Average Coefficient Of Determination Research Articles

Dynamic Modeling of Bacterial Cellulose Production Using Combined Substrate- and Biomass-Dependent Kinetics

In this work, kinetic models are assessed to describe bacterial cellulose (BC) production, substrate consumption, and biomass growth by K. xylinus in a batch-stirred tank bioreactor, under 700 rpm and 500 rpm agitation rates. The kinetic models commonly used for Acetobacter or Gluconacetobacter were fitted to published data and compared using the Akaike Information Criterion (AIC). A stepwise fitting procedure was proposed for model selection to reduce computation effort, including a first calibration in which only the biomass and substrate were simulated, a selection of the three most effective models in terms of AIC, and a calibration of the three selected models with the simulation of biomass, substrate, and product. Also, an uncoupled product equation involving a modified Monod substrate function is proposed for a 500 rpm agitation rate, leading to an improved prediction of BC productivity. The M2c and M1c models were the most efficient for biomass growth and substrate consumption for the combined AIC, under 700 rpm and 500 rpm agitation rates, respectively. The average coefficients of determination for biomass, substrate, and product predictions were 0.981, 0.994, and 0.946 for the 700 rpm agitation rate, and 0.984, 0.991, and 0.847 for the 500 rpm agitation rate. It is shown that the prediction of BC productivity is improved through the proposed substrate function, whereas the computation effort is reduced through the proposed model fitting procedure.

Strain distribution prediction in UHPC beams using deep learning model

AbstractOver the last three decades, ultra‐high‐performance concrete (UHPC) has emerged as a highly innovative cementitious engineering material. This paper utilizes a distributed fiber optic sensor based on optical frequency domain reflectometry (OFDR) combined with the deep learning model to monitor and predict the strain distribution in UHPC T‐beams subjected to two point bending experiment. These sensors are deployed on the side surfaces of the UHPC T‐beams to capture the strain distribution when subjected to vertical loads, facilitating the determination of crack locations based on strain distribution peaks. While time series modeling is widely used in civil engineering to monitor potential damage using sensor data, its application in tracking and predicting known cracks is less explored. To address this gap, a Long Short‐Term Memory (LSTM) neural network model is developed to forecast the increase in the peak value of the strain distribution prior to structural damage, thus predicting crack initiation. The accuracy of the proposed LSTM model in predicting the UHPC strain distribution was thoroughly investigated. The results demonstrate excellent agreement between the predicted strain distributions and those detected by the ODISI 6000 monitoring system. The root‐mean‐square errors of the model‐predicted strains are generally below 10 με, with an average coefficient of determination (R2) reaching up to 98%, indicating a high degree of accuracy.

An integrated BLS-Net optimized with dual PCA and improved PSO-CARS variable selection strategy to determine heavy metals in soil by XRF

Combining machine learning with X-ray fluorescence (XRF) spectroscopy is a promising solution for quantitatively analyzing heavy metal elements in soil. However, the implied linear and nonlinear interferences between spectral intensities and elemental concentrations are difficult to quantify by a single model, thus degrading the prediction performance for low-concentration elements. This paper presents a novel combined spectral variable selection and fusion modeling framework for quantitative analysis of heavy metal elements in soil XRF, which consists of the proposed Particle Swarm Optimization-based Competitive Adaptive Re-weighted Sampling (PSO-CARS) by adaptive decay strategy and Dual Principal Component Analysis-based broad learning system (BLS-Net). The proposed method is compared with advanced machine learning methods by predicting the concentrations of Cr, Cd, Cu, and Pb. The results show that PSO-CARS provides more efficient characteristic and auxiliary spectral variables for BLS-Net. BLS-Net precisely quantified concentrations of heavy metal elements by distinguishing between linear and nonlinear responses in spectra, resulting in average predictive coefficients of determination (R2) exceeding 0.995 of the different elements. The proposed method is a competitive spectral quantitative fusion modeling solution, offering a more precise and reliable analysis of heavy metal elements.

Quickly Calculating the Activity Coefficient of a NaCl Solution Based on Machine Learning Algorithms.

