Average Mean Absolute Percentage Error Research Articles

Hybrid algorithm for predicting the temperature-variation-induced wavelength drift of DBR semiconductor lasers.

In this paper, a hybrid algorithm to predict the wavelength drift induced by ambient temperature variation in distributed Bragg reflector semiconductor lasers is proposed. This algorithm combines the global search capability of a genetic algorithm (GA) and the supermapping ability of an extreme learning machine (ELM), which not only avoids the randomness of ELM but also improves its generalization performance. In addition, a tenfold cross-validation method is employed to determine the optimal activation function and the number of hidden layer nodes for ELM to construct the most suitable model. After applying multiple sets of test data, the results demonstrate that GA-ELM can quickly and accurately predict the wavelength drift, with an average rms error of 4.09×10-4nm and average mean absolute percentage error of 0.21 %. This model is expected to combine the temperature and current tuning models for a wavelength in follow-up research to achieve rapid tuning and high stability of a wavelength without additional devices.

Battery capacity trajectory prediction by capturing the correlation between different vehicles

Advances in battery technology and dwindling oil resources have greatly boosted the popularity of electric vehicles (EVs). Accurate prediction of battery capacity trajectory is critical to ensure safety and timely maintenance of EVs. However, present studies only based on laboratory data. To bridge this gap, this paper proposes a data-driven capacity prediction framework using the vehicle field data, which can automatically match the aging pattern (AP) of the vehicle and fully utilizes the correlation between vehicles and does not need to extract features. The real capacity of battery pack in vehicle is calculated by ampere-hour integration method combined with open circuit voltage correction. To assess the overall effectiveness of this diagnostic methodology, the target vehicle's accessible data is divided into three stages: early, middle, and late. It is demonstrated that after automatic matching of AP, the average mean absolute percentage error (MAPE) of capacity prediction based on the multioutput spectral mixture Gaussian process (MOSMGP) are reduced by 9.82%, 21.25%, and 26.92% respectively in three different aging stages. Compared with the other six machine learning methods, the MOSMGP has the highest accuracy in early prediction. Its average MAPE and average root mean squared error (RMSE) are only 1.39% and 1.92Ah, respectively.

Epidermal piezoresistive structure with deep learning-assisted data translation

Continued research on the epidermal electronic sensor aims to develop sophisticated platforms that reproduce key multimodal responses in human skin, with the ability to sense various external stimuli, such as pressure, shear, torsion, and touch. The development of such applications utilizes algorithmic interpretations to analyze the complex stimulus shape, magnitude, and various moduli of the epidermis, requiring multiple complex equations for the attached sensor. In this experiment, we integrate silicon piezoresistors with a customized deep learning data process to facilitate in the precise evaluation and assessment of various stimuli without the need for such complexities. With the ability to surpass conventional vanilla deep regression models, the customized regression and classification model is capable of predicting the magnitude of the external force, epidermal hardness and object shape with an average mean absolute percentage error and accuracy of <15 and 96.9%, respectively. The technical ability of the deep learning-aided sensor and the consequent accurate data process provide important foundations for the future sensory electronic system.

Medium- to long-term nickel price forecasting using LSTM and GRU networks

Recently, nickel is a critical metal for manufacturing stainless steel, rechargeable electric vehicle batteries, and alloys utilized in the state-of-the-art technologies. The use of more environmentally friendly electric vehicles has become widespread and brought tackling climate change to forefront, especially for reducing greenhouse gas emissions. Therefore, the demand for rechargeable batteries that power electric vehicles and the need for the nickel in the production of these batteries will increase as well. In addition to those, nickel prices significantly impact mine investment decisions, mine planning, economic development of nickel companies, and countries that depend on nickel resources. However, there is uncertainty about how the nickel price will trend in the future, and the solution to this problem attracts the attention of researchers. For forecasting nickel price, this paper proposes recurrent neural networks-based on long short-term memory (LSTM) and gated recurrent unit (GRU) networks, classified as deep learning algorithms. Mean absolute percentage error (MAPE) was used as the performance measure to compute the accuracy of the proposed techniques. As a result, it has been determined that the LSTM and GRU networks are very useful and successful in forecasting the nickel price variations owing to having average MAPE values of 7.060% and 6.986%, respectively. Furthermore, it has been observed that GRU networks surpassed the LSTM networks by 33% in terms of average computational time.

