Wyniki wyszukiwania - BazTech

1

Passenger traffic in Polish seaports in the face of the COVID-19 pandemic

Sobczuk Sebastian, Borucka Anna

Zeszyty Naukowe Politechniki Morskiej w Szczecinie

|

2025

|

nr 81 (153)

57--66

EN

The outbreak of the COVID-19 pandemic had a profound impact on the global economy and disrupted daily life across many regions of the world. Restrictions imposed at the time, such as the closure of national borders and restrictions on mobility, led to unprecedented challenges for the transportation sector and related tourism services compared with any prior crisis. This disruption also affected maritime passenger transport in Poland. This article aims to assess the impact of the COVID-19 pandemic on passenger traffic in Polish seaports and to develop mathematical models that could support management in the event of future epidemic threats. Three different models are proposed, which showed that the epidemic crisis resulted in a significant decline in passenger traffic at Polish seaports. The most accurate proved to be the SARIMA model. The Holt-Winters model also demonstrated high fitting and predictive performance. In turn, the STL model offered intriguing insights with its time series decomposition, enabling a detailed analysis of individual components. A comparative analysis of the proposed models confirms their usefulness in forecasting passenger traffic in seaports in the face of disruptions such as the COVID-19 pandemic. These models can be an effective decision-support tool, helping to reduce the negative effects of future epidemic threats.

2

Forecasting electricity prices in the Polish Day-Ahead Market using machine learning models

Sowiński Rafał, Komorowska Aleksandra

Polityka Energetyczna

|

2025

|

T. 28, z. 2

211--230

EN

Given the constantly changing market situation for electricity prices, driven by shifts in the energy mix, regulatory reforms, and broader socio-economic factors, it is necessary to reassess the understanding of price forecasting periodically. Traditional statistical methods may struggle when faced with heightened volatility, nonlinear dependencies, and rapidly changing input features. In contrast, machine learning models, particularly Artificial Neural Networks (ANNs), can adapt more effectively to complex, non-stationary patterns in price time series. In this study, six distinct artificial neural network (ANN) architectures were developed and trained using eight years of historical Polish Day-Ahead Market electricity price data (2016–2024). Four of these were plain deep learning models: a Multilayer Perceptron (MLP), a Convolutional Neural Network (CNN), a Long Short-Term Memory (LSTM) model, and a Gated Recurrent Unit (GRU) model. Two others were hybrid models combining convolutional layers with recurrent layers. The hybrid architectures, namely CNN+LSTM and CNN+GRU, were designed to leverage the capacity of CNN to automatically extract features from narrower sliding windows of past prices and the LSTM/GRU layers’ ability to capture long-term temporal dependencies. The models’ performances were evaluated using three metrics: Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and the coefficient of determination (R2). The top-performing CNN+LSTM achieved an MAE of 75.21 PLN/MWh, an RMSE of 103.64 PLN/MWh, and an R2 of 0.59. Results were also compared against several models previously reported in the literature. These results may be used to improve price forecasting by indicating the optimal pathways for building forecasting models and, in extension, lead to more efficient power system planning.

PL

Ze względu na stale zmieniające się ceny energii elektrycznej, spowodowane zmianami w miksie energetycznym, regulacyjnymi i innymi czynnikami społeczno-ekonomicznymi, konieczne staje się okre sowe weryfikowanie podejścia do prognozowania cen. Tradycyjne metody statystyczne mogą zawodzić w warunkach nasilonej zmienności, nieliniowych zależności i często zmieniających się cech wejściowych. Modele uczenia maszynowego, a zwłaszcza Sztuczne Sieci Neuronowe (SSN), potrafią skutecznie dostosowywać się do złożonych, niestacjonarnych wzorców w szeregach czasowych. W niniejszym badaniu opra cowano i wytrenowano sześć różnych modeli SSN, korzystając z danych historycznych z Polskiego Rynku Dnia Następnego z lat 2016–2024. Cztery z tych modeli to czyste modele głębokiego uczenia: wielowar stwowy perceptron (MLP), sień konwolucyjna (CNN), długa pamięć krótkotrwała (LSTM) oraz bramkowa jednostka rekurencyjna (GRU). Dwa pozostałe to architektury hybrydowe, oznaczone jako CNN+LSTM i CNN+GRU, łączą zdolność CNN do wychwytywania cech z węższych okien czasowych i umiejętność warstw rekurencyjnych do uczenia się zależności długoterminowych. Wydajność modeli oceniano na pod stawie trzech miar: średniego błędu bezwzględnego (MAE), pierwiastka ze średniego błędu kwadratowego (RMSE) i współczynnika determinacji (R2). Najlepsze wyniki osiągnęła architektura CNN+LSTM, uzy skując MAE na poziomie 75,21 zł/MWh, RSME równe 103,64 zł/MWh i R2 wynoszące 0,59. Wyniki te mogą zostać wykorzystane do usprawnienia procesów prognozowania cen energii elektrycznej poprzez wskazanie wytycznych dotyczących projektowania modeli prognostycznych opartych na uczeniu maszy nowym, co z kolei może wiązać się z wydajniejszym planowaniem działania systemu energetycznego.

