Lowest Mean Absolute Percentage Error Research Articles

Rainfall Projections for the Brazilian Legal Amazon: An Artificial Neural Networks First Approach

Rainfall in the Brazilian Legal Amazon (BLA) is vital for climate and water resource management. This research uses spatial downscaling and validated rainfall data from the National Water and Sanitation Agency (ANA) to ensure accurate rain projections with artificial intelligence. To make an initial approach, Recurrent Neural Networks (RNNs) with Long Short-Term Memory (LSTM) were employed to forecast rainfall from 2012 to 2020. The RNN model showed strong alignment with the observed patterns, accurately predicting rainfall seasonality. However, median comparisons revealed fair approximations with discrepancies. The Root Mean Square Error (RMSE) ranged from 6.7 mm to 11.2 mm, and the coefficient of determination (R2) was low in some series. Extensive analyses showed a low Wilmott agreement and high Mean Absolute Percentage Error (MAPE), highlighting limitations in projecting anomalies and days without rain. Despite challenges, this study lays a foundation for future advancements in climate modeling and water resource management in the BLA.

Accuracy of Wrist-Worn Heart Rate Monitors: A Comprehensive Review of Smartwatches in Exercise Monitoring

Wrist-worn heart rate monitors have become integral in personal health monitoring, particularly during exercise and daily activities. This review evaluates the precision of various commercially available smartwatches in assessing heart rate, focusing on their accuracy across different physical activities, such as walking, running, cycling, and elliptical training. A systematic analysis of 29 peer-reviewed studies, encompassing over 900 participants, was conducted to compare devices like the Apple Watch, Fitbit, Garmin, and others. The results demonstrate significant variation in accuracy depending on the device, type of activity, and individual factors such as body mass index and skin tone. The Apple Watch consistently outperformed other devices, showing the lowest mean absolute percentage error (MAPE), particularly during moderate-intensity activities. In contrast, devices such as the Fitbit Blaze exhibited greater error rates, especially during high-intensity exercises or activities involving significant arm movement. Although wrist-worn heart rate monitors provide users with convenient real-time data, limitations such as motion artifacts, device placement, and external factors may affect their reliability.

Advancing Geotechnical Evaluation of Wellbores: A Robust and Precise Model for Predicting Uniaxial Compressive Strength (UCS) of Rocks in Oil and Gas Wells

This study examines the efficacy of various machine learning models for predicting the uniaxial compressive strength (UCS) of rocks in oil and gas wells, which are essential for ensuring wellbore stability and optimizing drilling operations. The investigation encompasses Linear Regression, ensemble methods (including Random Forest, Gradient Boosting, XGBoost, and LightGBM), support vector machine-based regression (SVM-SVR), and multilayer perceptron artificial neural network (MLP-ANN) models. The results demonstrate that XGBoost and Gradient Boosting offer superior predictive accuracy for UCS in drillability, as indicated by low Mean Absolute Percentage Error (MAPE) values of 3.87% and 4.18%, respectively, and high R2 scores (0.8542 for XGBoost). These models emerge as optimal choices for UCS prediction focused on drillability, offering increased accuracy and reliability in practical engineering scenarios. Ensemble methods and MLP-ANN emerge as frontrunners, providing valuable tools for improving wellbore stability assessments, optimizing drilling parameter selection, and facilitating informed decision-making processes in oil and gas drilling operations. Moreover, this study lays a foundation for further research in drillability-centred predictive modelling for geotechnical parameters, advancing our understanding of rock behaviour under drilling conditions.