Development of An Artificial Neural Network Based Time-Series Fault Predictive Model for Power System

Electrical power is one of the most important forms of energy which is needed in almost every field of human endeavour. However, increase in size of electrical power structure without proper planning has negative effects on power system supplied to end users, thereby increasing the fault level of the network. This research paper therefore, developed an Artificial Neural Network based Time Series (ANN-TS) fault predictive model for forecasting of fault levels in power system. In this paper, ANN-TS model was trained with three years (2015-2017) outage fault frequency and fault duration data obtained from Ayede 132/33 kV transmission substation of Transmission Company of Nigeria (TCN) in Ibadan using Resilient Back-Propagation (RBP) algorithm and simulation was carried out in MATLAB environment using Mean Absolute Percentage Error (MAPE) as performance metric. The model was used to predict yearly fault frequency and fault duration for twenty-three years (2018-2040). The results of the fault frequency forecast showed that monthly forecast graphs were overlapped. Yearly MAPE varied between 0.004 % and 25 %, and the feeders’ average MAPE was between 6 % and 10 %. In fault duration, the graphs followed the same pattern in nearly all the paths of the graphs, the yearly MAPE varied between 0.001 % and 25.54 % and the feeders’ average MAPE varied between 6 % and 11 %. The model produced a fairly accurate forecast according to the criteria of MAPE. The average overall MAPE of each feeder was between 6 % and 10 % which indicated between 90 % and 94 % accuracy of the model. Therefore the ANN-TS model is effective for fault prediction in reference time series. Keywords :Electrical Power, Fault, Artificial Neural Network based Time Series, Resilient Back-Propagation, Mean Absolute Percentage Error, Forecasting, Transmission substation. DOI: 10.7176/JETP/12-3-01 Publication date: August 31 st 2022

Artificial Neural Network Approach to Guarantee the Positioning Accuracy of Moving Robots by Using the Integration of IMU/UWB with Motion Capture System Data Fusion.

This study presents an effective artificial neural network (ANN) approach to combine measurements from inertial measurement units (IMUs) and time-of-flight (TOF) measurements from an ultra-wideband (UWB) system with OptiTrack Motion Capture System (OptiT-MCS) data to guarantee the positioning accuracy of motion tracking in indoor environments. The proposed fusion approach unifies the following advantages of both technologies: high data rates from the MCS, and global translational precision from the inertial measurement unit (IMU)/UWB localization system. Consequently, it leads to accurate position estimates when compared with data from the IMU/UWB system relative to the OptiT-MCS reference system. The calibrations of the positioning IMU/UWB and MCS systems are utilized in real-time movement with a diverse set of motion recordings using a mobile robot. The proposed neural network (NN) approach experimentally revealed accurate position estimates, giving an enhancement average mean absolute percentage error (MAPE) of 17.56% and 7.48% in the X and Y coordinates, respectively, and the coefficient of correlation R greater than 99%. Moreover, the experimental results prove that the proposed NN fusion is capable of maintaining high accuracy in position estimates while preventing drift errors from increasing in an unbounded manner, implying that the proposed approach is more effective than the compared approaches.

From Policy to Prediction: Forecasting COVID-19 Dynamics Under Imperfect Vaccination

Understanding the joint impact of vaccination and non-pharmaceutical interventions on COVID-19 development is important for making public health decisions that control the pandemic. Recently, we created a method in forecasting the daily number of confirmed cases of infectious diseases by combining a mechanistic ordinary differential equation (ODE) model for infectious classes and a generalized boosting machine learning model (GBM) for predicting how public health policies and mobility data affect the transmission rate in the ODE model (Wang et al. in Bull Math Biol 84:57, 2022). In this paper, we extend the method to the post-vaccination period, accordingly obtain a retrospective forecast of COVID-19 daily confirmed cases in the US, and identify the relative influence of the policies used as the predictor variables. In particular, our ODE model contains both partially and fully vaccinated compartments and accounts for the breakthrough cases, that is, vaccinated individuals can still get infected. Our results indicate that the inclusion of data on non-pharmaceutical interventions can significantly improve the accuracy of the predictions. With the use of policy data, the model predicts the number of daily infected cases up to 35 days in the future, with an average mean absolute percentage error of 20.15%, which is further improved to 14.88% if combined with human mobility data. Moreover, the most influential predictor variables are the policies of restrictions on gatherings, testing and school closing. The modeling approach used in this work can help policymakers design control measures as variant strains threaten public health in the future. Supplementary InformationThe online version contains supplementary material available at 10.1007/s11538-022-01047-x.

