Driving analysis and prediction of COD based on frequency division

doi:10.21203/rs.3.rs-4716541/v1

Download PDF

Research Article

Driving analysis and prediction of COD based on frequency division

https://doi.org/10.21203/rs.3.rs-4716541/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

COD (Chemical Oxygen Demand) is an important indicator to measure organic pollution of water body. To strengthen in-depth analysis and prediction of COD, a new method was proposed in this paper. A frequency division method, Variational Mode Decomposition (VMD) was used to complete time domain decomposition of COD data before model simulation. The original data was separated into five signals with different frequency bands, IMF1, IMF2, IMF3, IMF4 and IMF5, with which the influence of meteorological factors and water quality factors on COD were explored. The long-term COD content is mainly driven by nutrient factors phosphorus and nitrogen, while the immediate fluctuation characteristics exhibit relatively stability. Random Forest, Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) were used to predict COD with the original data and the signal data processed by VMD. It is found that frequency division can improve simulation stability and accuracy of GRU and LSTM more significantly than Random Forest. VMD-GRU and VMD-LSTM models can be used reliably for COD analyzation and prediction in Chengdu area.

frequency division

driving analysis

prediction model

COD

COD is considered an important indicator for characterizing the presence of organic pollutants in the water environment. The higher COD value indicates more severe pollution by organic pollutants in the water. Organic pollutants are often transferred to water bodies through atmospheric dry and wet deposition, agricultural runoff and wastewater discharge (Arfaeinia et al. 2017). Bioaccumulation may be caused by these pollutants when they are deposited into the atmosphere and enter water bodies, posing potential toxicity to both biota and humans (Agrell et al. 2002). A large amount of organic substances have been found to exist in the aqueous phase or particle extract during rainstorm sampling, expanding the scope of pollution as they migrate with the runoff (Glaser et al. 2023). A study examining organic compounds in 38 rivers of the United States were examined, which showed the 10 most widespread anthropogenic organisms are strongly associated with agricultural runoff (Bradley et al. 2017). By time-of-flight mass spectrometry (GC × GC-TOF-MS) combined with two-dimensional gas chromatography, organic pollutants in Yongding River basin were analyzed. It was found that the types of organic pollutants in downstream rivers were relatively similar to those in the effluent of sewage treatment plants (Jiang et al. 2023), which showed that wastewater treatment plant effluents are a non-negligible source of organic pollutants in surface runoff. A survey of 46 wastewater treatment plant effluents in China were found to contain 302 trace organic pollutants, accounting for 59 percent of the total organic matter (Liu et al. 2024). Organic pollutants pose a serious threat to the environment due to their toxic characteristics over prolonged period (Kumar et al. 2022). Organic pollutants can suffocate aquatic organisms by disrupting photosynthesis of aquatic plants and leading to significant reduction in water quality which can indirectly affect human health (Muralikrishna and Manickam 2017). Mussels farmed in the coast of the Gulf of Naples and Domitio had been found to pose a high food safety risk in terms of heavy metals and persistent organic pollutants (POPs) (Esposito et al. 2020). POPs pose a greater risk in these environments because they are difficult to be broken down and can easy to be accumulated in the air, water, soil, and fat cells of animals (Shetty et al. 2023; Zaynab et al. 2021; Lan et al. 2024; Mallah et al. 2022; Yua et al. 2018). Due to the toxicity of some organic pollutants to human health, its characteristics and toxicity have become the focus of global attention (Li et al. 2017). Researches by scholars not only focus on individual organic pollutants but also emphasize the measurement of organic content. Measuring and analyzing COD content has also been a hot topic of research. In China, a lot of work has been taken out to measure, analyze and control COD. But there are still many challenges (Li et al. 2012). Table 1 shows the status of COD emission and control in China.

Table 1

Status of COD emission and control in China
Items	China
Formulation	Mainly national standards, A small quantity local standards
Implementation	1) National standard specified uniform limits for similar dischargers 2) The control items are mainly centralized control. The total quantity control is comprehensively implemented. 3) Local standards outperform national standards in execution
Control level	1) Compared to the lower treatment rate of ammonia nitrogen, COD treatment rate reaches over 90%. 2) The monitoring and supervision system is largely established, but the development of the shared system is uneven.
Limits of major pollutants	1)National standard: Standard A: 50mg/L, Standard B: 60mg/L.
Limits of major pollutants	2)Sichuan Province Standards: Urban sewage treatment plant༚30mg/L