The activity coefficient represents the deviation between an actual solution and an ideal solution, influencing the solubility and diffusion of CO2 within a saltwater layer. Consequently, it serves as a crucial parameter for numerical simulations of CO2 storage in deep saltwater layers. However, in numerical simulations of CO2 geological storage, the majority of studies rely on the Helgeson-Kirkham-Flowers (HKF) equation to compute activity coefficients, which necessitates obtaining Debye-Hückel (DH) parameters. The conventional method calculates the DH parameters via an interpolation algorithm, which requires a long computation time during the numerical simulation. Therefore, developing a method to quickly and accurately calculate activity coefficients is vital for the overall model efficiency. This study employed machine learning algorithms to train DH parameters derived from the IAPWS-95 method. It could establish empirical formulas for DH parameters as functions of temperature and pressure, which were then substituted into the HKF equation to swiftly compute activity coefficients. The results demonstrate that the activity coefficients obtained using this method exhibit a small relative deviation from experimental values, with an average coefficient of determination of 0.9463 and an average relative error of 2.28%. Furthermore, the computational speed was improved by 48%. This approach reduces the calculation time for activity coefficients in geochemical reaction modeling, enabling DH parameters to be calculated solely based on temperature and pressure, which is easy to use and has high accuracy. It facilitates rapid calculation of activity coefficients for solutions within a temperature range of 0 to 300 °C and a pressure range of 0 to 200 MPa. Ultimately, this study holds significant importance for the numerical simulation of geochemical reactions.

Combining residual convolutional LSTM with attention mechanisms for spatiotemporal forest cover prediction

Understanding spatiotemporal variations in forest cover is crucial for effective forest resource management. However, existing models often lack accuracy in simultaneously capturing temporal continuity and spatial correlation. To address this challenge, we developed ResConvLSTM-Att, a novel hybrid model integrating residual neural networks, Convolutional Long Short-Term Memory (ConvLSTM) networks, and attention mechanisms. We evaluated ResConvLSTM-Att against four deep learning models: LSTM, combined convolutional neural network and LSTM (CNN-LSTM), ConvLSTM, and ResConvLSTM for spatiotemporal prediction of forest cover in Tasmania, Australia. ResConvLSTM-Att achieved outstanding prediction performance, with an average root mean square error (RMSE) of 6.9% coverage and an impressive average coefficient of determination of 0.965. Compared with LSTM, CNN-LSTM, ConvLSTM, and ResConvLSTM, ResConvLSTM-Att achieved RMSE reductions of 31.2%, 43.0%, 10.1%, and 6.5%, respectively. Additionally, we quantified the impacts of explanatory variables on forest cover dynamics. Our work demonstrated the effectiveness of ResConvLSTM-Att in spatiotemporal data modelling and prediction.

Integrating material flow analysis into hydrological model for water environment management of large-scale urban-rural mixed catchment

Simultaneous simulation of urban and rural hydrological processes is important for water environment management of mixed land-uses catchments. However, the discharge paths of pollution in the urban drainage system are not described in traditional catchment hydrological models. In this study, an urban-rural water environment (URWE) model is developed through incorporating the material flow analysis (MFA) and the soil and water assessment tool (SWAT) into a general framework. The URWE model is an advancement with respect to traditional hydrological models in terms of simultaneously simulating the urban organized and rural decentralized discharges of pollution. Due to the low data requirement and high computational efficiency of MFA, URWE model is applicable to large-scale catchment with wide urban area. The URWE model is applied to a typical urban-rural mixed catchment, the Dianchi Catchment (China), where the pollution characteristics are analyzed and the pollution control measures are investigated. Results indicate that the URWE model outperforms the conventional SWAT model for both water quantity and quality simulations, with an 8.5 % improvement in average coefficient of determination (R2) and a 67.4 % improvement in average Nash coefficient (NSE). Rural best management practice, rainwater-sewage separation, and storage capacity expansion are identified as the most cost-effective measures for COD, TN, and TP reduction, respectively. Contributions of this study are to improve the accuracy of water environment simulation in urban-rural mixed catchment, as well as to help decision-makers develop synergistic urban-rural water environment management measures.