Short-Term Forecast of OD Passenger Flow Based on Ensemble Empirical Mode Decomposition

The development of metro systems can be a good solution to many problems in urban transport and promote sustainable urban development. A metro system plays an important role in urban public transit, and the passenger-flow forecasting is fundamental to assisting operators in establishing an intelligent transport system (ITS). In order to accurately predict the passenger flow of urban metros in different periods and provide a scientific basis for schedule planning, a short-term metro passenger-flow prediction model is constructed by integrating ensemble empirical mode decomposition (EEMD) and long short-term memory neural network (LSTM) to solve the problem that the existing empirical mode decomposition (EMD) is prone to modal aliasing. According to the processed metro-card data, the time series of historical OD data of metro passenger flow is obtained. After EEMD modal decomposition, several intrinsic mode functions sub-items and residual items are obtained. Then, an LSTM network is constructed for prediction. The time step of the network is decided according to the partial autocorrelation functions. The prediction results of intrinsic mode function (IMF) and residual items are integrated to obtain prediction results. The station is classified according to the land types around the station, and the model is tested by using the metro automatic fare collection (AFC) data. In order to test the actual prediction, a different number of training set samples are selected to predict. The measured data of the next day is continuously added to the original training set to compare the prediction accuracy. The results show that the mean absolute percentage error (MAPE) and root mean square error (RMSE) of the EEMD-LSTM model are better than the EMD-LSTM in predicting the OD value of commercial–residential stations and scenic–residential stations. Compared with the EMD-LSTM model, the EEMD-LSTM model showed an average reduction by 3.112% in MAPE values and 15.889 in RMSE, indicating that the EEMD-LSTM has higher prediction accuracy, and EEMD-LSTM model has higher accuracy in short-term metro passenger-flow prediction. The average MAPE for the 35-to-42-day historical data sample decreased from 13.02% to 10.39% with a decreasing trend. It shows that the prediction accuracy keeps improving as the training set samples increase.

Bagging–XGBoost algorithm based extreme weather identification and short-term load forecasting model

Accurate short-term load forecasting of distribution transformer in extreme weather will effectively assist power dispatching and enable safe and stable operation of power grid. Therefore, this paper proposes the Bagging–XGBoost algorithm based extreme weather identification and short-term load forecasting model, which can warn the time period and detailed value of peak load in advance. Firstly, based on Extreme Gradient Boosting (XGBoost) algorithm, the idea of Bagging is introduced to reduce the output variance and enhance the generalization ability of the algorithm. Then, the mutual information (MI) between weather influencing factors and load is analyzed to adjust the input weight of the model and improve its ability to track weather changes. Next, considering the load, weather and time factors, the extreme weather identification model is established to determine the occurrence range of peak load. Finally, the specialized training set is selected based on the weighted similarity, and high-accuracy short-term load forecasting model is established. Compared with the traditional model, the model proposed in this paper reduces the average Mean Absolute Percentage Error (MAPE) of peak load by 3% to 10%.

Forecasting COVID-19 Epidemic Trends by Combining a Neural Network with Rt Estimation.

