Short-term Prediction Of Wind Power And Photovoltaic Power Generation

With the rapid development of smart grid, the distributed energy integration which using for the operation of micro-grid and traditional power grid has become a hot research topic. As renewable resources, wind power and solar power have entered a rapid progress stage and attract more and more attention due to inexhaustible and environment-friendly. But wind power and solar power have the disadvantages of unstable and intermittent, which will bring serious challenge for the integration operation of micro-grid and traditional power grid, and also pose a major threat for the stabilization and quality of power system. And then restrict the scale of wind power and photovoltaic power development. The key to solving above problems is to accurately predict the short-term power generation. In this context, this paper put the short-term power prediction of wind power and photovoltaic power as research content.Firstly, a summary of prediction methods of wind power and photovoltaic power were described, the advantages and disadvantages of these methods were also analyzed. Secondly, this paper constructed two short-term power prediction models for wind power and photovoltaic power respectively, the two models were based on empirical mode decomposition and artificial neural network. For wind power prediction model, the raw power signal was decomposed into several components with different characteristics by empirical mode decomposition for reducing the influence of non-stationary and non-linear. Then building forecast models for each component based on radial basis function neural network and aggregating all the predictive values to obtain the ultimate predictive result. For photovoltaic power forecasting model, also using empirical mode decomposition to decompose raw power signal into several eigenmode components and a residue, then constructed prediction models for these components based on adaptive genetic algorithm and error back propagation neural network and aggregated all the predictive values to obtain the predictive value of photovoltaic power. For the shortcomings of traditional training algorithms of BP network, this paper purposed a new training method based on adaptive genetic algorithm and error back propagation algorithm. The study of this paper was based on the real data of two power farms from the institute of Zhejiang Electric Power Test&Research, and the results of simulation show the proposed prediction models were effective short-term prediction methods for both wind power and photovoltaic power, the conclusions have actual meanings.