Elsevier

Measurement

Volume 170, January 2021, 108743
Measurement

On-line ammonia nitrogen measurement using generalized additive model and stochastic configuration networks

https://doi.org/10.1016/j.measurement.2020.108743Get rights and content

Highlights

  • We propose an online soft measurement method for the ammonia nitrogen.

  • We establish an aquaculture system in our laboratory to evaluate the novel method.

  • The online sensor combines the mechanism model and data driven compensation.

  • The proposed ammonia nitrogen model is verified by simulation experiments.

Abstract

On-line measuring of ammonia nitrogen in seawater is essential to the intensive aquaculture process. Its concentration has an important effect on the growth and development of the breeding organisms. The popular Nessler's reagent method suffers from large time delays. It can only be applied off-line. To realize on-line measurement, we propose a hybrid soft measurement model for ammonia nitrogen. This new model combines mechanism analysis and data driven methods, the generalized additive model and stochastic configuration networks are used respectively. To demonstrate the effectiveness of the proposed modelling techniques, an intensive aquaculture system is designed in laboratory to realize this soft measurement method. The accuracy of the test data performed in generalized additive model is 0.073, the accuracy after error compensation is 0.064. Compared with other methods, the stochastic configuration networks compensation model has the shortest running time which is 0.5 s. These experimental results show that the proposed hybrid method performs better than other existing data driven modeling methods, and this work will be the foundation for the practical application.

Introduction

With the continuous improvement of aquaculture technology, large-scale and intensive aquaculture has flourished. In order to increase the density of breeding and yield per unit area, the amount of artificially fed bait has been increasing, and the ecological balance of water becomes the key to the success of aquaculture [1]. The concentration of ammonia nitrogen in water is one of the important indicators of aquaculture water quality. Its toxic components directly affect the healthy growth of breeding organisms and determine the yield and quality of breeding organisms, so the real-time detection of ammonia nitrogen concentration is an urgent problem to be solved in aquaculture.

At present, the main methods for the detection of ammonia nitrogen concentration include Nessler's reagent, molecular absorption spectroscopy, hypobromite oxidation, electrochemical methods and so on [2]. These methods can effectively detect the ammonia nitrogen concentration, but they are all offline, and can’t be used for real-time monitoring [3]. Although there are some multi-parameter water quality analyzers, such as America’s YSI 6820/6920, German WTW Multi 9630 and so on, they still have problems such as high cost and high maintenance cost, and most instruments are suitable for fresh water instead of seawater [4]. In recent years, soft measurement methods have been used to monitor the ammonia nitrogen concentration in sewage [5], [6] and achieved better results, intelligence methods have been used, such as support vector machine [7] and fuzzy neural networks [8]. The aquaculture water environment is a complex ecosystem. Changes in water quality are often the result of a combination of many factors. While paying attention to the quality and safety of breeding organism products, it is necessary to ensure the safety of breeding organisms and avoid unnecessary losses.

Researches also tried to use soft measurement method in aquaculture. Deng et al [9] used radial basis function (RBF) neural networks to establish the ammonia nitrogen concentration prediction model by using water temperature, pH, dissolved oxygen, nitrate nitrogen content and nitrite nitrogen content as auxiliary variables. As the content of nitrate and nitrite are also hard to be measured online. Therefore, the practicality of this method is not strong. To improve this method, Li et al. and Yu et al. used water temperature, pH, conductivity and so on as the auxiliary variables and artificial intelligence methods, such as support vector machines (SVM) [10], stochastic configuration networks (SCN) [11] and Random vector functional-link (RVFL) [12] to predict ammonia nitrogen. As most of these models were based on data driven methods and due to the “black box” characteristics, the established models do not have the ability to interpret and are also difficult to guarantee accuracy. So far there is no successful on-line sensors for the ammonia nitrogen measurement.

Hybrid modeling with part of the model from first principles and part of the model inferred from data, is a good way to take advantages of making full use of mechanistic knowledge and apply the data-driven method to compensate model error for domain problem solving [13].

The mechanism analysis uses the generalized additive model (GAM). The compensation model uses the random neural network SCN. This method can effectively use the relationship between the auxiliary variables and the dominant variable, and the established model has certain explanatory power. This model can overcome the difficulties of the back propagation (BP) and RBF networks, such as slow learning process, falling to the local minimum, and poor generalization capability.

In order to create an on-line soft measurement method of the ammonia nitrogen, especially for the aquaculture water quality control, in this paper, we make the following contributions:

  • (1)

    We propose the on-line sensor of the ammonia nitrogen for aquaculture water quality. It uses the mechanism model and data driven compensation.

