Research of dissolved oxygen prediction in recirculating aquaculture systems based on deep belief network

https://doi.org/10.1016/j.aquaeng.2020.102085Get rights and content

Highlights

  • The approach uses a BP neural network to improve the deep neural network structure and optimize a prediction model.

  • The approach uses a VMD algorithm to reduce the noise in the original data, improve the quality of the model input data.

  • The VMD-DBN model displays the highest simulation accuracy and stability compared with different predictive models.

Abstract

Recirculating aquaculture has received more and more attention because of its high efficiency of treatment and recycling of aquaculture wastewater. The content of dissolved oxygen is an important indicator of control in recirculating aquaculture, its content and dynamic changes have great impact on the healthy growth of fish. However, changes of dissolved oxygen content are affected by many factors, and there is an obvious time lag between control regulation and effects of dissolved oxygen. To ensure the aquaculture production safety, it is necessary to predict the dissolved oxygen content in advance. The prediction model based on deep belief network has been proposed in this paper to realize the dissolved oxygen content prediction. A variational mode decomposition (VMD) data processing method has been adopted to evaluate the original data space, it takes the data which has been decomposed by the VMD as the input of deep belief network (DBN) to realize the prediction. The VMD method can effectively separate and denoise the raw data, highlight the relations among data features, and effectively improve the quality of the neural network input. The proposed model can quickly and accurately predict the dissolved oxygen content in time series, and the prediction performance meets the needs of actual production. When compared with bagging, AdaBoost, decision tree and convolutional neural network, the VMD-DBN model produces higher prediction accuracy and stability.

Introduction

The crucial feature or advantage of a recirculating aquaculture system should be frequently and regularly recycling used water through a series of relative electrical approaches and technologies, such as physical and chemical filtering, disinfecting, increasing oxygen and adjusting the temperature, to automatically purify the water before flowing it back into ponds with the help of some equipment (Xiao et al., 2018). So, it’s better to have pre-knowledge of the condition of water in order to make it in better control during the coming processes. In that case, our research mainly concentrates on the predication of recirculating water condition, especially the quantity of dissolved oxygen, through using the model that combines the VMD and the DBN. And a series of experiments show that the combined model has good accuracy and excellent stability.

The normal in-door recirculating aquaculture systems are mostly so shielded from the outside factors that only several important indicators, such as dissolved oxygen content, ammonia nitrogen content, temperature, pH, conductivity, and turbidity has influences on the system. Among them, dissolved oxygen content as a key indicator can directly reflect the aquatic environment, so it is the focus of our research. As Wang’s article reads(Wang et al., 2013), the effect of the operation of oxygen increasing equipment in a recirculation aquaculture system would function after an obvious period of time lag when the real-time concentration of dissolved oxygen detector can detect the change. Moreover, in order to address the potential concern about lopsided balance between economical profits and the prediction accuracy, we would have a significant improvement to current quantity of prediction models to manage and control the water condition more effectively.

Also, besides the importance, the difficulty of collecting such crucial water condition detecting factors can also be patent. On the other hand, the recirculating aquaculture system always is a complex high-dimensional nonlinear data space, which means it contains various kinds of complex noise. All of these inhibit solving the challenging prediction problem (Jain and Srinivasulu, 2004). Common prediction methods are: expert evaluation method, mathematical measurement method, water quality modeling method, chaos theory method, support vector machine method and neural network method. Each of the above methods has its own characteristics and limitations.

The expert evaluation method refers to the method for the professional to judge the prediction by observing the color and odor of the water body based on the experience of the industry. It can only target the specific environment and the prediction accuracy fluctuates drastically. The mathematical measurement rule is based on a large amount of data collection and analysis, which is costly and time-sensitive. The water quality modeling rule mainly relies on the traditional linear model to predict the biochemical sampling indicators of water bodies, but the linear model cannot fully capture the nonlinear relationship between complex high-dimensional data variables, and thus cannot accurately predict the special situation except the average water quality. Based on the linear regression model of chaotic phase space, the chaos theory mainly grasps the inherent correlation between nonlinear space factors through phase space reconstruction.Which simplifies the multi-input and multi-output complex system and finally realizes the prediction of short-term prediction of dissolved oxygen and other biochemical indicators. However, this method assumes the infiniteness of the time series and the noise-free influence of the application environment, which is too ideal in the actual application process. Support Vector Machine (SVM) is a supervised learning method used in machine learning to deal with regression and fitting problems. It can well predict the output of nonlinear complex space (Qiu et al., 2014). In recent years, it has been widely used in the field of dissolved oxygen prediction in aquaculture. However, there is no effective method to obtain the optimal parameter combination of SVM accurately.

