Abstract: The dynamic characteristics of carbon emissions from coal-fired power units in Zhejiang Power Grid, particularly under varying load and seasonal conditions, remain unclear, posing challenges for low-carbon dispatch. This study develops a real-time emission monitoring method and investigates the minute-resolution dynamic characteristics of units across different technological generations and capacity grades. Based on energy and mass balance principles, a real-time carbon emission monitoring model is established that circumvents the fuel sampling lag by back-calculating the as-fired coal heating value from the thermal balance. Seven units, covering subcritical, supercritical, and ultra-supercritical parameter grades with capacities ranging from 300 MW to 1000 MW, are monitored on four typical days representing spring, summer, autumn, and winter of 2025. The results show a significant negative correlation between power-generation carbon emission intensity and load rate. During minute-resolution load-ramping processes, the carbon emission intensity lags behind load changes by approximately 2–4 minutes. High-parameter units exhibit a clear carbon efficiency advantage at medium-to-high loads, whereas low-parameter units show a larger increase in emission intensity at low loads. A clear seasonal pattern is observed, with higher emission intensities in summer and winter and lower values in spring and autumn; however, the carbon efficiency advantage of high-parameter units is weakened in summer and winter due to frequent and large-amplitude load fluctuations. Therefore, when implementing low-carbon dispatch or deep peak shaving, grid operators should consider the synergistic effects of unit technical grade, load interval, and seasonal operating characteristics to optimize unit commitment and achieve overall carbon emission minimization.