On 31 December 2019, a cluster of pneumonia cases of unknown etiology was reported in Wuhan (China). The cases were declared to be Coronavirus Disease 2019 (COVID-19) by the World Health Organization (WHO). COVID-19 has been defined as SARS Coronavirus 2 (SARS-CoV-2). Some countries, e.g., Italy, France, and the United Kingdom (UK), have been subjected to frequent restrictions for preventing the spread of infection, contrary to other ones, e.g., the United States of America (USA) and Sweden. The restrictions afflicted the evolution of trends with several perturbations that destabilized its normal evolution. Globally, has been used to estimate time-varying reproduction numbers during epidemics. Methods: This paper presents a solution based on Deep Learning (DL) for the analysis and forecasting of epidemic trends in new positive cases of SARS-CoV-2 (COVID-19). It combined a neural network (NN) and an estimation by adjusting the data produced by the output layer of the NN on the related estimation. Results: Tests were performed on datasets related to the following countries: Italy, the USA, France, the UK, and Sweden. Positive case registration was retrieved between 24 February 2020 and 11 January 2022. Tests performed on the Italian dataset showed that our solution reduced the Mean Absolute Percentage Error (MAPE) by 28.44%, 39.36%, 22.96%, 17.93%, 28.10%, and 24.50% compared to other ones with the same configuration but that were based on the LSTM, GRU, RNN, ARIMA (1,0,3), and ARIMA (7,2,4) models, or an NN without applying the as a corrective index. It also reduced MAPE by 17.93%, the Mean Absolute Error (MAE) by 34.37%, and the Root Mean Square Error (RMSE) by 43.76% compared to the same model without the adjustment performed by the . Furthermore, it allowed an average MAPE reduction of 5.37%, 63.10%, 17.84%, and 14.91% on the datasets related to the USA, France, the UK, and Sweden, respectively.

Daily Model Calibration with Water Loss Estimation and Localization Using Continuous Monitoring Data in Water Distribution Networks

Due to the increasing deployment of sensors in water distribution networks (WDNs) for continuous monitoring, hydrodynamic data are readily available for engineers to improve the daily operations of WDNs. In collaboration with Public Utility Board (PUB), Singapore’s National Water Agency, an alternative model calibration approach using continuously monitored data is proposed to facilitate PUB’s smart water grid (SWG). The generic approach was developed as an integrated solution methodology for practical engineers to conduct a series of systematic analyses daily, namely, (1) estimating the system’s daily nonrevenue water (NRW) volume and NRW time series; (2) adjusting the available pumps’ operational curves and control statuses, followed by calibrating the system’s net demand pattern to fulfill the flow balance accounting for the actual consumption; (3) identifying and rectifying possible offsets in the monitored pressure head values for each sensor station; (4) performing model calibration with anomaly localization analysis when the system’s total NRW volume exceeds its assumed background NRW volume; and (5) calibrating other physical properties to fulfill the system’s energy balance, especially for the peak demand hours. The effectiveness of our proposed approach was then tested and verified using a real-world WDN having total pipe length of >650 km with available monitoring data. Key findings from the case study analysis include (1) anomaly events localized including, but not limited to, five out of six reported leaks for the selected week to within 400 m with lead time of 1–2 days; (2) the system’s initial flow imbalance addressed by estimating the daily NRW volume and localizing the possible anomaly events; and (3) pipe roughness values calibrated to further improve the energy balance in the system, especially during the peak demand hours, by attaining an average mean absolute percentage error (MAPE) score of 2.5%.

Hyperspectral reconstruction method for optically complex inland waters based on bio-optical model and sparse representing

