任妮 鲍彤 刘杨 荀广连 蒋永年

摘要： 为准确预测蟹塘溶解氧质量浓度，及时掌握溶解氧质量浓度的变化趋势，提前采取防控措施从而降低河蟹养殖风险，提出了一种基于粒子群优化算法（PSO）和长短时记忆神经网络（LSTM）的蟹塘溶解氧质量浓度预测模型，采用PSO算法优化LSTM模型参数后对蟹塘溶解氧质量浓度进行预测。结果表明，PSO-LSTM模型不仅整体优于ARIMA模型，相较于其他LSTM模型也有更高的预测精度，在连续10个时间点的预测中相比于LDO-LSTM、LSTM和ARIMA模型平均百分误差分别降低了2.55%、1.891%和4.055%。说明PSO-LSTM模型在蟹塘溶解氧质量浓度预测中具有良好的准确性和稳定性，可以为河蟹养殖中水质精准预测与调控提供参考。

关键词： 溶解氧预测;河蟹养殖;粒子群优化算法;长短时记忆神经网络

中图分类号： S126  文献标识码： A  文章编号： 1000-4440（2021）02-0426-09

Abstract： To predict the mass concentration of dissolved oxygen in Chinese mitten crab ponds accurately， grasp the changing trend of the mass concentration of dissolved oxygen timely and take preventive and control measures in advance to reduce the risk in Chinese mitten crab culturing， a model for predicting the mass concentration of dissolved oxygen in Chinese mitten crab ponds based on particle swarm optimization （PSO） and long short-term memory （LSTM） neural networks was proposed. The mass concentration of dissolved oxygen in Chinese mitten crab ponds was predicted after optimizing LSTM model parameters by PSO algorithm. The results showed that the PSO-LSTM model was not only superior to the ARIMA model， but also had higher prediction accuracy compared with other LSTM models. In the predictions at 10 consecutive time points， the average percentage error of the PSO-LSTM model reduced by 2.55%， 1.891% and 4.055% respectively， compared with the LDO-LSTM， LSTM and ARIMA models. It can be seen that the PSO-LSTM model has good accuracy and stability in the prediction of the mass concentration of dissolved oxygen in Chinese mitten crab ponds， and can provide a reference for accurate prediction and regulation of water quality in Chinese mitten crab culturing.

Key words： prediction of dissolved oxygen;culturing of Chinese mitten crab;particle swarm optimization algorithm;long short-term memory neural networks

河蟹，學名中华绒螯蟹，俗称大闸蟹。河蟹养殖是中国很多地区实施精准扶贫、拉动经济增长、促进农民增收的重要突破口。溶解氧（Dissolved oxygen，DO）即溶解于水中的分子态氧，是集约化河蟹养殖成功与否的关键因素之一，其含量多少关乎河蟹的生长速度、发病率、死亡率，以及蟹塘中饲料的利用率和有害物质的产生量等。准确预测蟹塘中溶解氧的含量，有利于及时掌握溶解氧的变化趋势，提前采取防控措施，从而降低河蟹养殖风险，增加养殖经济效益，同时还对水质监测和疾病防控等生态问题具有预警意义。

近年来，随着机器学习和深度学习等技术的发展，越来越多的研究者将此类方法应用于水体溶解氧的预测研究中。……