基于长短期记忆神经网络的罗非鱼生长预测模型

Tilapia growth prediction model based on Long Short-term Memory neural network

【目的】通过测定吉富罗非鱼生长指标,建立其生长的长短期记忆神经网络(Long Short-term Memory neural network model,LSTM)模型,分析模型的拟合度和准确度,为罗非鱼的育种和养殖提供参考。【方法】以罗非鱼生长阶段的生长时间、投喂量及水槽编号3个指标数据作为输入量,通过Dropout和one-hot的方法建立LSTM模型。【结果】模型在训练开始后迅速下降,100次迭代左右,误差下降速度开始逐步减缓,在1000次迭代后,误差开始收敛,数值趋于稳定,稳定值误差在0.0036左右。训练完成的模型对测试集的预测结果相对误差随真实值变大而逐渐变小,真实值较大且稳定时,相对误差较小,整体拟合程度较好。【建议】生长预测模型满足基本生产需求的同时,需增加样本数据的记录采集,建立生长数据库;结合信息平台等技术获取多影响影子数据,增加输入变量,使模型更加完善合理;选择合适的模型,结合预测数据与生产,合理规划上市时间及安排投饲方案等,使养殖利益最大化。

【Objective】The purpose of this study was to determine the growth index of GIFT tilapia and establish the Long Short-term Memory neural network model(LSTM)for its growth.The fitting and accuracy of this model were analyzed to provide references for the breeding and aquaculture of tilapia.【Method】The data of growth time,feeding amount,and sink number of tilapia growth stage were used as input variables and the LSTM was established through Dropout and one-hot methods.【Result】The model declined rapidly after the training started.When it had about 100 iterations,the rate of error reduction began to gradually slow down,while after 1000 iterations,the error began to converge and the value tended to stabilize.The stability value error was about 0.0036.The relative error of the prediction results of the test set by the trained model gradually became smaller as the true value became larger.When the actual value was larger and stable,the relative error was smaller and the overall model fitting degree was better.【Suggestion】While the growth prediction model satisfies the basic production requirements,it is necessary to increase the record collection of sample data and establish a growth database.It is suggested to combine information platform and other technologies to obtain multiple impact shadow data,increase input variables,and make the model more improved and reasonable;select the appropriate model,combine with forecast data and production,rationally plan the time-to-market and schedule feeding programs so as to maximize the benefits of aquaculture.