  • (2)

    We establish an intensive aquaculture system in our laboratory to evaluate our soft measurement method. Based on this system, aquaculture water parameters can be measured. The proposed ammonia nitrogen model is verified by simulation experiments.

The remainder of this paper is organized as follows: Section 2 describes the intensive aquaculture process and mechanism of ammonia nitrogen reaction in aquaculture water. Section 3 introduces the hybrid modelling strategy of ammonia nitrogen on-line measurement, details the mechanism analysis model and error compensation model. Section 4 reports the intensive aquaculture system in laboratory and our experimental results, and Section 5 concludes this paper.

Section snippets

Intensive aquaculture process

The intensive recirculating aquaculture system uses biological reaction and physical filtration methods to remove residual bait feces, ammonia nitrogen, nitrite and other harmful substances contained in the aquaculture water, and recycle the purified water by disinfection, aeration, temperature adjustment and so on to reuse [14]. It involves intervention in the growing process, such as water aeration and supplemental feeding. The process mainly includes biological filtration, disinfection and

Strategy of on-line measurement

From the above mechanism analysis, we can see that there is a nonlinear relationship between the ammonia nitrogen concentration and each influence factor, and the overall constitutes a logarithmic relationship. GAM is based on nonparametric regression and smoothing techniques. It consists of nonparametric additive terms. Each additive term is estimated using a single smoothing function and can explain how the response variable changes with the explanatory variable. Therefore, based on the

Experiment system

In order to realize the proposed ammonia nitrogen soft measurement model, an intensive aquaculture hardware measuring system is developed as shown in Fig. 4. It is mainly used to control the circulation of water and monitoring the water quality parameters. This system uses liquid level sensor, pH sensor, temperature and conductivity integrated sensor to collect data. The obtained signal is transmitted to the PLC controller, then to the computer which is used to implement the proposed soft

Conclusions

In this paper, we solve the on-line measurement problem of the ammonia nitrogen concentration. We use hybrid modelling strategy, which uses the mechanism model and data driven compensation model. The laboratory intensive circulating aquaculture system is designed and developed. The relevant water quality parameters are obtained by the proposed method. From the experimental results, we can see the mechanism model can get the internal relation and data driven model can improve the performance

CRediT authorship contribution statement

Wei Wang: Conceptualization, Methodology, Writing - original draft. Yao Jia: Methodology. Wen Yu: Supervision, Writing - review & editing. Hongshuai Pang: Software. Kewei Cai: Software, Visualization.

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.

References (32)

  • Hong-Gui Han et al.

    Intelligent modeling approach to predict effluent quality of wastewater treatment process

    Wastewater Water Qual.

    (2018)
  • Muhammed Enes Bayrakdar

    Cost effective smart system for water pollution control with underwater wireless sensor networks: A simulation study

    Comput. Syst. Sci. Eng.

    (2020)
  • Jinwu Ju et al.

    Analysis of ammonia nitrogen content in water based on weighted least squares support vector machine algorithm

    J. Software Eng. Appl.

    (2016)
  • Hongbiao Zhou et al.

    Soft sensing of effluent ammonia nitrogen using rule automatic formation-based adaptive fuzzy neural network

    Desalin. Water Treat.

    (2019)
  • Changhui Deng, Deyan Kong, Yanhong Song, et al. A soft-sensing approach to on-line predicting ammonia nitrogen based on...
  • Li Kang, Wang Wei, Lin shaohan, Soft measurement of ammonia nitrogen concentration based on GA-SCN, 2018 IEEE...
  • Cited by (15)

    • Portable device for on-site detection of ammonia nitrogen

      2022, Information Processing in Agriculture
      Citation Excerpt :

      Hence, in the real environment, it is no need to consider the measurement effect of the device when the pH change. The research has shown that the measurement of the ammonia nitrogen concentration is theoretically influence by temperature [34]. Hence, it is necessary to study the exact relationship between its stability results and ambient temperature.

    • Broad stochastic configuration network for regression

      2022, Knowledge-Based Systems
      Citation Excerpt :

      The input parameters of hidden layer nodes are configured by the supervision mechanism and the output parameters are calculated by Moore–Penrose inverse. SCN, with a series of advantages of lower human intervention, higher learning efficiency and stronger generalization ability, has been applied in the fields of online measurement problems [7], industrial process modeling [8], vibration signal fault diagnosis [9] and power big data processing [10]. Since it was demonstrated in 2017, a lot of improved models and variants have been developed.

    View all citing articles on Scopus
    View full text