Time Series Data Anomaly Detection And Prediction

With the development of artificial neural networks,a neural network structure that can process time series,namely recurrent neural networks(RNNs),has become popular,which brings new ideas for processing time series data.Based on the existing research results,this thesis focuses on time series data anomaly detection and prediction.An anomaly detection mechanism with low resource consumption and a financial time series prediction model are designed.At the same time,the accuracy and stability of the anomaly detection model are optimized.The main works of the thesis include:1.Firstly,the research background of this project is introduced,including the origin of the topic and the main research contents,and the innovation points of this thesis are given.Then the basic knowledge and basic working principle of the RNN are described.Finally,the problems existing in time series data anomaly detection and prediction are summarized,and the corresponding solutions are proposed.2.Mass spectral substance detection methods are proposed,which employ long short-term memory(LSTM)recurrent neural networks to classify the time-series mass spectrometry data and can accurately detect the substances class.As the LSTM has the excellent understanding ability for the historical information and classification capability for the time series data,a high detection rate is obtained.In addition,by combining the difference operation and the ReliefF algorithm,the network size is reduced by 28% and the training speed is increased by 35%.An LSTM-based substance detection system can achieve the tradeoff between high detection rate and low computational resource consumption,which is beneficial to the devices with constrained computing resources,such as low-cost embedded hardware systems.3.By inheriting the excellent memory ability of the RNN to time series,the gated recurrent unit(GRU)neural network overcomes the long-term dependence problem.The GRU extension is applied to financial time series prediction.A financial time series prediction model based on the differential operation and GRU neural network is proposed.The experimental results show that the differencing operation can improve the generalization ability and the prediction accuracy of the GRU neural network,and the proposed method can perform a better prediction for the financial time series than the conventional approach,with a relatively low computing overhead.4.Ensemble learning is used to optimize the performance of the designed low resource consumption anomaly detection model.At the same time,the differential evolution algorithm is used to dynamically assign the weight of the ensemble learning base classifier.The experimental results show that ensemble learning combined with differential evolution algorithm can effectively improve the accuracy and stability of the low resource consumption anomaly detection model.