Research On Prediction Model Of Airport Noise Based On I

With the sustainable development of china’s civil aviation, air transportation has brought us convenience and prosperity, however, a series of environmental problems arise, especially the airport noise pollution. In order to better arrange airport schedule and design airport layout, and prevent noise pollution more effectively, we should find out the dynamic variation rule of airport noise and build models to predict the new noise. Most of the existing airport noise prediction methods lack ability of learning and generalization, so the model cannot be revised and optimized with respect to the airport noise real-time data. Therefore, we aim to solve this problem and study airport noise prediction model based on incremental learning. The main research contents are as follows:First, this paper summarizes and analyzes the traditional methods of reducing redundant data in the incremental process, and proposes a new method based on AGM theory according to belief revision. It could reduce the storage cost of massive high-dimensional noise data, and continuously improve the quality of the training set, so that the prediction accuracy and robustness of the model are enhanced.Second, an online incremental model based on SVR is proposed for real-time analysis and processing of airport noise data. It can not only reduce the cost of data storage, but also adjust the current model in accordance with new data in a real-time manner to ensure the prediction accuracy of the model. Whether the current prediction model should be adjusted or not depends on the characteristics of the new samples. The sample similarity measure is introduced to ensure the quality of the training samples, and to realize the deletion of history samples with the help of the sample labeling and the principle of error-driven.Third, a batch incremental model based on SVR is proposed to deal with cumulative real-time data for the lack of analysis and processing capabilities or system fault. Different from the traditional model-dependent incremental learning algorithm, in our model, similar sample sets are extracted from the historical samples and appropriate belief sets are selected to establish the prediction model according to the characteristics of belief sets and new incremental data set. Then, the similar sample sets are revised and the current epistemic state is adjusted to adapt the new environment. Meanwhile, the theory of similar situations emphasizes the importance of the surrounding samples in the proposed prediction model.