3

Optimising the Operation of Ships with Artificial Intelligence Systems

Bolkvadze Nugzar, Verdzadze Ruslan, Bazaian Aram, Bolkvadze Levan, Gabedava George

Polish Maritime Research

|

2025

|

nr 4

119--131

EN

The objective of this study was to conduct an empirical evaluation of the effectiveness of artificial intelligence systems in optimising the operation of commercial maritime vessels. The methodology involved collecting and processing telemetry from 58 synchronised onboard measurement channels, including temperatures, vibration metrics, gyroscopic data, trim and heel angles, and data from automatic identification systems, differential global positioning systems, and radar signals. Data were sampled at intervals of 1 s, filtered using the Hampel method, and aggregated into frames of 3 min. A hybrid deep learning model was developed to forecast vessel speed, fuel usage, and stability. Experiments were conducted on 16 vessels: six container carriers (3,000 20-foot equivalent units class) and 10 Handymax bulk carriers (40,000–55,000 deadweight tons). These vessels completed 97 voyages between March 2023 and February 2024, 45% of which took place in the Black Sea and 55% in the North Sea. A validation campaign comprising 9,230 h of simulator trials and real-world deployment was carried out to test the artificial intelligence model under variable sea states and in scenarios involving disruptions to automatic identification and differential global positioning systems. The results showed a 12.4% reduction in average fuel consumption and an 8.2% decrease in voyage duration. Ship stability improved, with a 22% reduction in roll amplitude. Predictive maintenance algorithms achieved 95% accuracy, enabling early fault detection and reducing unscheduled downtime. Only three manual interventions were recorded during deployment, and course deviations remained below 1.3°. An environmental analysis revealed a 4.2% improvement in carbon intensity, demonstrating compliance with the International Maritime Organization Carbon Intensity Indicator standards.

4

Shaping a Safe Working Environment in the Coal Production Process: Utilizing Model Studies to Assess Methane Hazard Formation

Tutak Magdalena, Brodny Jarosław, Romanyshyn Lyubomyr

Management Systems in Production Engineering

|

2025

|

nr 2 (33)

311--320

EN

The establishment of a safe working environment is one of the key challenges in the implementation of various production processes. This is particularly relevant to underground coal production, where the primary operations (exploitation) take place in an underground environment. In this context, the paper presents the results of a study on methane hazard formation during the coal production process. The analysis was conducted through model studies that included driven dog headings, longwall workings with and without auxiliary ventilation equipment, and collapsing goafs. The research methodology, mining region models, and validation of the obtained results were further supported by tests conducted under real conditions. The findings highlight the significant potential of model studies based on structural models for assessing ventilation hazards in the examined regions and the phenomena occurring within them. Based on these results, the identification and assessment of methane hazard levels in the studied regions were carried out. This, in turn, opens avenues for their practical application in enhancing both safety and efficiency in the mining production process. The developed methodology and models are universal in nature, offering a broad range of applications for studying various ventilation states, both in steady and unsteady conditions. Additionally, they allow for comprehensive predictions of methane concentration distributions, forming a critical basis for preventive measures and improvements in underground mining safety.

5

Application of neural networks for drought forecasting based on the standardised precipitation index

Fathiyana Rana Z., Haryadi Deny, Ananda Naufal, Hidayat Dinda J., Wicaksana Haryas Subyantara, Azahra Rahma Wafii, Saputra Hakim Giraldi

Journal of Water and Land Development

|

2025

|

no. 65

113--121

EN

Based on data from the National Disaster Management Agency, South Sumatra is one of the provinces with a reasonably large drought-affected area, totalling 8,853,691.009 ha. Drought is a hydrometeorological disaster, characterised by anomalous rainfall below normal levels. Reduced rainfall can lead to decreased soil moisture, reduced river flows, and a general scarcity of water, which limits availability of water both on the surface and in the soil. To anticipate and mitigate the impacts of drought, an accurate forecasting system is essential for effective disaster management and mitigation. This research focuses on forecasting drought using the standardised precipitation index (SPI) based on Long Short-Term Memory (LSTM) and Multilayer Perceptron (MLP) algorithms. It compares LSTM and MLP algorithms by integrating rainfall data from the FY-4A satellite and observational rain gauges, which are processed to generate SPI values. These data are employed to train and test MLP and LSTM models in predicting future drought conditions. The results indicate that drought can be effectively predicted using both MLP and LSTM. However, the MLP outperforms the LSTM, as reflected by a higher Nash-Sutcliffe efficiency (NSE) value, a lower error rate, and a predicted date trend that more closely aligns with actual observations.

