This dissertation studies autocovariance change point problems in heavy-tailed and high-dimensional time series under both offline and online scenarios. In the first project, we consider the offline multiple autocovariance change point detection problems in high-dimensional and heavy-tailed time series. First, we introduce an element-wise truncated autocovariance estimator for high dimensional and nonstationary time series. Next, we introduce a moving sum statistic and a binary segmentation segmentation algorithm to detect the number and locations of change points. Detection consistency is guaranteed under mild moments, dependence and signal-to-noise ratio conditions. Simulation study demonstrates superior performance of the proposed approach. The second project of this dissertation involves autocovariance change point problems in a online manner. Besides the element-wise truncated estimator, we introduce a spectrum-wise truncated estimator of which the nonasymptotic property is provided. Next, we construct CUSUM-type statistics with the two estimators to run through the data sequence as new observation arrives concurrently and detect the change point as soon as it occurs. We show that, under mild moments, dependence and signal-to-noise ratio conditions, with appropriate threshold of certain order, false alarm rate of the scheme can be controlled, and detection delay is upper bounded with high probability. We introduce a more efficient algorithm with linear computational cost, which preserve the same theoretical guarantees as before. In addition, we study the delay and provide a minimax lower bound from independent Gaussian setting. The proposed online approach is evaluated by two experiments in the end. The last project is concerned about application of proposed change point detection approach to pandemic time series data. We conduct retrospective analysis with offline method and identify several change points which might relate to several important events over the course of Covid-19 pandemic. Besides that, we apply the online method in the monitoring case and provide a hybrid model with the assistance of change points to improve the forecasting. Results shows the proposed model moderately improves the accuracy and dramatically boosts the efficiency.