Normalized Root Mean Research Articles

State of Charge Estimation Method of Energy Storage Battery Based on Multiple Incremental Features

Accurately estimating the state of charge (SOC) is crucial for energy storage battery management systems as it ensures battery performance and extends lifespan. However, existing deep learning-based methods often overlook the dynamic process information during battery charging and discharging, which compromises the accuracy of SOC estimation. To address this limitation, this paper proposes a novel SOC estimation method. First, we employ differential processing on the collected voltage, current, and temperature data to capture dynamic feature changes. Next, all features are normalized to ensure they are on the same scale. Finally, the processed data is divided into sliding windows and input into the TCN-BiLSTM-Attention Net (TBANet) model for SOC estimation. The results show that compared with traditional deep learning based SOC estimation methods, adding incremental features to TBANet improves the estimation accuracy by 15.8%. The average absolute error and root mean square error of the experimental results are 0.72% and 0.91%, respectively. In addition, this approach adopts transfer learning methods to verify the strong adaptability of the proposed method on different datasets, which highlights the robustness of TBANet and its potential for wide applicability in real-world scenarios.

The solvation thermodynamics of rose acetate in four mixed solvents: Molecular dynamics simulation, molecular conformation analysis, machine learning and solubility prediction

The solubility of rose acetate (alpha-(trichloromethyl) benzyl acetate) in four binary solvent mixtures was measured by static gravimetric method at the temperature ranging from 288.15 K to 328.15 K under atmosphere pressure. The dissolution behavior and solubility prediction were performed through simulations and calculations, using Materials Studio software and Back Propagation Neural Network. The modified Apelblat equation, λh model, Van’t Hoff equation, and NRTL model were employed to correlate the solubility data respectively. And the modified Apelblat equation provided the best fitting performance, corresponding to the lowest average value of average relative deviation and root mean square deviation. After analyzing the solvent properties, it was obsersved that the solubility of rose acetate in selected solvents was influenced by the polarity of the solvent. The mixing thermodynamic properties including ΔmixH, ΔmixS and ΔmixG indicated that the dissolution process of rose acetate was exothermic and spontaneous.Molecular dynamics (MD) simulations were utilized to analyze the interaction mechanism between rose acetate and binary solvent mixtures. The results revealed that the hydrogen bonding interaction played an insignificant role in the dissolution process. Moreover, the solute molecules were characterized in both crystal and solution state based on their conformation and molecular energies. A back-propagation (BP) neural network was constructed to predict the solubility of rose acetate, and the predicted data exhibited good agreement with the measured valuses.

Convformer: A Model for Reconstructing Ocean Subsurface Temperature and Salinity Fields Based on Multi-Source Remote Sensing Observations

Observational data on ocean subsurface temperature and salinity are patently insufficient because in situ observations are complex and costly, while satellite remote-sensed measurements are abundant but mainly focus on sea surface data. To make up for the ocean interior data shortage and entirely use the abundant satellite data, we developed a data-driven deep learning model named Convformer to reconstruct ocean subsurface temperature and salinity fields from satellite-observed sea surface data. Convformer is designed by deeply optimizing Vision Transformer and ConvLSTM, consisting of alternating residual connections between multiple temporal and spatial attention blocks. The input variables consist of sea surface temperature (SST), sea surface salinity (SSS), sea surface height (SSH), and sea surface wind (SSW). Our results demonstrate that Convformer exhibits superior performance in estimating the temperature-salinity structure of the tropical Pacific Ocean. The all-depth average root mean square error (RMSE) of the reconstructed subsurface temperature (ST)/subsurface salinity (SS) is 0.353 °C/0.0695 PSU, with correlation coefficients (R²) of 0.98663/0.99971. In the critical thermocline, although the root mean square errors of ST and SS reach 0.85 °C and 0.121 PSU, respectively, they remain smaller compared to other models. Furthermore, we assessed Convformer’s performance from various perspectives. Notably, we also delved into the potential of Convformer to extract physical and dynamic information from a model mechanism perspective. Our study offers a practical approach to reconstructing the subsurface temperature and salinity fields from satellite-observed sea surface data.

