Short-Term Photovoltaic Power Forecasting Using an LSTM Neural Network and Synthetic Weather Forecast

Abstract

In this paper, a forecasting algorithm is proposed to predict photovoltaic (PV) power generation using a long short term memory (LSTM) neural network (NN). A synthetic weather forecast is created for the targeted PV plant location by integrating the statistical knowledge of historical solar irradiance data with the publicly available type of sky forecast of the host city. To achieve this, a K-means algorithm is used to classify the historical irradiance data into dynamic type of sky groups that vary from hour to hour in the same season. In other words, the types of sky are defined for each hour uniquely using different levels of irradiance based on the hour of the day and the season. This can mitigate the performance limitations of using fixed type of sky categories by translating them into dynamic and numerical irradiance forecast using historical irradiance data. The proposed synthetic weather forecast is proved to embed the statistical features of the historical weather data, which results in a significant improvement in the forecasting accuracy. The performance of the proposed model is investigated using different intraday horizon lengths in different seasons. It is shown that using the synthetic irradiance forecast can achieve up to 33% improvement in accuracy in comparison to that when an hourly categorical type of sky forecast is used, and up to 44.6% in comparison to that when a daily type of sky forecast is used. This highlights the significance of utilizing the proposed synthetic forecast, and promote a more efficient utilization of the publicly available type of sky forecast to achieve a more reliable PV generation prediction. Moreover, the superiority of the LSTM NN with the proposed features is verified by investigating other machine learning engines, namely the recurrent neural network (RNN), the generalized regression neural network (GRNN) and the extreme learning machine (ELM). INDEX TERMS PV power forecasting, machine learning, LSTM, neural network, deep learning, synthetic weather forecast.

Short-Term Forecasting Photovoltaic Solar Power for Home Energy Management Systems

“…The work presented in [33] also employs the MAPE criterion. For summer months, for 6, 12, and 24 h the values obtained were 28.6%, 38.5% and 37.8%.…”

“…To conclude the discussion, the authors would like to highlight that this predictive model is only one of the models necessary for our HEMS. Models of electric consumption were developed by the authors using the same methodology [30,33,44] and will be integrated into a MBPC HEMS scheme as the one described in [54]. Furthermore, in future studies the effect of dust on the PV panels will also to be considered, following the guidelines of [55], where an experimental analysis was developed for different dust types to evaluate their impact on the power output of the modules.…”

“…Hossain and Mahmood [33] employ an LSTM neural network, using as exogeneous variable a synthetic weather forecast. A synthetic weather forecast is created for the targeted PV plant location by integrating the statistical knowledge of historical solar irradiance data with the publicly available sky forecast of the host city.…”

“…Many (several hundred) variables are acquired, either with a sampling time of 1 s or 1 min. The data acquisition system is described in the authors' previous work and may be found in [33]. The residence employed is a detached house, with two floors and with 20 different spaces.…”

Forecast Quality of Physics-Based and Data-Driven PV Performance Models for a Small-Scale PV System

“…Due to the non-linear and time-varying characteristics of the PV output generation, machine learning approaches like neural networks (Hossain and Mahmood, 2020) or ensemble learning techniques (Ahmad et al, 2018) Figure 2 illustrates the development process of the forecasting models based on historical weather forecasts and historical power production measurements. The process can be summarized in the following steps:…”