Interpretable attention-based multi-encoder transformer based QSPR model for assessing toxicity and environmental impact of chemicals.

The rising demand from consumer goods and pharmaceutical industry is driving a fast expansion of newly developed chemicals. The conventional toxicity testing of unknown chemicals is expensive, time-consuming, and raises ethical concerns. The quantitative structure-property relationship (QSPR) is an efficient computational method because it saves time, resources, and animal experimentation. Advances in machine learning have improved chemical analysis in QSPR studies, but the real-world application of machine learning-based QSPR studies was limited by the unexplainable 'black box' feature of the machine learnings. In this study, multi-encoder structure-to-toxicity (S2T)-transformer based QSPR model was developed to estimate the properties of polychlorinated biphenyls (PCBs) and endocrine disrupting chemicals (EDCs). Simplified molecular input line entry systems (SMILES) and molecular descriptors calculated by the Dragon 6 software, were simultaneously considered as input of QSPR model. Furthermore, an attention-based framework is proposed to describe the relationship between the molecular structure and toxicity of hazardous chemicals. The S2T-transformer model achieved the highest R2 scores of 0.918, 0.856, and 0.907 for logarithm of octanol-water partition coefficient (Log KOW), octanol-air partition coefficient (Log KOA), and bioconcentration factor (Log BCF) estimation of PCBs, respectively. Moreover, the attention weights were able to properly interpret the lateral (meta, para) chlorination associated with PCBs toxicity and environmental impact.

Soft sensor for predicting indoor PM2.5 concentration in subway with adaptive boosting deep learning model.

Public health depends on indoor air quality (IAQ), hence soft measurement techniques must be implemented in the subway environment for more precise and reliable monitoring of indoor particulate matter concentration levels. Adaptive boosting (AdaBoost), an ensemble learning technique, is simple to code and less prone to overfitting. Compared to a single model, it is better able to take into consideration the intricate elements included in air quality data. It is suggested to use an adaptive boosting of long short-term memory (AdaBoost-LSTM) model and kernel principal component analysis (KPCA) for ensemble learning. The kernel function and PCA are first coupled to create KPCA, which is a nonlinear dimensionality reduction method for IAQ. This removes the negative impacts of noise interference. The learning performance of LSTM is then enhanced using AdaBoost as an ensemble learning technique. The KPCA-AdaBoost-LSTM model can deliver higher modeling performance, according to the results. The R2 reached 0.9007 and 0.8995 when predicting PM2.5 in the hall and platform. SHapley Additive exPlanations (SHAP) analysis was used to interpret the input contributions of the model, enhancing the interpretability and transparency of the proposed soft sensor.

Sludge bulking monitoring in industrial wastewater treatment plants through graphical methods: A dynamic graph embedding and Bayesian networks approach.

Sludge bulking is a prevalent issue in wastewater treatment plants (WWTPs) that negatively impacts effluent quality by hindering the normal functioning of treatment processes. To tackle this problem, we propose a novel graph-based monitoring framework that employs advanced graph-based techniques to detect and diagnose sludge bulking events. The proposed framework utilizes historical datasets under normal operating conditions to extract pertinent features and causal relationships between process variables. This enables operators to trigger alarms and diagnose the root cause of the bulking event. Sludge bulking detection is carried out using the dynamic graph embedding (DGE) method, which identifies similarities among process variables in both temporal and neighborhood dependencies. Consequently, the dynamic Bayesian network (DBN) computes the prior and posterior probabilities of a belief, updated at each time step. Variations in these probabilities indicate the potential root cause of the sludge bulking event. The results demonstrate that the DGE outperforms other linear and non-linear feature extraction methods, achieving a detection rate of 99%, zero false alarms, and less than one percent incorrect detections. Additionally, the DBN-based diagnostic method accurately identified the majority of sludge bulking root causes, primarily those resulting from sudden drops in COD concentration, with an accuracy of 98% an improvement of 11% over state-of-the-art techniques.

Smart solutions for urban health risk assessment: A PM2.5 monitoring system incorporating spatiotemporal long-short term graph convolutional network.