For better use of well-performed water quality parameter estimation models and the comprehensive use of multi-source remote sensing data, hyperspectral reconstruction is urgently needed in the remote sensing of optically complex inland waters. In this study, we proposed a bio-optical–based hyperspectral reconstruction (BBHR) algorithm to generate hyperspectral above-surface remote-sensing reflectance (Rrs) data ranging in wavelength from 400 to 800 nm. One core advantage of the BBHR method is its in situ data independency, which theoretically renders the algorithm universal. The other advantage is its ability to reconstruct hyperspectral Rrs for the 400–800 nm spectral range, which facilitates the construction of more high accuracy chlorophyll-a concentration (Cchla) estimation models for optically complex waters. The reconstruction was tested by employing six widely used multispectral sensors: the Medium Resolution Imaging Spectrometer, (MERIS), Sentinel-3 Ocean and Land Color Instrument (S3 OLCI), Sentinel-2 Multispectral Instrument (S2 MSI), Geostationary Ocean Color Imager (GOCI), Visible Infrared Imaging Radiometer Suite (VIIRS), and Moderate Resolution Imaging Spectroradiometer (MODIS). The model performance was validated by using a ASD FieldSpec spectroradiometer-measured hyperspectral dataset containing 1396 samples and a satellite-in-situ match-up dataset with 66 samples. The results show that the proposed BBHR method exhibits satisfactory performance. The average mean absolute percentage error (MAPE), root mean square error (RMSE), R2 and bias indices of the BBHR-reconstructed Rrs over all spectral bands of the six multispectral sensors were 3.27%, 8.86 × 10−4 sr−1, 0.98, and − 6.53 × 10−5 sr−1, respectively. In the field Cchla estimation experiment that contained 391 samples (mean Cchla is 25.42 ± 16.37 μg/L), the BBHR algorithm improved the MAPE and RMSE indices of multispectral data from 0.47 and 12.80 μg/L to 0.42 and 10.16 μg/L, respectively. For the satellite image match-up dataset (66 samples), the BBHR method decreased the MAPE and RMSE indices of multispectral images from 0.51 and 12.94 μg/L to 0.32 and 8.01 μg/L, respectively. The proposed algorithm outperformed the other two high-accuracy models in terms of spectral fidelity and Cchla estimation. In addition, the BBHR method shows great potential for the multi-source monitoring of inland water bodies. This could improve the accuracy and robustness of the reconstruction when semi-synchronized multi-source data are input and increase the consistency of multi-source data when non-synchronized multi-source data are provided. Our results revealed that BBHR is a trustworthy algorithm that offers hyperspectral Rrs data and facilitates the remote monitoring of turbid inland waterbodies.

A novel decomposition-ensemble forecasting system for dynamic dispatching of smart grid with sub-model selection and intelligent optimization

Short-Term Electricity Load Forecasting (STLF) has drawn increasing attention due to its considerable function on improving the dispatch and management of power system. The path of load forecasting has been widely explored in the past studies. However, most forecasting methods ignore the importance of sub-model selection and interval forecasting, causing poor forecasting performance. In this paper, a hybrid ensemble forecasting scheme, constructed by sub-model selection, data decomposition, multi-objective optimization, point prediction, and interval prediction, is proposed to enrich the current load forecasting system. In the load forecasting system design, the original load data is adaptively decomposed into some sub-series by using the Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN). The most proper sub-model for decomposed series is selected from nine widely-used forecasting methods based on sub-model selection. The forecasting results of each sub-series are ensembled by the Multi-Objective Grasshopper Optimization Algorithm (MOGOA) to estimate the final point forecasting values. By integrating point forecasting errors and distribution fitting strategy, the interval forecasting results are further estimated, which investigates the effectiveness of the proposed system. Three numerical experiments are carried out based on the real load data obtained from three observation sites. The experimental results indicate that the hybrid ensemble system can provide better point and interval forecasting performance with the average Mean Absolute Percentage Error (MAPE) values of 0.3275%, 0.4419%, and 0.7024% and the mean Coverage Width-based Criteria (CWC) values (α=0.05) of 0.0668, 0.1335, and 0.3178 from one-step to three-step forecasting, respectively.

Multi-Step Ahead Prediction of Reheat Steam Temperature of a 660 MW Coal-Fired Utility Boiler Using Long Short-Term Memory

With increases in the penetration of renewables in grids, there is an increasing demand for coal-fired power plants to operate flexibly. Regulation of reheat steam temperature is of great importance for the safe and efficient operation of coal-fired power plants. However, the difficulty of reheat steam temperature regulation increases largely during flexible operation due to the large delay and nonlinear properties, especially those units designed to shoulder base load and with limited regulating strategy. A multistep prediction model on the reheat steam temperature of a 660-MW coal-fired utility boiler was developed based on long short-term memory. The results show that the multistep prediction model performs well. The average root mean square error and mean absolute percentage error values of the five-step prediction results are less than 0.52°C and 0.07%, respectively. The correlation coefficients of the five-step predictions are all greater than 0.95. With a sample interval of 30 s, the model provides an accurate prediction of reheat steam temperature within 2.5 min, which could supply an important reference for the reheat steam temperature regulation.