Effect of preprocessing on performances of machine learning-based mineral composition analysis on gas hydrate sediments, Ulleung Basin, East Sea

Gas hydrate (GH) is an unconventional resource estimated at 1,000−120,000 trillion m3 worldwide. Research on GH is ongoing to determine its geological and flow characteristics for commercial production. After two large-scale drilling expeditions to study the GH-bearing zone in the Ulleung Basin, the mineral composition of 488 sediment samples was analyzed using X-ray diffraction (XRD). Because the analysis is costly and dependent on experts, a machine learning model was developed to predict the mineral composition using XRD intensity profiles as input data. However, the model's performance was limited because of improper preprocessing of the intensity profile. Because preprocessing was applied to each feature, the intensity trend was not preserved even though this factor is the most important when analyzing mineral composition. In this study, the profile was preprocessed for each sample using min-max scaling because relative intensity is critical for mineral analysis. For 49 test data among the 488 data, the convolutional neural network (CNN) model improved the average absolute error and coefficient of determination by 41% and 46%, respectively, than those of CNN model with feature-based preprocessing. This study confirms that combining preprocessing for each sample with CNN is the most efficient approach for analyzing XRD data. The developed model can be used for the compositional analysis of sediment samples from the Ulleung Basin and the Korea Plateau. In addition, the overall procedure can be applied to any XRD data of sediments worldwide.

A dynamic state evaluation method for current sensors based on reference values and time series

Abstract Due to the complex and ever-changing power environment, when designing dynamic state evaluation methods for current sensors, there is often a problem of inaccurate calculation of state evaluation values, leading to poor evaluation accuracy of the methods. A dynamic state evaluation method for current sensors based on reference values and time series is proposed. Firstly, we normalize the time series data of the current sensor, construct a covariance matrix, calculate the degree of correlation between the time series data, and construct a higher quality current sensor state dataset. Then, we select multiple indicators as state evaluation indicators and quantify them. Then, the feature vector of the evaluation indicators and the state evaluation value are calculated. Under the influence of the reference value, we set the threshold of the current sensor in different states to divide the current sensor into different states. Through the above design, the dynamic state evaluation method for current sensors is completed. The experimental results show that the designed dynamic state evaluation method for current sensors based on reference values and time series has an average coefficient of determination of 0.95 in practical applications, with high evaluation accuracy.

Precise solar radiation forecasting for sustainable energy integration: A hybrid CEEMD-SCM-GA-LGBM model for day-ahead power and hydrogen production

In response to the critical need for sustainable energy solutions, this study introduces a groundbreaking hybrid model for enhancing solar radiation forecasting, which is crucial for optimizing solar power integration and hydrogen production. Leveraging a sophisticated amalgamation of Ensemble Empirical Mode Decomposition (EEMD), Sample Entropy-based System Clustering Method (SCM), Genetic Algorithms (GA), and Light Gradient Boosting Machine (LGBM), the model sets a new benchmark in forecasting accuracy. Tested across diverse seasonal data from Jiangsu province, China, it outperforms conventional models like Multi-Layer Perceptron (MLP) and Long Short-Term Memory (LSTM) networks, achieving exceptional accuracy with an average coefficient of determination (R2) of 0.99, root mean square error (RMSE) of 31.17, and mean absolute error (MAE) of 24.03. The model's practical efficacy was validated through a simulated system comprising a photovoltaic array, a DC-DC converter, and an Alkaline electrolyzer, demonstrating a peak hydrogen production of 0.25 kg at 2:30 p.m. on June 18. This indicates the model's superior capability to forecast solar irradiance accurately, which is essential for enhancing the efficiency of solar energy systems and hydrogen generation. The CEEMD-SCM-GA-LGBM hybrid model's adoption is advocated for solar radiation forecasting, promising significant advancements in the integration of solar energy and the optimization of hydrogen production, aligning with global sustainability objectives.