Prakiraan Jumlah Curah Hujan Berdasarkan Data Parameter Cuaca Tahun 2011-2021 di Kota Pangkalpinang

Tropical weather prevails in the Pangkalpinang city region, complemented with a changeable rainy season that experiences monthly variations. This phenomena results in a considerable amount of precipitation in the city of Pangkalpinang, making the area known for its extreme rainfall. Using multiple linear regression approaches to anticipate monthly precipitation is the aim of this work. Rainfall (RR), air temperature (T), air humidity (RH), and air pressure are all part of the dataset that was collected over an 11-year period (2011–2021). The information was obtained from the meteorological station in Depati Amir Pangkalpinang. This study uses the multiple linear regression method. The rainfall forecast results are compared to the observed actual rainfall amounts. Based on air temperature and air humidity as predictors, the results show that the estimation of cumulative monthly precipitation in the Pangkalpinang city region for the year 2021 produced an average root mean square error (RSME) value of 112.6 mm/month. Furthermore, the T-test analysis produced a 0.9 result. With air temperature, humidity, and pressure as predictors, the total monthly rainfall for 2021 was predicted with an average root mean square error (RMSE) value of 107.9 mm/month. A T-test was also performed, and the result was 0.4. The results of this study show that the monthly rainfall forecast model for the city of Pangkalpinang that includes factors related to temperature, humidity, and pressure produces better results than the model that only includes these variables. November and December are the months with the greatest variations.

Impact of Climate Change on the Phenology of Winter Oilseed Rape (Brassica napus L.)

In our investigation, we have developed innovative statistical models tailored to predict specific phenological stages of winter oilseed rape (WOSR) cultivation in Luxembourg. Leveraging extensive field observations and meteorological data, our modeling approach accurately forecasts critical growth stages of WOSR, including inflorescence emergence (BBCH 51), initial flowering (BBCH 60), and cessation of flowering (BBCH 69), capitalizing on accumulated heat units. Our findings challenge conventional assumptions surrounding base temperatures, advocating for a specific base temperature of 3 °C for winter oilseed rape emergence, consistent with prior research. Validation via leave-one-out cross-validation yields promising outcomes, with average Root Mean Square Error (RMSE) values below 1, surpassing analogous studies. Particularly noteworthy is our model’s performance in predicting crucial growth stages, notably BBCH 60, pivotal for pest control. Despite advancements, hurdles persist in forecasting late-stage phenological events influenced by leaf senescence and anticipated climate change impacts, likely accelerating WOSR development and introducing new risks. In response, cultivar selection strategies informed by individual development rates and temperature sensitivities emerge as vital mitigation measures. As climate variability intensifies, precision agriculture assumes paramount importance in optimizing resource allocation and ensuring sustainable WOSR cultivation practices. Our study advocates for proactive integration of predictive modeling into adaptive management frameworks, empowering stakeholders to make informed decisions taking climatic dynamics into account.

Lithium-Ion Battery Capacity Prediction with GA-Optimized CNN, RNN, and BP

Over the last 20 years, lithium-ion batteries have become widely used in many fields due to their advantages such as ease of use and low cost. However, there are concerns about the lifetime and reliability of these batteries. These concerns can be addressed by obtaining accurate capacity and health information. This paper proposes a method to predict the capacity of lithium-ion batteries with high accuracy. Four key features were extracted from current and voltage data obtained during charge and discharge cycles. To enhance prediction accuracy, the Pearson correlation coefficient between these features and battery capacities was analyzed and eliminations were made for some batteries. Using a genetic algorithm (GA), the parameter optimization of Convolutional Neural Network (CNN), Backpropagation (BP), and Recurrent Neural Network (RNN) algorithms was performed. The parameters that provide the best performance were determined in a shorter time using GA, which includes natural selection and genetic processes instead of a trial-and-error method. The study employed five metrics—Mean Square Error (MSE), Root Mean Square Error (RMSE), Normalized Root Mean Square Error (NRMSE), Mean Absolute Error (MAE), and Squared Correlation (R2)—to evaluate prediction accuracy. Predictions based on NASA experimental data were compared with the existing literature, demonstrating superior accuracy. Using 100 training data, 68 data predictions were made with a Root Mean Square Error (RMSE) of 0.1176%. This error rate represents an accuracy level 2.5 times higher than similarly accurate studies in the literature.