The continuous determination of COD with long time series can be helpful to obtain a lot of necessary data for the analysis and prediction of organic matter in watershed. In order to achieve this goal, many methods for COD measurement were developed continuously and the regional, conventional and continuous monitoring methods were gradually realized. Traditional methods for COD measurement, such as potassium dichromate, potassium permanganate, and UV-vis spectroscopy, Ultraviolet-visible (UV-vis) spectroscopy combined with stoichiometric tools was used to determine COD content. This method is particularly suitable for real-time and rapid determinations while a large number of sample data is required for modeling (Li et al. 2018). Carbon point fluorescent capillary sensors are also used to determine COD. This method has the advantages of low detection limit, good linear range and reliability, portability and low cost. But it is easy to be limited by interference protection and environmental impact (Zhang et al. 2023). A nano-lead dioxide-composite electrochemical sensor has been used for COD determination offering fast response times, simple instrumentation, low cost, high detection sensitivity and wide linear range. However, sampling limitations and sensor stability can pose challenges (Wang et al. 2022). Currently, COD monitoring has been further developed to automatic monitoring. The CL system with flow injection was applied for COD determination, which has the obvious advantages of much shorter analysis time, simple operation except for a lower detection rate (Li et al. 2003). A new reagent-free method for measuring chemical oxygen demand (COD) was proposed based on ultraviolet absorption spectroscopy (UV-AS) which enables high-precision and long-range COD measurement by automatically selecting the wavelength and analyzing the full spectrum of the data. This method can therefore be used for in situ and online environmental monitoring (Wang et al. 2019). The development of automated monitoring has significantly improved the efficiency and frequency of COD monitoring. It can realize the simultaneous determination of COD over a wide area and provide technical and methodological support for regional COD monitoring research.

To better understand the changing patterns and drivers of organic matter in water bodies, a variety of models have been established by scientists to analyze and predict COD. These models input actual monitoring data and predicted values of COD. The dynamics of COD in water bodies are studied to analyze and manage the flow patterns and long-term series of organic matter. The removal of COD from polluted solutions has been simulated and predicted using artificial neural networks (ANN) (Elmolla et al. 2010; Masouleh et al. 2022). One main advantage of ANN is its ability to handle complex relationships between input and output variables (Ataei et al. 2021). However, ANN techniques require large datasets to train the model and are computationally expensive (Khanmohammadi et al. 2024). Faster convergence, higher convergence accuracy, and better pattern recognition are achieved by recurrent neural networks (RNN) compared to ANN (Al-Qaili et al. 2024; Gholami et al. 2023). RNNs are more suitable for time series prediction. However, RNNs suffer from the problem of long-term dependence, i.e., gradient vanishing and gradient explosion will be met when RNNs learns long sequences, which causes its difficulty to understand nonlinear relationships for long periods of time (Hochreiter 1991). Therefore, the use of convolutional neural networks (CNNs) was investigated by some researchers for efficient feature extraction to extract data features. A correlation between COD concentration and spectral reflectance in urban rivers was found and COD concentration was accurately predicted by using a one-dimensional convolutional neural network (1D-CNN) (Cai et al. 2022). The Long Short-Term Memory (LSTM) network, introduced by Hochreiter and Schmidhuber (1997), is an extension of the RNN. Long-term dependencies can be learned by LSTMs, avoiding the exploding or vanishing gradient problem that affects traditional RNNs (Xu et al. 2020). For monitoring and predicting the performance of COD in wastewater treatment, a novel LSTM-based soft sensor was developed (Xu et al. 2023). The LSTM method can mine the potential information between different water quality indicators at different time scales improving prediction accuracy. Major pollutants such as Biological Oxygen Demand(BOD), Chemical Oxygen Demand(COD), Total Nitrogen(TN), Total Phosphorus(TP), and Ammonia Nitrogen(AN)were predicted and demonstrated to have a high degree of correlation with each other using an integrated multivariate LSTM network(Wang, Xue et al. 2024). Although LSTMs can effectively process and predict events with long intervals and delays (Yousfi et al. 2017). A large number of parameters was needed and low convergence rate is existed. For this reason, GRU based LSTM was proposed (Cho et al. 2014)to simplify the internal cell architecture of LSTM and reduce the network training time in order to guarantee the prediction accuracy. Combining machine learning with sensor networks, multiple machine learning algorithms are employed to predict COD emissions. Comparative results showed that GRU is better than LSTM (Miao et al. 2021).

