Demand Forecasting Research Articles (Page 1)

Demand forecasting and PV/EV-integrated energy system modeling for achieving a sustainable island

Seasonal forecasting of peak electricity demand using spectral analysis

Safe-AutoSAC: AutoML-enhanced safe deep reinforcement learning for integrated energy system scheduling with multi-channel informer forecasting and electric vehicle demand response

ML-Based Demand Forecasting in Fuzzy Inventory Models with Emission Costs: A Memory Effect Analysis

e-values based continuous-time model selection for residential electricity demand forecasts

This paper proposes a profit optimization method for the natural gas industry chain driven by demand forecasting. The method mainly consists of two core components: the construction of a natural gas demand forecasting model and the solution of an industry chain profit optimization model. In the forecasting stage, three models are trained using historical natural gas demand data, and the optimal model is selected based on performance evaluation indicators to predict natural gas demand for the coming month. In the optimization stage, the physical and operational characteristics of key components in the natural gas pipeline network are fully considered, and a nonlinear programming model is formulated with the objective of maximizing the overall profit of the industry chain. The model is validated using historical data. Finally, the demand forecast results are incorporated into the optimization model to calculate the expected industry chain profit for the next month. The findings of this study can provide theoretical foundations and quantitative decision-making support for natural gas suppliers to develop more economically efficient gas supply strategies.

China's economy relies heavily on its diverse industries, and efficient economic management plays a vital role in national development and improving people's livelihoods. The application of massive data in economic management not only enhances the level of financial control but also lays a solid foundation for subsequent optimization and research on financial management technologies. This paper analyzes the application of big data in economic management, discusses its development, management practices, and existing challenges, and highlights the important role of data-driven decision-making in optimizing financial control. A new economic management model based on data monitoring, consumer demand forecasting, and product price information analysis is proposed. The proposed algorithm achieves a 22.98% higher accuracy in predicting consumer purchasing propensity compared to the traditional support vector machines algorithm.

BiLSTM-attention for electricity demand forecasting and grid stability: supporting Morocco’s 2030 energy transition

In the era of digital transformation, traditional banking institutions are increasingly adopting Artificial Intelligence (AI), Machine Learning (ML), and Deep Learning (DL) to enhance their products, services, and customer engagement strategies. This study presents a comprehensive analysis of the products and services offered by Bank of Baroda (BoB), with a focus on how AI-driven technologies can be applied to improve service delivery, operational efficiency, and customer satisfaction.The research begins with a traditional evaluation of BoB's core banking offerings, including savings and current accounts, fixed deposits, credit cards, loans, insurance, investment services, and digital banking platforms. Using publicly available data and simulated customer feedback, we implement ML models such as decision trees and clustering algorithms to categorize and predict customer preferences across different demographic segments. Further, DL models—particularly neural networks—are employed to analyze transactional data patterns and forecast customer behavior, such as loan repayment tendencies, usage of digital services, and likelihood of product adoption.The findings reveal that integrating AI, ML, and DL into product and service analysis enables more accurate customer segmentation, demand forecasting, and service personalization. For example, predictive models can identify which customers are more likely to benefit from personal loans or digital banking features, thereby enabling targeted marketing and resource optimization. This AIpowered approach not only helps Bank of Baroda adapt to changing consumer behavior but also positions the bank as a proactive, data-driven institution in India’s competitive financial ecosystem

Background Red blood cells (RBCs) infusion is very important for the treatment of hematology patients, but how to maintain a balanced state between the supply and demand of RBCs is still a major challenge. Objective This study aimed to explore the feasibility of seasonal autoregressive integrated moving average (SARIMA) model and exponential smoothing (ES) model in predicting the clinical demand of RBCs for hematology patients each month. Methods Our study collected the monthly RBCs usage data of hematology patients from January 2014 to December 2023 to establish the SARIMA model and ES model, respectively. Then, the optimal model was used to forecast the monthly usage of RBCs from January to June 2024, and we subsequently compared the data with actual values to evaluate the prediction effect of the model. Results The best fitting SARIMA model was SARIMA (2,1,0)(1,1,1) 12 , whose R 2 = 0.603, MAE = 37.092, MAPE = 13.693, BIC = 7.896. The best fitting ES model was Winters addition model, whose R 2 = 0.702, MAE = 32.617, MAPE = 12.138, BIC = 7.485. The mean relative errors of two models were 0.085 and 0.159, respectively. The SARIMA (2,1,0)(1,1,1) 12 model performed better in prediction. Conclusion Compared with the ES model, the SARIMA model has a smaller mean relative error in predicting RBCs usage in hematology patients. DM test also verify this result. But in the future, more similar research data are needed to make research more convincing.

