Mean Absolute Percentage Error Research Articles

Research and application of a novel grey multivariable model in port scale prediction under the impact of Free Trade Zone

Purpose Considering the impact of the Free Trade Zone (FTZ) policy on forecasting the port cargo throughput, this paper constructs a fractional grey multivariate forecasting model to improve the prediction accuracy of port cargo throughput and realize the coordinated development of FTZ policymaking and port construction.Design/methodology/approach Considering the effects of data randomization, this paper proposes a novel self-adaptive grey multivariate prediction model, namely FDCGM(1,N). First, fractional-order accumulative generation operation (AGO) is introduced, which integrates the policy impact effect. Second, the heuristic grey wolf optimization (GWO) algorithm is used to determine the optimal nonlinear parameters. Finally, the novel model is then applied to port scale simulation and forecasting in Tianjin and Fujian where FTZs are situated and compared with three other grey models and two machine learning models.Findings In the Tianjin and Fujian cases, the new model outperforms the other comparison models, with the least mean absolute percentage error (MAPE) values of 6.07% and 4.16% in the simulation phase, and 6.70% and 1.63% in the forecasting phase, respectively. The results of the comparative analysis find that after the constitution of the FTZs, Tianjin’s port cargo throughput has shown a slow growth trend, and Fujian’s port cargo throughput has exhibited rapid growth. Further, the port cargo throughput of Tianjin and Fujian will maintain a growing trend in the next four years.Practical implications The new multivariable grey model can effectively reduce the impact of data randomness on forecasting. Meanwhile, FTZ policy has regional heterogeneity in port development, and the government can take different measures to improve the development of ports.Originality/value Under the background of FTZ policy, the new multivariable model can be used to achieve accurate prediction, which is conducive to determining the direction of port development and planning the port layout.

Estimation of Groundwater River Availability in Leang Lonrong Cave Using ARIMA Model and Econophysics Valuation Approach

This quantitative descriptive study uses the ARIMA model approach to predict groundwater river water availability. It assesses its valuation with an econophysics orientation in the Leang Lonrong Cave area of TN Bantimurung Bulusaraung. Secondary data from 2010-2023, including Water Discharge and Water Level data from the River Basin Center of Pompengan Jeneberang (BBWS) and primary data on the Valuation of Willingness to Pay, were utilized. The analysis reveals that water discharge predictions (2024-2030) in the Gua Leang Lonrong area indicate consistent annual water availability, with the highest discharge projected in December 2030 (5.08 m3/s) and the lowest in September 2024 (1.12 m3/s). Valuation of utilizing Leang Lonrong River water through willingness to pay showed residents willing to pay Rp 42,142.85/month for clean water management. At the same time, visitors are willing to contribute Rp 21,596.77/visit to enhance water quality management. Elasticity analysis modeling water discharge with willingness to pay suggests a relatively elastic relationship, with a minor impact, indicating unresponsiveness to changes in water discharge percentages. Statistical and economic methods were employed to analyze daily water discharge patterns and evaluate residents' willingness to pay for underground river water utilization. The ARIMA model (0,1,1) forecasted future water discharge patterns with a mean absolute percentage error (MAPE) of 33.7%. The study surveyed 62 respondents, finding an average willingness to pay approximately IDR 21,596.77 per visit to maintain Leang Lonrong River water quality management.