6

Heterogeneous ensemble neural network for forecasting the state of multi-zone heating facilities

Yukhymchuk Maria, Dubovoi Volodymyr, Harbar Zhanna, Yeraliyeva Bakhyt

Informatyka, Automatyka, Pomiary w Gospodarce i Ochronie Środowiska

|

2025

|

T. 15, nr 2

94--99

EN

The research is aimed at increasing the accuracy of forecasting the state of multi-zone thermal facilities. Such facilities include multi-room premises, multi-zone greenhouses, tunnel kilns for brick production, and others. Thehigh inertia of such facilities reduces the effectivenessof "ad hoc control". Modern proactive control systems based on forecasting are mainly based on using neural network training. However, to forecastthe state of a specific multi-zone thermal facility, training the network requires a very large dataset, which is difficult to create and use. A combined neuro-structural method for forecasting the state of multi-zone thermal facilities is proposed, in which the structure of the neural model reflects the structureof the mutual influence of the facility zones. The research of the method has shown the possibility of ensuring sufficiently high forecast accuracywith a smaller size of the training dataset.

PL

Badania mają na celu zwiększenie dokładności prognozowania stanu wielostrefowych obiektów cieplnych. Obiekty takie obejmują obiekty wielopokojowe, wielostrefowe szklarnie, piece tunelowe do produkcji cegieł i inne. Duża bezwładność takich obiektów zmniejszaskuteczność "sterowania ad hoc". Nowoczesne proaktywne systemy sterowania oparte na prognozowaniu opierają się głównie na szkoleniu sieci neuronowych. Jednak w celu prognozowania stanu konkretnego wielostrefowego obiektu termicznego, szkolenie sieci wymaga bardzo dużego zbioru danych, który jest trudnydo utworzenia i wykorzystania. Zaproponowano połączoną neurostrukturalną metodę prognozowania stanu wielostrefowych obiektów cieplnych, w której struktura modelu neuronowego odzwierciedla strukturę wzajemnego wpływu stref obiektu. Badania metody wykazały możliwość zapewnienia wystarczająco wysokiej dokładności prognozy przymniejszym rozmiarze zbioru danych treningowych.

7

Development of a system for predicting failures of bagging machines

Huliieva Nataliia, Lishchyna Nataliia, Pasternak Viktoriya, Huliieva Zemfira

Informatyka, Automatyka, Pomiary w Gospodarce i Ochronie Środowiska

|

2025

|

T. 15, nr 4

10--13

EN

The reliabilityand effective operation of machines is a pressing problem for every enterprise, which requires labour intensivesystematizationof production processes.The goal is to develop an algorithm and a system for predicting failures of packaging machines based on the analysisof operational indicators.The scientific novelty lies in the integration of statistical data to assess the efficiency of machine operation and predict possible failures, which allows significantly improving maintenance processes and reducing the risks of unforeseen breakdowns.The practical valueis the development of a forecasting system that collects the necessary statistical data and performs forecasting. Based on the collected data, an assessment of the efficiency of work and forecasting of possible failures is carried out. The forecasting system is demonstrated on the example of packaging machines LEMO INTERmat ST-SA 850 of "Tatrafan" LLC.Two research methods were used: calculation (mathematical) and forecasting system (least squares method). The forecasting system provides two ways of presenting data: tabular and graphical. Tabular presentation of data allows filtering information according to various criteria, while graphical display is implemented in the formof diagrams showing the operating time and downtime of machines.The main results are the determined rangeof probable failure of LEMO INTERmat ST-SA 850 packaging machines, which lies in the range from 9090.5to 12736.5 hours of operation and almost coincides with the manufacturer's warranty period. With timely maintenance, it is possible to increase the lower limit of this interval.