The dual-channel structure resulted from manufacturer encroachment could alter the incentives of downstream retailer to ex ante communicate demand forecast. And different types of channel competition need to be investigated in this dual-channel information sharing scenario. This paper aims to investigate retailer’s ex ante imperfect demand information sharing strategy given that upstream manufacturer has set up direct sales channel (manufacturer encroachment). The imperfect information sharing means the demand information shared is uncertain and has some error relative to the real-world demand condition. It examines two types of channel competition: quantity competition and price competition. Additionally, this study discusses the encroaching manufacturer’s incentives for adjusting channel substitution. The paper adopts a stylized game theoretic model to describe interactions between retailer and the encroaching manufacturer. Contrary to conventional wisdom, the paper shows that under manufacturer encroachment, it is always possible for ex ante demand information sharing. Specifically, in the Cournot competition scenario where retailer channel and the encroaching manufacturer direct channel compete in quantity, the encroaching manufacturer could encourage demand information communication through side payment. Furthermore, in the Bertrand competition scenario, retailer may voluntarily share demand information. In addition, in either quantity or price competition, the encroaching manufacturer has incentives to adjust channel substitution for profit maximization.

Stage-adjusted forecasting of radiotherapy demand and outcome benefits across income groups: Estimating survival and local control gains by 2050.

To improve the intelligent and refined management level of power distribution systems in equipment operation and maintenance as well as emergency support, this work proposes an integrated “prediction-optimization” model that combines genetic algorithm (GA) with machine learning methods. This method uses GA to intelligently screen key features and optimize model parameters. It dynamically integrates the prediction link with inventory decisions, alleviating the problem of multi-objective coupling imbalance in traditional fragmented optimization. Compared with a single machine learning or heuristic algorithm, this model significantly reduces the unit prediction error under load fluctuations and extreme weather scenarios. Verification of model performance based on The European Network of Transmission System Operators for Electricity (ENTSO-E) dataset shows that the model achieves good results in the prediction stage. For example, in load time series data, the mean absolute percentage error is 3.41%, and the coefficient of determination reaches 0.942. In the inventory optimization stage, the model reduces the average inventory level to 42.63, controls the total cost per unit equipment at 92.37, and lowers the redundant inventory ratio to 9.42%. Its comprehensive performance is better than that of Temporal Fusion Transformer (TFT) and Neural Basis Expansion Analysis for Time Series Forecasting (N-BEATS). This work provides theoretical models and empirical support for research in the field of typical equipment prediction and inventory optimization in intelligent power distribution systems, and has certain practical value and promotion significance.

Biotechnology is a cross-functional field that involves the creation of biological products, process optimization, and sustainable use of resources. Deep learning (DL) has recently gained popularity in the biotechnology industry for solving problems such as production optimization, production demand forecasting, and process monitoring. However, classic DL methods suffer from overfitting, are inflexible to new data, and cannot handle contaminated or missing data. This research paper aims to address these shortcomings by proposing a robust predictive workflow based on an RNN-based framework and the best learning techniques. To resolve the issues, the proposed method, a combination of Transformer and Graph Attention Network (GAT), works together to facilitate the long-term prediction of temporal and interdependent features for biotechnology demand forecasting. Additionally, Self-supervised learning (SSL) is used to preprocess the data, allowing the model to learn the latent structure and improve the original dataset. Furthermore, AutoML then uses reinforcement learning (AutoMLRL) to select the most relevant demand-related features and remove redundancies. Finally, model training using RNN + Meta-Learning (MAML) is then applied as part of the workflow to learn temporal dependencies and adapt to different datasets. Finally, the presented method robustly forecasts demand, predicting future production rates, resource requirements, and operational conditions with a high accuracy of 93%.