Improving forest carbon sequestration through thinning strategies under soil conservation constraints: A case study in Shaanxi Province, China

Forest thinning is an effective measure to improve carbon sequestration (CS) to adapt to and mitigate global warming. However, reducing forest cover could negatively affect other ecosystem services such as soil conservation (SC), making it difficult to effectively implement thinning strategies. This problem is exacerbated by a lack of previous research. Accordingly, after assessing thinning for two typical trees (Robinia pseudoacacia L. (RP) and Quercus L. (QL)) in Shaanxi Province, we propose a forest management framework that provides detailed thinning strategies and improves CS using a process-based biogeochemical model (Biome-BGCMuSo) by considering future climate change and SC constraints. After calibration for Biome-BGCMuSo, the average root mean squared error and percentage bias for the simulated results and observed data of RP and QL decreased (39.2–54.5 % and 66.0–86.9 %) and the correlation coefficient and Nash–Sutcliffe efficiency improved (11.6–12.3 % and 114.7–153.0 %,) respectively. Comparison between current (2001–2020) and future (2081–2100) conditions predicted overall future CS increase, and future SC would display either increasing or decreasing trends in different climatic scenarios and subregions. Partial correlation analysis showed precipitation as the dominant factor positively influencing SC, while temperature, precipitation, and CO2 concentration were the dominant factors positively influencing CS. Based on the proposed framework, RP and QL should be thinned three times every 7–11 and 3–11 years during the non-growing season with intensity ranges of 2–27 % and 1–9 %, respectively. Although the average CS during 2023–2100 with thinning increased slightly (0.1–2.6 %) compared with that without thinning, a notable potential to increase CS (0.2–7.7 %) was observed for thinning during a period at the end of this century predicted to experience the most warming. The proposed framework facilitates maximizing CS and maintaining SC. As the framework is based on a process-based biogeochemical model, it could be applicable to other regions.

Improvements of 177Lu SPECT images from sparsely acquired projections by reconstruction with deep-learning-generated synthetic projections

BackgroundFor dosimetry, the demand for whole-body SPECT/CT imaging, which require long acquisition durations with dual-head Anger cameras, is increasing. Here we evaluated sparsely acquired projections and assessed whether the addition of deep-learning-generated synthetic intermediate projections (SIPs) could improve the image quality while preserving dosimetric accuracy.MethodsThis study included 16 patients treated with 177Lu-DOTATATE with SPECT/CT imaging (120 projections, 120P) at four time points. Deep neural networks (CUSIPs) were designed and trained to compile 90 SIPs from 30 acquired projections (30P). The 120P, 30P, and three different CUSIP sets (30P + 90 SIPs) were reconstructed using Monte Carlo-based OSEM reconstruction (yielding 120P_rec, 30P_rec, and CUSIP_recs). The noise levels were visually compared. Quantitative measures of normalised root mean square error, normalised mean absolute error, peak signal-to-noise ratio, and structural similarity were evaluated, and kidney and bone marrow absorbed doses were estimated for each reconstruction set.ResultsThe use of SIPs visually improved noise levels. All quantitative measures demonstrated high similarity between CUSIP sets and 120P. Linear regression showed nearly perfect concordance of the kidney and bone marrow absorbed doses for all reconstruction sets, compared to the doses of 120P_rec (R2 ≥ 0.97). Compared to 120P_rec, the mean relative difference in kidney absorbed dose, for all reconstruction sets, was within 3%. For bone marrow absorbed doses, there was a higher dissipation in relative differences, and CUSIP_recs outperformed 30P_rec in mean relative difference (within 4% compared to 9%). Kidney and bone marrow absorbed doses for 30P_rec were statistically significantly different from those of 120_rec, as opposed to the absorbed doses of the best performing CUSIP_rec, where no statistically significant difference was found.ConclusionWhen performing SPECT/CT reconstruction, the use of SIPs can substantially reduce acquisition durations in SPECT/CT imaging, enabling acquisition of multiple fields of view of high image quality with satisfactory dosimetric accuracy.

