The Research And Application Of Grid Connected PV Power Short-term Forecasting

Along with the development of industrialization, human being in the economic development and scientific and technological progress, is also facing the energy crisis and environmental pollution and other severe tests. Because of its inexhaustible, solar energy and no pollution characteristics is undoubtedly one of the ideal green energy. At present, photovolatic power generation has become the main way of solar energy utilization. Photovolatic power generation is divided into two kinds, which are grid connected and the network. The grid connected system is the developing trend of photovoltaic power generation in the world. However, the output power of photovoltaic generation system has the characteristics of intermittent and volatility due to the direct influence of the solar radiation intensity and meteorological factors. In order to provide reliable data information for power grid dispatching personnel and ensure the safe and statble operation of power network, the short-term power of photovoltaic power generation sysytem must be accurately predicted.At present, the prediction method of short-term power of photovolatic generation sysytem mainly has two kinds of indirect predictiong method and direct prediction method, which typical direct prediction method is based on support vector machine regression prediction model, based on Markov chain prediction model, based on BP neural network prediction model. The forecasting model based on BP neural network has the advantages of strong learning ability, organizationg, fault tolerance and so on. It has become a common method for the prediction of the output power of photovoltaic power generation system. The traditional BP neural network is easy to fall into local minimum and slow convergence. In this paper, we use LM algorithm to improve the BP neural network and set up the forecast model of the output power of photovoltaic power generation system. At the same time, in the process of selecting the historical power data, the predictiong model is obtained by using the method of similar days, and historical data is selected as the sample, which greatley reduces the prediction error and the convergence of the model.In addition, in order to meet the requirements of power grid security and stability, the PV short-term power forecasting is not only to give the forecast value, but also to make a reasonable assessment of the risks contained in the forecast. For this demand, a confidence interval estimation method based on the error distribution characteristic is proposed, and the Bootstrap method is used to construct the short-term photovoltaic power forecast interval. The test results based on real data show that the power interval prediction based on Bootstrap method can effectively characterize the variation of short-term photovoltaic power generation.At the last, combing the research results of this paper, we design and implement a power forecasting sysytem for photovoltaic power generation. Mainly includes four modules:system management, data management, LM(Levenberg-Marquart) based on improved BP neural network power forecasting and Bootstrap based method of power interval forecast.