Solar Power Forecast Based On Big Data And Machine Learning

In the process of photovoltaic power prediction and analysis,as photovoltaic panels are affected by external factors,the output power will change,thereby threatening the security of the power grid.This paper focuses on the innovative application of photovoltaic power influencing factors and algorithms.The experiment mainly completed the following work and research:(1)The SVM algorithm is a classic algorithm selected in this experiment to predict photovoltaic output power using machine learning.The study uses GBDT algorithm and SVM algorithm to make short-term prediction of photovoltaic power output.Because the obtained power plant data contains multiple sets of factors,the GBDT algorithm is used to analyze the importance of the factors in the data,and find the factors with high importance as the training set.During the training of the model,the data was divided into six groups,and each group of data models predicted the test set.The experiments show that the GBDT-PSO-SVR integrated prediction model proposed in this study does not need to consider the weather category of the data during the training process.The obtained prediction model has a relatively accurate prediction ability and strong applicability.(2)The method uses long-term and short-term memory networks combined with CNN and attention mechanism to predict photovoltaic power generation.The CNN algorithm extracts features from the input data,and then uses the attention mechanism to assign weights to the LSTM hidden layer output vectors.Finally,the results are obtained through the fully connected layer.The study uses two years of PV measured data for training.The entire experiment process uses three steps.Before performing these three steps,a correlation analysis is performed on the combination of multiple sets of factors in the data and the impact of photovoltaic power to find out The most relevant factor data were used as experimental data.The first step of the experiment is the clustering stage.The data is divided into four groups using a self-organizing mapping algorithm to establish prediction models.The second step is the training phase.CNN and LSTM algorithms are used to extract the data features.The obtained feature vectors are combined with the attention mechanism to assign weights,and finally the training model is constructed.The third step is the prediction phase.The training set model is selected for prediction based on the month of the test set.(3)Hadoop big data processing platform and Spark calculation module to predict photovoltaic power generation was researched.The experiment uses DBN algorithm in Py Spark MLlib and the single computer DBN algorithm to compare the PV prediction results.Because there is no support vector regression model code in Py Spark MLlib,the experiment uses the SVR code of Scikit-learn library to combine with Py Spark.During the model training process,the training data is put into the SVR training model code in the form of RDD data structure.This operation can greatly reduce the training time of the model.