Advancing short-term solar irradiance forecasting accuracy through a hybrid deep learning approach with Bayesian optimization

The optimization of solar energy integration into the power grid relies heavily on accurate forecasting of solar irradiance. In this study, a new approach for short-term solar irradiance forecasting is introduced. This method combines Bayesian Optimized Attention-Dilated Long Short-Term Memory and Savitzky-Golay filtering. The methodology is implemented to analyze data obtained from a solar irradiance probe situated in Douala, Cameroon. Initially, the unprocessed data is augmented by integrating distinctive solar irradiation variables, and the Savitzky-Golay filter with Bayesian Optimization is used to enhance its quality. Subsequently, multiple deep learning models, including Long Short-Term Memory, Bidirectional Long Short-Term Memory, Artificial Neural Networks, Bidirectional Long Short-Term Memory with Additive Attention Mechanism, and Bidirectional Long Short-Term Memory with Additive Attention Mechanism and Dilated Convolutional layers, are trained and evaluated. Out of all the models considered, the proposed approach, which combines the attention mechanism and dilated convolutional layers, demonstrates exceptional performance with the best convergence and accuracy in forecasting. Bayesian Optimization is further utilized to fine-tune the polynomial and window size of the Savitzky-Golay filter and optimize the hyperparameters of the deep learning models. The results show a Symmetric Mean Absolute Percentage Error of 0.6564, a Normalized Root Mean Square Error of 0.2250, and a Root Mean Square Error of 22.9445, surpassing previous studies in the literature. Empirical findings highlight the effectiveness of the proposed methodology in enhancing the accuracy of short-term solar irradiance forecasting. This research contributes to the field by introducing novel data pre-processing techniques, a hybrid deep learning architecture, and the development of a benchmark dataset. These advancements benefit both researchers and solar plant managers, improving solar irradiance forecasting capabilities.

A new stress tensor approach for application to the conductor surface

PurposeThe purpose of this paper is to derive a new stress tensor for calculating the Lorentz force acting on an arbitrarily shaped nonmagnetic conductive specimen moving in the field of a permanent magnet. The stress tensor allows for a transition from a volume to a surface integral for force calculation.Design/methodology/approachThis paper derives a new stress tensor which consists of two parts: the first part corresponds to the scaled Poynting vector and the second part corresponds to the velocity term. This paper converts the triple integral over the volume of the conductor to a double integral over its surface, where the subintegral functions are continuous through the different compartments of the model. Numerical results and comparison to the standard volume discretization using the finite element method are given.FindingsThis paper evaluated the performance of the new stress tensor computation on a thick and thin cuboid, a thin disk, a sphere and a thin cuboid containing a surface defect. The integrals are valid for any geometry of the specimen and the position and orientation of the magnet. The normalized root mean square errors are below 0.26% with respect to a reference finite element solution applying volume integration.Originality/valueTensor elements are continuous throughout the model, allowing integration directly over the conductor surface.

