基于数据驱动的机器学习方法是预测叶绿素a的一种重要的非机理方法，但是现有以神经网络模型为代表的叶绿素a预测方法较少考虑藻类生消对环境因子的时滞效应。以三峡库区叶绿素a质量浓度较高的典型支流香溪河为研究区域，采用相关性分析、主成分分析和灰色关联分析3种方法综合确定叶绿素a的主要贡献因子，利用交叉相关分析及Almon分布时滞模型筛选主要贡献因子中的时滞因子，并确定其最优滞后时间，在此基础上构建Almon-BP时滞神经网络模型进行叶绿素a趋势预测。结果表明：香溪河峡口主要时滞因子为气温、风速、太阳辐照、pH、溶解氧，最优滞后时间分别为4、2、3、7、3 d；香溪河平邑口主要时滞因子为水温、气温、风速、降雨量、太阳辐照、浊度、pH、溶解氧、三峡水位差、氧化还原电位、三峡水位，最优滞后时间分别为2、2、2、4、2、10、3、2、6、10、6 d；相比于只考虑叶绿素a主要贡献因子的BP神经网络模型（贡献因子-BP），Almon-BP时滞神经网络模型对香溪河峡口预测结果的均方误差（EMS）、均方根误差（ERMS）、平均绝对误差（EMA）、平均偏差（EMB）等误差指标分别降低44.4%、25.6%、31.3%、53.9%，对香溪河平邑口预测结果的EMS、ERMS、EMA、EMB等误差指标分别降低66.7%、42.1%、37.5%、45.8%。研究模型对叶绿素a预测具有较高的准确度，对叶绿素a早期预警具有重要意义。

Since the impoundment of the Three Gorges Reservoir, the nutrient enrichment status of tributary water bodies in the reservoir area significantly deteriorated, transitioning from riverine to lacustrine ecosystems. The reservoir impoundment resulted in reduced flow velocities of tributary water bodies, leading to increased water transparency and nutrient concentrations conducive to the proliferation of increasing in chlorophyll-a mass concentration. Consequently, typical tributaries such as the Xiangxi River in the Three Gorges Reservoir area experienced frequent outbreaks of increasing in chlorophyll-a mass concentration. These recurrent increase in chlorophyll-a mass concentration events not only degraded water quality and aquatic ecosystems but also posed constraints on the sustainable development of society.Correlation analysis, principal component analysis, and grey relational analysis were utilized to identify and validate significant contributors to increase in chlorophyll-a mass concentration. Subsequently, cross-correlation analysis and the Almon distributed lag model were employed to ascertain factors among the major contributors exhibiting time lags and to determine the optimal lag time. Building upon this analysis, an Almon-BP neural network model was developed to forecast the trends of chlorophyll-a mass concentration.The major contributing factors to increase in chlorophyll-a mass concentration at the Xiangxi River Xiakou included dissolved oxygen, pH, air temperature, solar radiation, wind speed, wind direction, turbidity, rainfall, Three Gorges water level difference, and water temperature. Similarly, significant factors at the Pingyikou of the Xiangxi River included dissolved oxygen, pH, air temperature, solar radiation, wind speed, wind direction, turbidity, rainfall, Three Gorges water level difference, water temperature, redox potential, and Three Gorges water level. Among environmental factors at the Xiakou of the Xiangxi River, air temperature, wind speed, solar radiation, pH, and dissolved oxygen exhibited lag effects on increased in chlorophyll-a mass concentration, with optimal lag times ranging from 2 to 7 days, while other environmental factors did not display time lags. Conversely, at the Pingyikou of the Xiangxi River, factors such as water temperature, air temperature, wind speed, rainfall, solar radiation, turbidity, pH, dissolved oxygen, Three Gorges water level difference, redox potential, and Three Gorges water level exhibited lageffects on chlorophyll-a mass concentration, with optimal lag times ranging from 2 to 10 days. Wind direction did not show lag effects.Comparative analysis of three prediction models the BP neural network model considering all environmental factors, the BP neural network model considering only major contributing factors, and the Almon-BP neural network model considering the optimal lag time of major contributors revealed that the Almon-BP neural network model outperformed the corresponding BP models in predicting chlorophyll-a mass concentration in the Xiangxi River, with lower prediction errors. This underscored the efficacy of the Almon-BP neural network model in enhancing the accuracy of chlorophyll-a mass concentration prediction, which was crucial for early warning and mitigating harmful algal bloom occurrences.