Design and Equivalent Circuit Model Extraction of a Fractal Slot-Loaded 3–40 GHz Super Wideband Antenna

In this paper, we present the design and equivalent circuit model (ECM) of a fractal slot-loaded super wideband (SWB) antenna for compact and high-performance applications operating in the 3–40 GHz range. The proposed antenna features a compact dimension of 40 × 35 × 1.57 mm³, a measured bandwidth ratio of 13:1, a peak gain of 9.7 dBi, an average radiation efficiency of 94%, and a low cross-polarization level across the entire bandwidth. The presented ECM is derived using transmission line theory and incorporates the individual behavior of each constituting element of the antenna. A dual sequential optimization approach is employed to determine the optimal element values. The ECM results show good agreement with both simulated and measured results in terms of the magnitude of reflection coefficient |S11| and both real and imaginary impedances with low mean absolute percentage errors of 4.9%, 7.5%, and 7.7%, respectively, demonstrating the model’s ability to accurately predict the antenna’s performance.

Machine learning prediction of microhardness and corrosion current density of Ni-Co alloys produced by scanning jet electrodeposition

ABSTRACT Scanning jet electrodeposition (SJE) is a promising method for electrodeposition. Integrated with automation and machine learning, intelligent electrodeposition can be achieved. In this study, a machine learning method was used to develop the relationship between the process parameters and properties of Ni-Co alloys produced by SJE. To evaluate the various models, the coefficient of determination (R 2), mean absolute percentage error (MAPE), and root mean square error (RMSE) were calculated as performance metrics. The multivariate nonlinear regression (MNR) model shows a higher R 2, lower MAPE and RMSE values, and better performance in both the training and testing sets, indicating superior approximation and prediction accuracies for microhardness. A back-propagation artificial neural network (BP-ANN) shows stable R 2 value of closest to 1, and the lowest MAPE and RMSE for corrosion current density and presents an exceptional generalisation ability. Finally, under the present experimental conditions, the highest microhardness and corresponding process parameters were determined using the MNR model, and the lowest corrosion current density was optimised using the BP-ANN model.

Effect of chemical modification on hydraulic conductivity of stratified porous media

ABSTRACT Understanding hydraulic conductivity is essential in porous media as it dictates the capacity of fluids to permeate through these substances. This research investigated the impact of chemical modification using fly ash on hydraulic conductivity. The coefficient of hydraulic conductivity along the bedding plane at different angles of orientations was predicted by employing machine learning techniques such as gene expression programming (GEP), artificial neural networks (ANNs), and regression tree (RT), considering various influencing parameters. As the amount of fly ash in the porous medium rises, the hydraulic conductivity decreases. The statistical results demonstrate that the proposed ANN model effectively forecasted the horizontal hydraulic conductivity. It achieved the highest R2 value (0.989), the lowest mean absolute percentage error value (5.772), the lowest scatter index value (0.040), and the lowest Akaike Information Criterion value (−2,359.914) compared to the GEP, RT, and previously established approaches. Moreover, a novel equation has been proposed for determining horizontal hydraulic conductivity, which can be used for any stratified medium, distinguishing this work from previous ones. These findings can inform global practices in mitigating groundwater contamination, enhancing irrigation efficiency, and improving the design of subsurface fluid flow systems, making it a valuable contribution to both academic and practical applications worldwide.

Forecasting Green Chilli Prices: Using Data Analytics to Gain Market Understanding

Green chilli is a commercially significant vegetable crop grown year-round due to its high demand for both nutritional and health benefits. India stands as the largest producer and consumer of chilli globally, with West Bengal leading in area under cultivation and ranking sixth in production. This study aims to compare and identify the most accurate model for forecasting green chilli prices in the Haldibari market of Cooch Behar district, West Bengal. Price data from January 2015 to May 2024, sourced from AGMARKNET, was used for model development, with 85% of the data allocated for training and 15% for validation. The models tested include the traditional Seasonal Auto-Regressive Integrated Moving Average (SARIMA) and machine learning models like Artificial Neural Networks (ANN) and Support Vector Regression (SVR). Among these, the ANN model is found to be the most accurate, with a low Mean Absolute Percentage Error (MAPE) of 0.14. The model is used to forecast prices for the next 12 months, up to May 2025. The study’s findings can aid farmers and policymakers in creating effective crop planning strategies, helping to boost local farm income.