Prediction results can be influenced by the model characteristics themselves. How to improve their performance has become a necessary research topic nowadays. Time-frequency domain transformation technique was tried to be combined with some models to enhance the simulation effects. A complex signal can be decomposed into a series of intrinsic modal functions (IMF) with different frequencies and amplitudes by frequency division. the complexity and strong nonlinearities in a time series can be effectively reduce while obtaining a relatively stable subsequence that contains several different frequency scales. This approach extracts the main data components, and representing them as subsequences with different frequency features, thus optimizing the simulation effect of some intelligent models. The Variational Mode Decomposition (VMD), an adaptive and fully non-recursive signal processing method, was proposed in 2014(Dragomiretskiy and Zosso 2014) with the advantage of determining the number of mode decompositions. The VMD was used to decompose river flow into IMF, and the mixed models RVFL_VMD, GRNN_VMD and RBFNN_VMD were established to predict river turbidity. The best performance was achieved by RVFL_VMD on the hourly time scale, while GRNN_VMD provided the best prediction on the daily time scale (Heddam et al. 2022). VMD can improve the accuracy of power consumption forecasts. The combination of Bi-directional Gated Recurrent Unit (BiGRU) and LSTM models with VMD showed excellent prediction accuracy on various assessment metrics, especially for short to medium term predictions (Ahmed et al. 2023). The VMD method can be used to study the periodicity of vegetation and its relationship with climate (Wang et al. 2023). A new wind power forecasting method called IVMD combining VMD and HFCM was developed to achieve more accurate wind power generation prediction and reduce the prediction error by extracting time series features and learning the weights using Bayesian ridge regression method (Qiao et al. 2022). The spectral characteristics of VMD can be used to sort iron ore in hyperspectral images (Nie et al. 2023).

In this paper, taking the rivers in Chengdu area of China were selected as the research focus, with COD as the primary research index. The frequency division technique of VMD was applied to process the raw COD data. Moreover, three models — Random Forest, LSTM and GRU were integrated with VMD to develop an optimized model framework. The research aimed to find a new method was hoped to be found for the prediction and analysis of organic matter in the Chengdu watershed.

2.1 Used data parameters

Data parameters include water quality data and meteorological data.

Water quality data includes WT (Water Temperature), PH (pH Value), DO (Dissolved Oxygen), COD_Mn (Potassium Permanganate Index), NH₃-N (Ammonia Nitrogen), TP (Total Phosphorus), TN (Total Nitrogen), EC (Electric Conductivity) and NTU (Turbidity). Data from 2020 to 2022 were obtained from daily monitoring of state-controlled cross sections. Monitoring was conducted six times a day at 0:00, 4:00, 8:00, 12:00, 16:00 and 20:00.

Meteorological data includes T (Mean Air Temperature), MT (Maximum Air Temperature), LT (Minimum Air Temperature), BP (Mean Air Pressure), MBP (Maximum Air Pressure), LBP (Minimum Air Pressure), RH (Relative Humidity), LH (Minimum Relative Humidity) and P (Precipitation). The daily meteorological data from 2020 to 2022 were obtained from two meteorological stations of Chengdu area, monitoring once a day.

2.2 Data processing

(1) Missing values: A small amount of missing data was found in COD_Mn, NH₃-N, TP and TN. Polynomial interpolation was used to make up the missing values.

(2) Anomalous values: A small amount of anomalous data was found in EC, NTU, P, LH, RH and LT. The anomalous values were filled in by polynomial interpolation.

2.3Variational mode decomposition

If the COD (Chemical Oxygen Demand) in the river water corresponds to an approximate function f (t) and it is the sum of the set of k (i.e., IMF) components, the constraint problem is expressed ass (Ling et al. 2021; Li et al. 2021b; Xu et al. 2020):

$$\:\begin{array}{rr}&\:\underset{{u}_{k},{\omega\:}_{k}}{\text{m}\text{i}\text{n}}\left\{\sum\:_{k=1}^{K}{∥{\partial\:}_{t}\left[\left(\delta\:\left(t\right)+\frac{j}{\pi\:t}\right)\text{*}{\mu\:}_{k}\left(t\right)\right]{e}^{-j{\omega\:}_{k}\left(t\right)}∥}_{2}^{2}\right.\\\:&\:such.that.\sum\:_{k=1}^{K}{\mu\:}_{k}=f\end{array}$$

where (t) is the time, δ (t) corresponds to the impulse function, i.e., the Dirac distribution and {u_k (t)¼ {u₁ (t), u₂ (t),…, u_k (t)} is the series of the decomposed K Intrinsic Mode Functions (IMF) modes, and the set of the center frequencies of each IMF is given as {ω_k (t)¼{ω₁ (t), ω₂ (t),…,ω_k (t)} (Xu et al., 2020). f is a given constant.

3.1 VMD decomposition results

Figure 1 shows the original data and the decomposed IMF1 to IMF5 signals from low to high frequencies. IMF1 usually corresponds to the low-frequency component of the data and represents the long-term trend. IMF2 to IMF5 representing the short-term abrupt changes in the data, are usually correspond to rapid changes, fluctuations or transient features in the data.

