Explore the vast range of titles available.

Data‐driven weather models have shown the potential to match the accuracy of state‐of‐the‐art numerical weather predictions (NWPs). However, existing data‐driven forecasting models still have limitations in operational applications. For example, most of them are predominantly trained via fifth‐generation climate reanalysis data (ERA5). However, in actual forecasting operations, the models are usually initiated by analysis fields instead of reanalysis data; this leads to a mismatch between the training data used by machine learning (ML) forecasting models and the actual operational data. To address this issue, we attempt to fine‐tune the data‐driven model with the initiation fields in operation. This study first develops a fine‐tuned Pangu Weather Model (PGW) by integrating forecasting system (IFS) analysis data from 2021 to 2022 and conducts a comprehensive evaluation of its performance. By comparing the fine‐tuned version (PGW_O) with the public version (PGW_P) against IFS models with different resolutions (IFS_L at 0.25° and IFS_H at 0.1°), this research highlights advancements in data‐driven forecasting methodologies. The models are tested on data from South China, a region with dense meteorological observation networks, over a three‐month period, encompassing a detailed case study of Tropical Cyclone Haikui (2023). The findings show that with the forecast activity (FA) level comparable to PGW_P, PGW_O significantly reduces the root mean square error (RMSE) and mean error (ME) across upper atmospheric variables and demonstrates superior accuracy in predicting surface elements. The operational relevance of these models is evaluated through both ERA5 reanalysis and surface observations, revealing that fine‐tuning with IFS data enhances PGW compatibility and forecasting precision, particularly for severe weather events. To improve the forecasting ability within the operational framework, the Pangu Weather Model (PGW) is fine‐tuned for the first time via integrated forecasting system (IFS) analysis data from 2021 to 2022. Compared with the public version of the PGW (PGW_P) and IFS models, the fine‐tuned Pangu Weather Model (PGW_O) has the best forecasting ability, particularly for the upper variables, where the fine‐tuning effects are especially pronounced.

We introduce a data-driven forecasting method for high-dimensional chaotic systems using long short-term memory (LSTM) recurrent neural networks. The proposed LSTM neural networks perform inference of high-dimensional dynamical systems in their reduced order space and are shown to be an effective set of nonlinear approximators of their attractor. We demonstrate the forecasting performance of the LSTM and compare it with Gaussian processes (GPs) in time series obtained from the Lorenz 96 system, the Kuramoto–Sivashinsky equation and a prototype climate model. The LSTM networks outperform the GPs in short-term forecasting accuracy in all applications considered. A hybrid architecture, extending the LSTM with a mean stochastic model (MSM–LSTM), is proposed to ensure convergence to the invariant measure. This novel hybrid method is fully data-driven and extends the forecasting capabilities of LSTM networks.

This research conducts a comparative analysis of nine Machine Learning (ML) models for temperature and humidity prediction in Photovoltaic (PV) environments. Using a dataset of 5,000 samples (80% for training, 20% for testing), the models—Support Vector Regression (SVR), Lasso Regression, Ridge Regression (RR), Linear Regression (LR), AdaBoost, Gradient Boosting (GB), Decision Tree (DT), Random Forest (RF), and eXtreme Gradient Boosting (XGBoost)—were evaluated based on Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and the Coefficient of Determination (R²). For temperature prediction, XGBoost demonstrated the best performance, achieving the lowest MAE of 1.544, the lowest RMSE of 1.242, and the highest R² of 0.947, indicating strong predictive accuracy. Conversely, SVR had the weakest performance with an MAE of 4.558 and an R² of 0.674. Similarly, for humidity prediction, XGBoost outperformed other models, achieving an MAE of 3.550, RMSE of 1.884, and R² of 0.744, while SVR exhibited the lowest predictive power with an R² of 0.253. This comprehensive study serves as a benchmark for the application of ML models to environmental prediction in PV systems, a research area that is relatively important. Notably, the results underscore the performance advantage of ensemble-based approaches, especially for XGBoost and RF compared to simpler, linear-based methods such as LR and SVR, when it comes to well-dispersed environmental interactions. The proposed machine-learning based power generation analysis approach shows significant improvements in predictive analytics capabilities for renewable energy systems, as well as a means for real-time monitoring and maintenance practices to improve PV performance and reliability.