Comparison of Triple Exponential Smoothing and ARIMA in Predicting Cryptocurrency Prices

Cryptocurrency has emerged as a prominent digital asset over the past decade, but its high price volatility presents significant challenges for investors. This study evaluates and compares the effectiveness of the Triple Exponential Smoothing (TES) and Autoregressive Integrated Moving Average (ARIMA) methods in forecasting the prices of five major cryptocurrencies: Bitcoin (BTC), Ethereum (ETH), Binance Coin (BNB), Solana (SOL), and Ripple (XRP). TES models trends and seasonality in time series data, while ARIMA captures autoregressive patterns and moving averages. The dataset is split into 80% for training and 20% for testing, with performance evaluated using Mean Absolute Percentage Error (MAPE) and Root Mean Squared Error (RMSE). TES outperforms ARIMA in predicting Bitcoin and Binance Coin, achieving MAPE values of 10.38% and 13.81%, and RMSE values of 3,985.55 and 41.28, respectively. However, ARIMA shows better performance for Ethereum, Solana, and Ripple, with MAPE ranging from 8.78% to 32.84% and RMSE between 0.08 and 204.59. Notably, Ethereum has the lowest MAPE at 8.78%, while Ripple exhibits the smallest RMSE at 0.08. These findings suggest that TES is more suitable for cryptocurrencies with relatively stable price patterns, while ARIMA is better adapted to forecasting highly volatile assets. This research underscores the importance of selecting forecasting models based on the specific characteristics of each cryptocurrency

Evaluation of Trend Models Performance and Forecasting Onion Production: A Comparative Study

Onions, a significant spice crop in Bangladesh, frequently experience price surges due to an imbalance between demand and supply, compelling the government to import onions. The volatility of the onion market in Bangladesh and the measures needed to stabilize the situation require intensive policy attention. Therefore, this study was carried out to forecast the onion production over time and to evaluate the efficacy of various trend models. The secondary data was collected from the Food and Agriculture Organization's Statistical Database (FAOSTAT) from 1972 to 2022. The autoregressive integrated moving average (ARIMA) model was used to predict the onion production over time. Additionally, metrics such as modified and the coefficient of determination were used to evaluate the effectiveness of eight trend models comprehensively. The findings indicate that the cubic model is superior to others since it has larger and adjusted values. The models' accuracy is also determined utilizing mean absolute percentage error (MAPE) and mean absolute deviation (MAD) metrics. The cubic trend model exhibits lower MAPE and MAD values of 29.43% and 80638.67, respectively. Therefore, it is concluded that the ARIMA (2, 2, 2) model is the best model for forecasting the annual production of onion in Bangladesh. A five-year trend analysis spanning from 2023-2027 indicates an upward trend. Such a result offers a critical insight for stakeholders and policymakers in formulating sustainable onion production strategies. J Bangladesh Agril Univ 22(3): 386-395, 2024

Modeling and Forecasting High Frequency Currency Exchange Rates. A Comparative Study of ARIMA, ANN, and Hybrid ARIMA-ANN Models

Aims/ Objectives: The study develops comparative results on the modeling and prediction performance of the autoregressive integrated moving average (ARIMA), artificial neural network (ANN) and the hybrid ARIMA-ANN time series models for high frequency data. Methodology: The study made use of ARIMA, ANN, and hybrid ARIMA-ANN models to forecasts the East Africa Community countries' daily currency exchange rates data which were obtained from the Central Bank of Kenya website and covered the period from January 2017 to December 2023. Stationarity of the time series data was established using the ADF test. The Ljung Box test and ACF plots were used to establish and compare the goodness-of-fit of the resultant models while the Mean Absolute Percentage Error (MAPE) and Root Mean Squared Error(RMSE) values were used to compare the prediction performance. Results: The study established that the hybrid ARIMA-ANN methodolgy provided better-fitting models for the currency exchange rates data compared to ARIMA and ANN modeling strategies since all the Lyung Box test statistics had p values greater than 5%. Comparatively, the hybrid methodology registered lower MAPE and RMSE values hence had better prediction accuracy compared to ARIMA and ANN methods. Conclusion: The Hybid methodology improves the modeling and forecasting accuracy over the ARIMA and ANN models for high frequency time series data due to its ability to captures both the linear and nonlinear patterns in the time series data.

