Application of an Improved GA-BPNN Algorithm for Wireless Sensor Network in Hydrological Forecasting

Hydrology is a variety of changes and motion laws of water in the nature. It is a key fundamental work for each country to understand the national geographic hydrology. It is particularly important to understand the hydrological data in the water conservancy project planning, management, drought prevention and flood control, as well as the protection and utilization of water resources. Due to the complexity of the influencing factors and the limit of the current level of science, hydrological forecasting accuracy is relatively low. How to improve the prediction accuracy attracts the concern of the hydrology scientists. Firstly, aiming at the difficulty of monitoring to the seawater hydrology, an online monitoring scheme composed of wireless sensor network and computer technology are proposed in this paper. Secondly, the neural network method is used to process the data collected by wireless sensors in order to forecast the related hydrological data. Thirdly, a hybrid algorithm combined with genetic algorithm and BP neural network is developed to improve the performance due to the defect of BP algorithm easily falling into local extremum in the process of training. In the section of simulation experiment, a designated wireless sensor networks for New York Harbor have been set up. This network is composed of a number of fixed wireless sensor network nodes in the upper reaches and the estuary of the river. This network combines the prediction model in this paper with the fixed-point data collection and data fitting to achieve the purpose of using wind velocity to predict hydrological information, which includes water level, water temperature, salinity, wave height and wave period. As can be seen from the experimental results, the improved BPNN algorithm is significantly better than the other two algorithms. It shows that the optimization of continuous space based on GA algorithm is very important for the number of hidden layers which affect the prediction accuracy of BPNN. It avoids the defects of the experience when the parameters are chosen, and the prediction accuracy is obviously improved.