Time Series Modeling Based On Filtering And One-way Attention Model

Time series are always produced in human activities and nature.The evolution law of complex system can be found by analyzing time series.Accurate prediction of output value is a reliable basis for decision-making in practical work.In the coding stage of time series modeling,the use of filtering algorithm can play the role of noise reduction,but also help people understand the internal information of time series;In the decoding stage of time series modeling,the attention mechanism can make the prediction model focus on the information of those important time periods,and the attention mechanism has achieved great success in solving the problem of long-distance dependence.In practical engineering application,the selection of filtering algorithm is difficult.Specifically,when faced with different types of time series data,it needs to spend a lot of energy to test the appropriate filtering method in this scenario,which makes the migration ability and versatility of the model based on a single filtering algorithm poor.To solve this problem,this paper designs a hybrid filtering method,which uses a variety of filtering algorithms to process the original time series data at the same time,and stitches the obtained signal components and the reconstructed signal into a matrix.Compared with a single filtering algorithm,the hybrid filtering method can provide more observation angles for the analysis of the original data.At the same time,aiming at the problem of noise interference in input vector,a feature optimization network based on gating technology is designed,which can play a role in screening noise and effective information.Aiming at the difficulty of decoding temporal relationship in time series data,a decoding module based on attention mechanism is designed,which mines the independent features of different feature subspaces by multi attention,enhances the ability of the model to process long time series data,avoids the local information uncorrelation problem of CNN and the problem of non-parallel operation in RNN layer,and makes the whole model easy to train.In this paper,a wind power prediction model based on MSE-OFSN-AM is proposed.Experiments on open wind power data sets show that the above modeling method is helpful to improve the prediction accuracy.