Long-term price guidance mechanism for integrated energy systems based on gated recurrent unit - vision transformer prediction and fractional-order stochastic dynamic calculus control

Against the backdrop of global climate change, renewable energy (RE) performs significant effects in reducing carbon emissions. Nevertheless, volatile RE sources disrupt the energy supply-demand balance (ESDB) in the energy market. Flexible energy (FE) can effectively mitigate the problem with high flexibility and controllability. FE sources are widely distributed in integrated energy systems (IESs). However, existing methods guide scattered FE in IESs with insufficient accuracy. To accurately guide FE, this study proposes an improved long-term price guidance mechanism (ILPGM) based on gated recurrent unit (GRU)- vision Transformer (ViT) prediction and fractional-order stochastic dynamic calculus (FSDC) control. The ILPGM combines GRU and ViT for baseline load forecasting and applies the FSDC method. The ILPGM can optimize the energy consumption of IESs and promote the ESDB while decreasing energy costs. The ILPGM is applied to the IES at Arizona State University-Tempe. The experimental results indicate that: the average coefficient of determination of electrical and heat energy under ILPGM is 97.69 %, realizing flexible regulation of energy consumption; the ILPGM can save 26.36 % actual cost and 94.62 % potential cost savings for the IES; compared to the previous long-term price guidance mechanism, the ILPGM provides an additional 5.87 % total cost savings and 28.65 % potential savings.

Experimental and numerical evaluation of soil water and salt dynamics in a corn field with shallow saline groundwater and crop-season drip and autumn post-harvest irrigations

In areas with shallow saline groundwater, soil salts inevitably accumulate in the root zone during the growth period due to irrigation and upward movement of salts from the groundwater. In Northern China, autumn irrigation (AIR) with large amounts of water is commonly employed post-harvest to mitigate soil salt stress on crop growth in the subsequent year. Optimizing the total irrigation depth during both crop-growth and non-growth periods is challenging because of the movement of soil salts, which is influenced by their two-dimensional distribution around drippers and the impact of the winter freeze-thaw cycles, significantly affecting water flow and solute transport during winter. In this study, the HYDRUS-1D and HYDRUS-2D models were integrated and calibrated using experimental data collected from 2021 to 2023 in China's Ordos south bank irrigation area. This model integration was conducted to assess soil water and salt dynamics during the non-growth and corn-growth periods under different irrigation strategies: a) AIR with high (AH) and low (AL) irrigation depths, and b) drip irrigation (DIR) with high (DH), medium (DM), and low (DL) irrigation depths. The results indicated that HYDRUS effectively modeled the electrical conductivity of the saturation paste extract (ECe) across different irrigation strategies, yielding an average coefficient of determination (R2) and the root mean square errors (RMSE) of 0.87 and 0.53 dS m−1, respectively. Generally, ECe increased during the growth period with DIR and decreased during the non-growth period with AIR. For the 0–40 cm soil layer, ECe decreased by 5.7 % and 12 % for every 100 mm increase in the AIR and DIR depths, respectively. Compared with the AHDM and AHDL treatments, reducing an AIR depth and increasing a DIR depth resulted in lower ECe in the 0–40 cm layer during the growth period and higher crop yield (CY) and irrigation water productivity (WPI). Specifically, the average ECe in the 0–40 cm layer decreased by 4.8 % during the growth period in the ALDH treatment compared to the AHDM treatment, and CY and WPI increased by 7.2 % and 10.3 %, respectively. Additionally, the irrigation strategy was the most effective in reducing ECe when AIR accounted for 35 % of the total irrigation. This study suggested that combining low AIR and high DIR could enhance water and field productivity.