Current spatial-temporal early warning systems aim to predict outdoor air quality in urban areas either at short or long temporal horizons. These systems implemented architectures without considering the geographical distribution of each air quality monitoring station, increasing the uncertainty of the forecasting framework. This study developed an integrated spatiotemporal forecasting architecture incorporating an extensive air quality PM2.5 monitoring network and simultaneously forecasts PM2.5 concentrations at all locations, allowing the monitoring of the health risk associated with exposure to these levels. First, this study uses a graph convolutional layer to incorporate the spatial relationship of the neighboring stations at their current state with real-time measurements. Then, it is coupled to a deep learning temporal model to form the long- and short-term time-series graph convolutional network (LSTGraphNet) model, anticipating high pollutant concentration events. This work tested the proposed model with a case study of an existing ambient air quality monitoring network in South Korea. LSTGraphNet model showed prediction performances of PM2.5 at multiple monitoring stations with a mean absolute error (MAE) of 1.82 µg/m3, 4.46 µg/m3, and 4.87 µg/m3 for forecasting horizons of one, three, and 6 h ahead, respectively. Compared to conventional sequential models, this architecture was superior among the state-of-the-art baselines, where the MAE decreased to 41%, respectively. The results of the study showed that the proposed architecture was superior to conventional sequential models and could be used as a tool for decision-making in smart cities by revealing hotspots of higher and lower PM2.5 concentrations in the long term.

Rethinking of the future sustainable paradigm roadmap for plastic waste management: A multi-nation scale outlook compendium.

The myriad consumption of plastic regularly, environmental impact and health disquietude of humans are at high risk. Along the line, international cooperation on a global scale is epitomized to mitigate the environmental threats from plastic usage, not limited to implementing international cooperation strategies and policies. Here, this study aims to provide explicit insight into possible cooperation strategies between countries on the post-treatment and management of plastic. First, a thorough cradle-to-grave assessment in terms of economic, environmental, and energy requirements is conducted on the entire life cycle across different types of plastic polymers in 6 main countries, namely the United States of America, China, Germany, Japan, South Korea, and Malaysia. Subsequently, P-graph is introduced to identify the integrative plastic waste treatment scheme that minimizes the economic, environmental, and energy criteria (1000 sets of solutions are found). Furthermore, TOPSIS analysis is also being adapted to search for a propitious solution with optimal balance between the dominant configuration of economic, environmental, and energy nexus. The most sustainable configuration (i.e., integrated downcycle and reuse routes in a closed loop system except in South Korea, which proposed another alternative to treat the plastic waste using landfill given the cheaper cost) is reported with 4.08 × 108 USD/yr, 1.76× 108 kg CO2/yr, and 2.73 × 109 MJ/yr respectively. To attain a high precision result, Monte-Carlo simulation is introduced (10,000 attempts) to search for possible uncertainties, and lastly, a potential global plastic waste management scheme is proposed via the PESTLE approach.

Data-Driven Prediction of Janus/Core-Shell Morphology in Polymer Particles: A Machine-Learning Approach.

The majority of research on Janus particles prepared by solvent evaporation-induced phase separation technique uses models based on interfacial tension or free energy to predict Janus/core-shell morphology. Data-driven predictions, in contrast, utilize multiple samples to identify patterns and outliers. Using machine-learning algorithms and explainable artificial intelligence (XAI) analysis, we developed a model based on a 200-instance data set to predict particle morphology. As model features, simplified molecular input line entry system syntax identifies explanatory variables, including cohesive energy density, molar volume, the Flory-Huggins interaction parameter of polymers, and the solvent solubility parameter. Our most accurate ensemble classifiers predict morphology with an accuracy of 90%. In addition, we employ innovative XAI tools to interpret system behavior, suggesting phase-separated morphology to be most affected by solvent solubility, polymer cohesive energy difference, and blend composition. While polymers with cohesive energy densities above a certain threshold favor the core-shell structure, systems with weak intermolecular interactions favor the Janus structure. The correlation between molar volume and morphology suggests that increasing the size of polymer repeating units favors Janus particles. Additionally, the Janus structure is preferred when the Flory-Huggins interaction parameter exceeds 0.4. XAI analysis introduces feature values that generate the thermodynamically low driving force of phase separation, resulting in kinetically stable morphologies as opposed to thermodynamically stable ones. The Shapley plots of this study also reveal novel methods for creating Janus or core-shell particles based on solvent evaporation-induced phase separation by selecting feature values that strongly favor a given morphology.

Fine Particulate Matter (PM2.5) Sources and Its Individual Contribution Estimation Using a Positive Matrix Factorization Model.