Prediction of motor and non-motor Parkinson’s disease symptoms using serum lipidomics and machine learning: a 2-year study

Identifying biological factors which contribute to the clinical progression of heterogeneous motor and non-motor phenotypes in Parkinson’s disease may help to better understand the disease process. Several lipid-related genetic risk factors for Parkinson’s disease have been identified, and the serum lipid signature of Parkinson’s disease patients is significantly distinguishable from controls. However, the extent to which lipid profiles are associated with clinical outcomes remains unclear. Untargeted high-performance liquid chromatography-tandem mass spectrometry identified >900 serum lipids in Parkinson’s disease subjects at baseline (n = 122), and the potential for machine learning models using these lipids to predict motor and non-motor clinical scores after 2 years (n = 67) was assessed. Machine learning models performed best when baseline serum lipids were used to predict the 2-year future Unified Parkinson’s disease rating scale part three (UPDRS III) and Geriatric Depression Scale scores (both normalised root mean square error = 0.7). Feature analysis of machine learning models indicated that species of lysophosphatidylethanolamine, phosphatidylcholine, platelet-activating factor, sphingomyelin, diacylglycerol and triacylglycerol were top predictors of both motor and non-motor scores. Serum lipids were overall more important predictors of clinical outcomes than subject sex, age and mutation status of the Parkinson’s disease risk gene LRRK2. Furthermore, lipids were found to better predict clinical scales than a panel of 27 serum cytokines previously measured in this cohort (The Michael J. Fox Foundation LRRK2 Clinical Cohort Consortium). These results suggest that lipid changes may be associated with clinical phenotypes in Parkinson’s disease.

Precise measurement of human heart rate based on multi-channel radar data fusion

To achieve non-contact measurement of human heart rate and improve its accuracy, this paper proposes a method for measuring human heart rate based on multi-channel radar data fusion. The radar data were firstly extracted by human body position identification, phase extraction and unwinding, phase difference, band-pass filtering optimized by power spectrum entropy, and fast independent component analysis for each channel data. After overlaying and fusing the four-channel data, the heartbeat signal was separated using frost-optimized variational modal decomposition. Finally, a chirp Z-transform was introduced for heart rate estimation. After validation with 40 sets of data, the average root mean square error of the proposed method was 2.35 beats per minute, with an average error rate of 2.39%, a Pearson correlation coefficient of 0.97, a confidence interval of [-4.78, 4.78] beats per minute, and a consistency error of -0.04. The experimental results show that the proposed measurement method performs well in terms of accuracy, correlation, and consistency, enabling precise measurement of human heart rate.

Deterministic Localization for Fully Automatic Operation: A Survey and Experiments.

With the rapid development of fully automatic operation (FAO) and location-based services, the evaluation criteria of average localization accuracy can no longer meet our demands, in favor of deterministic localization. However, most localization researches modeled localization performance function and enhanced it by minimizing average localization root mean square error (RMSE). The performance degradation in a small region was not considered. In this paper, we present a survey of deterministic localization and analyze the relationship between accuracy and certainty. In this paper, two common solutions of localization enhancement are presented and their localization certainties are discussed. Furthermore, we carry out related localization enhancement experiments in rail transit line and analyze their improvement on deterministic localization. The experimental results show that the overall localization performance is improved, while the deterministic localization requires the stricter solution to promote.

Temperature-dependent hysteresis model for Li-ion batteries

The increasing importance of accurate battery state estimations in advanced Battery Management Systems (BMS) underscores the need for precise modelling of battery behaviour and characteristics. While equivalent circuit models are widely utilized for their low computational demands, they face challenges in maintaining precision and adaptability during dynamic conditions, posing a persistent concern for future advancements. This study focuses specifically on the battery hysteresis effect, a complicating factor in the modelling and estimation processes. Open circuit voltage (OCV) measurements and parameter identification for equivalent circuit models were conducted on prevalent Li-ion battery technologies, namely nickel-manganese–cobalt (NMC) and lithium-iron-phosphate (LFP). The experimental results indicate the hysteresis effect becomes more significant with lower temperatures. In this paper, a battery model covering the temperature influence on the hysteresis effect is proposed. The proposed model exhibits an average root mean square error of less than 13 mV. The model holds promise for application in modern battery management systems, offering an enhancement to state-of-charge estimation methodologies.

