Nowadays, plenty of data is continuously pouring from the PhotoVoltaic Power Plants (PV) monitoring systems and sensors that could be successfully handled by big data technologies. This paper proposes a methodology that automatically collects the data logs from sensors installed on PV arrays, inverters and weather stations, checks the health status of the PV components, forecasts the generated power for each inverter based on its real operating conditions and the predicted irradiance and finally provides useful insights of the PV system based on the Key Performance Indicators (KPI) using big data technologies. The Ultra-Short-Term Forecast (USTF) algorithm provides the estimations of irradiance and generated power for the next 30 min and is applied on a sliding time window interval. The algorithm uses a Feed-Forward Artificial Neural Network (FF-ANN) and, to significantly reduce the number of iterations, we propose a backtracking adjustment of the learning rate that enables faster convergence reducing the computational time that is essential for USTF. Two data sets from PV Agigea 0.5 MW and PV Giurgiu 7.5 MW, located in the South-East and South of Romania, that consist in data logs from inverters and arrays, are used for simulation. The exhaustive analyses are performed for PV Agigea (including KPI calculation), while PV Giurgiu data set was mainly used to check the scalability and replicability of the algorithm.

如今，光伏电站（PV）监控系统和传感器不断涌入大量数据，这些数据可以通过大数据技术成功处理。本文提出了一种方法，该方法可自动从安装在光伏阵列，逆变器和气象站上的传感器收集数据日志，检查光伏组件的运行状况，根据其实际运行状况和预测的辐照度预测每个逆变器的发电量，最后使用大数据技术基于关键绩效指标（KPI）提供了有关光伏系统的有用见解。超短期预报（USTF）算法提供了下一个30分钟的辐照度和发电功率的估计值，并应用于滑动时间窗口间隔。该算法使用前馈人工神经网络（FF-ANN），为了显着减少迭代次数，我们提出了对学习率的回溯调整，以实现更快的收敛速度，从而减少了USTF必不可少的计算时间。位于罗马尼亚东南和南部的PV Agigea 0.5 MW和PV Giurgiu 7.5 MW的两个数据集用于仿真，这些数据集包含来自逆变器和阵列的数据日志。对PV Agigea进行了详尽的分析（包括KPI计算），而PV Giurgiu数据集主要用于检查算法的可伸缩性和可复制性。位于罗马尼亚东南和南部的PV Agigea 0.5 MW和PV Giurgiu 7.5 MW的两个数据集用于仿真，这些数据集包含来自逆变器和阵列的数据日志。对PV Agigea进行了详尽的分析（包括KPI计算），而PV Giurgiu数据集主要用于检查算法的可伸缩性和可复制性。位于罗马尼亚东南和南部的PV Agigea 0.5 MW和PV Giurgiu 7.5 MW的两个数据集用于仿真，这些数据集包含来自逆变器和阵列的数据日志。对PV Agigea进行了详尽的分析（包括KPI计算），而PV Giurgiu数据集主要用于检查算法的可伸缩性和可复制性。