Utilization and Verification of Inverse Distance Weighting (IDW) Interpolation Technology for Predicting Solar Radiation of Photovoltaic System

Purpose: Weather conditions around solar photovoltaic (PV) systems have a direct effect on the amount of generated power. Therefore, highly correlated meteorological environmental variables are used to accurately predict the power generation capacity of PV systems. However, it is difficult to install observation equipment due to high installation costs and maintenance. Therefore, public data from the Korea Meteorological Administration are often used. However, uncertainty may arise due to the large distance between the stations of the Korea Meteorological Administration. Therefore, in this study, inverse distance weighting (IDW) was used to predict the solar radiation of an arbitrary area to improve the uncertainty caused by the large distance between the meteorological stations. Method: In this study, IDW was used as a spatial statistical technique to predict the monthly average of daily accumulation solar radiation in South Korea. Moreover, four cases of the automated synoptic observing system (ASOS) were used as study cases to evaluate the prediction technology. The actual solar radiation observed and the predicted solar radiation was compared and analyzed in each study case, and the accuracy was evaluated as an indicator of the mean absolute percentage error (MAPE). Additionally, the MAPE was compared with the prediction results of the multiple regression model. Result: Although the results differ for each case, the average MAPE was approximately 11.06% using the inverse distance weighting interpolation method, and the predicted value by the multiple regression method yielded an average MAPE of approximately 16.12%.

Applying Clustered KNN Algorithm for Short-Term Travel Speed Prediction and Reduced Speed Detection on Urban Arterial Road Work Zones

This study developed and verified a travel speed prediction model based on the travel speed and work zone statistics collected from the advanced traffic management system (ATMS) real-time data in Daegu, South Korea. A clustered K-nearest neighbors (CKNN) algorithm was used to predict travel speed, resulting in a 6.9% average mean absolute percentage error (MAPE) using the data from 1,815 work zones. Furthermore, road network impact due to road work was calculated by comparing the travel speed prediction results obtained from the historical speed data. The predicted travel speed data in a work zone generated from this study is expected to allow drivers to select optimized paths and use them for traffic management strategies to operate in a work zone efficiently.

Incorporation of near-real-time hospital occupancy data to improve hospitalization forecast accuracy during the COVID-19 pandemic.

Public health decision makers rely on hospitalization forecasts to inform COVID-19 pandemic planning and resource allocation. Hospitalization forecasts are most relevant when they are accurate, made available quickly, and updated frequently. We rapidly adapted an agent-based model (ABM) to provide weekly 30-day hospitalization forecasts (i.e., demand for intensive care unit [ICU] beds and non-ICU beds) by state and region in North Carolina for public health decision makers. The ABM was based on a synthetic population of North Carolina residents and included movement of agents (i.e., patients) among North Carolina hospitals, nursing homes, and the community. We assigned SARS-CoV-2 infection to agents using county-level compartmental models and determined agents’ COVID-19 severity and probability of hospitalization using synthetic population characteristics (e.g., age, comorbidities). We generated weekly 30-day hospitalization forecasts during May–December 2020 and evaluated the impact of major model updates on statewide forecast accuracy under a SARS-CoV-2 effective reproduction number range of 1.0–1.2. Of the 21 forecasts included in the assessment, the average mean absolute percentage error (MAPE) was 7.8% for non-ICU beds and 23.6% for ICU beds. Among the major model updates, integration of near-real-time hospital occupancy data into the model had the largest impact on improving forecast accuracy, reducing the average MAPE for non-ICU beds from 6.6% to 3.9% and for ICU beds from 33.4% to 6.5%. Our results suggest that future pandemic hospitalization forecasting efforts should prioritize early inclusion of hospital occupancy data to maximize accuracy.

