Uncertainty Knowledge Representation And Prediction For Time Series Data

As an effective way to deal with data uncertainty,uncertainty knowledge representation model is one of research hotspots in the field of time series data prediction.However,the most of ucertainty knowledge representation models are focus on deal with the uncertainty of the data coverage and boundaries,and lack of consideration for the uncertainty caused by limited data cognition.The reasons are as follows: 1)uncertainty knowledge representation and prediction of time series data with uncertainty values need to be further strengthened;2)the prediction algorithm based on knowledge representation is still in the theoretical stage,and its application area needs to be further explored.Aiming at the above problems,the grey theory is introduced to describe the time series data with uncertainty values in this thesis.The knowledge representation and prediction based on gery thory and the possible application fields of prediction models is explored and studied.1.The optimized prediction model based on continuous grey number uncertainty knowledge representation is proposed.In order to reduce the uncertainty in the process of whitening,the thesis uses the kernel and information diffusion as the sequence development factors to avoid the whitening process.At the same time,the development factors are used to construct the multivariable prediction model directly,and the trend of data development is grasped from the view of the system,which reflects the interaction between sequence development factors.2.A prediction model based on discrete grey number with element irregularity uncertainty knowledge representation(MS-DGM)is proposed.The traditional prediction method based on discrete grey number knowledge representation are extracts kernel and grey cell area sequences,directly.However,when elements are not homogeneous,traditional methods can not extract the grey cell area sequences.In order to solve this problem,the Grubbs algorithm is used to test and mark the missing values(outliers),and the adjacent mean value is used to fill the missing values(outliers).These means widening the application field of the prediciton model.3.The application practice of the MS-DGM model.In this thesis,the uncertainty knowledge representation is used to solve the problem of data fusion in wireless senser networks(discrete grey number is an effective tool to describe the spatio-temporalcorrelation between the sensor data).Based on this,the discrete grey number are used to describe the spatiotemporal correlation of multi-sensor at first.Then the Grubbs test's method is used to detect and mark the outliers or missing data to reduce the impact of abnormal data on prediction model.After that,the adjacent mean value is used to fill the outliers and missing values,and the sequence development factors are extracted.Finally,a grey discrete single variable prediction model is constructed based on the sequence development factors,which is used to predict the data at the next time.