Enhancing hydrological data completeness: A performance evaluation of various machine learning techniques using probabilistic fusion imputer with neural networks for streamflow data reconstruction

The present-day accessibility of streamflow data, particularly in the developing countries, is often marked by a multitude of data shortfalls or distortions. This study investigates the estimate of missing streamflow data using machine learning approaches, including K-nearestneighbour (KNN), Predictive Mean Matching (PMM), Random Forest (RF) and a novel technique of Probabilistic Fusion Imputer with Neural Networks (PROFINN). This study tackles the issue of data insufficiency in such time series considering the inclusion of numerous hydrological parameters by means of RF feature selector, to determine the most significant ones among them. The study explores the efficacy of selected models on various data gaps under different hydrological scenarios, including diverse flow characteristics (mean, high and low flows) and gap lengths (long and short gaps, continuous and discontinuous gaps) and presents a pattern of ranking system that assesses the level of suitability of each technique for various data gaps. This study underscore PROFINN’s remarkable performance across all scenarios, yielding an average Root Mean Square Error (RMSE) of 0.91 and an average Nash-Sutcliffe Efficiency (NSE) value of 0.93 when applied for an intermittent river system of Pamba in Southern Kerala, India. RF follows PROFINN in the imputation of extreme flows as well as long and short gap scenarios. KNN closely follows PROFINN for the imputation of continuous and discontinuous gap scenarios. This study augments the significance of tailored machine learning techniques in enhancing the integrity of hydrological datasets, offering valuable insights for effective decision-making in water resource management and related fields. Furthermore, the success of PROFINN in streamflow data imputation suggests its potential extension to other hydrological research domains, like historical data reconstruction assisting climate change studies and future water resources planning, emphasizing the broader relevance and applicability of this study’s findings.

Detection of the low-velocity layer using a convolutional neural network on passive surface-wave data: An application in Hangzhou, China

Passive surface-wave methods using dense seismic arrays have gained growing attention in near-surface high-resolution imaging in urban environments. Deep learning (DL) in the extraction of dispersion curves and inversion can release a tremendous workload brought by dense seismic arrays. We presented a case study of imaging shear-wave velocity (Vs) structure and detecting low-velocity layer (LVL) in the Hangzhou urban area (eastern China). We used traffic-induced passive surface-wave data recorded by dense linear arrays. We extracted phase-velocity dispersion curves from noise recordings using seismic interferometry and multichannel analysis of surface waves. We adopted a convolutional neural network to estimate near-surface Vs models by inverting Rayleigh-wave fundamental-mode phase velocities. To improve the accuracy of the inversion, we utilized the sensitivities to weight the loss function. The average root mean square error from the weighted inversion is 46% lower than that from the unweighted DL inversion. The estimated pseudo-2D Vs profiles correspond to the velocities obtained from downhole seismic measurements. Compared with an investigation on the same survey area, our inversion results are more consistent with the Vs provided by downhole seismic measurements within 50–60 m where the LVL exists. The trained neural network successfully identified that the LVL is located at 50–60 m deep. To check the applicability of the trained neural network, we applied it to a nearby passive surface-wave survey and the inversion results agree with the existing investigation results. The two applications demonstrate the accuracy and efficiency of delineating near-surface Vs structures with the LVL from traffic-induced noise using the DL technique. The DL inversion has great potential for monitoring subsurface medium changes in urban areas.