As data-driven weather forecasting models increasingly come in operational use, questions of fairness and equitable access to forecast improvements are gaining urgency. This paper introduces a conceptual framework for evaluating outcome-based fairness in global data-driven forecasts, drawing on principles from the algorithmic fairness literature. Specifically, we focus on two criteria: statistical parity (i.e. comparable improvements across protected groups) and conditional independence (i.e. no dependence of improvements on protected variables). Using ECMWF’s AIFS model as a case study and IFS HRES as a baseline, we assess whether forecast improvements are equitably distributed across different income groups and population densities. We find that although AIFS provides substantial overall improvements in forecast skill, these gains are unevenly distributed: on average, wealthier and more densely populated areas are more likely to experience forecast improvements, violating group fairness and conditional independence criteria. We conclude by discussing how fairness-aware loss functions could be incorporated into data-driven weather forecasting systems and argue for a broader integration of fairness considerations into model development and evaluation.

Among natural hazards, floods pose the greatest threat to lives and livelihoods. To reduce flood impacts, short-term flood forecasting can contribute to early warnings that provide communities with time to react. This manuscript explores how machine learning (ML) can support short-term flood forecasting. Using two methods [strengths, weaknesses, opportunities, and threats (SWOT) and comparative performance analysis] for different forecast lead times (1–6, 6–12, 12–24, and 24–48 h), we evaluate the performance of machine learning models in 94 journal papers from 2001 to 2023. SWOT reveals that the best short-term flood forecasting was produced by hybrid, random forest (RF), long short-term memory (LSTM), artificial neural network (ANN), and adaptive neuro-fuzzy inference system (ANFIS) approaches. The comparative performance analysis, meanwhile, favors convolutional neural network, ANFIS, multilayer perceptron, k-nearest neighbors algorithm (KNN), hybrid, LSTM, ANN, and support vector machine (SVM) at 1–6 h; hybrid, ANFIS, ANN, and LSTM at 6–12 h; SVM, hybrid, and RF at 12–24 h; and hybrid and RF at 24–48 h. In general, hybrid approaches consistently perform well across all lead times. Trends such as hybridization, model selection, input data selection, and decomposition seem to improve the accuracy of models. Furthermore, effective stand-alone ML models such as ANN, SVM, RF, genetic algorithm, KNN, and LSTM, provide better outcomes through hybridization with other ML models. By including different machine learning models and parameters such as environmental, socio-economical, and climatic parameters, the hybrid system can produce more accurate flood forecasting, making it more effective for early warning operational purposes.

In recent decades, the traditional monopolistic energy exchange market has been replaced by deregulated, competitive marketplaces in which electricity may be purchased and sold at market prices like any other commodity. As a result, the deregulation of the electricity industry has produced a demand for wholesale organized marketplaces. Price predictions, which are primarily meant to establish the market clearing price, have become a significant factor to an energy company’s decision making and strategic development. Recently, the fast development of deep learning algorithms, as well as the deployment of front-end metaheuristic optimization approaches, have resulted in the efficient development of enhanced prediction models that are used for electricity price forecasting. In this paper, the development of six highly accurate, robust and optimized data-driven forecasting models in conjunction with an optimized Variational Mode Decomposition method and the K-Means clustering algorithm for short-term electricity price forecasting is proposed. In this work, we also establish an Inverted and Discrete Particle Swarm Optimization approach that is implemented for the optimization of the Variational Mode Decomposition method. The prediction of the day-ahead electricity prices is based on historical weather and price data of the deregulated Greek electricity market. The resulting forecasting outcomes are thoroughly compared in order to address which of the two proposed divide-and-conquer preprocessing approaches results in more accuracy concerning the issue of short-term electricity price forecasting. Finally, the proposed technique that produces the smallest error in the electricity price forecasting is based on Variational Mode Decomposition, which is optimized through the proposed variation of Particle Swarm Optimization, with a mean absolute percentage error value of 6.15%.