Performance evaluation for tool wear prediction based on Bi-directional, Encoder–Decoder and Hybrid Long Short-Term Memory models

PurposeExcessive tool wear is responsible for damage or breakage of the tool, workpiece, or machining center. Thus, it is crucial to examine tool conditions during the machining process to improve its useful functional life and the surface quality of the final product. AI-based tool wear prediction techniques have proven to be effective in estimating the Remaining Useful Life (RUL) of the cutting tool. However, the model prediction needs improvement in terms of accuracy.Design/methodology/approachThis paper represents a methodology of fusing a feature selection technique along with state-of-the-art deep learning models. The authors have used NASA milling data sets along with vibration signals for tool wear prediction and performance analysis in 15 different fault scenarios. Multiple steps are used for the feature selection and ranking. Different Long Short-Term Memory (LSTM) approaches are used to improve the overall prediction accuracy of the model for tool wear prediction. LSTM models' performance is evaluated using R-square, Mean Absolute Error (MAE), Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE) parameters.FindingsThe R-square accuracy of the hybrid model is consistently high and has low MAE, MAPE and RMSE values. The average R-square score values for LSTM, Bidirection, Encoder–Decoder and Hybrid LSTM are 80.43, 84.74, 94.20 and 97.85%, respectively, and corresponding average MAPE values are 23.46, 22.200, 9.5739 and 6.2124%. The hybrid model shows high accuracy as compared to the remaining LSTM models.Originality/valueThe low variance, Spearman Correlation Coefficient and Random Forest Regression methods are used to select the most significant feature vectors for training the miscellaneous LSTM model versions and highlight the best approach. The selected features pass to different LSTM models like Bidirectional, Encoder–Decoder and Hybrid LSTM for tool wear prediction. The Hybrid LSTM approach shows a significant improvement in tool wear prediction.

Improving Clustering-Based Forecasting of Aggregated Distribution Transformer Loadings With Gradient Boosting and Feature Selection

Load forecasting is more important than ever to enable new monitor and control functionalities of distribution networks aiming to mitigate the impact of the energy transition. Load forecasting at medium voltage (MV) level is becoming more challenging, because these load profiles become more stochastic due to the increasing penetration of photovoltaic (PV) generation in distribution networks. This work combines medium to low voltage (MV/LV) transformer loadings measured with advanced metering infrastructure (AMI) and machine learning (ML) algorithms to propose a new clustering based day-ahead aggregated load forecasting approach. This four-step approach improves the day-ahead load forecast of a city. First, MV/LV transformer loadings are clustered based on the shape of their load pattern. Second, a gradient boosting algorithm is used to forecast the load of each cluster and calculate the related feature importance. Third, feature selection is applied to improve the forecast accuracy of each cluster. Finally, the day-ahead load forecast of all clusters are aggregated. The case study presented uses 519 measured MV/LV transformer loadings in a city to perform 30 day-ahead load forecasts. Compared against the day-ahead aggregated load forecast without clustering, the average normalized root mean squared error (NRMSE) reduced 12.7 &#x0025;, the average mean absolute percentage error (MAPE) reduced 18.2 &#x0025; and the average Pearson Correlation Coefficient (PCC) increased 0.37 &#x0025;. The 95 &#x0025; confidence interval of the difference between the average NRMSE, MAPE and PCC without clustering and with the proposed method indicates a statistically significant improvement in accuracy.

Hierarchical Multiobjective Distributed Deep Learning for Residential Short-Term Electric Load Forecasting

Short-term load forecasting plays an essential role in appliance control in households and demand response at the neighborhood or community level. When load forecasting is simultaneously performed at different levels in a smart community, a central method that directly aggregates raw household data may incur a security issue. However, an individual method that builds separate models for each level requires additional data collection. We propose a distributed load forecasting framework that is secure, data efficient, and resource efficient. Our proposed method shares the latent variables, the reduced expression of the input data, of the household models without revealing the raw data. The latent variables are aggregated and fed to the neighborhood model, and during inference, input data collection can be abbreviated in the neighborhood. Because household models provide informative latent variables, the number of model parameters for the neighborhood model can be reduced. Our proposed method is evaluated using real electricity consumption data for six different models under 24 h and 1 h ahead forecasting. Our proposed method with the long short-term memory, multi-layer perceptron, and convolution neural network-based models presents the degradation in average mean absolute percentage error 3.68% and 5.16% at most for household and neighborhood forecasting, respectively, compared to the individual method. At the same time, it reduces the neighborhood&#x2019;s model size by approximately 53%.