主成分分析和长短时记忆神经网络预测水产养殖水体溶解氧

为了提高水产养殖溶解氧预测的精度,提出了基于主成分分析(principal component analysis,PCA)和长短时记忆神经网络(long short-term memory,LSTM)的水产养殖溶解氧预测模型。首先通过主成分分析提取水产养殖溶解氧的关键影响因子,消除了原始变量之间的相关性,降低了模型输入向量维度;然后,在Tensorflow深度学习框架的基础上建立LSTM神经网络的水产养殖溶解氧预测模型;最后,利用该模型对浙江省淡水水产养殖研究所综合实验基地某池塘溶解氧进行验证。试验结果表明:该模型与BP神经网络等其他浅层模型相比,模型评价指标平均绝对误差、均方根误差和平均绝对误差分别为0.274、0.089和0.147,均优于传统的预测方法;该模型具有良好的预测性能和泛化能力,能够满足水产养殖溶解氧精确预测的实际需要,可以为水产养殖水质精准调控提供参考。

Principal component analysis and long short-term memory neural network for predicting dissolved oxygen in water for aquaculture

China has the largest aquaculture industry, accounting for almost 70% of the aquaculture production in the world. The dissolved oxygen in aquaculture directly affects the quality and safety of aquatic products. The dissolved oxygen is susceptible to many factors such as temperature, wind speed, wind direction, etc. So it is significant to understand timely and accurately the change of the dissolved oxygen content which can prevent water quality deterioration, disease outbreaks and optimize aquaculture management. The traditional methods in dissolved oxygen prediction have problems such as low prediction accuracy and poor robustness, with shortcomings like limited ability to express complex functions under limited amount of sample data as well as poor generalization ability for complicated problems. In order to improve the prediction accuracy of the dissolved oxygen in aquaculture, a hybrid model based on principal component analysis (PCA) and long short-term memory (LSTM) neural network was proposed to forecast the dissolved oxygen content in aquaculture. First, the key impact factors of dissolved oxygen in aquaculture were extracted by PCA, which can eliminate the correlations of original variable and reduce the input dimension. Therefore, the key impact factors selected were water temperature, solar radiation, wind speed, wind direction, soil temperature and soil moisture, respectively. Then, a LSTM network model was built based on Tensorflow deep learning framework to construct the nonlinear prediction model between the dissolved oxygen and these key impact factors. Finally, based on the presented prediction model of PCA-LSTM, the dissolved oxygen content was predicted for an experimental pond during July 8th, 2017 to August 8 th, 2017 in the Research Institute of Freshwater Aquaculture, Zhejiang province. In the model accuracy analysis process, a 5-fold cross validation method was used to evaluate the approximation accuracy. The experimental results showed that the proposed prediction model of PCA-LSTM had better prediction performance than BP neural network (BPNN), particle swarm optimization BP neural network (PSO-BP), extreme learning machine (ELM) and least squares support vector machine (LSSVM). In the case of the same data set, the MAE, MAPE and RMSE of the PCA-LSTM were 0.274, 0.089 and 0.147, respectively; the MAE, MAPE and RMSE of LSTM were 0.354, 0.103 and 0.288, respectively; the MAE, MAPE and RMSE of PCA-LSSVM were 0.338, 0.100 and 0.297, respectively; the relative MAE, MAPE and RMSE of PCA-ELM were 0.419, 0.130 and 0.343, respectively; the relative MAE, MAPE and RMSE of PCA-PSO-BP were 0.377, 0.133 and 0.280, respectively; and the relative MAE, MAPE and RMSE of PCA-BP were 0.414, 0.141 and 0.335, respectively. It was clear that the presented prediction model was more accurate than BP algorithm, PSO-BP algorithm and ELM algorithm, slightly higher than LSSVM algorithm. The dissolved oxygen prediction model based on PCA-LSTM network exhibited best prediction accuracy and generalization performance when compared with other traditional forecasting models. Therefore, the presented model based on PCA-LSTM network can meet the actual demand of accurate forecasting of dissolved oxygen and provide a reference for water quality control in aquaculture. As well as it also can help farmers make decisions and reduce farming risks.