Prediction model of shrimp feeding amount based on self-attention mechanism and CNN-LSTM deep learning
HE Jinmin, ZHANG Lizhen
1.College of Engineering Science and Technology, Shanghai Ocean University, Shanghai 201306, China; 2.Shanghai Engineering Research Center of Marine Renewable Energy, Shanghai 201306, China
摘要 为提高对虾饲料的利用率，减少养殖成本，提高养殖效益，提出了一种基于自注意力机制(self-attention，ATTN)和卷积神经网络(convolutional neural network，CNN)-长短期记忆网络(long short term memory，LSTM)的对虾投饵量预测模型(CNN-LSTM-ATTN)，以水温、溶解氧、对虾的数量与质量作为预测模型的输入数据，通过CNN挖掘输入数据间的内在联系，提取出数据特征信息，利用LSTM的长期记忆能力保存数据特征信息，使用ATTN突出不同时间节点数据特征的重要性，进一步提升模型的性能。结果表明，本研究中提出的CNN-LSTM-ATTN预测模型的均方根误差、平均绝对误差和平均绝对百分误差分别为0.816、0.681和0.018，均小于BP(back propagation)神经网络、LSTM和CNN-LSTM 3个基准模型，其模型预测能力和稳定性优于其他模型。研究表明，本研究中构建的模型能较好地实现对虾投饵量的准确预测，可为对虾养殖投饵量的管理调控提供参考依据。
Abstract: In order to reduce production cost, and improve the utilization rate of feed, and efficiency of breeding, a method—CNN-LSTM-ATTN used to predict shrimp feeding amount was proposed based on a self-attention mechanism(ATTN)and convolutional neural network(CNN)-long short term memory(LSTM), with water temperature, dissolved oxygen, quantity and quality of shrimp as input.Firstly, the CNN module was applied to mine the potential connections between variables and extract data features, and then the long-term memory ability of LSTM was used to save the data feature information, and ATTN was used to highlight the importance of data features at different time nodes, and enhance the performance in prediction.The results showed that the CNN-LSTM-ATTN prediction model proposed here had mean square error of 0.816, mean absolute error of 0.681, and mean absolute percentage error of 0.018, far smaller than those of BP neural network(BP), LSTM and CNN-LSTM.Its prediction ability and stability had been greatly improved and proved.The findings indicate that this prediction model can well realize the prediction of feeding amount for shrimp and can provide a reference with the management and control of shrimp feeding amount.
何津民, 张丽珍. 基于自注意力机制和CNN-LSTM深度学习的对虾投饵量预测模型[J]. 大连海洋大学学报, 2022, 37(2): 304-311.
HE Jinmin, ZHANG Lizhen. Prediction model of shrimp feeding amount based on self-attention mechanism and CNN-LSTM deep learning. Journal of Dalian Ocean University, 2022, 37(2): 304-311.
2022, Vol. 37 
