The solar energy has a high theoretical potential to meet the increasing global demand for energy. And when in real application, the solar energy’s variability and availability must be taken into account. Meanwhile, energy management is also vital for the future economic prosperity, and load consumption plays a key role in energy management system, not only for normal resident, but also for commercial building or manufacturing enterprises. Thus the utilization of prediction on photovoltaic power generation and load consumption can alleviate the uncertainty of photovoltaic system, benefit the development of solar energy and reduce the cost in electrical part.

In this thesis, an Elman Neural Network is developed for prediction the daily generation of photovoltaic system and l...[ Read more ]

The solar energy has a high theoretical potential to meet the increasing global demand for energy. And when in real application, the solar energy’s variability and availability must be taken into account. Meanwhile, energy management is also vital for the future economic prosperity, and load consumption plays a key role in energy management system, not only for normal resident, but also for commercial building or manufacturing enterprises. Thus the utilization of prediction on photovoltaic power generation and load consumption can alleviate the uncertainty of photovoltaic system, benefit the development of solar energy and reduce the cost in electrical part.

In this thesis, an Elman Neural Network is developed for prediction the daily generation of photovoltaic system and load consumption. With the history data and weather information, the Elman Neural Network has powerful calculation and strong robustness, which can solve the time-varying characteristic and accuracy fluctuation in photovoltaic generation model and load model. During the simulation process, with the condition of enough data, the error between true value and prediction value on daily photovoltaic power generation can be [-5%, 5%]. And even under the lack of data, the error between true value and prediction value on photovoltaic power generation and load consumption of every hour in one day can be no more than 20%, which means those prediction result can be taken full advantage in the energy management system. Thus, simulation results and real prototype in Nansha illustrate that the model has high accuracy for the predicted day in short term, and testify the model's ability as a feasible and convenient method to forecast the photovoltaic power generation and load consumption.