文章摘要
李祥铜 1,2,曹 亮 1,2,李湘丽 2,3,刘双印 1,2,4,徐龙琴 1,2,呼 增 1,2,黄运茂 2,尹 航 1,2.基于 WTD-LSTM 的对虾养殖水温组合预测模型[J].广东农业科学,2021,48(2):153-160
查看全文    HTML 基于 WTD-LSTM 的对虾养殖水温组合预测模型
Prediction Model of Water Temperature Combination for Prawn Cluture Based on WTD-LSTM
  
DOI:10.16768/j.issn.1004-874X.2021.02.020
中文关键词: 对虾  水温  预测  小波阈值降噪  长短时记忆神经网络
英文关键词: prawn  water temperature  prediction  wavelet threshold denoising(WTD)  long short-term memory (LSTM)neural network
基金项目:国家自然科学基金(61871475);广东省科技计划项目(2017B0101260016);广州市创新平台建设计划项目(201905010006);广东省农业技术研发项目(2018LM2168)
作者单位
李祥铜 1,2,曹 亮 1,2,李湘丽 2,3,刘双印 1,2,4,徐龙琴 1,2,呼 增 1,2,黄运茂 2,尹 航 1,2 1. 仲恺农业工程学院信息科学与技术学院广东 广州 510225 2. 广东省高校智慧农业工程技术研究中心 / 广州市农产品质量安全溯源信息技术重点实验室 广东 广州 5102253. 仲恺农业工程学院图书馆广东 广州 510225 4. 石河子大学机械电气工程学院新疆 石河子 832000 
摘要点击次数: 103
全文下载次数: 0
中文摘要:
      【目的】提高对虾养殖水温预测精度,及时掌握水产养殖水温变化规律。【方法】提出基于小波 阈值降噪(Wavelet threshold denoising,WTD)和长短时记忆神经网络(Long short-term memory,LSTM)的水产 养殖水温预测模型,利用 WTD 方法消除原变量间的相关性,减少数据噪声干扰并增强信号数据平滑性,进一步 利用预测能力极强的 LSTM 进行预测。【结果】WTD-LSTM 模型评价指标平均绝对误差(MAPE)、均方根误差(RMSE) 及平均绝对误差(MAE)分别为 0.0104、0.0382 和 0.0288,与标准 BP 神经网络、标准 ELM、标准 LSTM 等 3 种 模型进行对比,评价指标 MAPE、RMSE、MAE 分别降低了 64.85%、59.62%、64.62%, 63.64%、61.18%、60.12%, 47.48%、37.07%、46.27%;从可视化分析来看,WTD-LSTM 预测模型预测结果贴近真实值曲线,相比其他 3 种 模型,能很好地拟合养殖水温非线性时间序列变化趋势。【结论】WTD-LSTM 模型具有良好的预测性能和泛化 能力,可以满足对虾养殖水温精确预测的实际需求,能为对虾养殖水质预测预警提供决策。
英文摘要:
      【Objective】The study was conducted to improve the prediction accuracy of water temperature in prawn culture and grasp the change rules of aquaculture timely【Method】An prediction model of aquaculture water temperature based on Wavelet Threshold Denoising(WTD) and Long Short-term Memory(LSTM)neural network was proposed. The WTD method was used to eliminate the correlation between the original variables, reduce noise interference and enhance the smoothness of signal data. Furtherly, the LSTM with strong predictive power was used to predict the signals.【Result】 The mean absolute error(MAPE), root mean square error(RMSE), and absolute error(MAE)of WTD-LSTM were 0.0104, 0.0382 and 0.0288, respectively. Compared with standard BP neural network, standard ELM and standard LSTM, the evaluation indicators of MAPE, RMSE and MAE decreased by 64.85%, 59.62%, 64.62%; 63.64%, 61.18%, 60.12%; and 47.48%, 37.07%, 46.27%, respectively. According to the visual analysis, compared with the other three models, the prediction result of WTDLSTM was close to the true curve value, which could well fit for the nonlinear time series trend of aquaculture water temperature. 【Conclusion】The model has good prediction performance and generalization ability, which can meet the actual demand for accurate prediction of water temperature in prawn culture and provide decision-making for water quality prediction and early warning of prawn culture.
  查看/发表评论  下载PDF阅读器

手机扫一扫看