A hybrid machine learning forecasting model for photovoltaic power

The increasing use of photovoltaic (PV) power generation presents a significant opportunity for global energy transformation. However, accurately forecasting PV power remains a challenge. This study proposes a hybrid approach that combines variational mode decomposition (VMD), whale optimization algorithm (WOA), and long short-term memory neural network (LSTM) to forecast photovoltaic (PV) power accurately. The model decomposes the time series of PV power using VMD to address the non-stationary nature of the time series. The VMD parameters are optimized using WOA. Subsequently, and the PV power time series data is then decomposed using the optimized VMD parameters to obtain multiple intrinsic mode functions (IMFs) components and a residual component. These components are then reconstructed with meteorological parameters to obtain the reconstructed IMF components and residuals. Finally, multiple LSTM sub-models are built, with each of them taking the IMF components and residual from the previous reconstruction as inputs. The sub-models are optimized using the WOA method to determine their hyperparameters and then constructed with these optimized hyperparameters. The predicted values of each IMF and residual are output and added for sequence reconstruction to derive the final predicted value of PV power. The model's effectiveness was verified for one-hour-ahead forecasting at the 1.8 MW solar system in Yulara, central Australia. The test results show that the proposed model outperforms other benchmark models, with mean absolute error (MAE), root mean square error (RMSE), mean absolute percentage error (MAPE), and R-squared value (R2) of 15.247 kW, 19.753 kW, 4.405 % and 0.997, respectively. Compared to the LSTM, MAE, RMSE, and MAPE decreased by 84.942 %, 86.746 %, and 82.611 %, respectively, while R2 increased by 23.438 %. The proposed model has better predictive performance for both stable power changes and large fluctuations, essential for effectively integrating renewable energy sources into the power grid.

Investigating a hybrid extreme learning machine coupled with Dingo Optimization Algorithm for modeling liquefaction triggering in sand-silt mixtures

Liquefaction is a devastating consequence of earthquakes that occurs in loose, saturated soil deposits, resulting in catastrophic ground failure. Accurate prediction of such geotechnical parameter is crucial for mitigating hazards, assessing risks, and advancing geotechnical engineering. This study introduces a novel predictive model that combines Extreme Learning Machine (ELM) with Dingo Optimization Algorithm (DOA) to estimate strain energy-based liquefaction resistance. The hybrid model (ELM-DOA) is compared with the classical ELM, Adaptive Neuro-Fuzzy Inference System with Fuzzy C-Means (ANFIS-FCM model), and Sub-clustering (ANFIS-Sub model). Also, two data pre-processing scenarios are employed, namely traditional linear and non-linear normalization. The results demonstrate that non-linear normalization significantly enhances the prediction performance of all models by approximately 25% compared to linear normalization. Furthermore, the ELM-DOA model achieves the most accurate predictions, exhibiting the lowest root mean square error (484.286 J/m3), mean absolute percentage error (24.900%), mean absolute error (404.416 J/m3), and the highest correlation of determination (0.935). Additionally, a Graphical User Interface (GUI) has been developed, specifically tailored for the ELM-DOA model, to assist engineers and researchers in maximizing the utilization of this predictive model. The GUI provides a user-friendly platform for easy input of data and accessing the model's predictions, enhancing its practical applicability. Overall, the results strongly support the proposed hybrid model with GUI serving as an effective tool for assessing soil liquefaction resistance in geotechnical engineering, aiding in predicting and mitigating liquefaction hazards.

Fast and High-Resolution T2 Mapping Based on Echo Merging Plus k-t Undersampling with Reduced Refocusing Flip Angles (TEMPURA) as Methods for Human Renal MRI.

