This paper proposes a new model for short-term forecasting power generation capacity of large-scale solar power plant (SPP) in Vietnam considering the fluctuations of weather factors when applying the Long Short-Term Memory networks (LSTM) algorithm. At first, a configuration of the model based on the LSTM algorithm is selected in accordance with the weather and operating conditions of SPP in Vietnam. Not only different structures of LSTM model but also other conventional forecasting methods for time series data are compared in terms of error accuracy of forecast on test data set to evaluate the effectiveness and select the most suitable LSTM configuration. The most suitable configuration has been selected and applied on Thanh Thanh Cong No 1 (TTC) SPP with 2 input cases: real historical weather data and forecasted weather data. The results show that second case gives a much larger Mean Absolute Percentage Error (MAPE) than that of first case (10.857% versus 3.491%). Based on above experiment, new additional features are proposed to improve the selected LSTM model precision and cope with the problem of error due to weather forecast data. The result of the application of the new prediction model for TTC solar plant indicates that the MAPE is reduced from 10.857% to 9.881%.

本文提出了一种在应用长短期记忆网络（LSTM）算法时考虑天气因素波动的越南大型太阳能发电厂（SPP）短期发电量预测新模型。首先，根据越南SPP的天气和运行条件，选择基于LSTM算法的模型配置。不仅LSTM模型的不同结构，而且其他传统的时间序列数据预测方法在测试数据集的预测误差精度方面进行比较，以评估有效性并选择最合适的LSTM配置。最合适的配置已被选择并应用于 Thanh Thanh Cong No 1 (TTC) SPP，有 2 个输入案例：真实历史天气数据和预测天气数据。结果表明，第二种情况的平均绝对百分比误差 (MAPE) 比第一种情况大得多（10.857% 对 3.491%）。在上述实验的基础上，提出了新的附加特征，以提高所选 LSTM 模型的精度，并解决由于天气预报数据导致的错误问题。TTC太阳能电站新预测模型的应用结果表明，MAPE从10.857%降低到9.881%。