Figure 2 shows the Fourier spectrum of the original data and the decomposed signals. The distribution of the signal in the frequency domain in Fig. 2 indicates the intensity or amplitude of the different frequency components contained in the signal. Each wave peak corresponds to the intensity or energy of a particular frequency component. The higher the crest, the stronger the energy of the corresponding frequency component, indicating the greater the contribution of that frequency component in the signal. As can be seen from Fig. 2, there is basically no frequency entanglement between the decomposed signals.

3.2Driving analysis of COD change

Pearson's correlation analysis was used to analyze the correlation between variables driving the COD change and COD content. Positive value of the correlation coefficient indicates positive correlation, negative value indicates negative correlation. The larger the absolute value of the value, the stronger the correlation. The correlation judgment criteria are shown in Table 2(Jia 2018). The correlation values of COD_Mn with water quality data and hydrological data in the original data and the decomposed signals IMF1-IMF5 are shown in Table 3.

Table 2

Correlation criterion (Jia et al. 2018)
Absolute value of the correlation value	Relevance
0.0-0.3	No relevant
0.3–0.8	relatively strong correlated
0.8-1.0	Strong correlation

Table 3

Correlation between the original data, decomposed signals and COD_Mn.
Factor	Original Data	IMF1	IMF2	IMF3	IMF4	IMF5
WT	0.28	0.38	0.11	0.00	0.00	0.01
PH	-0.44	-0.41	-0.58	-0.01	-0.00	-0.02
DO	-0.57	-0.60	-0.33	-0.01	-0.00	-0.01
NH₃ -N	0.57	0.55	0.42	0.01	0.01	0.04
TP	0.69	0.70	0.71	0.02	0.10	0.45
TN	0.55	0.54	0.34	0.02	0.05	0.34
EC	0.20	0.26	0.04	-0.00	0.00	0.01
NTU	0.30	0.32	0.17	0.01	0.02	0.09
T	0.22	0.36	-0.19	-0.14	0.00	0.00
MT	0.20	0.37	-0.33	-0.20	0.01	0.00
BP	-0.25	-0.41	0.11	0.01	0.00	0.01
MBP	-0.24	-0.41	0.12	0.01	0.00	0.01
LBP	-0.26	-0.44	0.12	0.01	0.00	0.01
RH	0.03	-0.10	0.15	-0.00	-0.00	-0.00
LH	0.03	-0.11	0.19	-0.02	-0.01	-0.00
P	0.15	0.13	0.28	0.08	-0.01	0.00

In the long term, the absolute correlation value between TP, DO, NH₃ -N, TN, LBP, PH, BP, MBP, WT, T, NTU and COD is between 0.3–0.6, indicating a relatively strong correlation. Among these, the correlation values for range from 0.54 to 0.70, showing that COD content is more influenced by water quality factors than by climate factors. Nitrogen and phosphorus are the main indicators of eutrophication in water bodies, suggesting that long-term changes in COD are strongly positively correlated with eutrophication. That is, the more serious the pollution of organic matter in water bodies, the more serious the eutrophication of water bodies. The change in DO also positively correlates with eutrophication, further verifying the strong correlations between eutrophication and COD. When there is an excess input of nutrients (e.g., nitrogen, phosphorus, etc.) to a water body, eutrophic substances increase the concentration of COD and cause other changes. Eutrophic waters usually contain more organic waste and organic material, resulting in an increase in COD_Mn content in the water bodies as well. This relation between eutrophic substances with COD provides a simplified idea for analyzing the long-term pollution characteristics of water bodies in Chengdu. When data on nitrogen, phosphorus, or COD are missing or abnormal, the strong correlations among these indicators can be used to infer the missing values. Characteristics changes of one indicator can also be used to infer the changes in other indicators.

For the relatively longer short-term change characteristics of COD, the eutrophication index nitrogen and phosphorus and PH, DO, MT still have some influence. But except the nitrogen and phosphorus, the influence of other factors basically disappeared with the gradual elimination of the signals of the influencing factors in the short time series. It is reasonable to speculate that short-term changes in COD are largely unaffected by water and climatic parameters, demonstrating the relative stability of the instantaneous characteristics of COD.

3.3 Simulation results with or without frequency division

Firstly, Random Forest, LSTM, and GRU models were used to predict COD with the original data. Then, the original data was decomposed, and the signals with a correlation of less than 0.3 with the predicted target COD were eliminated from the variable modal decomposition signals. Thus, Random Forest, LSTM, and GRU models were used again to predict COD with the filtered signal data. The predicted results with and without frequency division were compared by two methods. One is Scatter plots of the predicted value and the true values, the other is three kinds of error values, namely mean absolute error (MAE), root mean square error (RMSE) and symmetric mean absolute percentage error (SMAPE).

