丰满电站下游松花江水文站河段分布有一系列鱼类产卵场,电站拟通过分层取水调控下泄水温,改善下游鱼类生存环境.该文基于大量实测数据分析,建立了松花江站水温的人工神经网络预报模型,通过输入上游吉林水文站的水温与流量,以及地区气象条件,可计算出下游松花江站2日后的水温变化.根据中长期天气预报数据与电站泄流计划,采用该模型通过2日递推的方法,可预测出下游鱼类产卵场的水温变化过程.运用2006-2013年实测数据对网络模型进行训练,然后对2014年松花江站水温变化过程进行计算,计算值与实测值的变化过程甚为吻合,相关系数为0.992,水温平均误差为0.51 ℃.在水温生态调度运行期间,根据产卵场水温变化的预报数据,可适当调控电站下泄水温,保持适宜的鱼类产卵条件.%In this study, the water temperature regulation were carried out through the selective intake facilities in Fengman Hydropower station to improve the downstream living environment. The power plant released flow first reaches Jilin hydrologic station at 20 km downstream, and then through the 160 km long reach, arrive at the Songhuajiang hydrologic station, where there are a series of spawning sites of black carp, grass carp, silver carp, etc. The field data analysis showed that, there was a strong correlation between the water temperature of the power plant and Jilin Station, so the empirical relationship has been established based on the statistical analysis of the measured data in earlier research. However, there was obvious difference and poor correlation between the water temperature of Jilin Station and Songhuajiang station. The main reason was that the heat exchange between the channel water and the surrounding environment led to a significant change in water temperature. Firstly, by analyzing the correlation coefficients between all the hydrological and meteorological factors with the water temperature of Songhuajiang station, the six external influence factors were identified, including the flow and water temperature of Jilin Station, and the air temperature, relative humidity, wind speed and sunshine duration of Changchun meteorological station. Then, based on the field data, the water temperature prediction model of Songhuajiang station was established by using a RBF (radial basis function) neural network, which can automatically select the sample vectors with maximum error as a new neuron until to finally reach the required precision. It took about 2 days to flow from Jilin to Songhuajiang station, so the model predictors had temporal and spatial attributes. The flow and water temperature of Jilin station should be the values of the first day, the climate conditions of Changchun station were of the next day, and the water temperature of Songhuajiang station was of the third day. Therefore the neural network model actually reflected a heat exchange process within two days. According to the medium or long term weather forecast data and power station discharge plan, the neural network model can be used to predict the time course of the water temperature at the spawning sites by using the above two day recursive method. The model was trained by the field data in 2006 - 2013, and to predict the temperature time course in 2014, the time variation of the calculated and measured water temperatures were in good agreement, the average deviation was 0.51 ℃, and the correlation coefficient was 0.992. In May 2010 to August, for example, the average temperature increased from Jilin to Songhuajiang station was 4.6 ℃. When the released water temperature upstream rose 3.2 ℃ by regulation, because of the decrease of the heat exchange between the channel water and the surrounding environment, the temperature increased between the two stations dropped 3.3 ℃. It was proved that this model can better reflect the influence of heat exchange along the river on the water temperature of downstream spawning field. During the water temperature regulation, the water temperature at spawning sites will be predicted, and the releasing discharge of power plant is adjusted properly, to provide suitable spawning conditions.
展开▼