Modelling of Steady-State Seepage of an Embankment Dam Using Teaching-Learning Based Optimization Algorithm

The goal of the this study is to investigate the applicability of the teaching-learning based optimization (TLBO) algorithm for modeling seepage in embankment dams. The input parameters selected for the models to be built are the values of permeability (ks), van Genuchten's suitability parameters α and n, whose effect on seepage has been investigated over the years due to their uncertainties. The validity of the TLBO was compared with that of conventional regression analysis (CRA) methods. Both methods were utilized with different regression forms. The parameters chosen as input are modeled as random variables with a log-normal distribution, and total discharge (Q) was obtained. Four statistical indices, that is, root mean square error, mean absolute error, average relative error and coefficient of determination, were used to evaluate the performance of the models. The equations obtained using TLBO algorithms can predict the total discharge in embankment dams better than CRA. In addition, the reliability of TLBO has been demonstrated by conducting analyses using the outputs of CRA as a benchmark.

Neural decoding and feature selection methods for closed-loop control of avoidance behavior

Objective.Many psychiatric disorders involve excessive avoidant or defensive behavior, such as avoidance in anxiety and trauma disorders or defensive rituals in obsessive-compulsive disorders. Developing algorithms to predict these behaviors from local field potentials (LFPs) could serve as the foundational technology for closed-loop control of such disorders. A significant challenge is identifying the LFP features that encode these defensive behaviors.Approach.We analyzed LFP signals from the infralimbic cortex and basolateral amygdala of rats undergoing tone-shock conditioning and extinction, standard for investigating defensive behaviors. We utilized a comprehensive set of neuro-markers across spectral, temporal, and connectivity domains, employing SHapley Additive exPlanations for feature importance evaluation within Light Gradient-Boosting Machine models. Our goal was to decode three commonly studied avoidance/defensive behaviors: freezing, bar-press suppression, and motion (accelerometry), examining the impact of different features on decoding performance.Main results.Band power and band power ratio between channels emerged as optimal features across sessions. High-gamma (80-150 Hz) power, power ratios, and inter-regional correlations were more informative than other bands that are more classically linked to defensive behaviors. Focusing on highly informative features enhanced performance. Across 4 recording sessions with 16 subjects, we achieved an average coefficient of determination of 0.5357 and 0.3476, and Pearson correlation coefficients of 0.7579 and 0.6092 for accelerometry jerk and bar press rate, respectively. Utilizing only the most informative features revealed differential encoding between accelerometry and bar press rate, with the former primarily through local spectral power and the latter via inter-regional connectivity. Our methodology demonstrated remarkably low training/inference time and memory usage, requiring<310 ms for training,<0.051 ms for inference, and 16.6 kB of memory, using a single core of AMD Ryzen Threadripper PRO 5995WX CPU.Significance.Our results demonstrate the feasibility of accurately decoding defensive behaviors with minimal latency, using LFP features from neural circuits strongly linked to these behaviors. This methodology holds promise for real-time decoding to identify physiological targets in closed-loop psychiatric neuromodulation.

Prediction of the minimum miscibility pressure for CO2 flooding based on a physical information neural network algorithm

Abstract The minimum miscibility pressure (MMP) is a crucial parameter in assessing the miscibility of CO2 displacement and evaluating the effectiveness of oil displacement. Traditional methods for calculating MMP are intricate and time-consuming, involving numerous related parameters. Therefore, precise and efficient determination of MMP is highly significant in the development of CO2-driven reservoirs. This study first utilized the Pearson correlation coefficient to analyse the correlation factor mechanism of 36 sets of fine-tube experimental data. Subsequently, the physical information neural network prediction model was employed with reservoir temperature, crude oil composition, and injected gas type as input parameters. The PRI state equation and Glaso correlation equation drove the model, with parameter optimization and training conducted under both physical and data driving. The model demonstrates high prediction accuracy and strong generalization ability. Finally, Validation of the model was performed using fine-tube experimental data from 5 other wells, revealing a relatively small relative deviation between calculated and experimental values, with an average coefficient of determination of 0.95 and an average relative error of 4.42%. The prediction accuracy was improved by about 75% compared to other machine learning algorithms. This model holds potential for application in on-site reservoir development, enhancing the measurement accuracy of the minimum miscible pressure of pure CO2 flooding, greatly shortening the design cycle of reservoir development, expediting the process of reservoir development, and providing technical guidance for improving oil and gas recovery and pure CO2 flooding exploration and development.