PL

Niezawodność i efektywna praca maszyn stanowi palący problem każdego przedsiębiorstwa, wymagający pracochłonnego usystematyzowania procesów produkcyjnych. Celem jest opracowanie algorytmu i systemu prognozowania awarii maszyn pakujących na podstawie analizy wskaźników eksploatacyjnych. Nowością naukową jest integracja danych statystycznych w celu oceny efektywności pracy maszyn i przewidywania ewentualnych awarii, co pozwala znacząco usprawnić procesy utrzymania ruchu i ograniczyć ryzyko nieprzewidzianych awarii. Znaczenie praktyczne ma opracowanie systemu prognostycznego, który będzie zbierał niezbędne dane statystyczne i wykonywał prognozowanie. Na podstawiezebranych danych przeprowadzana jest ocena efektywności pracy i prognozowanie ewentualnych awarii. System prognozowania zaprezentowano na przykładzie maszyn pakujących LEMO INTERmat ST-SA 850 firmy Tatrafan LLC.W badaniach zastosowano dwie metody: obliczeniową (matematyczną) i prognostyczną (metodę najmniejszych kwadratów). System prognozowania umożliwia prezentację danych na dwa sposoby: w formie tabelarycznej i graficznej. Prezentacja danych w formie tabelarycznej pozwala na filtrowanie informacji według różnych kryteriów, natomiast prezentacja graficzna realizowana jest w formie diagramów, obrazujących pracę i przestoje maszyn.Głównymi wynikami jest określony zakres prawdopodobnej awarii maszyn pakujących LEMO INTERmat ST-SA 850, który mieści się w przedziale od 9090,5 do 12736,5 godzin pracy i niemal pokrywa się z okresem gwarancji producenta. Dzięki terminowej konserwacji możliwe jest podwyższenie dolnej granicy tego przedziału.

8

Trends and challenges of forecasting in the airline industry research

Dewi Oki Anita Candra, Masruroh Nur Aini, Wibowo Budhi Sholeh

Engineering Management in Production and Services

|

2025

|

Vol. 17, no. 2

23--36

EN

This study aims to comprehensively review aviation forecasting research by identifying its bibliometric trends, evolving research areas, and thematic developments. It focuses on understanding the aviation industry’s research gaps, highlighting emerging trends, and offering insights into future forecasting innovations. A systematic literature review in the Scopus database used Preferred Reporting Items for Systematic Review and Meta-Analyses (PRISMA) and bibliometric analysis. It identified key patterns, influential publications, and emerging topics. A science mapping analysis was executed to pinpoint research trends in airline forecasting using Biblioshiny to visualise the network analysis and thematic evolution keywords mapping. The study categorised research trends and identified underexplored areas for future investigation. The findings reveal significant shifts in aviation forecasting research, with three distinct phases of publication growth and a surge in output from 2016 onwards. Passenger demand forecasting remains the most researched topic, though its growth has stabilised. Emerging issues such as customer behaviour, financial forecasting, and dynamic pricing have gained prominence, driven by advancements in machine learning and big data analytics. The study also highlights transitioning from traditional statistical methods to more advanced predictive techniques, emphasising real-time decision-making and operational efficiency. Established research areas, such as air cargo forecasting and f leet scheduling, have become more standardised, reducing the need for further innovation.

9

Enhancing flash flood risk prediction: A case study from the Assaka watershed, Guelmim region, Southwestern Morocco

Talha Soukaina, Akhssas Ahmed, Aarab Abdellatif, Aabi Ayoub, Berkat Badr

Ecological Engineering & Environmental Technology

|

2025

|

Vol. 26, iss. 1

8--28

EN

Since the onset of the Industrial Revolution, significant climatic shifts have led to various environmental imbalances globally, notably increasing the frequency of flash floods, especially in vulnerable regions like the Assaka watershed in southwestern Morocco. This study aims to enhance flash flood risk prediction by integrating Machine Learning (ML) algorithms with Geographic Information System (GIS) technology. The Random Forest (RF) algorithm was employed to analyze over eight million data points, using fourteen predictors categorized into topographic (e.g., Altitude, Slope, Topographic Wetness Index (TWI)), climatic (e.g., Land Surface Temperature (LST), Soil Moisture Index (SMI)), and geological factors (e.g., Drainage Density, Soil Type, Lithology). These variables were derived from remotely sensed data and geospatial analyses. The RF model classified the Assaka watershed into five flood susceptibility levels: lowest, low, medium, high, and highest. The results indicated that the most vulnerable areas are near the watershed outlet and the main tributaries, Essayed and Oum Laachar Wadis. These regions are characterized by high land surface temperatures, low drainage density, poor soil moisture, and specific geological conditions, all of which contribute to heightened flood risk. The model's performance was evaluated using multiple metrics, achieving Precision (0.968), Recall (0.967), Accuracy (0.967), F1 Score (0.965), Kappa Statistic (0.839), and an AUC of 1.0, highlighting its robustness and predictive capabilities. The originality of this study lies in its comprehensive integration of ML with GIS to develop a highly reliable flood susceptibility map for the Assaka watershed. This framework addresses existing gaps in flood risk assessment, offering a significant advancement over traditional methods through its use of advanced data-driven modeling techniques. The findings provide essential insights for prioritizing conservation and flood management strategies, contributing to better preparedness against flash floods in the Guelmim region and potentially other similar environments globally.