Investigation of operational settings, environmental conditions, and faults on the gas turbine performance

Abstract Gas turbine engines are complex mechanical marvels widely employed in diverse applications such as marine vessels, aircraft, power generation, and pumping facilities. However, their intricate nature renders them susceptible to numerous operational faults, significantly compromising their performance and leading to excessive emissions, consequently incurring stringent penalties from environmental regulatory bodies. Moreover, the deterioration of gas turbine performance is exacerbated by variations in working conditions based on operational settings and environmental conditions. Past studies have focused on certain working conditions that limit effectiveness in real-world applications where operational settings and environmental conditions vary during operations. The influence of these working conditions on the performance of gas turbines also needs to be assessed, as they can lead to different fault patterns resulting in unplanned maintenance, unnecessary maintenance costs, unsafe conditions and stringent penalties. This study uses the gas turbine simulation program to simulate a high-bypass turbofan engine inspired by Pratt & Whitney PW-4056, analysing the combined effects of operational settings and environmental conditions on engine performance while also incorporating simulations of common gas turbine faults like fouling and erosion in various locations and severities along the gas path. The model’s accuracy is confirmed by low mean absolute percentage errors of 0.004% of thrust at the cycle reference point and 0.15% and 0.28% at 2 km and 7 km altitudes, respectively, demonstrating the model’s robustness across varying operational scenarios. In conclusion, this research highlights the significant effects of operational settings and environmental factors on gas turbine performance, particularly impacting specific fuel consumption and thrust. The study reveals that operational settings and environmental factors significantly impact fuel consumption and thrust. Specifically, compressor fouling and low-pressure turbine erosion increase nitrogen oxide (NOx) emissions by 4.5% and 11.1%, while fouling of nozzle guide vanes and high-pressure turbine erosion raise unburned hydrocarbon by 10.0% and 20.2%, and carbon monoxide (CO) by 3.2% and 5.2%, respectively, compared to a healthy engine. These insights highlight the importance of component-specific degradation in influencing gas turbine performance and emissions.

Entropy optimization on Casson nanofluid flow with radiation and Arrhenius activation energy over different geometries: A numerical and statistical approach

This study employs numerical and statistical approaches to investigate the entropy optimization of steady Casson nanofluid flow over three different geometries subject to boundary conditions induced by convective flow. Multiple linear regression is employed to statistically examine. The present model incorporates several novel elements, such as Arrhenius activation energy, Brownian motion, the Cattaneo-Christov dual flux, thermophoresis, thermal radiation, and so on. Moreover, a comparison is presented between Newtonian and non-Newtonian fluids. By applying the proper similarity transformations, ordinary differential equations (ODEs) are obtained by converting foundational partial differential equations (PDEs). The Runge-Kutta fourth-order method is utilised to solve the obtained ODEs along with the shooting technique. The outcomes are visually depicted via tables and graphs. The velocity drops with increasing Grashof number, and the magnetic field becomes progressively more forceful as the suction parameter increases. The temperature gets reduced with the increase of the suction parameter, solute Grashof number increases with the magnetic field, thermophoresis, and radiation parameters. The entropy is observed to rise with the increase of the effective parameters (magnetic field, Brinkmann number and radiation). The MAD (mean absolute deviation), MSE (mean squared error), and RMSE (root mean square error) values are approaching zero, indicating that the derived outcomes are highly accurate. A lower MAPE (mean absolute percentage error) suggests that the model has a higher level of precision. Therefore, the outcomes of the present model are more precise and reliable. This study has various potential applications such as power plant heat exchangers, material processing industries, and solar thermal energy systems.

Predicting the fetal weight by ultrasonography for isolated polyhydramnios: Comparison of 14 formulas.