To develop a highly accelerated multi-echo spin-echo method, TEMPURA, for reducing the acquisition time and/or increasing spatial resolution for kidney T2 mapping. TEMPURA merges several adjacent echoes into one k-space by either combining independent echoes or sharing one echo between k-spaces. The combined k-space is reconstructed based on compressed sensing theory. Reduced flip angles are used for the refocusing pulses, and the extended phase graph algorithm is used to correct the effects of indirect echoes. Two sequences were developed: a fast breath-hold sequence; and a high-resolution sequence. The performance was evaluated prospectively on a phantom, 16 healthy subjects, and two patients with different types of renal tumors. The fast TEMPURA method reduced the acquisition time from 3-5 min to one breath-hold (18 s). Phantom measurements showed that fast TEMPURA had a mean absolute percentage error (MAPE) of 8.2%, which was comparable to a standardized respiratory-triggered sequence (7.4%), but much lower than a sequence accelerated by purely k-t undersampling (21.8%). High-resolution TEMPURA reduced the in-plane voxel size from 3 × 3 to 1 × 1 mm2, resulting in improved visualization of the detailed anatomical structure. In vivo T2 measurements demonstrated good agreement (fast: MAPE = 1.3%-2.5%; high-resolution: MAPE = 2.8%-3.3%) and high correlation coefficients (fast: R = 0.85-0.98; high-resolution: 0.82-0.96) with the standardized method, outperforming k-t undersampling alone (MAPE = 3.3-4.5%, R = 0.57-0.59). TEMPURA provides fast and high-resolution renal T2 measurements. It has the potential to improve clinical throughput and delineate intratumoral heterogeneity and tissue habitats at unprecedented spatial resolution.

Short-term power load forecasting for integrated energy system based on a residual and attentive LSTM-TCN hybrid network

In the context of Integrated Energy System (IES), accurate short-term power demand forecasting is crucial for ensuring system reliability, optimizing operational efficiency through resource allocation, and supporting effective real-time decision-making in energy management. However, achieving high forecasting accuracy faces significant challenges due to the inherent complexity and stochastic nature of IES’s short-term load profiles, resulting from diverse consumption patterns among end-users and the intricate coupling within the network of interconnected energy sources. To address this issue, a dedicated Short-Term Power Load Forecasting (STPLF) framework for IES is proposed, which relies on a newly developed hybrid deep learning architecture. The framework seamlessly combines Long Short-Term Memory (LSTM) with Temporal Convolutional Network (TCN), enhanced by an attention mechanism module. By merging these methodologies, the network leverages the parallel processing prowess of TCN alongside LSTM’s ability to retain long-range temporal information, thus enabling it to dynamically concentrate on relevant sections of time series data. This synergy leads to improved prediction accuracy and broader applicability. Furthermore, the integration of residual connections within the network structure serves to deepen its learning capabilities and enhance overall performance. Ultimately, results from a real case study of a user-level IES demonstrate that the Mean Absolute Percentage Error (MAPE) of the proposed framework on the test set is 2.35%. This error rate is lower than the averages of traditional methods (3.43%) and uncombined single submodules (2.80%).

A multi-scale fuel cell degradation prediction method based on isometric convolution block and long short-term memory networks

Performance degradation prediction is considered an effective method to enhance the durability of Proton Exchange Membrane Fuel Cells (PEMFCs). However, accurate prediction under dynamic conditions remains a challenge. The MSD-ICB-LSTM model is proposed to improve the accuracy of the performance degradation prediction, which incorporates the information from both long-term and short-term modes within the PEMFC history operating data. Firstly, redundant variables are removed from the data, and the noise is eliminated using locally weighted scatterplot smoothing (LOWESS) to raise the quality of the input samples. Secondly, the Multi-Scale Decomposition (MSD) method is used to decompose the pre-processed data into long-term and short-term modes. Thirdly, a combined module containing encoding, internal decomposition and a double-layer Isometric Convolution Block (ICB) is used to deeply explore the nonlinear degradation characteristics in the short-term mode, while a Long-Short-Term Memory Network (LSTM) is utilized to estimate the long-term mode. The final result of prediction is obtained by combining the predictions of two modes. The experimental results show that the prediction accuracy of the proposed MSD-ICB-LSTM model is 99.74% in R2，and the Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE) are 0.0034 and 0.77%, respectively. Thus, the method can efficiently explore the multi-scale features and enhance the accuracy of PEMFC performance degradation prediction.

