Water quality soft-sensor prediction in anaerobic process using deep neural network optimized by Tree-structured Parzen Estimator

Anaerobic process is regarded as a green and sustainable process due to low carbon emission and minimal energy consumption in wastewater treatment plants(WWTPs).However,some water quality metrics are not measurable in real time,thus influencing the judgment of the operators and may increase energy consumption and carbon emission.One of the solutions is using a soft-sensor prediction technique.This article introduces a water quality soft-sensor prediction method based on Bidirectional Gated Recurrent Unit(BiGRU)combined with Gaussian Progress Regression(GPR)optimized by Tree-structured Parzen Estimator(TPE).TPE automatically optimizes the hyperparameters of BiGRU,and BiGRU is trained to obtain the point prediction with GPR for the interval prediction.Then,a case study applying this prediction method for an actual anaerobic process(2500 m^(3)/d)is carried out.Results show that TPE effectively optimizes the hyperparameters of BiGRU.For point prediction of CODeff and biogas yield,R^(2)values of BiGRU,which are 0.973 and 0.939,respectively,are increased by 1.03%–7.61%and 1.28%–10.33%,compared with those of other models,and the valid prediction interval can be obtained.Besides,the proposed model is assessed as a reliable model for anaerobic process through the probability prediction and reliable evaluation.It is expected to provide high accuracy and reliable water quality prediction to offer basis for operators in WWTPs to control the reactor and minimize carbon emission and energy consumption.

Machine learning and Bayesian optimization for performance prediction of proton-exchange membrane fuel cells

Proton-exchange membrane fuel cells (PEMFCs) are critical components of zero-emission electro-hydrogen generators. Accurate performance prediction is vital to the optimal operation management and preventive maintenance of these generators. Polarization curve remains one of the most important features representing the performance of PEMFCs in terms of efficiency and durability. However, predicting the polarization curve is not trivial as PEMFCs involve complex electrochemical reactions that feature multiple nonlinear relationships between the operating variables as inputs and the voltage as outputs. Herein, we present an artificial-intelligence-based approach for predicting the PEMFCs’ performance. In that way, we propose first an explainable solution for selecting the relevant features based on kernel principal component analysis and mutual information. Then, we develop a machine learning approach based on XGBRegressor and Bayesian optimization to explore the complex features and predict the PEMFCs’ performance. The performance and the robustness of the proposed machine learning based prediction approach is tested and validated through a real industrial dataset including 10 PEMFCs. Furthermore, several comparison studies with XGBRegressor and the two popular machine learning-based methods in predicting PEMFC performance, such as artificial neural network (ANN) and support vector machine regressor (SVR) are also conducted. The obtained results show that the proposed approach is more robust and outperforms the two conventional methods and the XGBRegressor for all the considered PEMFCs. Indeed, according to the coefficient of determination criterion, the proposed model gains an improvement of 6.35%, 6.8%, and 4.8% compared with ANN, SVR, and XGBRegressor respectively.

Impact of Autotuned Fully Connected Layers on Performance of Self-supervised Models for Image Classification

With the recent advancements of deep learning-based methods in image classification,the requirement of a huge amount of training data is inevitable to avoid overfitting problems.Moreover,supervised deep learning models require labelled datasets for train-ing.Preparing such a huge amount of labelled data requires considerable human effort and time.In this scenario,self-supervised models are becoming popular because of their ability to learn even from unlabelled datasets.However,the efficient transfer of knowledge learned by self-supervised models into a target task,is an unsolved problem.This paper proposes a method for the efficient transfer of know-ledge learned by a self-supervised model,into a target task.Hyperparameters such as the number of layers,the number of units in each layer,learning rate,and dropout are automatically tuned in these fully connected(FC)layers using a Bayesian optimization technique called the tree-structured parzen estimator(TPE)approach algorithm.To evaluate the performance of the proposed method,state-of-the-art self-supervised models such as SimClr and SWAV are used to extract the learned features.Experiments are carried out on the CIFAR-10,CIFAR-100,and Tiny ImageNet datasets.The proposed method outperforms the baseline approach with margins of 2.97%,2.45%,and 0.91%for the CIFAR-100,Tiny ImageNet,and CIFAR-10 datasets,respectively.