3.3.1 Simulation results with Random Forest and VMD-Random Forest

The prediction results with the Random Forest and VMD-Random Forest model were shown in Fig. 3.

The scatter plot showing the fitting degree between the true values and the predicted values was shown in Fig. 4. The simulation effect before and after frequency division was evaluated by mean absolute error (MAE), root mean square error (RMSE), and symmetric mean absolute percentage error (SMAPE) were shown in Table 4.

Table 4

Quantitative metrics for Random Forest and VMD- Random Forest
MODEL	RMSE	MAE	SMAPE
Random Forest	0.397	0.285	0.14
VMD-Random Forest	0.293	0.224	0.104

By Fig. 4 and Table 4, it can be concluded that the simulation effect is improved because of VMD processing. The improvement rates are 26.20%, 21.40%, and 25.71%, respectively. It can be found that VMD can improve the simulation effect of Random Forest to some extent.

3.3.2 Simulation results with GRU and VMD-GRU

The simulation results with GRU and VMD-GRU models are shown in Fig. 5.

Scatter plot was shown in Fig. 6. The values of MAE, RMSE and SMAPE were shown in Table 5.

Table 5

Quantitative metrics for GRU and VMD-GRU
MODEL	RMSE	MAE	SMAPE
GRU	0.502	0.395	0.175
VMD-GRU	0.14	0.098	0.046

After the original data was processed by VMD, the values of RMSE, MAE and SMAPE decreased, and the degree of dispersion shown in Fig. 6 was reduced. This indicates that VMD can significantly improve the prediction accuracy of the GRU model. The simulation performance of GRU was improved by 72.11%, 75.19%, and 73.14% respectively. This demonstrates that VMD frequency division can more effectively enhance the GRU model's ability to understand and capture data by eliminating interference factors, thereby improving the simulation effect of the GRU.

3.3.3 Simulation results with LSTM and VMD-LSTM

The simulation results with LSTM and VMD-LSTM were shown in Fig. 7.

Scatter plot was shown in Fig. 8. The error values of MAE, RMSE and SMAPE were shown in Table 6.

Table 6

Quantitative metrics for the LSTM and VMD-LSTM
MODEL	RMSE		MAE	SMAPE
LSTM	0.772		0.578	0.24
VMD-LSTM	0.205		0.156	0.073

The values of LSTM RMSE, MAE and SMAPE reduced after VMD processing, indicating that VMD-LSTM model achieves lower error and higher accuracy compared to the original LSTM model. VMD frequency division significantly improves the prediction performance. The improvement rate of VMD to LSTM model is 73.45%, 73.01%, and 69.58% respectively.

COD is often used as an important indicator to evaluate the organic pollution in water environments. Although there have been many researches on water quality prediction models, the prediction accuracy of these models using raw data often falls short of ideal results. This is because water quality is affected by many factors from weather, human activities, and the water environment. There are complex interaction relations among these factors, which is difficult to be captured by traditional water quality prediction models so as to reduce the prediction accuracy of the model.

In this paper, a new COD prediction method was proposed. Taking water system in Chengdu area of China as the research object, a kind of frequency division method, VMD method was tried to complete the time domain decomposition of COD data before model simulation. the frequency division method was used to separate the COD original data into signals of different frequency bands, namely IMF1, IMF2, IMF3, IMF4 and IMF5. Using these different signal data, the influence of COD on meteorological factors and water quality factors was explored. Based on the correlation between signals of different frequency bands and COD, it is found that the long-term COD content can be affected by eutrophication factors such as phosphorus and nitrogen relatively easily. The short-term immediate change characteristics of COD are relatively stable and are less affected by any water quality factors and meteorological factors.

Using the original data and the signal data without weak correlated factors, the simulation effects of GRU, LSTM and Random Forest models on COD were compared. By the frequency division, the original data was decomposed to signals with different frequency, and the signals with a correlation of less than 0.3 were eliminated further from the variable modal decomposition signals. Removing the interference factors of COD is conducive to improving the simulation effect. The simulation results verified this. When combining Random Forest, GRU or LSTM with variational mode decomposition in COD prediction, the frequency division is helpful to improve the stability and accuracy of the simulation of these models. It has better promotion effects on GRU and LSTM with higher improvement rates. The improvement rate can even amount to more than 70% after VMD processing, especially for LSTM and GRU simulation. The VMD-GRU model and the VMD-LSTM model can be used as more reliable tools for water quality analyzation and management in Chengdu area.

