Abstract
weather is the most important factor affecting the photovoltaic power generation.In this paper, the irradiance data of a photovoltaic power station in crodora in 2020 are collected, and the daily out of ground irradiance and the measured irradiance curve of that day are compared and observed, then the weather of that year is classified by human work, and then the daily irradiance data records are counted for the relevant indicators, with the maximum third order Based on the attributes of difference value, discrete difference and normalized variance, it is unified with the classified weather type.Then, the SVM prediction model of weather category is established based on radial basis function, and the optimal model parameters are determined by cross validation, so that a large number of historical date weather categories can be classified and predicted.This is obviously different from the traditional prediction method based on linear statistical theory, and the results show that it has a good effect.
Highlights
The above methods are faced with some problems,such as the network parameters are difficult to be determined reasonably,easy to fall into local minimum,the network learning process is prone to concussion,and the generalization ability is not strong.Based on the support vector machine, it has a strong classification function and good generalization performance.It can obtain a strong generalization ability when there are few training samples,which depends on learning a large number of training samples.The neural network which can obtain generalization ability is incomparable,and is suitable for solving high-dimensional and nonlinear problems.,it has a very wide application prospect in classification prediction
In this paper,we use the method based on historical data and support vector machine to build the weather classification prediction model.First,the classifier is constructed by analyzing or "learning" from the training set.The training set consists of database tuples aCorresponding author: 523405800@qq.com © The Authors, published by EDP Sciences
Among the 365 samples in the selected database,there are 178 in the first category,108 in the second category and 79 in the third category,with category labels of 1, 2 and 3 respectively.In this paper,70% of the weather samples of each category are selected as training samples,and the remaining 30% of the corresponding data are used as test sets to test the accuracy of classification.Figure 7 is the classification result chart of the final forecast.It can be seen from Figure 7 that the accuracy of Support vector machine (SVM) model with 70% training samples is 95.4128%(104/109).There are 5 sample classification errors,3 belong to the second category,and the results are classified into the first category.The other two samples belong to the third category,and the results are divided into the second category.But the accuracy of the whole model is better,and it achieves the expected effect
Summary
In today's society,energy shortage and environmental pollution are increasingly prominent.People are actively looking for renewable clean energy as a substitute for fossil energy.Compared with the traditional energy,solar energy has been widely concerned because of its rich reserves,no transportation,clean and pollution-free. The output power of photovoltaic power station mainly depends on weather factors such as irradiance[12].For the photovoltaic power prediction[3],the historical weather can be divided into several types first,so as to establish the corresponding historical model,and the forecast weather types can be mapped to the abovementioned different divisions,and the relevant factors in the latest period of the same weather type can be extracted from the related database to achieve the power prediction of the photovoltaic power station.Such prediction accuracy ratio is not The prediction accuracy of classification is much higher.,the weather classification is the premise of power prediction.In essence,weather classification is a problem of pattern recognition,which includes two basic aspects: feature selection and classification.At present,the main methods of classification and prediction are as follows[4]:1.Bayesian classification;.Back propagation neural network learning algorithm;.K-nearest classification. (represented by n-dimensional attribute vectors) and their corresponding class numbers.Assuming that each tuple belongs to a predefined class,the learning model can be provided in the form of classification rules,decision trees or mathematical formulas.Use the model to classify the future or unknown objects.First,evaluate the prediction accuracy of the model:for each test sample,compare the known class number with the learning model class prediction of the sample.The accuracy of the model on a given test set is the percentage of the correctly classified test samples.The test set should be independent of the training sample set,or "over fitting" will occur Condition
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