Retrieval algorithm of chlorophyll-a concentration for coastal waters based on ridge regression

Abstract. In this paper, we use the multiple scattering near-infrared aerosol correction model of SeaDAS to execute the atmospheric correction for Terra/Aqua MODIS remote sensing data, and also use three standard operational chlorophyll-a concentration inversion algorithms built in SeaDAS, named OC2, OC3, and OC4, to carry out chlorophyll-a concentration inversion in the Yellow Sea and East China Sea. We validate the inversion results using in-situ measured chlorophyll-a concentration data collected from the Yellow Sea and East China Sea in 2003. The results show that the inversion results of OC3 and OC4 algorithm are significantly larger than in-situ measured values, and the results of the OC2 algorithm are closer to the measured values. In view of the shortcomings of these algorithms, we proposes a chlorophyll-a concentration inversion model based on ridge regression, and carry out chlorophyll-a concentration inversion test using MODIS images covering the Yellow Sea and East China Sea 2003. The results indicate that, the new inversion model can effectively overcome the deficiencies of OC2, OC3, and OC4 algorithms. The inversion model can effectively overcome the covariance problem of OC2, OC3 and OC4 algorithms on the multivariate linear regression model, and the model passed the F-test (F=25.893, p=0.000<0.05), the mean absolute percentage error (MAPE) between the inversion values and the in-situ measured values was 21.8%, the root mean squared error (RMSE) was 0.325, and the coefficient of determination (R2) was 0.847. The accuracy and the fit degree of the new model were significantly better than those of the OC2, OC3 and OC4 algorithms. Therefore, the chlorophyll-a concentration inversion model based on ridge regression can effectively invert the chlorophyll-a concentration in the offshore.

Evaluating geometric measurement accuracy based on 3D model reconstruction of nursery tomato plants by Agisoft photoscan software

Measuring the vegetation structure of a plant with the help of 3D images is a non-destructive method that can be used to monitor the plant’s status during the growing season. Structure from motion (SfM) is a popular optical distance measurement method that uses 2D images to create 3D models. Identifying factors that influence the precision of SfM is important due to its increasing use in phenotyping. The primary objectives of this study were, i) to scrutinize the impact of imaging parameters—including, color checkerboard location, overlap size of images, and background filtering—on the quality indices of 3D models, and ii) to attain precise measurements of leaf area, stem diameter, and height through the combination of images captured from different camera shooting angles. The three nursery tomato plants were photographed at every five-degree rotation of the turntable, five different locations of color checkerboards, and five camera shooting angles. Results indicated the color checkerboard’s presence and position affect the quality indices of the 3D models. Visually and by the lowest mean keypoint size value, the least noisy 3D model was obtained when the wall of the plant pot was covered with a checkerboard. When the base-to-distance (b/y) ratio as the image overlap index increased from 0.1 to 0.2, the generation time of the 3D models decreased by more than two times. Optimal Mean Absolute Percentage Error (MAPE) values of 1.08 %, 1.01 %, and 1.38 % were achieved for the estimation of leaf area, stem diameter, and height, respectively. This was recorded by using images taken at 0°, 30°, and 105° vertical azimuth angles. Constructing a 3D model by integrating images captured from other angles or more azimuth angles represents numerous visual challenges and leads to problems of point mismatch, which affects the accuracy of measurements.

Calibration Methods for Low-Cost Particulate Matter Sensors Considering Seasonal Variability

Many countries use low-cost sensors for high-resolution monitoring of particulate matter (PM2.5 and PM10) to manage public health. To enhance the accuracy of low-cost sensors, studies have been conducted to calibrate them considering environmental variables. Previous studies have considered various variables to calibrate seasonal variations in the PM concentration but have limitations in properly accounting for seasonal variability. This study considered the meridian altitude to account for seasonal variations in the PM concentration. In the PM10 calibration, we considered the calibrated PM2.5 as a subset of PM10. To validate the proposed methodology, we used the feedforward neural network, support vector machine, generalized additive model, and stepwise linear regression algorithms to analyze the results for different combinations of input variables. The inclusion of the meridian altitude enhanced the accuracy and explanatory power of the calibration model. For PM2.5, the combination of relative humidity, temperature, and meridian altitude yielded the best performance, with an average R2 of 0.93 and root mean square error of 5.6 µg/m3. For PM10, the average mean absolute percentage error decreased from 27.41% to 18.55% when considering the meridian altitude and further decreased to 15.35% when calibrated PM2.5 was added.