This simulation method constructed in this paper can also be considered for exploring other water environment indicators in different regions, potentially providing a more reliable tool for water quality analysis and management.

Funding

This work was supported by [the National Natural Science Foundation] (Grant numbers [52050410328] and [62266014]) and [the Guangxi Natural Science Foundation] (Grant number 2021GXNSFAA220056).

Author Contributions

[Mei Li] was involved in the initial draft and editing; [Kexing Chen] contributed to the investigation and methodology; [Deke Wang] was responsible for data curation; [Rui Xu] was involved in the review and overall supervision.

Data Availability Statement

Some or all data, models, or code that support the findings of this study are available from the corresponding author upon reasonable request: (1) original data of COD, (2) models, and (3) codes.

Conflict of interest

The authors declare they have no conflict of interest.

Ahmed AAM, Bailek N, Abualigah L, Bouchouicha K, Kuriqi A, Sharifi A, Sareh P (2023) Global control of electrical supply: A variational mode decomposition- aided deep learning model forenergy consumption prediction. Energy Rep 10:2152–2165. https://doi.org/10.1016/j.egyr.2023.08.076
Agrell C, Larsson P, Okla L, Agrell J (2002) PCB congeners in precipitation, wash out ratios and depositional fluxes within the Baltic Sea region, Europe. Atmos Environ 36(2):371–383. https://doi.org/10.1016/S1352-2310(01)00228-X
Al-Qaili AH, Al-Mansour AI, Al-Solieman H, AlSharabi K (2024) RNN-based pavement moduli prediction for flexible pavement design enhancement. Case Stud Constr Mater 20. https://doi.org/10.1016/j.cscm.2023.e02811
Arfaeinia H, Asadgol Z, Ahmadi E, Seifi M, Moradi M, Dobaradaran S (2017) Characteristics, distribution and sources of polychlorinated biphenyls (PCBs) in coastal sediments from the heavily industrialized area of Asalouyeh, Iran. Water Sci Technol 76(11–12):3340–3350. https://doi.org/10.2166/wst.2017.500
Bradley PM, Journey CA, Romanok KM, Barber LB, Buxton HT, Foreman WT, Furlong ET, Glassmeyer ST, Hladik ML (2017) Expanded Target-Chemical Analysis Reveals Extensive Mixed-Organic-Contaminant Exposure in U.S. Streams. Environ Sci Technol 51(9):4792–4802. https://doi.org/10.1021/acs.est.7b00012
Cai J, Meng L, Liu H, Chen J, Xing Q (2022) Estimating Chemical Oxygen Demand in estuarine urban rivers using unmanned aerial vehicle hyperspectral images. Ecol Indic 139:108936. https://doi.org/10.1016/j.ecolind.2022.108936
Cho K, van Merrienboer B, Bahdanau D, Bengio Y (2014) On the Properties of Neural Machine Translation: Encoder-Decoder Approaches. In Proceedings of SSST@ EMNLP (pp. 103–111). https://doi.org/10.1016/j.jconhyd.2023.104287
Dragomiretskiy K, Zosso D (2014) Variational Mode Decomposition. IEEE Trans Signal Process 62(3):531–544. https://doi.org/10.1109/TSP.2013.2288675
Elmolla ES, Chaudhuri M, Eltoukhy MM (2010) The use of artificial neural network (ANN) for modeling of COD removal from antibiotic aqueous solution by the Fenton process. J Hazard Mater 179(1):127–134. https://doi.org/10.1016/j.jhazmat.2010.02.068
Esposito M, Canzanella S, Lambiase S, Scaramuzzo A (2020) Organic pollutants (PCBs, PCDD/Fs, PAHs) and toxic metals in farmed mussels from the Gulf of Naples (Italy): Monitoring and human exposure. Reg Stud Mar Sci 40:101497. https://doi.org/10.1016/j.rsma.2020.101497
Gholami H, Mohammadifar A, Golzari S, Song Y (2023) Interpretability of simple RNN and GRU deep learning models used to map land susceptibility to gully erosion. Sci Total Environ 904:166960. https://doi.org/10.1016/j.scitotenv.2023.166960
Glaser C, Escher BI, Engelhardt M, Liu Y, Krauss M (2023) Particle-associated organic contaminant and cytotoxicity transport in a river during storm events. J Hydrol 621:129600. https://doi.org/10.1016/j.jhydrol.2023.129600
Hochreiter S (1991) Investigations on dynamic neural networks. Dissertation, Johannes Kepler University, Linz
Hochreiter S, Schmidhuber J (1997) Long Short-Term Memory. Neural Comput 9(8):1735–1780. https://doi.org/10.1162/neco.1997.9.8.1735
