ABSTRACT

Accurate prediction of dissolved oxygen (DO) concentration is important for managing healthy aquatic ecosystems. This study investigates the comparative potential of the emotional artificial neural network-genetic algorithm (EANN-GA), and three ensemble techniques, i.e. emotional artificial neural network (EANN), feedforward neural network (FFNN), and neural network ensemble (NNE), to predict DO concentration in the Kinta River basin of Malaysia. The performance of EANN-GA, EANN, FFNN, and NNE models in predicting DO was evaluated using statistical metrics and visual interpretation. Appraisal of the results revealed a promising performance of the NNE-M3 model (Nash-Sutcliffe efficiency (NSE) = 0.8743/0.8630, correlation coefficient (CC) = 0.9351/0.9113, mean square error (MSE) = 0.5757/0.6833 mg/L, root mean square error (RMSE) = 0.7588/0.8266 mg/L, and mean absolute percentage error (MAPE) = 20.6581/14.1675) during the calibration/validation period compared to EANN-GA, EANN, and FFNN models in DO prediction in the study basin.

摘要

准确预测溶解氧 (DO) 浓度对于管理健康的水生生态系统很重要。本研究调查了情感人工神经网络-遗传算法 (EANN-GA) 和三种集成技术的比较潜力，即情感人工神经网络 (EANN)、前馈神经网络 (FFNN) 和神经网络集成 (NNE)，预测马来西亚近打河流域的溶解氧浓度。使用统计指标和视觉解释评估 EANN-GA、EANN、FFNN 和 NNE 模型在预测 DO 方面的性能。对结果的评估表明 NNE-M3 模型具有良好的性能（纳什-萨特克利夫效率 (NSE) = 0.8743/0.8630，相关系数 (CC) = 0.9351/0.9113，均方误差 (MSE) = 0.5757/0.6​​833 mg/L） , 均方根误差 (RMSE) = 0.7588/0。