Machine learning based data governance methods for demand response databases

With the increasing level of grid intelligence and the related demand response database expanding, it is important to study a compound problem data governance method for demand response, while the traditional data governance methods have problems such as not considering data temporality and ignoring the impact of noise and duplicate data on data repair. As a result, this project will develop an anomaly data extraction and repair model based on two-way long and short memory networks, and repair the anomaly data by respective noise smoothing, missing data filling, and duplicate data cleaning. The paper also provides an adaptive moment estimation approach for optimisation to raise the model’s accuracy. The outcomes demonstrated that the study model’s precision for anomalous data extraction was 100% and its recall rate was 80%, which was a significant improvement over the previous state. In terms of anomalous data repair, the research model had the root mean square error value and lowest mean absolute percentage error value when compared with related models, at 0.0049 MPa and 1.375% respectively. Both the abnormal data extraction and repair performance of the research model are greatly improved over the related models, and have important value in the abnormal data governance of demand response databases.

Research on the application of deep learning techniques in stock market prediction and investment decision-making in financial management

Introduction: This paper introduces a deep learning approach based on Convolutional Neural Networks (CNN), Bidirectional Long Short-Term Memory Networks (BiLSTM), and attention mechanism for stock market prediction and investment decision making in financial management. These methods leverage the advantages of deep learning to capture complex patterns and dependencies in financial time series data. Stock market prediction and investment decision-making have always been important issues in financial management.Methods: Traditional statistical models often struggle to handle nonlinear relationships and complex temporal dependencies, thus necessitating the use of deep learning methods to improve prediction accuracy and decision effectiveness. This paper adopts a hybrid deep learning model incorporating CNN, BiLSTM, and attention mechanism. CNN can extract meaningful features from historical price or trading volume data, while BiLSTM can capture dependencies between past and future sequences. The attention mechanism allows the model to focus on the most relevant parts of the data. These methods are integrated to create a comprehensive stock market prediction model. We validate the effectiveness of the proposed methods through experiments on real stock market data. Compared to traditional models, the deep learning model utilizing CNN, BiLSTM, and attention mechanism demonstrates superior performance in stock market prediction and investment decision-making.Results and Discussion: Through ablation experiments on the dataset, our deep learning model achieves the best performance across all metrics. For example, the Mean Absolute Error (MAE) is 15.20, the Mean Absolute Percentage Error (MAPE) is 4.12%, the Root Mean Square Error (RMSE) is 2.13, and the Mean Squared Error (MSE) is 4.56. This indicates that these methods can predict stock market trends and price fluctuations more accurately, providing financial managers with more reliable decision guidance. This research holds significant implications for the field of financial management. It offers investors and financial institutions an innovative approach to better understand and predict stock market behavior, enabling them to make wiser investment decisions.

Non-invasive monitoring of microalgae cultivations using hyperspectral imager

AbstractHigh expectations are placed on microalgae as a sustainable source of valuable biomolecules. Robust methods to control microalgae cultivation processes are needed to enhance their efficiency and, thereafter, increase the profitability of microalgae-based products. To meet this need, a non-invasive monitoring method based on a hyperspectral imager was developed for laboratory scale and afterwards tested on industrial scale cultivations. In the laboratory experiments, reference data for microalgal biomass concentration was gathered to construct 1) a vegetation index-based linear regression model and 2) a one-dimensional convolutional neural network model to resolve microalgae biomass concentration from the spectral images. The two modelling approaches were compared. The mean absolute percentage error (MAPE) for the index-based model was 15–24%, with the standard deviation (SD) of 13-18 for the different species. MAPE for the convolutional neural network was 11–26% (SD = 10–22). Both models predicted the biomass well. The convolutional neural network could also classify the monocultures of green algae by species (accuracy of 97–99%). The index-based model was fast to construct and easy to interpret. The index-based monitoring was also tested in an industrial setup demonstrating a promising ability to retrieve microalgae-biomass-based signals in different cultivation systems.