10

Simulation analysis of artificial neural network and XGBoost algorithms in time series forecasting

Sroka Łukasz

Zeszyty Naukowe. Organizacja i Zarządzanie / Politechnika Śląska

|

2024

|

z. 195

561--575

EN

Purpose: The aim of the article was to prepare a simulation analysis of artificial neural network and XGBoost algorithm with determining which of the method was characterized by a lower level of forecast errors for time series predictions. Design/methodology/approach: The objective of the article was reached by applying, a simulation study on a sample of 1000 artificially generated time series. The analyzed XGBoost algorithm and the artificial neural network ANN model were intended to prepare forecasts for five periods ahead. These forecasts were compared with the actual implementations of the time series and proposed forecast error measures. Findings: It is possible to use simulated time series to check which of the presented algorithms were characterized by a lower forecast error. The study showed that applying of the artificial neural networks ANN to forecast future observations generated a lower level of MAPE, MAE and RMSE errors than in the case of the XGBoost algorithm. It was found that both methods generate a lower level of forecast error for time series characterized by a high level of mean value, standard deviation and variance, and levels of kurtosis and skewness close to 0. Practical implications: The research results can be used by both investors and enterprises to better adjust their business decisions to changing market prices by using a model with a lower forecast bias. Originality/value: The original contribution of this article is a comprehensive comparison of forecasts generated by the XGBoost and ANN algorithm, along with determining for which types of time series of the algorithms forecast future values with less error. Moreover, due to the use of simulated artificial time series, it was possible to test each algorithm for various market conditions.

11

Forecasting the number of road accidents in Poland using trend models depending on the days of the week

Gorzelanczyk Piotr

Technical Sciences / University of Warmia and Mazury in Olsztyn

|

2024

|

Vol. 27

175--192

EN

Every year a very large number of people die on the roads. Although the number decreases year by year, it remains high. The pandemic has reduced the number of road accidents, but the value is still very high. For this reason, it is necessary to know on which days the highest number of traffic accidents occur, and to know the forecast of accidents by day of the week for the coming years, so that we can do everything possible to minimize the number of traffic accidents. The purpose of the article is to make a forecast of the number of road accidents in Poland according to the day of the week. The research was divided into two parts. The first was the analysis of annual data from the Police statistics on the number of road accidents in Poland in 2000-2021, and on this basis the forecast of the number of road accidents for 2022-2031 was determined. The second part of the research, dealt with monthly data from 2000-2021. Again, the analyzed forecast for the period January 2022 – December 2023 was determined. The results of the study indicate that we can still expect a decline in the number of accidents in the coming years, which is particularly evident when analyzing annual data. It is worth noting that the prevailing pandemic distorts the results obtained. The research was conducted in MS Excel, using selected trend models.

12

A Machine Learning Approach for Predicting Road Accidents

Budzyński Artur, Federowicz Mikołaj, Jabłoński Adam, Hasan Wesam, Gorszanów Jan, Nikitishyn Taras

Safety & Defense

|

2024

|

2

60--70

EN

Road accidents pose significant challenges to public safety and necessitate proactive measures to mitigate them. This paper introduces a machine-learning approach for predicting road accident incidences, leveraging diverse datasets encompassing traffic patterns, weather conditions, and historical accident records. The proposed model integrates feature engineering techniques to capture the multifaceted nature of variables influencing accidents. Through the application of advanced machine learning algorithms, such as ensemble methods and neural networks, the model aims to discern complex patterns within the data, facilitating accurate predictions of accident likelihood. The study also explores the interpretability of the model outputs, providing insights into the key predictors and their interactions. Validation and performance assessment involve rigorous testing on diverse datasets to ensure the generalizability and robustness of the predictive model. The outcomes of this research hold promise for the development of proactive road safety strategies and the implementation of targeted interventions, ultimately contributing to reducing road accidents and their associated societal impacts.

