Research On Solar Power Generation Prediction Based On Deep Learning

Solar energy is a kind of renewable clean energy,which has many advantages,such as rich resources,small environmental pollution and so on.It has gradually become the focus of the world’s attention.Using photovoltaic motor to convert solar energy into electric energy is the main form of solar energy utilization.However,photovoltaic power generation has the characteristics of randomness,intermittence and volatility.With the increasing scale of photovoltaic power generation,it is more and more important to predict the power of solar power generation in practical application.At present,some physical methods and statistical learning methods have the disadvantages of high modeling cost,large data demand and so on,while some traditional machine learning methods have the problem of low prediction accuracy.For the above problems,in this paper,two power prediction models based on deep learning are proposed.One is the Multi Meteorological Factors Convolution Neural Network,which is recorded as MMF-CNN prediction model,the other is the Long Short-Term Memory Network with Self Attention,which is recorded as SA-LSTM prediction model.Using these two prediction models effectively improves the prediction effect.The main work of this paper is as follows:(1)The influence of different meteorological factors on the power of solar power generation is analyzed in detail,including solar radiation,temperature,relative humidity and wind speed.The multiple meteorological factors and the historical output power of photovoltaic motor are embedded into Convolutional Neural Network as input features,and MMF-CNN prediction model is constructed.Finally,The prediction effect is improved.(2)The attention mechanism is applied to the field of solar power generation power prediction,combined with the Long-Short Term Memory Network,to give full play to the advantages of the network in dealing with time series problems.The self attention module is added to the existing two Long Short-Term Memory Networks,and the SALSTM prediction model is constructed,which improves the prediction effect and greatly improves the training speed of the model.(3)Two photovoltaic motors from two different regions in the data set provided by the Desert Knowledge Australia Solar Centre were used for experimental verification,and the experimental results were compared and analyzed using a variety of evaluation indicators to verify the effectiveness of the proposed prediction method.