Jiang J, Zhao J, Zhao G, Liu L (2023) Recognition, possible source, and risk assessment of organic pollutants in surface water from the Yongding River Basin by non-target and target screening. Environ Pollut 331:121895. https://doi.org/10.1016/j.envpol.2023.121895
Khanmohammadi M, Rahmani F, Shahrouzi JR, Sene RA (2024) Insightful properties-performance study of Ti–Cu–O heterojunction sonochemically embedded in mesoporous silica matrix for efficient tetracycline adsorption and photodegradation: RSM and ANN-based modeling and optimization. Chemosphere 141223. https://doi.org/10.1016/j.chemosphere.2024.141223
Kolb M, Bahadir M, Teichgräber B (2017) Determination of chemical oxygen demand (COD) using an alternative wet chemical method free of mercury and dichromate. Water Res 122:645–654. https://doi.org/10.1016/j.watres.2017.06.034
Kumar R, Gupta K, Bordoloi N (2022) The potential of engineered endophytic bacteria to improve phytoremediation of organic pollutants. In: Advances in Microbe-assisted Phytoremediation of Polluted Sites, 477–496. Elsevier. https://doi.org/10.1016/B978-0-12-823443-3.00014-4
Lan Y, Gao X, Xu H, Li M (2024) 20 years of polybrominated diphenyl ethers on toxicity assessments. Water Res 249:121007. https://doi.org/10.1016/j.watres.2023.121007
Li B, Zhang Z, Wang J, Chunli (2003) Chemiluminescence system for automatic determination of chemical oxygen demand using flow injection analysis. Talanta 61(5):651–658. https://doi.org/10.1016/S0039-9140(03)00326-6
Li J, Li F, Liu Q (2017) PAHs behavior in surface water and groundwater of the Yellow River estuary: Evidence from isotopes and hydrochemistry. Chemosphere 178:143–153. https://doi.org/10.1007/s11356-011-0572-7
Li J, Tong Y, Guan L, Wu S, Li D (2018) Optimization of COD determination by UV–vis spectroscopy using PLS chemometrics algorithms. Optik 174:591–599. https://doi.org/10.1016/j.ijleo.2018.08.111
Li R, Chen X, Balezentis T, Streimikiene D, Niu Z (2021) Multi-step least squares support vector machine modeling approach for forecasting short-term electricity demand with application. Neural Comput Appl 33:301–320. https://doi.org/10.1007/s00521-020-04996-3
Li WW, Sheng GP, Zeng RJ, Liu XW, Yu HQ (2012) China’s wastewater discharge standards in urbanization. Environ Sci Pollut Res 19(5):1422–1431. https://doi.org/10.1016/j.chemosphere.2017.03.052
Ling Q, Zhang Q, Zhang J, Kong L, Zhang W, Zhu L (2021) Prediction of landslide displacement using multikernel extreme learning machine and maximum information coefficient based on variational mode decomposition: a case study in Shaanxi, China. Nat Hazards 108(3):1–22. https://doi.org/10.1007/s11069-021-04713-w
Liu J, Yang F, Cai Y, Lu G, Li Y, Li M, Fan L, Gao L (2024) Unveiling the existence and ecological hazards of trace organic pollutants in wastewater treatment plant effluents across China. Eco-Environment Health 3(1):21–29. https://doi.org/10.1016/j.eehl.2023.09.006
Mallah MA, Changxing L, Mallah MA, Noreen S, Liu Y (2022) Polycyclic aromatic hydrocarbon and its effects on human health: An overview. Chemosphere 296:133948. https://doi.org/10.1016/j.chemosphere.2022.133948
Masouleh SY, Mozaffarian M, Dabir B, Ramezani SF (2022) COD and ammonia removal from landfill leachate by UV/PMS/Fe2 + process: ANN/RSM modeling and optimization. Process Saf Environ Prot 159:716–726. https://doi.org/10.1016/j.psep.2022.01.031
Miao S, Zhou C, AlQahtani SA, Alrashoud M (2021) Applying machine learning in intelligent sewage treatment: A case study of chemical plant in sustainable cities. Sustain Cities Soc 72:103009. https://doi.org/10.1016/j.scs.2021.103009
Muralikrishna IV, Manickam V (2017) Chapter One - Introduction. In: Environmental Management, Muralikrishna IV, Manickam V (eds) Butterworth-Heinemann, pp 1–4. https://doi.org/10.1016/B978-0-12-811989-1.00001-4
Nie C, Jiang J, Deng J, Li K, Jia L, Sun T (2023) Predicting TFe content and sorting iron ores from hyperspectral image by variational mode decomposition-based spectral feature. J Clean Prod 429:139629. https://doi.org/10.1016/j.jclepro.2023.139629