Recently, several papers have been published regarding the use of preprocessing models, such as Discrete Wavelet, in Data-Driven Forecasting Frameworks (DDFF). However, these models face unresolved issues, including the use of future data, boundary-affected data, and incorrect selection of decomposition level and wavelet filter, which can lead to inaccurate results. In contrast, the Wavelet-based Data-Driven Forecasting Framework (WDDFF) overcomes these problems. To address the first two issues, we can use Maximal Overlap Discrete Wavelet Transform (MODWT) and a-trous algorithm (AT). Although there is currently no definitive solution for selecting the decomposition level and wavelet filter, we propose a novel approach using Entropy to address these issues. By utilizing the concept of predictability of time series using entropy, we can determine the optimal decomposition level and suitable filter to develop the Maximal Overlap Discrete Wavelet-Entropy Transform (MODWET) and apply it to WDDFF accurately. This study, demonstrates the effectiveness of MODWET through three real-world case studies on the CAMELS data set. In these studies, we will forecast the streamflow of specific stations one month ahead to prove the effectiveness of using preprocessing algorithms for forecasting models. The proposed model combines Input Variable Selection (IVS), preprocessing model, and Data-Driven Model (DDM). We will conclude that MODWET-ANN is the most effective model and highlight how entropy can accurately identify the optimal decomposition level and filter, resolving the concerns associated with using WDDFF in hydrological forecasting problems.

This study investigates the capability of Gaussian process regression (GPR) models in the probabilistic forecasting of water flow and depth in a combined sewer system. Traditionally, deterministic methods have been implemented in sewer flow forecasting and anomaly detection, two crucial techniques for a good wastewater network and treatment plant management. However, with the uncertain nature of the factors impacting on sewer flow and depth, a probabilistic approach which takes uncertainties into account is preferred. This research introduces a novel use of GPR in sewer systems for real-time control and forecasting. To this end, a composite kernel is designed to capture flow and depth patterns in dry- and wet-weather periods by considering the underlying physical characteristics of the system. The multi-input, single-output GPR model is evaluated using root mean square error (RMSE), coverage, and differential entropy. The model demonstrates high predictive accuracy for both treatment plant inflow and manhole water levels across various training durations, with coverage values ranging from 87.5% to 99.4%. Finally, the model is used for anomaly detection by identifying deviations from expected ranges, enabling the estimation of surcharge and overflow probabilities under various conditions.

Background Post bariatric hypoglycaemic (PBH) is a late complication of weight loss surgery, characterised by critically low blood glucose levels following meal-induced glycaemic excursions. The disabling consequences of PBH underline the need for the development of a decision support system (DSS) that can warn individuals about upcoming PBH events, thus enabling preventive actions to avoid impending episodes. In view of this, we developed various algorithms based on linear and deep learning models to forecast PBH episodes in the short-term. Methods We leveraged a dataset obtained from 50 patients with PBH after Roux-en-Y gastric bypass, monitored for up to 50 days under unrestricted real-life conditions. Algorithms’ performance was assessed by measuring Precision, Recall, F1-score, False-alarms-per-day and Time Gain (TG). Results The run-to-run forecasting algorithm based on recursive autoregressive model (rAR) outperformed the other techniques, achieving Precision of 64.38%, Recall of 84.43%, F1-score of 73.06%, a median TG of 10 min and 1 false alarm every 6 days. More complex deep learning models demonstrated similar median TG but inferior forecasting capabilities with F1-score ranging from 54.88% to 64.10%. Conclusions Real-time forecasting of PBH events using CGM data as a single input imposes high demands on various types of prediction algorithms, with CGM data noise and rapid postprandial glucose dynamics representing the key challenges. In this study, the run-to-run rAR yielded most satisfactory results with accurate PBH event predictive capacity and few false alarms, thereby indicating potential for the development of DSS for people with PBH.

This paper proposes novel forecasting models for fractional-order chaotic oscillators, such as Duffing’s, Van der Pol’s, Tamaševičius’s and Chua’s, using feedforward neural networks. The models predict a change in the state values which bears a weighted relationship with the oscillator states. Such an arrangement is a suitable candidate model for out-of-sample forecasting of system states. The proposed neural network-assisted weighted model is applied to the above oscillators. The improved out-of-sample forecasting results of the proposed modeling strategy compared with the literature are comprehensively analyzed. The proposed models corresponding to the optimal weights result in the least mean square error (MSE) for all the system states. Further, the MSE for the proposed model is less in most of the oscillators compared with the one reported in the literature. The proposed prediction model’s out-of-sample forecasting plots show the best tracking ability to approximate future state values.