13

Predicting Road Accident Counts in Poland and the Czech Republic Using Neural Network Models

Gorzelańczyk Piotr, Ližbetinová Lenka, Pečman Jan

Rocznik Ochrona Środowiska

|

2024

|

Tom 26

603--615

EN

Every year, there is a decline in the number of car accidents reported in Poland, the Czech Republic, and globally. While recent trends due to the pandemic have influenced these figures, the overall rate remains significant. Therefore, it is crucial to take measures aimed at reducing this number. The primary focus of this article is to analyze the traffic accident statistics for Poland and the Czech Republic. Annual data regarding traffic incidents in both countries has been scrutinized to achieve this. Projections for 2024 to 2030 have been developed based on police reports. Various neural network models were utilized to forecast the number of accidents. The findings indicate that the number of traffic incidents is likely to stabilize. This stabilization can be viewed in the context of the increasing number of vehicles on the roads and the expansion of new highways. Additionally, selecting sample sizes for training, testing, and validation is crucial in influencing the results. Forecasting the number of traffic accidents is important for environmental protection, as accidents can lead to air and water pollution and increase noise, negatively affecting human health and ecosystems.

14

Data-driven model for long-term forecasting of 5G throughput and delay per network slice with the context of cell configuration

Dulas Dominik, Witulska Justyna, Wyłomańska Agnieszka, Walkowiak Krzysztof

Przegląd Telekomunikacyjny + Wiadomości Telekomunikacyjne

|

2024

|

nr 4

419--424

EN

The primary innovation and contribution of this study is the evaluation of a multidimensional VAR-MAX model, utilizing real traffic and environmental network data combined with cell configuration during the training phase, to accurately forecast telecommunication metrics, which are crucial in the dimensioning of 5G networks with network slicing. The results show that this technique is effective in predicting delay and throughput, which significantly influence the slice quality of service, over a longterm horizon of approximately 3 months.

PL

Podstawowym wkładem tej pracy jest ocena wielowymiarowego modelu VARMAX, wykorzystującego rzeczywiste dane o ruchu i środowisku sieci w połączeniu na etapie uczenia z danymi o konfiguracji komórek, w celu dokładnego prognozowania wskaźników telekomunikacyjnych, które są kluczowe przy wymiarowaniu sieci 5G z zastosowaniem plastrowania. Wyniki pokazują, że technika ta jest skuteczna w długoterminowym przewidywaniu opóźnień i przepustowości, które znacząco wpływają na jakość usługi w plastrze, w horyzoncie 3 miesięcy.

15

Forecasting the number of road accidents in Poland by province

Gorzelanczyk Piotr

Przegląd Komunikacyjny

|

2024

|

R. 79, nr 4/5

13--22

PL

Każdego roku na polskich drogach ginie bardzo duża liczba osób. Z roku na rok wartość ta spada, ale liczba ta nadal jest bardzo wysoka. Pandemia znacznie zmniejszyła liczbę wypadków drogowych, ale wartość ta nadal jest bardzo wysoka. Z tego powodu należy dowiedzieć się, w których województwach dochodzi do największej liczby wypadków drogowych oraz poznać prognozę wypadków na najbliższe lata, aby móc zrobić wszystko, aby tę liczbę zminimalizować. Celem artykułu jest sporządzenie prognozy liczby wypadków drogowych w Polsce w podziale na województwa. W tym celu przeanalizowano miesięczne dane dotyczące liczby wypadków w Polsce w latach 2007-2021 pochodzące ze statystyk Policji oraz dokonano prognozy na lata 2022-2024. Na podstawie uzyskanych danych można stwierdzić, że pandemia spowodowała spadek liczby wypadków drogowych w Polsce średnio o 21%. Rozrzut w zależności od województwa waha się w przedziale: 10% dla województwa lubuskiego do prawie 53% dla województwa lubelskiego. Spadek jest najbardziej zauważalny w województwach lubelskim, wielkopolskim i małopolskim. Ponadto prognozy pokazują, że w obecnej sytuacji możemy spodziewać się dalszego spadku liczby wypadków drogowych w Polsce. Wyniki badania pokazują, że nadal możemy spodziewać się podobnego poziomu wypadków drogowych jak przed pandemią z minimalnym spadkiem na polskich drogach, ale panująca pandemia zniekształca uzyskane wyniki. Do prognozowania liczby wypadków drogowych wykorzystano szeregi czasowe i modele wykładnicze.