To recalculate the estimated fetal weight (EFW) based on ultrasound measurements in patients complicated with isolated polyhydramnios, using 14 current formulas to observe which formula better predicts the EFW. This study examined pregnant women who gave birth in the hospital between January 2015 and January 2020. Maximum vertical pocket (MVP) was classified as, mild, moderate, and severe polyhydramnios, and the patients' measurements were reanalyzed using 14 formulas. The estimation of birth weight (EBW) alongside observed birth weight (OBW) facilitated the computation of statistical indices, namely the mean absolute percentage error (MAPE) expressed as [(EBW - OBW)/OBW × 100], the mean percentage error (MPE) denoted as (EBW - OBW)/(OBW × 100), and their corresponding 95% confidence intervals. A total of 564 polyhydramnios patients were included in the study. When looking at the MAPE, the lowest rate (7.65) was found in the Hadlock 2 formula. Hadlock 1, Hadlock 3, and Shinozuka formulas demonstrated MAPE values most closely aligned with Hadlock 2. Weiner I and Thurnau were the formulas with the highest MAPE values. When the cut-off values for MAPE were taken as 10%, 4/14 of the formulas (Weiner I-II, Vintzleos and Thurnau) gave results above 10%. Among 14 formulas, 3 (21.4%) had positive (sonographic overestimation) (Hadlock 3, Shinozuka, and Vintzleos) and the other 11 (78.6%) had negative MPE (sonographic underestimation). The Hadlock 2 formula had the lowest MAPE in predicting birth weight in patients with polyhydramnios, closely followed by the Hadlock 1, Hadlock 3, and Shinozuka formulas.

Artificial intelligence-implemented prediction and cost-effective optimization of micropollutant photodegradation using g-C3N4/Bi2O3 heterojunction

Micropollutants pose a formidable hazard to human health, however, conventional water treatment fails to efficiently and economically eliminate them. The imperative need for real-time concentration monitoring of these micropollutants, vital for effective elimination and cost optimization, has proven elusive without the intervention of artificial intelligence. To bridge these critical gaps, g-C3N4/Bi2O3 heterojunction was proposed as an efficient catalyst to eliminate recalcitrate micropollutants under visible light in this study, demonstrating a 1.8-fold increase in catalytic efficiency compared to g-C3N4, and furthermore, artificial intelligence models, i.e., the proposed Long short-term memory combined with Gated residual network (ALG) and Deep Deterministic Policy Gradient (DDPG), were employed for real-time concentration prediction of micropollutants (R2 = 0.9541) and their treatment processing optimization, respectively. The predominant active species for degrading micropollutants were identified as O2•− and 1O2. We proved that HOO• serves as an intermediate to transform HO• and H2O2 generated in-situ to O2•− and 1O2, further enhancing micropollutant elimination. Compared to other models, the ALG demonstrated significantly lower root mean square error (RMSE = 0.0432) and mean absolute percentage error (MAPE = 4.8273). Weight analysis revealed that Catalyst Type (31.80 %) is the most critical factor. Finally, the proposed DDPG model dynamically adjusted its treatment strategy, aiming to discover an optimal balance among cost, reaction time, and removal efficiency. This study provides mechanistic insights into the pivotal roles of artificial intelligence in prediction and cost-effective optimization of micropollutant elimination.

Forecasting the incidence frequencies of schizophrenia using deep learning

Mental disorders are becoming increasingly prevalent worldwide, and accurate incidence forecasting is crucial for effective mental health strategies. This study developed a long short-term memory (LSTM)-based recurrent neural network model to predict schizophrenia in inpatients in Taiwan. Data was collected on individuals aged over 20 years and diagnosed with schizophrenia between 1998 and 2015 from the National Health Insurance Research Database (NHIRD). The study compared six models, including LSTM, exponential smoothing, autoregressive integrated moving average, particle swarm optimization (PSO), PSO-based support vector regression, and deep neural network models, in terms of their predictive performance. The results showed that the LSTM model had the best accuracy, with the lowest mean absolute percentage error (2.34), root mean square error (157.42), and mean average error (154,831.70). This finding highlights the reliability of the LSTM model for forecasting mental disorder incidence. The study's findings provide valuable insights that can help government administrators devise clinical strategies for schizophrenia, and policymakers can use these predictions to formulate healthcare education and financial planning initiatives, fostering support networks for patients, caregivers, and the public.