The effective management and regulation of fine particulate matter (PM2.5) is essential in the Republic of Korea, where PM2.5 concentrations are very high. To do this, however, it is necessary to identify sources of PM2.5 pollution and determine the contribution of each source using an acceptance model that includes variability in the chemical composition and physicochemical properties of PM2.5, which change according to its spatiotemporal characteristics. In this study, PM2.5 was measured using PMS-104 instruments at two monitoring stations in Bucheon City, Gyeonggi Province, from 22 April to 3 July 2020; the PM2.5 chemical composition was also analyzed. Sources of PM2.5 pollution were then identified and the quantitative contribution of each source to the pollutant mix was estimated using a positive matrix factorization (PMF) model. From the PMF analysis, secondary aerosols, coal-fired boilers, metal-processing facilities, motor vehicle exhaust, oil combustion residues, and soil-derived pollutants had average contribution rates of 5.73 µg/m3, 3.11 µg/m3, 2.14 µg/m3, 1.94 µg/m3, 1.87 µg/m3, and 1.47 µg/m3, respectively. The coefficient of determination (R2) was 0.87, indicating the reliability of the PMF model. Conditional probability function plots showed that most of the air pollutants came from areas where PM2.5-emitting facilities are concentrated and highways are present. Pollution sources with high contribution rates should be actively regulated and their management prioritized. Additionally, because automobiles are the leading source of artificially-derived PM2.5, their effective control and management is necessary.

A machine learning-driven spatio-temporal vulnerability appraisal based on socio-economic data for COVID-19 impact prevention in the U.S. counties.

A mature and hybrid machine-learning model is verified by mature empirical analysis to measure county-level COVID-19 vulnerability and track the impact of the imposition of pandemic control policies in the U.S. A total of 30 county-level social, economic, and medical variables and a timeline of the imposed policies constitutes a COVID-19 database. A hybrid feature-selection model composed of four machine-learning algorithms is developed to emphasize the regional impact of community features on the case fatality rate (CFR). A COVID-19 vulnerability index (COVULin) is proposed to measure the county's vulnerability, the effects of model's parameters on mortality, and the efficiency of control policies. The results showed that the dense counties in which minority groups represent more than 45% of the population and those with poverty rates greater than 24% were the most vulnerable counties during the first and the last pandemic peaks, respectively. Highly-correlated CFR and COVULin scores indicated a close agreement between the model outcomes and COVID-19 impacts. Counties with higher poverty and uninsured rates were the most resistant to government intervention. It is anticipated that the proposed model can play an essential role in identifying vulnerable communities and help reduce damages during long-term alike disasters.

Dynamic calibration of process-wide partial-nitritation modeling with airlift granular for nitrogen removal in a full-scale wastewater treatment plant.

A main challenge in rapid nitrogen removal from rejected water in wastewater treatment plants (WWTPs) is growth of biomass by nitrite-oxidizing bacteria (NOB) and ammonia-oxidizing bacteria (AOB). In this study, partial nitritation (PN) coupled with air-lift granular unit (AGU) technology was applied to enhance nitrogen-removal efficiency in WWTPs. For successful PN process at high-nitrogen-influent conditions, a pH of 7.5-8 for high free-ammonia concentrations and AOB for growth of total bacterial populations are required. The PN process in a sequential batch reactor (SBR) with AGU was modeled as an activated sludge model (ASM), and dynamic calibration using full-scale plant data was performed to enhance aeration in the reactor and improve the nitrite-to-ammonia ratio in the PN effluent. In steady-state and dynamic calibrations, the measured and modeled values of the output were in close agreement. Sensitivity analysis revealed that the kinetic and stoichiometric parameters are associated with growth and decay of heterotrophs, AOB, and NOB microorganisms. Overall, 80% of the calibrated data fit the measured data. Stage 1 of the dynamic calibration showed NO2 and NO3 values close to 240 mg/L and 100 mg/L, respectively. Stage 2 showed NH4 values of 200 mg/L at day 30 with the calibrated effluent NO2 and NO3 value of 250 mg/L. In stage 3, effluent NH4 concentration was 200 mg/L at day 60.

Non-Gaussian multivariate statistical monitoring of spatio-temporal wind speed frequencies to improve wind power quality in South Korea.

