Five series of physico-chemical data(1986—1996), including annual precipitation (p), average water temperature from July to August (T), the ratio of inflow and storage in July and August, the ratio of outflow and storage in July and August and phosphorus (PO4) from Dahuofang Reservoir was used to developed for predicting timing and magnitudes for phytoplankton and four series (1980—1989) of the ratio of inflow and outflow in July and August, average water temperature from July to August (T), phosphorus (PO4) and total nitrogen (TN) from Dahuofang Reservoir was trained to developed for predicting timing for Cyanophyta dominant by artificial neural network model, respectively. These models were successful in estimating the output in two years (model 1 in 1997—1998 and model 2 in 1990—1991), with the average relative errors of 2% and 1% for calculated and observed data, respectively. The study indicates the potential of artificial neural network as predictive tool for highly non-linear phenomena, such as phytoplankton dynamics in reservoirs, better than classical statistical models.%根据辽宁大伙房水库1980—1997年的水文和湖沼学观测资料，分别建立浮游植物丰度和蓝藻优势度人工神经网络模型。将年降雨量、7—9月平均水温、7—8月入库水量与7—8月库容之比和磷酸盐作为输入，浮游植物生物量和丰度作为输出，建立浮游植物群落消长的人工神经网络模型；将7—9月平均水温、7—8月入出库水量之比、磷酸盐和总氮作为输入，蓝藻优势度作为输出，建立浮游植物演替的人工神经网络预测模型，并进行检验，其模拟值与观测值平均相对误差分别为2%和1%。结果表明，人工神经网络方法优于传统的统计学模型，可进行水库浮游植物群落动态的预测预报，并具有较高的精度。
展开▼