Machine Learning Prediction and Evaluation for Structural Damage Comfort of Suspension Footbridge

To investigate the impact of structural damages on the comfort level of suspension footbridges under human-induced vibrations, this study addresses the limitations of traditional manual testing, which often entails significant manpower and material resources. The aim is to achieve rapid estimation and health monitoring of comfort levels during bridge operation. To accomplish this, the study combines finite-element simulation results to establish a data-driven library and introduces three distinct machine learning algorithms. Through comparative analysis, a machine learning-based method is proposed for quick evaluation of bridge comfort levels. Focusing on the Yangjiadong Suspension Bridge, the study evaluates and researches the comfort level of the structure under the influence of human-induced vibrations. The findings revealed a relatively low base frequency and high flexibility. Additionally, when considering the mass of individuals, peak acceleration decreased. The predictive performance of the Artificial Neural Network (ANN) model was found to be superior when accounting for multi-parameter damages, yielding root mean square error (RMSE), mean absolute percentage error (MAPE), and R-squared (R2) values of 0.03, 0.02, and 0.98, respectively. Moreover, the error ratio of the generalization performance analysis was below 5%. Furthermore, the study identified a damage coefficient of 0.13 for the bridge’s main cable, hanger, and steel longitudinal beam. Under a crowd density of 0.5 people per square meter, the predicted peak acceleration was 1.098 m/s2, with a model error of less than 10% compared to the observed value of 1.004 m/s2. These results underscore the model’s effectiveness in swiftly evaluating bridge comfort levels, thereby offering valuable insights for the health monitoring of bridge comfort levels.

Predicting hourly heating load in residential buildings using a hybrid SSA–CNN–SVM approach

This study proposes a hybrid prediction model using sparrow search algorithm (SSA) to optimize the convolutional neural network (CNN) and support vector machine (SVM), in order to perform accurate prediction of secondary supply temperature (Ts2). The historical operation data of Weifang residential building thermal station was adopted and reasonable data preprocessing was performed to suppress the interference of abnormal data. The input variables of the prediction model were screened using the correlation analysis method, taking the influence of the hysteresis effect into consideration. The SSA–CNN–SVM model was then developed for prediction. The performance of the model was evaluated by the root mean square error, mean absolute error, mean absolute percentage error (MAPE), and absolute value of relative error of each time step. The results obtained demonstrated that the SSA–CNN–SVM model has high prediction accuracy. The MAPE values of the two heat exchange stations were between 2.28 % and 2.4 %. The indoor temperature significantly affected the prediction accuracy of Ts2. After the introduction of the indoor temperature, the MAPE of the predicted values of the hybrid model was reduced by 0.35 %. The maximum reduction in MAPE of the SSA–CNN–SVM model was 1.5 % compared with other prediction models.

An ANN Hardness Prediction Tool Based on a Finite Element Implementation of a Thermal–Metallurgical Model for Mild Steel Produced by WAAM

WAAM has emerged as a promising technique for manufacturing medium- and large-scale metal parts due to its high material deposition efficiency and automation level. However, its high heat accumulation and complex thermal evolution strongly affect the resulting microstructures and mechanical properties. The heterogeneous and unpredictable nature of these properties hinder the widespread application of WAAM in the steel construction industry. In this study, an artificial neural network (ANN) hardness model is developed, based on a thermal–metallurgical model for mild steel. The objective is to establish non-linear relationships between the input process parameters and the desired output, i.e., hardness. The thermal–metallurgical model utilizes a well-distributed heat source model, a death-and-birth algorithm, and a metallurgical model to simulate the temperature field and to calculate the microstructure phase fraction. The temperature prediction errors at four thermocouple positions are mostly below 20%. Because of the limited experimental data, twenty-five simulation experiments are performed using the L25 orthogonal array based on the Taguchi method. The analysis of variance (ANOVA) reveals that the travel speed has the greatest impact on hardness. With the dataset from the thermal–metallurgical model, an ANN model to predict hardness is developed. A comparison to experimental data shows excellent performance and accuracy, with the Mean Absolute Percentage Error (MAPE) of ANN predictions within 10% of the targeted hardness.