The spatial and temporal variability of renewable energy resources, particularly wind energy, should be statistically evaluated to achieve sustainable economic development to mitigate climate change. In this study, a non-Gaussian multivariate statistical monitoring approach is proposed to investigate the wind speed frequencies across different regions of South Korea. Anemometer data were first collected in 11 different provinces of South Korea with hourly resolution for one year. The best-of-fit for the corresponding distribution function was identified to characterize the behavior of the wind speed frequency at each region among more than 60 candidate functions using the chi-squared test. Furthermore, a non-Gaussian multivariate statistical monitoring method based on the Hotelling T2 chart was developed to spatially and temporally analyze the physical patterns of the wind speed frequencies using the estimated distribution parameters. Then determination rule of cut-in and cut-out speeds of wind turbine was suggested to improve the wind power quality across the regions. The results indicated that Weibull and Gamma distributions are best-of-fit functions of each province in South Korea; the physical patterns of wind including the average wind speed and gale can be identified by distribution parameters. Furthermore, the proposed non-Gaussian multivariate monitoring approach can elucidate the spatial and temporal variability of the regional wind speed frequencies, including the average wind speeds and extreme wind events across South Korea. Based on the statistically identified variability of wind behavior, the wind power quality of wind turbines can be improved by 12% on average by adjusting the cut-in and cut-off speed. Thus, the proposed non-Gaussian multivariate monitoring approach can provide practical guidelines for manufacturers to achieve reliable wind energy generation by considering the spatial and temporal wind behavior.

Missing data imputation and sensor self-validation towards a sustainable operation of wastewater treatment plants via deep variational residual autoencoders.

Missing data imputation and automatic fault detection of wastewater treatment plant (WWTP) sensors are crucial for energy conservation and environmental protection. Given the dynamic and non-linear characteristics of WWTP measurements, the conventional diagnosis models are inefficient and ignore potential valuable features in the offline modeling phase, leading to false alarms and inaccurate imputations. In this study, an inclusive framework for missing data imputation and sensor self-validation based on integrating variational autoencoders (VAE) with a deep residual network structure (ResNet-VAE) is proposed. This network structure can automatically extract complex features from WWTP data without the risk of vanishing gradients by learning the potential probability distribution of the input data. The proposed framework is intended to increase the reliability of faulty sensors by imputing missing data, detecting anomalies, identifying failure sources, and reconstructing faulty data to normal conditions. Several metrics were utilized to assess the performance of the suggested approach in comparison with other different methods. The VAE-ResNet approach showed superiority to detect (DRSPE = 100%), reconstruct faulty WWTP sensors (MAPE = 15.41%-5.68%) and impute the missing values (MAPE = 10.44%-3.98%). Lastly, the consequences of faulty data, missing data, reconstructed and imputed data were evaluated considering electricity consumption and resilience to demonstrate the ResNet-VAE model's superior performance for WWTP sustainability.

Transfer learning driven sequential forecasting and ventilation control of PM2.5 associated health risk levels in underground public facilities.

Particulate matter with aerodynamic diameter less than 2.5 µm (PM2.5) has become a major public concern in closed indoor environments, such as subway stations. Forecasting platform PM2.5 concentrations is significant in developing early warning systems, and regulating ventilation systems to ensure commuter health. However, the performance of existing forecasting approaches relies on a considerable amount of historical sensor data, which is usually not available in practical situations due to hostile monitoring environments or newly installed equipment. Transfer learning (TL) provides a solution to the scant data problem, as it leverages the knowledge learned from well-measured subway stations to facilitate predictions on others. This paper presents a TL-based residual neural network framework for sequential forecast of health risk levels traced by subway platform PM2.5 levels. Experiments are conducted to investigate the potential of the proposed methodology under different data availability scenarios. The TL-framework outperforms the RNN structures with a determination coefficient (R2) improvement of 42.84%, and in comparison, to stand-alone models the prediction errors (RMSE) are reduced up to 40%. Additionally, the forecasted data by TL-framework under limited data scenario allowed the ventilation system to maintain IAQ at healthy levels, and reduced PM2.5 concentrations by 29.21% as compared to stand-alone network.

An autonomous operational trajectory searching system for an economic and environmental membrane bioreactor plant using deep reinforcement learning.

Optimal operation of membrane bioreactor (MBR) plants is crucial to save operational costs while satisfying legal effluent discharge requirements. The aeration process of MBR plants tends to use excessive energy for supplying air to micro-organisms. In the present study, a novel optimal aeration system is proposed for dynamic and robust optimization. Accordingly, a deep reinforcement learning (DRL)-based optimal operating system is proposed, so as to meet stringent discharge qualities while maximizing the system's energy efficiency. Additionally, it is compared with the manual system and conventional reinforcement learning (RL)-based systems. A deep Q-network (DQN) algorithm automatically learns how to operate the plant efficiently by finding an optimal trajectory to reduce the aeration energy without degrading the treated water quality. A full-scale MBR plant with the DQN-based autonomous aeration system can decrease the MBR's aeration energy consumption by 34% compared to other aeration systems while maintaining the treatment efficiency within effluent discharge limits.