A Novel TCN-LSTM Hybrid Model for sEMG-Based Continuous Estimation of Wrist Joint Angles.

Surface electromyography (sEMG) offers a novel method in human-machine interactions (HMIs) since it is a distinct physiological electrical signal that conceals human movement intention and muscle information. Unfortunately, the nonlinear and non-smooth features of sEMG signals often make joint angle estimation difficult. This paper proposes a joint angle prediction model for the continuous estimation of wrist motion angle changes based on sEMG signals. The proposed model combines a temporal convolutional network (TCN) with a long short-term memory (LSTM) network, where the TCN can sense local information and mine the deeper information of the sEMG signals, while LSTM, with its excellent temporal memory capability, can make up for the lack of the ability of the TCN to capture the long-term dependence of the sEMG signals, resulting in a better prediction. We validated the proposed method in the publicly available Ninapro DB1 dataset by selecting the first eight subjects and picking three types of wrist-dependent movements: wrist flexion (WF), wrist ulnar deviation (WUD), and wrist extension and closed hand (WECH). Finally, the proposed TCN-LSTM model was compared with the TCN and LSTM models. The proposed TCN-LSTM outperformed the TCN and LSTM models in terms of the root mean square error (RMSE) and average coefficient of determination (R2). The TCN-LSTM model achieved an average RMSE of 0.064, representing a 41% reduction compared to the TCN model and a 52% reduction compared to the LSTM model. The TCN-LSTM also achieved an average R2 of 0.93, indicating an 11% improvement over the TCN model and an 18% improvement over the LSTM model.

Damage evolution analysis of C/SiC screwed/bonded hybrid joints based on in-situ micro-CT technique

Due to own good high temperature mechanical properties, the C/SiC screwed/bonded hybrid joints are considered as an important development direction of vehicle connection structures. However, the difficulty of directly monitoring the interior of the hybrid joints during the bearing presents a potential challenge to the structural damage assessment. In this study, the damage process of the hybrid joints subjected to tensile loading was monitored and characterized based on the in-situ micro-CT technique. The intrinsic relationship between the mechanical responses of the hybrid joints and the local damage evolution was elucidated by observing the variation of the deposited SiC void volume at the overlap interface. The dual inhomogeneity of SiC bonding layer was quantitatively characterized utilizing the average failure rate and determination coefficient. The coupling influence mechanism of prefabricated bottom hole size and online connection process on the final assembly clearance was revealed.

PML_30: A high resolution (30 m) estimates of evapotranspiration based on remote sensing model with application in an arid region

Accurate high-resolution evapotranspiration (ET) estimates are essential for effective local-scale water resource management. This task is often challenging due to the complexities involved in parameterizing surface conditions, which is crucial for driving various ET models. To address this challenge, we presented a novel approach to acquire long-term surface parameters by synthesizing observations from multiple Landsat satellites, and furthermore integrate the Penman-Monteith-Leuning model, leading to the PML_30 model designed for ET estimations at an impressive 30 m resolution. At the site scale, comparison with observations from 42 FLUXNet sites across diverse arid and semi-arid regions globally showed the reliable performance of the PML_30 model, boasting average coefficient of determination (R2) of 0.79, root mean square error (RMSE) of 0.52 mm/day, and bias of 0.39 mm/day. At the regional scale, the PML_30 model was applied to estimate monthly ET from 2000 to 2020 over a typical arid region in Northwest China. The mean annual ET estimated by the PML_30 model closely aligned with the reference based on regional water balance. Additionally, the model accurately described the spatial distribution of ET, which showed a notable improvement over existing products. In summary, our proposed PML_30 model emerges as a reliable tool for ET estimations at a specific high spatial resolution, particularly beneficial for arid and semi-arid regions.