Experimental Validation of Electrothermal and Aging Parameter Identification for Lithium-Ion Batteries

Modeling and predicting the long-term performance of Li-ion batteries is crucial for the effective design and efficient operation of integrated energy systems. In this paper, we introduce a comprehensive semi-empirical model for Li-ion cells, capturing electrothermal and aging features. This model replicates the evolution of cell voltage, capacity, and internal resistance, in relation to the cell actual operating conditions, and estimates the ongoing degradation in capacity and internal resistance due to the battery use. Thus, the model articulates into two sub-models, an electrothermal one, describing the battery voltage, and an aging one, computing the ongoing degradation. We first propose an approach to identify the parameters of both sub-models. Then, we validate the identification procedure and the accuracy of the electrothermal and aging models through an experimental campaign, also comprising two real cycle load tests at different temperatures, in which real measurements collected from real Li-ion cells are used. The overall model demonstrates good performances in simulating battery characteristics and forecasting degradation. The results show a Mean Absolute Percentage Error (MAPE) lower than 1% for battery voltage and capacity, and a maximum absolute error on internal resistance that is on par with the most up-to-date empirical models. The proposed approach is therefore well-suited for implementation in system modeling, and can be employed as an informative tool for enhancing battery design and operational strategies.

Minimizing Temperature Deviations in Rubber Mixing Process using Artificial Neural Networks

ABSTRACT Rubber mixing is a complex manufacturing process that poses challenges for process control due to the high number of control variables including mixing parameter settings, rheological behaviour, compound viscosity and batch-dependent material variations. Already small deviations from the control variables can influence the compound properties, leading to increased scrap rates. To address these challenges, this paper introduces an Artificial Intelligence (AI)-based approach to enhance process control in rubber mixing by predicting mixing temperatures from input variables. The proposed method utilizes Feedforward Neural Networks (FFN) to enable early identification of batch-specific temperature deviations, enabling systematic improvements with each new application. The FFN was trained on a diverse dataset encompassing various rubber recipes and batches. Post-training, the FFN demonstrated remarkable accuracy, achieving a Mean Absolute Percentage Error (MAPE) of 1.00% on the training dataset and 1.44% on the validation dataset, thereby showcasing its efficacy in predicting temperature fluctuations within the mixing process. Consequently, the FFN can determine the relevant input variables necessary to achieve specific mixing temperatures, providing a foundation for an automated control system in rubber mixing process. This paper outlines the system architecture of the FFN tailored for rubber mixing and provides a comprehensive overview of the experimental results.

Geographic heterogeneity of activation functions in urban real-time flood forecasting: Based on seasonal trend decomposition using Loess-Temporal Convolutional Network-Gated Recurrent Unit model