Population response modeling and habitat suitability of Cobitis choii fish species in South Korea for climate change adaptation.

Endangered species ecosystems require appropriate monitoring for assessing population growth related to the emerging pollutants in their habitat conditions. The response of population growth of Cobitis choii, an endangered fish species, under the exposure to emerging pollutants present in the Geum River Basin of South Korea was studied. Toxicity models of concentration addition (CA), independent action (IA), and concentration addition-independent action (CAIA) were implemented utilizing the concentration of a set of 25 chemicals recorded in the study area. Thus, a population-level response analysis was developed based on the abundance of Cobitis choii for period 2011-2015. The results were compared showing that the CA and IA models were the most conservative approaches for the prediction of growth rate. Further, a standard abnormality index (SAI) and habitat suitability (HS) indicators based on the climate, habitat, and abundance data were presented to completely analyze the population growth of the species. Suitability of the species growth was most probable for year 2015 for the variables of air temperature and land surface temperature. A spatial analysis was complementarily presented to visualize the correlation of variables for the best suitability of the species growth. This study presents a methodology for the analysis of the ecosystem's suitability for Cobitis choii growth and its assessment of the chemicals present in Geum River stream.

Multivariate statistical monitoring of subway indoor air quality using dynamic concurrent partial least squares.

To maintain the health level of indoor air quality (IAQ) in subway stations, the data-driven multivariate statistical method concurrent partial least squares (CPLS) has been successfully applied for output-relevant and input-relevant sensor faults detection. To cope with the dynamic problem of IAQ data, the augmented matrices are applied to CPLS (DCPLS) to achieve the better performance. DCPLS method simultaneously decomposes the input and output data spaces into five subspaces for comprehensive monitoring: a joint input-output subspace, an output principal subspace, an output-residual subspace, an input-principal subspace, and an input-residual subspace. Results of using the underground IAQ data in a subway station demonstrate that the monitoring capability of DCPLS is superior than those of PLS and CPLS. More specifically, the fault detection rates of the bias of PM10 and PM2.5 using DCPLS can be improved by approximately 13% and 15%, respectively, in comparison with those of CPLS.

Deep learning driven QSAR model for environmental toxicology: Effects of endocrine disrupting chemicals on human health.

Over 80,000 endocrine-disrupting chemicals (EDCs) are considered emerging contaminants (ECs), which are of great concern due to their effects on human health. Quantitative structure-activity relationship (QSAR) models are a promising alternative to in vitro methods to predict the toxicological effects of chemicals on human health. In this study, we assessed a deep-learning based QSAR (DL-QSAR) model to predict the qualitative and the quantitative effects of EDCs on the human endocrine system, and especially sex-hormone binding globulin (SHBG) and estrogen receptor (ER). Statistical analyses of the qualitative responses indicated that the accuracies of all three DL-QSAR methods were above 90%, and greater than the other statistical and machine learning models, indicating excellent classification performance. The quantitative analyses, as assessed using deep-neural-network-based QSAR (DNN-QSAR), resulted in a coefficient of determination (R2) of 0.80 and predictive square correlation coefficient (Q2) of 0.86, which implied satisfactory goodness of fit and predictive ability. Thus, DNN was able to transform sparse molecular descriptors into higher dimensional spaces, and was superior for assessment qualitative responses. Moreover, DNN-QSAR demonstrated excellent performance in the discipline of computational chemistry by handling multicollinearity and overfitting problems.

An insight into nanocellulose as soft condensed matter: Challenge and future prospective toward environmental sustainability.

