Application of neural networks and ANFIS in estimating groundwater level of paddy field

• Main Authors: Cia-Syun Huang, 黃洽訓
• Other Authors: 洪念民
• Format: Others
• Language: zh-TW
• Online Access: http://ndltd.ncl.edu.tw/handle/79319657564730416229

碩士 === 僑光科技大學 === 資訊科技研究所 === 98 === Groundwater is the main important environmental resources in Taiwan, but a lot of factors affect the water table, including: weather conditions, groundwater recharge and groundwater pumping volume. The most significant influence on long-term changes in groundwater levels is the impact of human extraction of groundwater. For the effective management of groundwater, pumping volume can be pumped through the model for changes in volume and may be employed to assess the groundwater level, but some areas have more parameters. Due to the lack of cultural conditions, the construction of groundwater model is more difficult and requires considerable information. Tainan County is the model used in this study and will be used to judge groundwater as a viable source of irrigation. Furthermore the past characteristics of groundwater usage and reduced agricultural water demand are factored in to ascertain the relationship between the two in order to establish the regional response to prolonged groundwater usage. In this study Artificial Neural Network (ANN) and Adaptive Network-Based Fuzzy Inference System (ANFIS) are used to establish prediction models in the future. These methods are used in this study to predict the groundwater level in Tainan for the initial ten days by considering depth, accumulated water demand, rainfall and groundwater levels. These estimates from the cumulative output are used to calculate and reduce discharge volume, using data from 2000 to 2002 as a training model, and from 2003 to 2005 as a test model. Results show that feed-forward neural network prediction of one single well depth results in predicted values and feed-forward neural network in more depth in the single well and multi-well multi-depth results in inaccurate data. Using the ANFIS predictions and more than one single well depth resulted in the best outcome and thus can predict the trend of groundwater level fluctuations, with the actual water level divergence being quite small and also even less variance in predicting depth. ANFIS and ANN systems combine efficiently and effectively and both have the advantages that can make up for their own deficiencies. However with the neural network comparison, the neural network prediction results are not satisfactory, and the error fluctuated considerably, and forecast results unstable and less reliable. To predict accurate forecasts of groundwater levels the use of ANFIS was found to be the best choice in making predictions.