Urban real-time flood forecasting is crucial for flood prevention and sustainable development, but it poses challenges due to data inputs and activation functions selection in data-driven models without sufficient focus of geographic heterogeneity. In this study, a novel Seasonal Trend Decomposition using Loess (STL)-Temporal Convolutional Network (TCN)-Gated Recurrent Unit (GRU) model is proposed to improve urban real-time flood forecasting accuracy, twenty-one different activation functions are considered for geographic heterogeneity. Experiments are conducted at six urban drainage system locations in Odense, Denmark. The results show that: (1) STL effectively prepares data for forecasting using TCN and GRU models, leading improved performance compared to single models. STL-TCN-GRU deep learning model demonstrates strong applicability in urban forecasting, achieving an overall accuracy of 0.0079, 0.0140, 0.0482 and 0.9581 in Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE) and Nash-Sutcliffe efficiency coefficient (NSE), respectively. (2) Softsign emerges as the best activation function for forecasting in lower drainage system locations, with average accuracy of 0.0094, 0.0018, 0.0049 and 0.9938 in MAE, RMSE, MAPE and NSE, respectively. Furthermore, Softsign proves to be the best activation function for forecasting in middle drainage system locations, with average accuracy of 0.0047, 0.0081, 0.0309 and 0.9547 in MAE, RMSE, MAPE and NSE, respectively. Swish is the best activation function for forecasting in upper drainage system locations, with average accuracy of 0.0052, 0.0080, 0.0627 and 0.9060 in MAE, RMSE, MAPE and NSE, respectively. This study provides valuable insights for urban real-time flood forecasting modeling with high accuracy and evidence for activation function selection in data-driven models like STL-TCN-GRU for geographic heterogeneity.

Accuracy of the Apple Watch Series 4 and Fitbit Versa for Assessing Energy Expenditure and Heart Rate of Wheelchair Users During Treadmill Wheelchair Propulsion: Cross-sectional Study.

The Apple Watch (AW) Series 1 provides energy expenditure (EE) for wheelchair users but was found to be inaccurate with an error of approximately 30%, and the corresponding error for heart rate (HR) provided by the Fitbit Charge 2 was approximately 10% to 20%. Improved accuracy of estimated EE and HR is expected with newer editions of these smart watches (SWs). This study aims to assess the accuracy of the AW Series 4 (wheelchair-specific setting) and the Fitbit Versa (treadmill running mode) for estimating EE and HR during wheelchair propulsion at different intensities. Data from 20 manual wheelchair users (male: n=11, female: n=9; body mass: mean 75, SD 19 kg) and 20 people without a disability (male: n=11, female: n=9; body mass: mean 75, SD 11 kg) were included. Three 4-minute wheelchair propulsion stages at increasing speed were performed on 3 separate test days (0.5%, 2.5%, or 5% incline), while EE and HR were collected by criterion devices and the AW or Fitbit. The mean absolute percentage error (MAPE) was used to indicate the absolute agreement between the criterion device and SWs for EE and HR. Additionally, linear mixed model analyses assessed the effect of exercise intensity, sex, and group on the SW error. Interclass correlation coefficients were used to assess relative agreement between criterion devices and SWs. The AW underestimated EE with MAPEs of 29.2% (SD 22%) in wheelchair users and 30% (SD 12%) in people without a disability. The Fitbit overestimated EE with MAPEs of 73.9% (SD 7%) in wheelchair users and 44.7% (SD 38%) in people without a disability. Both SWs underestimated HR. The device error for EE and HR increased with intensity for both SWs (all comparisons: P<.001), and the only significant difference between groups was found for HR in the AW (-5.27 beats/min for wheelchair users; P=.02). There was a significant effect of sex on the estimation error in EE, with worse accuracy for the AW (-0.69 kcal/min; P<.001) and better accuracy for the Fitbit (-2.08 kcal/min; P<.001) in female participants. For HR, sex differences were found only for the AW, with a smaller error in female participants (5.23 beats/min; P=.02). Interclass correlation coefficients showed poor to moderate relative agreement for both SWs apart from 2 stage-incline combinations (AW: 0.12-0.57 for EE and 0.11-0.86 for HR; Fitbit: 0.06-0.85 for EE and 0.03-0.29 for HR). Neither the AW nor Fitbit were sufficiently accurate for estimating EE or HR during wheelchair propulsion. The AW underestimated EE and the Fitbit overestimated EE, and both SWs underestimated HR. Caution is hence required when using SWs as a tool for training intensity regulation and energy balance or imbalance in wheelchair users.