Vision foundation model for agricultural applications with efficient layer aggregation network

Agricultural production is transitioning from traditional tools to IoT-connected automation devices. The integration of computer vision and agricultural automation is becoming closer with the rapid development of deep learning technology. Agricultural vision tasks, nevertheless, are relatively simple and less diverse compared to large-scale vision tasks, and there is a scarcity of plant data. This is also the reason why researchers need to make significant adjustments when directly applying advanced algorithms to the agricultural domain. In practice, downstream vision models should be designed to be concise and efficient. In this paper, we meticulously designed a deep convolutional network called TasselELANet for agricultural vision applications. It features a concise and efficient global architecture, including a 16-fold downsampling layer encoder and a decoder utilizing only 2 feature layers, which is greatly different from the general approaches. At its core is the efficient layer aggregation network, which utilizes the cross-stage fusion strategy to effectively optimize the gradient propagation path, thereby enhancing the learning ability and inference speed of the network. We validated the performance of TasselELANet on three challenging publicly available plant detection and counting datasets: maize tassels, wheat ears, and rice panels. The experimental tests achieved an average Accuracy of 0.899 and an average determination coefficient R2 of 0.867, remarkably outperforming other advanced computer vision methods in terms of accuracy and efficiency, and demonstrating sufficient generalizability. We firmly believe that TasselELANet can serve as a powerful and reliable vision tool for agricultural practitioners while also providing a solid foundation for future research. The code, datasets can be accessed at https://github.com/Ye-Sk/TasselELANet.git.

Prediction of product properties and identification of key influencing parameters in microwave pyrolysis of microalgae using machine learning

Exploring the effects of operating parameters and microalgae composition on product distribution and properties using machine learning is an effective and economical approach. However, there is currently a lack of research utilizing machine learning to comprehensively describe the microwave pyrolysis of microalgae. In this study, machine learning was employed to accurately model microalgae microwave pyrolysis based on microalgae composition, operating parameters, and the yields and properties of three-phase products. Using the collected data set, machine learning models, including Support Vector Regressor, Random Forest Regressor, Gradient Boosting Regressor, and eXtreme Gradient Boosting, were employed to model microalgae microwave pyrolysis. Results showed that the Gradient Boosting Regressor model outperformed the other three models, with average coefficients of determination (R2) of 0.998 and 0.762 in the training and testing phases, respectively. SHapley Additive exPlanations analysis revealed that reaction temperature, reaction time, and microwave power significantly influenced the predictions of the yields and properties of three-phase products. The findings of this study can effectively save costs and time in the future study on microwave pyrolysis of microalgae, and serve as an important tool for guiding subsequent experiments and process optimization.

Explainable highway performance degradation prediction model based on LSTM

With the dense and huge highway network in China, the highway maintenance management system has become the most concerned issue for Chinese highway managers in recent years. Therefore, based on the highway network in Guizhou Province of China, this paper established the semi-rigid asphalt pavement performance multi-output Long Short-Term Memory (LSTM) prediction model with regional applicability, including Pavement Surface Condition Index (PCI), Pavement Rutting Depth Index (RDI), Pavement Riding Quality Index (RQI) and Pavement Skidding Resistance Index (SRI) and the hyperparameters of the model were optimized by Bayesian optimization algorithm and the casual analysis of the model was conducted based on SHapley Additive exPlanations (SHAP), providing a reference for China's highway network-level preventive maintenance decision-making. The results prove that the established model can effectively predict the performance of the following year through the data of the previous two years. The established model has a better comprehensive prediction effect in comparison to similar multi-output models. The average coefficient of determination (R2) of the four prediction indexes can reach 0.823. Besides, the prediction effect among the indexes is stable and multiple pavement performance indexes can be effectively and reliably predicted at the same time. In addition, the contribution of input feature variables to output is quantified based on SHAP. According to quantified the contribution, the association relationship of the neural network model is set-valued mapped to the causality relationship in the post hoc. The input feature variables with the main contribution are extracted for causal analysis. The causal analysis shows that the degradation of the four pavement performances all have obvious time memory characteristics and have significant differences, but their pavement performance degradation is the result of nonlinear changes caused by the coupling effects of traffic load, road age, climate, and pavement structure.