A direct comparison of multi-energy x-ray and proton CT for imaging and relative stopping power estimation of plastic and ex-vivo phantoms

Objective.Proton therapy administers a highly conformal dose to the tumour region, necessitating accurate prediction of the patient's 3D map of proton relative stopping power (RSP) compared to water. This remains challenging due to inaccuracies inherent in single-energy computed tomography (SECT) calibration. Recent advancements in spectral x-ray CT (xCT) and proton CT (pCT) have shown improved RSP estimation compared to traditional SECT methods. This study aims to provide the first comparison of the imaging and RSP estimation performance among dual-energy CT (DECT) and photon-counting CT (PCCT) scanners, and a pCT system prototype.Approach.Two phantoms were scanned with the three systems for their performance characterisation: a plastic phantom, filled with water and containing four plastic inserts and a wood insert, and a heterogeneous biological phantom, containing a formalin-stabilised bovine specimen. RSP maps were generated by converting CT numbers to RSP using a calibration based on low- and high-energy xCT images, while pCT utilised a distance-driven filtered back projection algorithm for RSP reconstruction. Spatial resolution, noise, and RSP accuracy were compared across the resulting images.Main results.All three systems exhibited similar spatial resolution of around 0.54 lp/mm for the plastic phantom. The PCCT images were less noisy than the DECT images at the same dose level. The lowest mean absolute percentage error (MAPE) of RSP,(0.28±0.07)%, was obtained with the pCT system, compared to MAPE values of(0.51±0.08)%and(0.80±0.08)%for the DECT- and PCCT-based methods, respectively. For the biological phantom, the xCT-based methods resulted in higher RSP values in most of the voxels compared to pCT.Significance.The pCT system yielded the most accurate estimation of RSP values for the plastic materials, and was thus used to benchmark the xCT calibration performance on the biological phantom. This study underlined the potential benefits and constraints of utilising such a novelex-vivophantom for inter-centre surveys in future.

Multivendor comparison of quantification accuracy of effective atomic number by Dual-Energy CT: A phantom study

PurposeOur study aimed to compare the accuracy of the effective atomic number (Zeff) of five dual-energy CT (DECT) from three vendors and different generations under different scanning parameters. MethodsZeff accuracy of five DECT scanners with twelve tube voltage configurations was evaluated by using the TomoTherapy cheese phantom. The potential dose dependence of the Zeff was investigated using three radiation dose (5, 15, and 25 mGy), and the robustness of Zeff was simulated for different organs of the body by placing the inserts at different positional depths. Bias and mean absolute percentage error (MAPE) were used to characterize the accuracy of Zeff. Data underwent analysis using one-way ANOVA, followed by the Turky and LSD post hoc tests, simple linear regression, and linear mixed models. ResultsAll tube voltage configurations had a bias of less than 1. Dual layer detector DECT (dl-DECT) −140 kV has the lowest MAPE (1.79 %±1.93 %). The third generation dual source DECT (ds-DECT) and the second generation rapid switch DECT (rs-DECT) have higher MAPE than their predecessor DECT. The results of the linear mixed model showed that tube voltage configuration (F=16.92, p < 0.001) and insert type (F=53.26, p < 0.001) significantly affect the MAPE. In contrast, radiation dose only has a significant effect on the MAPE of rs-DECT. The inserts position does not affect the final MAPE. ConclusionWhen scanning different inserts, Zeff accuracy varies by vendor and DECT generation. Of all the scanners, dl-DECT had the highest Zeff accuracy. Upgrading DECT generation doesn’t lead to higher accuracy, or even lower.

