TRLLD: Load Level Detection Algorithm Based on Threshold Recognition for Load Time Series

Load time series analysis is critical for resource management and optimization decision-making, particularly automated analysis techniques. Existing research insufficiently interprets the holistic characteristics of samples, resulting in markedly divergent load level detection conclusions for samples with different characteristics (trend, seasonality, periodicity). Achieving automated, feature-adaptive, and quantifiable analysis remains a challenge. This paper proposes a Threshold Recognition-based Load Level Detection algorithm (TRLLD) that, based on sample characteristics, effectively identifies regions of different load levels in samples of arbitrary scale and distribution type. The algorithm classifies data points using distribution density uniformity, ultimately yielding normalized load values. In the feature identification step, the algorithm employs a Difference-based Uniformity Index of Density (DUID), High Load Level Concentration (HLLC), and Low Load Level Concentration (LLLC) to assess sample characteristics; these metrics are independent of specific load values, thereby providing a standardized feature perspective and ensuring high efficiency and strong interpretability. Compared with traditional approaches, the proposed method demonstrates superior adaptivity and real-time analysis capability. Experimental results show that the method can effectively identify high-load and low-load regions across 16 time series samples with diverse load characteristics and produce highly interpretable results. The correlation between DUID and sample density distribution uniformity reaches 98.08%. When noise at 10% MAD intensity is introduced, the maximum relative error is 4.72%, demonstrating high robustness. Moreover, the method exhibits notable advantages in both general-sample and low-sample scenarios.