Daily Solar Direct Radiation Exposure Online Prediction Based On The Process Wavelet Neural Network

Daily direct solar exposure prediction is very important for the output situationanalysis of grid connected photovoltaic system. It is widely researched by domesticand foreign scholars. The nonlinear, large scale intermittent non-stationary, processtime properties and chaotic characteristics make it difficult to establish its accurateprediction model. Wavelet neural network and Fourier transform were combined tobuild the prediction model of daily direct solar exposure. This method provides a newway to the construction of neural network and improves the prediction accuracy ofdaily direct solar exposure.The main works in this paper are as follows:1. Analyzed the non-linear, non-stationary, Process time properties and Chaoscharacteristics of daily direct solar exposure sequence. The experiments alsodemonstrated the energy sequence of daily solar radiation and meteorologicalenvironment also have a very close relationship.2. Fourier transform and wavelet neural network were used to build thepredictive model of daily direct solar exposure. The reconstructed daily direct solarexposure data was used to train the model and made prediction of daily direct solarexposure. This method get ride of the shackles of trail and error method and provide anew way of building neural network.3. Phase space reconstruction embedding dimension of daily direct solarexposure was used to determine the input layer neurons. Wavelet analysis theory wasused to determine the hidden layer neurons. The number of Fourier expansion of theinput data is determined by the results of experiment.4.2006-2010daily direct solar exposure data that form U.S. NationalAeronautics and Space Administration Hefei，Sanya and Harebing were used in thispaper. Matlab R2009b simulation software was selected to make verificationexperiment of daily direct solar exposure wavelet neural network prediction model.Results show that the method is feasible and effective.