EN

Every year a very large number of people die on Polish roads. From year to year, the value decreases, but the number is still very high. The pandemic has significantly reduced the number of road accidents, but the value is still very high. For this reason, it is necessary to find out which provinces have the highest number of traffic accidents and to know the accident forecast for the coming years, so that we can do everything possible to minimize this number. The purpose of the article is to make a forecast of the number of road accidents in Poland by province. For this purpose, monthly data on the number of accidents in Poland in 2007-2021 from the statistics of the Police were analyzed, and a forecast was made for 2022-2024. Based on the data obtained, it can be said that the pandemic caused a decrease in the number of road accidents in Poland by an average of 21%. The spreads depending on the province sniff in the range: 10% for Lubuskie Voivodeship to almost 53% for Lubelskie Voivodeship. The decrease is most noticeable in the Lubelskie, Wielkopolskie and Małopolskie provinces. In addition, forecasts show that in the current situation we can expect a further decrease in the number of road accidents in Poland. The results of the study show that we can still expect a similar level of road accidents as before the pandemic with a minimal decrease on Polish roads, but the prevailing pandemic distorts the results obtained. Time series and exponential models were used to forecast the number of traffic accidents.

16

Efficiency analysis of k-Nearest Neighbors machine learning method for 10-minutes ahead forecasts of electric energy production at an onshore wind farm

Rutyna Inajara, Piotrowski Paweł

Przegląd Elektrotechniczny

|

2024

|

R. 100, nr 1

147--150

EN

This paper presents tests of the effectiveness of the K-Nearest Neighbors (KNN) machine learning technique for short-term forecasting of energy production at an onshore wind farm with a horizon of 10 minutes. The tests were performed for several variants of input variables to KNN models (only backward variables of the forecasted time series and the use of additional exogenous input variables - meteorological data). For each of the variants, the selection of an appropriate number of k was performed using the cross-validation method, separately for each of the distance measures tested. Analyses were performed of the found k values depending on the variant of the input variables and the distance measure. Conclusions and observations of the performed tests were formulated.

PL

W artykule przedstawiono testy skuteczności techniki uczenia maszynowego k najbliższych sąsiadów (K-Nearest Neighbors - KNN) do krótkoterminowego prognozowania produkcji energii na farmie wiatrowej lądowej z horyzontem 10 minut. Badania wykonano dla kilku wariantów zmiennych wejściowych do modeli KNN (tylko zmienne cofnięte prognozowanego szeregu czasowego oraz zastosowanie dodatkowych zmiennych wejściowych egzogenicznych – dane meteorologiczne). Dla każdego z wariantów wykonano dobór właściwej liczby k metodą walidacji krzyżowej, osobno dla każdej z testowanych miar odległosci. Wykonano analizy znalezionych wartości k w zależności od wariantu zmiennych wejściowych oraz miary odległości. Sformułowano wnioski i spostrzeżenia z wykonanych badań.

17

Analiza wybranych aspektów pracy KSE pod kątem zapotrzebowania mocy

Malska Wiesława

Przegląd Elektrotechniczny

|

2024

|

R. 100, nr 7

121--126

PL

W artykule zaprezentowano analizę wybranych aspektów pracy Krajowego Systemu Elektroenergetycznego pod kątem zapotrzebowania mocy. Przedstawiono wyniki analiz i obliczeń z wykorzystaniem programu komputerowego Statistica dla dobowej prognozy zapotrzebowania mocy i rzeczywistego zapotrzebowania mocy, a także zaprezentowano model wyrównywania wykładniczego i predykcji.

EN

The article presents an analysis of selected aspects of the operation of the National Power System in terms of power demand. The results of analysis and calculations using the Statistica computer software for daily power demand forecast and actual power demand are presented, and an exponential equalization and forecasting model is presented.

18

Mid-term forecasting of crude oil prices using the hybrid CEEMDAN and CNN_LSTM deep learning model

Gandhi Herry Kartika

Polityka Energetyczna

|

2024

|

T. 27, z. 4

19--38

EN

Forecasting crude oil prices has always been a matter of discussion among energy experts. Due to a significant dependence of the global economy on crude oil, the volatility of the spot price can impact the supply and demand of the market. Moreover, crude oil is still the primary energy for transportation worldwide. Although renewable energy sources have developed significantly, crude oil has been dominant in transportation fuels in the last few decades. This study focuses on mid-term multi-step forecasting and provides a forecasting model that provides a robust prediction for 60 to 90 steps ahead. Our main objective is to develop a forecasting model that can maintain high accuracy and low errors. Our analysis uses a hybrid Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) and the Convolutional Neural Network, Long Short-Term Memory (CNN_LSTM) deep learning model. These three techniques, which have different advantages, are put together, and the combination of them is able to identify features (trend and seasonality) in historical data learning and perform high prediction accuracy for next-term prediction. We compared the proposed model with other decomposition and deep learning techniques. The proposed model shows lower Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE) values than other benchmark models for Brent and crude West Texas Intermediate (WTI) oil prices – the proposed model’s Mean Absolute Percentage Error (MAPE) results in better forecasting with MAPE values between 4 to 10. The simulation with box plot analysis also gives a quartile range value below 0.2, which shows the stability of the model in each iteration. Finally, the proposed model can provide a robust forecasting model for multi-step mid-term forecasting.