Research on reconstruction of the global sound speed profile combining partial underwater prior information

The sound speed profile (SSP) is an important factor affecting the acoustic propagation characteristics of the ocean, making the accurate acquisition of SSP a crucial step in the interdisciplinary research of oceanography and underwater acoustics. Limited by the cost of in-situ measurement and the performance of the instrument itself, direct measurement of SSP inevitably leads to insufficient depth or even missing information. In this paper, we propose using partial underwater prior information (UWPI) only including underwater sound speed to obtain preliminary reconstruction results of global SSP for the first time. The empirical orthogonal function (EOF) reconstruction algorithm is optimized by employing assimilated SSP as the background SSP to further reduce reconstruction errors. The maximum global average reconstruction error and root mean square error (RMSE) after optimization decrease by >51% and 71%, respectively, which indicates that the performance of the optimized algorithm combined with partial UWPI is further improved. Finally, the performance of the optimized algorithm is discussed from the perspective of acoustic propagation. This research provides a reliable technical approach for SSP reconstruction under incomplete depth conditions, which can be applied in underwater sound field prediction and acoustic detection in the future.

An Integrated LSTM-Rule-Based Fusion Method for the Localization of Intelligent Vehicles in a Complex Environment.

To improve the accuracy and robustness of autonomous vehicle localization in a complex environment, this paper proposes a multi-source fusion localization method that integrates GPS, laser SLAM, and an odometer model. Firstly, fuzzy rules are constructed to accurately analyze the in-vehicle localization deviation and confidence factor to improve the initial fusion localization accuracy. Then, an odometer model for obtaining the projected localization trajectory is constructed. Considering the high accuracy of the odometer's projected trajectory within a short distance, we used the shape of the projected localization trajectory to inhibit the initial fusion localization noise and used trajectory matching to obtain an accurate localization. Finally, the Dual-LSTM network is constructed to predict the localization and build an electronic fence to guarantee the safety of the vehicle while also guaranteeing the updating of short-distance localization information of the vehicle when the above-mentioned fusion localization is unreliable. Under the limited arithmetic condition of the vehicle platform, accurate and reliable localization is realized in a complex environment. The proposed method was verified by long-time operation on the real vehicle platform, and compared with the EKF fusion localization method, the average root mean square error of localization was reduced by 66%, reaching centimeter-level localization accuracy.

Capacity estimation of lithium-ion battery with multi-task autoencoder and empirical mode decomposition

Capacity estimation of lithium-ion batteries is a commonly used method in health diagnosis and management. Its mainstream method involves using data-driven time series forecasting models to learn the patterns of changes in capacity. However, capacity regeneration poses a challenge for training time series forecasting models. Therefore, we propose a hybrid method that applies empirical mode decomposition and a multi-task autoencoder. In detail, empirical mode decomposition is applied to decompose the time series of capacity into intrinsic mode functions and a residual. Then, a multi-task autoencoder based on diagonal state space models is applied to estimate intrinsic mode functions while support vector regression is utilized for the residual. Experimental results show that the method outperforms seven baselines on three datasets, with an average root mean square error of 0.0103, 0.0111, and 0.0004. Furthermore, it is capable of performing an inference on the CPU in 3.57 ms with 0.69 MB of memory usage.

Novel natural gradient boosting-based probabilistic prediction of physical properties for polypropylene-based composite data

Accurately predicting the physical properties of polypropylene composites is challenging because they are highly complex due to the numerous combinations of materials used in their production. Therefore, several challenges must be addressed to develop a predictive model suitable for applied in commercial industry. This study proposes a probabilistic prediction model for the physical properties of polypropylene composites using NGBoost with categorization. The proposed model simultaneously addressed the data imbalance and uncertainty problems using the NGBoost with categorization method. In the data preprocessing step, categorization was used to classify data for each recipe type so that all types of data could be learned. In the model training step, the model performance was improved through hyperparameter optimization. Subsequently, the probabilistic possibility of the predicted value was presented using the probability density function of NGBoost. As a result, the performance of the model had a R2 of 0.9 or more and a normalized root mean square error of 0.1 or less in all four properties. Therefore, the proposed model simultaneously suggests a predicted physical property and its probability with a high performance using real commercial data. In addition, this model is useful for industrial workers because it saves time and cost for the manufacturing and property testing of PPCs.