Nanocellulose, a structural polysaccharide that has caught tremendous interests nowadays due to its renewability, inherent biocompatibility and biodegradability, abundance in resource, and environmental friendly nature. They are promising green nanomaterials derived from cellulosic biomass that can be disintegrated into cellulose nanofibrils (CNF) or cellulose nanocrystals (CNC), relying on their sensitivity to hydrolysis at the axial spacing of disordered domains. Owing to their unique mesoscopic characteristics at nanoscale, nanocellulose has been widely researched and incorporated as a reinforcement material in composite materials. The world has been consuming the natural resources at a much higher speed than the environment could regenerate. Today, as an uprising candidate in soft condensed matter physics, a growing interest was received owing to its unique self-assembly behaviour and quantum size effect in the formation of three-dimensional nanostructured material, could be utilised to address an increasing concern over global warming and environmental conservation. In spite of an emerging pool of knowledge on the nanocellulose downstream application, that was lacking of cross-disciplinary study of its role as a soft condensed matter for food, water and energy applications toward environmental sustainability. Here we aim to provide an insight for the latest development of cellulose nanotechnology arises from its fascinating physical and chemical characteristic for the interest of different technology holders.

Sequential prediction of quantitative health risk assessment for the fine particulate matter in an underground facility using deep recurrent neural networks.

Particulate matter with aerodynamic diameter less than 2.5 µm (PM2.5) in indoor public spaces such as subway stations, has represented a major public health concern; however, forecasting future sequences of quantitative health risk is an effective method for protecting commuters' health, and an important tool for developing early warning systems. Despite the existence of several predicting methods, some tend to fail to forecast long-term dependencies in an effective way. This paper aims to implement a multiple sequences prediction of a comprehensive indoor air quality index (CIAI) traced by indoor PM2.5, utilizing different structures of recurrent neural networks (RNN). A standard RNN (SRNN), long short-term memory (LSTM) and a gated recurrent unit (GRU) structures were implemented due to their capability of managing sequential, and time-dependent data. Hourly indoor PM2.5 concentration data collected in the D-subway station, South Korea, were utilized for the validation of the proposed method. For the selection of the most suitable predictive model (i.e. SRNN, LSTM, GRU), a point-by-point prediction on the PM2.5 was conducted, demonstrating that the GRU structure outperforms the other RNN structures (RMSE = 21.04 µg/m3, MAPE = 32.92%, R2 = 0.65). Then, this model is utilized to sequentially predict the concentration and quantify the health risk (i.e. CIAI) at different time lags. For a 6-h time lag, the proposed model exhibited the best performance metric (RMSE = 29.73 µg/m3, MAPE = 29.52%). Additionally, for the rest of the time lags including 12, 18 and 24 h, achieved an acceptable performance (MAPE = 29-37%).

A real time QSAR-driven toxicity evaluation and monitoring of iron containing fine particulate matters in indoor subway stations.

A fine particulate matter less than 2.5 µm (PM2.5) in the underground subway system are the cause of many diseases. The iron containing PMs frequently confront in underground stations, which ultimately have an impact on the health of living beings especially in children. Hence, it is necessary to conduct toxicity assessment of chemical species and regularized the indoor air pollutants to ensure the good health of children. Therefore, in this study, a new indoor air quality (IAQ) index is proposed based on toxicity assessment by quantitative structure-activity relationship (QSAR) model. The new indices called comprehensive indoor air toxicity (CIAT) and cumulative comprehensive indoor air toxicity (CCIAT) suggests the new standards based on toxicity assessment of PM2.5. QSAR based deep neural network (DNN) exhibited the best model in predicting the toxicity assessment of chemical species in particulate matters, which yield lowest RMSE and QF32 values of 0.6821 and 0.8346, respectively, in the test phase. After integration with a standard concentration of PM2.5, two health risk indices of CIAT and CCIAT are introduced based on toxicity assessment results, which can be use as the toxicity standard of PM2.5 for detail IAQ management in a subway station. These new health risk indices suggest more sensitive air pollutant level of iron containing fine particulate matters or molecular level contaminants in underground spaces, alerting the health risk of adults and children in "unhealthy for sensitive group".

Modeling of subway indoor air quality using Gaussian process regression.

Soft sensor modeling of indoor air quality (IAQ) in subway stations is essential for public health. Gaussian process regression (GPR), as an efficient nonlinear modeling method, can effectively interpret the complicated features of industrial data by using composite covariance functions derived from base kernels. In this work, an accurate GPR soft sensor with the sum of squared-exponential covariance function and periodic covariance function is proposed to capture the time varying and periodic characteristics in the subway IAQ data. The results demonstrate that the prediction performance of the proposed GPR model is superior to that of the traditional soft sensors consisting of partial least squares, back propagation artificial neural networks, and least squares support vector regression (LSSVR). More specifically, the values of root mean square error, mean absolute percentage error, and coefficient of determination are improved by 12.35%, 9.53%, and 40.05%, respectively, in comparison with LSSVR.