<p>Time series prediction is widely used in fields, such as power, weather, and transportation. Currently, deep learning-based methods have achieved significant success in time series forecasting. However, most of these methods model non-stationary time series data directly, rendering the extraction of crucial complex patterns within them a challenging task. To address this challenge, a Time Series Long-Short-Term Trend Analysis Model (TLSAM) is proposed in this paper. This model combines convolutional neural network (CNN) and multi-layer perceptron (MLP) to capture the essential features of the time series data. TLSAM consists of four main components: a time series periodicity decoupler (TPD), a time series long-term trend cross-scale analyzer (TLTCA), a time series short-term trend analyzer (TSTA), and a long-short trend integrator (LSTI). Initially, TPD detects periodicity in time series data, decomposing non-stationary series into relatively stationary sub-sequences. Next, TLTCA extracts the long-term trend of each sub-sequence, revealing the long-term variation patterns within the time series data. Then, TSTA captures the short-term trend features of the overall time series data, uncovering inherent short-term variation patterns. Finally, LSTI fuses long-short-term features for comprehensive, accurate predictions. Experimental evaluations on seven real-world datasets show that TLSAM surpasses other methods in both prediction accuracy and computational efficiency.</p>