A Study of Suitable Forecasting Model for the Export Quantity of Fresh or Chilled Fish in Thailand

The aim of this study was to find a suitable forecasting model for the export quantity of fresh or chilled fish in Thailand using seven statistical methods including Box-Jenkins method, Holt’s exponential smoothing method, Brown’s exponential smoothing method, damped trend exponential smoothing method, simple seasonal exponential smoothing method, Winters’ additive exponential smoothing method, and Winters’ multiplicative exponential smoothing method. The average monthly export quantity from the website of the Office of Agricultural Economics was used for data analysis. The data set used to create the models was the export quantity from January 2011 to December 2020 for 120 months. The data set used to compare the accuracy was the export quantity from January to December 2021 for 12 months using the lowest mean absolute percentage error. The results indicated that the most accurate method was the damped trend exponential smoothing method.

Using partial discharge data to identify highly sensitive electrochemical parameters of aged lithium-ion batteries

Highly sensitive (HS) electrochemical parameters can serve as battery aging indicators, deserving thorough examination. To identify these parameters and determine their cycle evolution across a wide range of battery state of health, this study conducts a three-stage analysis. First, after a detailed evaluation, a global sensitivity analysis is performed to evaluate the sensitivity values of ten selected parameters across five intervals of discharge voltage curves simulated through the Doyle-Fuller-Newman theory. Such analysis facilitates the classification of these parameters into HS and low sensitive (LS) groups. Next, a deep learning (DL) model is developed for parameter identification, employing as minimal data as possible, specifically only partial discharge curves and their corresponding first-order derivatives, as inputs, while the electrochemical parameters are generated as outputs. It is found that utilizing the model built with the HS parameters achieves a remarkably low mean absolute percentage error (MAPE) of 0.25 % for output prediction. On the contrary, adopting the LS parameters yields a MAPE of 13.30 %, highlighting the significant impact of parameter sensitivity on the model outputs. Lastly, the well-trained DL model is directly applied to identify the five HS parameters pertaining to real-world commercial batteries, whose state of health spans approximately from 97 % down to 55 %. With the experimentally recorded partial discharge voltage data, the model generates the cycle evolutions of the HS parameters, revealing a noticeable decrease in their values ranging from 12 % to 40 % with aging.

Short-Term Energy Forecasting to Improve the Estimation of Demand Response Baselines in Residential Neighborhoods: Deep Learning vs. Machine Learning

Promoting flexible energy demand through response programs in residential neighborhoods would play a vital role in addressing the issues associated with increasing the share of distributed solar systems and balancing supply and demand in energy networks. However, accurately identifying baseline-related energy measurements when activating energy demand response events remains challenging. In response, this study presents a deep learning-based, data-driven framework to improve short-term estimates of demand response baselines during the activation of response events. This framework includes bidirectional long-term memory (BiLSTM), long-term memory (LSTM), gated recurrent unit (GRU), convolutional neural networks (CNN), deep neural networks (DNN), and recurrent neural networks (RNN). Their performance is evaluated by considering different aggregation levels of the demand response baseline profile for 337 dwellings in the city of La Rochelle, France, over different time horizons, not exceeding 24 h. It is also compared with fifteen traditional statistical and machine learning methods in terms of forecasting accuracy. The results demonstrated that deep learning-based models, compared to others, significantly succeeded in minimizing the gap between the actual and forecasted values of demand response baselines at all different aggregation levels of dwelling units over the considered time-horizons. BiLSTM models, followed by GRU and LSTM, consistently demonstrated the lowest mean absolute percentage error (MAPE) in most comparison experiments, with values up to 9.08%, 8.71%, and 9.42%, respectively. Compared to traditional statistical and machine learning models, extreme gradient boosting (XGBoost) was among the best, with a value up to 11.56% of MAPE, but could not achieve the same level of forecasting accuracy in all comparison experiments. Such high performance reveals the potential of the proposed deep learning approach and highlights its importance for improving short-term estimates of future baselines when implementing demand response programs in residential neighborhood contexts.