detecting anomalous behaviors in the energy generated by the facilities over time. Specifically, the incorrect and correct behaviors of the photovoltaic facilities have been modeled through the use of different membership mappings. From these mappings, a decision support system based on OWA operators informs of the performances of the facilities per day, by using natural language. Moreover, a state machine is also designed to determine the stage of each facility based on the stages and the performances from previous days. The main advantage of the designed system is that it solves the problem of "constant loss of energy production", without the consideration of meteorological conditions and being able to be more profitable. Moreover, the system is also scalable and portable, and complements previous works in energy production optimization. Finally, the proposed mechanism has been tested with real data, provided by Grupo Energético de Puerto Real S.A. which is an enterprise in charge of the management of six photovoltaic facilities in Puerto Real, Cádiz, Spain, and good results have been obtained for faulting detection.

Figures (4)

• Figure 1: Membership function indicating the degree of suitable performance of the facility $\mathcal{I}_i$ with respect to $\mathcal{I}_k$.
• Figure 2: Membership function indicating the degree of suitable performance of the facility $\mathcal{I}_i$ with respect to $\mathcal{I}_k$, when we do not have information about the incorrect days neither $\mathcal{I}_i$ nor $\mathcal{I}_k$.
• Figure 3: State machine for every facility
• Figure 4: Total performance of energy production of $\mathcal{I}_i$ in June 2020.