PL

Prognozowanie cen ropy naftowej zawsze było przedmiotem dyskusji wśród ekspertów ds. energii. Ze względu na znaczną zależność światowej gospodarki od ropy naftowej, zmienność ceny spot może mieć wpływ na podaż i popyt na rynku. Ponadto ropa naftowa jest nadal podstawową energią dla transportu na całym świecie. Chociaż odnawialne źródła energii znacznie się rozwinęły, ropa naftowa dominuje w paliwach transportowych w ciągu ostatnich kilku dekad. Niniejsze badanie koncentruje się na prognozowaniu wieloetapowym w średnim okresie i dostarcza model prognostyczny, który zapewnia solidną prognozę na 60 do 90 kroków do przodu. Głównym celem jest opracowanie modelu prognostycznego, który może utrzymać wysoką dokładność i niskie błędy. Niniejsza analiza wykorzystuje hybrydowy model uczenia głębokiego Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) i model uczenia głębokiego Convolutional Neural Network, Long Short-Term Memory (CNN_LSTM). Dzięki połączeniu tych trzech różnych technik jesteśmy w stanie identyfikować cechy (trend i sezonowość) w uczeniu się danych historycznych i zapewniać wysoką dokładność prognozowania w przypadku prognozowania na następny okres. W artykule porównano proponowany model z innymi technikami dekompozycji i głębokiego uczenia. Proponowany model wykazuje niższe wartości średniego błędu bezwzględnego (MAE) i średniego błędu kwadratowego (RMSE) niż inne modele referencyjne dla cen ropy Brent i ropy West Texas Intermediate (WTI) – średni błąd procentowy bezwzględny proponowanego modelu (MAPE) skutkuje lepszym prognozowaniem z wartościami MAPE od 4 do 10. Symulacja z analizą wykresu pudełkowego daje również wartość zakresu kwartylowego poniżej 0,2, co pokazuje stabilność modelu w każdej iteracji. Wreszcie, proponowany model może zapewnić solidny model prognostyczny do wieloetapowego prognozowania średnioterminowego.

19

Does a meta-combining method lead to more accurate forecasts in the decision-making process?

Gulay Emrah, Aras Serkan

Operations Research and Decisions

|

2024

|

Vol. 34, No. 3

101--124

EN

To improve forecasting accuracy, researchers employed various combination techniques for a long time. When researchers deal with time series data by using dissimilar models, the combined forecasts of these models are expected to be superior. Deriving a weighting scheme performing better than simple but hard−to−beat combining methods has always been challenging. In this study, a new weighting method based on the hybridisation of combining algorithms is proposed. Five popular datasets were utilised to demonstrate the effectiveness of the proposed method in an out-of-sample context. The results indicate that the proposed method leads to more accurate forecasts than other combining techniques used in the study.

20

Analysing and forecasting the energy consumption of healthcare facilities in the short and medium term : a case study

Koç Ali, Seçkiner Serap Ulusam

Operations Research and Decisions

|

2024

|

Vol. 34, No. 3

165--192

EN

Healthcare facilities consist of multiple large buildings with complex energy systems and high energy consumption, resulting in high carbon emissions. The increasing trend in energy consumption of these facilities and the process of selecting an energy supplier from the open market requires reliable and robust energy forecasting studies. This situation calls for the use of reliable and accurate energy consumption prediction models for the energy needs of healthcare buildings. The aim of this study is to present a prediction framework based on historical energy consumption at different time intervals using six supervised regression algorithms, three linear single, one non-linear single and two non-linear ensembles. The approach adopted for predicting hospital energy consumption involves five steps: data acquisition, data pre-processing, data prediction, hyper-parameter optimisation and feature analysis. Furthermore, all regression algorithms have undergone hyper-parameter optimisation using random search, grid search and Bayesian optimisation to achieve the minimum prediction errors represented by different metrics. The results displayed that the two ensemble models, Extreme Gradient Boosting and Random Forest, outperformed single models in hourly, daily, and monthly energy load prediction. Nevertheless, when considering the computational time for all regression models, the single models have better computational times, although the error metrics are not as good as for the ensemble models. In addition, grid search and Bayesian optimisation performed better than random search in finding optimal hyperparameter values for all datasets. Finally, thanks to feature importance analysis, the most influential features under the hourly, daily, and monthly electrical and monthly natural gas prediction were identified.