Qiao B, Liu J, Wu P, Teng Y (2022) Wind power forecasting based on variational mode decomposition and high-order fuzzy cognitive maps. Appl Soft Comput 129:109586. https://doi.org/10.1016/j.asoc.2022.109586
Shetty SS, Deepthi D, Harshitha S, Sonkusare S, Naik PB, Kumari SN, Madhyastha H (2023) Environmental pollutants and their effects on human health. Heliyon 9(9). https://doi.org/10.1016/j.heliyon.2023.e19496
Wang C, Li W, Huang M (2019) High precision wide range online chemical oxygen demand measurement method based on ultraviolet absorption spectroscopy and full-spectrum data analysis. Sens Actuators B Chem 300:126943. https://doi.org/10.1016/j.snb.2019.126943
Wang H, Kang C, Tian Z, Zhang A, Cao Y (2023) Vegetation periodic changes and relationships with climate in Inner Mongolia Based on the VMD method. Ecol Indic 146:109764. https://doi.org/10.1016/j.ecolind.2022.109764
Wang J, Xue B, Wang Y, Wang AY G (2024) Identification of pollution source and prediction of water quality based on deep learning techniques. J Contam Hydrol 261:104287. https://doi.org/10.1016/j.jconhyd.2023.104287
Wang X, Wu D, Yuan D, Wu X (2022) A nano-lead dioxide-composite electrochemical sensor for the determination of chemical oxygen demand. J Environ Chem Eng 10(3):107464. https://doi.org/10.1016/j.jece.2022.107464
Xu B, Pooi CK, Tan KM, Huang S, Shi X, Ng HY (2023) A novel long short-term memory artificial neural network (LSTM)-based soft-sensor to monitor and forecast wastewater treatment performance. J Water Process Eng 54:104041. https://doi.org/10.1016/j.jwpe.2023.104041
Xu W, Jiang Y, Zhang X, Li Y, Zhang R, Fu G (2020) Using long short-term memory networks for river flow prediction. Hydrol Res 51(6):1358–1376. https://doi.org/10.2166/nh.2020.026
Yuan J, Liu Y, Wang J, Zhao Y, Li K, Jing Y, Zhang X (2018) Long-term Persistent Organic Pollutants Exposure Induced Telomere Dysfunction and Senescence- Associated Secretary Phenotype. J Gerontol Biol Sci Med Sci 73(8):1027–1035. https://doi.org/10.1093/gerona/gly002
Yousefi S, Berrani S-A, Garcia C (2017) Contribution of recurrent connectionist language models in improving LSTM-based Arabic text recognition in videos. Pattern Recognit 64:245–254. https://doi.org/10.1016/j.patcog.2016.11.011
Zhang R, Li YS, Luo YX, Zhang XY, Wen R (2023) A carbon-dot fluorescence capillary sensor for the determination of chemical oxygen demand. Microchem J 187:108445. https://doi.org/10.1016/j.microc.2023.108445
Zhang R, Li YS, Luo YX, Zhang XY, Wen R (2023) A carbon-dot fluorescence capillary sensor for the determination of chemical oxygen demand. Microchem J 187:108445. https://doi.org/10.1016/j.microc.2023.108445
Zaynab M, Fatima M, Sharif Y, Sughra K, Sajid M, Khan KA, Sneharani AH, Li S (2021) Health and environmental effects of silent killers Organochlorine pesticides and polychlorinated biphenyl. J King Saud Univ Sci 33(6):101511. https://doi.org/10.1016/j.jksus.2021.101511
Zhang R, Li YS, Luo YX, Zhang XY, Wen R (2023) A carbon-dot fluorescence capillary sensor for the determination of chemical oxygen demand. Microchem J 187:108445. https://doi.org/10.1016/j.microc.2023.108445

No competing interests reported.

Download PDF

Editorial decision: Revision requested
10 Jul, 2024
Editor assigned by journal
10 Jul, 2024
Submission checks completed at journal
10 Jul, 2024
First submitted to journal
10 Jul, 2024

You are reading this latest preprint version

Driving analysis and prediction of COD based on frequency division

Status:

Version 1

Abstract

Figures

1 Introduction

2 Methodology

2.1 Used data parameters

2.2 Data processing

2.3Variational mode decomposition

3 Results and Discussion

3.1 VMD decomposition results

3.2Driving analysis of COD change

3.3 Simulation results with or without frequency division

3.3.1 Simulation results with Random Forest and VMD-Random Forest

3.3.2 Simulation results with GRU and VMD-GRU

3.3.3 Simulation results with LSTM and VMD-LSTM

4 Conclusions

Declarations

References

Additional Declarations

Status:

Version 1