It can be seen that the traditional idea has always tried to theoretically optimize the prediction model itself. It is rarely involved in the processing of data. Due to the complex relationship between various water quality parameters, the noise between data is large, the data is first subjected to noise reduction processing, the relationship between data is highlighted, the law of data change is revealed, and the input data of prediction model is effectively improved. The quality of the prediction model improves the accuracy of the model. In this paper, a dissolved oxygen prediction model based on variational mode decomposition(VMD) is proposed.

This paper believes that it is also an effective idea to improve the prediction effect from the perspective of data processing. The variational mode decomposition algorithm is a decomposition technique that relies on variational solution to interpret signal features. It has good robustness and can successfully separate two pure harmonic signals with similar frequencies (Seo et al., 2018a,b). At the same time, the neural network has been tried to apply to the prediction of water biochemical indicators as early as 1994, since the structure is mainly simpler ANN and MLP-NN, there are disadvantages such as accurate prediction of the average situation, slow convergence rate and high training cost. In recent years, various neural network-based variants and optimization algorithms have been continuously developed (Dawson and Wilby, 2001). New methods to predict dissolved oxygen are being proposed constantly (Liu et al., 2019; Dabrowski et al., 2018; Rahman et al., 2019). Deep belief network(DBN) is one of the main algorithms for deep learning, with its powerful feature extraction and function representation ability and processing of high complexity nonlinear data. The advantages of image classification, speech recognition and fault diagnosis have achieved excellent results. Therefore, this paper solves the problem of dissolved oxygen timing prediction in recirculating aquaculture by establishing a deep belief network prediction model. In summary, this paper proposes a recirculating water dissolved oxygen prediction model algorithm combining VMD and DBN. That is to use VMD to perform variational mode decomposition on the original acquisition data space to separate the eigenmodes that are decisive for feature learning and filter out noise. Then use DBN to perform deep learning and regression analysis on the eigenmode quantities, and finally achieve the purpose of predicting dissolved oxygen content.

During the pre-analysis experiment, our proposed model achieved MAE as 0.225, RMSE as 0.71, MAPE as 4.46 % and R2 as 0.9336 as opposed to the normal model with MAE as 0.383, RMSE as 0.791, MAPE as 7.27 % R2 as 0.8692. This empirical results establish that our model has an obvious improvement in precision and stability. When compared with other common prediction models like Bagging, AdaBoost, Decision Tree and CNN, the empirical results support our model’s improvement.

Section snippets

Variational mode decomposition–deep belief network

This paper presents a VMD-DBN model for the prediction of dissolved oxygen in recirculated aquaculture water. This model consists of two parts: a data preprocessing module and a dissolved oxygen prediction module. First, water quality data preprocessing is performed through VMD and the noise reduction of the original data, and the preprocessed data are used as the input of the deep belief network model. Second, a DBN is constructed, and relevant parameters, such as the model bias and weights

Water quality data and data preprocessing

The quality of water in recirculation aquaculture ponds directly effects on the growth condition of organisms. When referring to collecting the characteristic data of water to follow water changes, it is much easier to detect some specific physical and chemical factors, such as dissolved oxygen, pH, water temperature, turbidity, ammonia nitrogen and water level than other factors that are too difficult to trace and expensive. We collected our experimental data from MingBo Aquatic Products Co.,

Conclusion

Real-time prediction of dissolved oxygen content is the premise and basis for achieving precise control of recirculating aquaculture. To solve the problems of traditional prediction model that low accuracy, poor stability, noise in the local characteristic data collected from water quality, the model which based on VMD-DBN has been proposed in this paper.

Because the content of dissolved oxygen in recirculating aquaculture is influenced by many water quality factors, it takes pH, ammonia

Author statement

We promise that the manuscript has not been previously published in any other form and is not under consideration for publication elsewhere. Prof. Dr. Yaoguang Wei takes responsibility on behalf of all authors for the authorship, authenticity and integrity of this manuscript, and affirms that all authors and acknowledged contributors have read and approved this manuscript. We completely comply with the policy of this journal.

CRediT authorship contribution statement

Qin Ren: Writing - original draft. Xuanyu Wang: Software. Wenshu Li: Software. Yaoguang Wei: Writing - review & editing. Dong An: Writing - review & editing.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgments

This study was supported by National Key Research and Development Program of China2017YFD0701702.

References (19)

There are more references available in the full text version of this article.

Cited by (58)

  • Applications of deep learning in water quality management: A state-of-the-art review

    2022, Journal of Hydrology
    Citation Excerpt :

    With the aid of the Wiener filter, the VMD is not prone to background noise, thus outperforming the EMD in many applications (Maji and Pal, 2016; Uzunoglu, 2018). Ren et al. (2020) reported that the VMD significantly reduced the noise and highlighted key data features for better DO prediction in aquaculture ponds. In another study, Fijani et al. (2019) proposed a two-staged decomposition method.

View all citing articles on Scopus
View full text