Temporal pattern recognition has become increasingly critical for predictive analytics in various domains, particularly in demand forecasting where accurate predictions directly impact business operations and profitability. Neural network (NN) architectures have demonstrated remarkable capabilities in capturing complex temporal dependencies within sequential data, outperforming traditional statistical methods in numerous applications. This review examines the evolution and application of neural network approaches specifically designed for temporal pattern recognition, with emphasis on their utilization in demand forecasting and predictive analytics. The paper provides a comprehensive analysis of recurrent neural networks (RNNs), long short-term memory (LSTM) networks, gated recurrent units (GRUs), convolutional neural networks (CNNs), and transformer-based architectures in the context of time series forecasting. Furthermore, this review explores the integration of attention mechanisms, the emergence of spatiotemporal graph neural networks (STGNNs), and hybrid model architectures that combine multiple approaches to enhance forecasting accuracy. The evaluation metrics commonly employed to assess model performance, including mean absolute error (MAE), root mean squared error (RMSE), and mean absolute percentage error (MAPE), are discussed alongside benchmark datasets utilized in the field. Through systematic examination of recent literature spanning from 2019 to 2025, this review identifies key architectural innovations, practical applications in retail and supply chain management, and emerging trends that define the current state of temporal pattern recognition. The findings reveal that while transformer-based models have gained significant attention for long-sequence forecasting, simpler linear architectures and hybrid approaches often demonstrate competitive or superior performance depending on dataset characteristics and application requirements. This comprehensive review serves as a foundation for researchers and practitioners seeking to understand the landscape of neural network methodologies for temporal pattern recognition and their practical deployment in demand forecasting systems.

Forecasting room utilization based on indoor environmental conditions offers a novel approach, which improves energy efficiency and also delivers the personalized indoor comfort. This study investigates whether parameters such as CO2 concentration, illumination, humidity, and temperature which can reliably predict the room occupancy. This work introduces a new deep learning-augmented predictive energy modeling (DL-PEM) framework combined with multi-objective particle swarm optimization (MOPSO) for real-time, occupancy-based energy management in intelligent buildings. In contrast to conventional linear predictive or rule-based systems, DL-PEM utilizes a deep feedforward neural network (DNN) that can extract non-linear relationships between indoor environmental variables (CO2, lighting, humidity, temperature) and occupation patterns. Using a Pareto-optimal approach to manage trade-offs, this real-time system minimizes energy consumption (kWh), reduces CO2 concentration (ppm), and maximizes the occupant thermal comfort index. It does this by adjusting adaptive HVAC and lighting control via MOPSO. The combined framework illustrates enhanced adaptability under changing conditions and scalability towards wider deployment. Experimental results indicate that the method proposed here attains 99.8% accuracy and at most 85% optimization efficiency, outperforming KNN, DT, AO-ANN, and even LSTM baselines for prediction and control tasks. Empirical evaluation using real building data demonstrates that the proposed DL-PEM-MOPSO framework significantly outperforms traditional models, enhances decision-making transparency, and offers a scalable, future-ready solution for smart building energy management. It improves occupancy data analysis and adapts to seasonal fluctuations, which also optimizes the thermal comfort and enables the accurate power demand forecasting, and enhances overall energy utilization.

Background: Healthcare systems worldwide face growing challenges in anticipating and managing patient surges, particularly in times of public health crises, natural disasters, or seasonal peaks. The ability of healthcare organisations to forecast and respond to such demand fluctuations—referred to as organisational readiness for patient capacity surge—has become a critical determinant of service continuity and patient outcomes. Despite the urgency, there remains a lack of consolidated evidence on how healthcare authorities measure, evaluate, and operationalise this readiness. This systematic review aims to identify and synthesise existing literature that presents case studies, methodologies, and strategic frameworks used to evaluate organisational preparedness for patient surge capacity. It also explores resource allocation mechanisms, hospital capacity planning algorithms, and temporary facility strategies documented in healthcare settings. Methods: The review was conducted across two major scientific repositories, i.e., PubMed and Web of Science (WoS). A set of four structured search queries were formulated to capture the breadth of the topic, focusing on demand forecasting, hospital capacity planning, workforce models, and resource management within the context of healthcare surge demand. The search was limited to publications from the last 10 years (2014–2024) to ensure the inclusion of contemporary practices and technologies. Resutls: A total of 142 articles were selected for detailed analysis. The articles were categorised into six thematic groups: (i) empirical case studies on healthcare surge management; (ii) hospital resources and capacity scaling; (iii) ethical frameworks guiding surge response; (iv) IT-driven algorithms and forecasting tools; (v) policy evaluations and actionable lessons learned; and (vi) existing systematic reviews in related domains. Notably, several articles provided evidence-based frameworks and simulation models supporting predictive planning, while others highlighted real-world implementation of temporary care facilities and staff redeployment protocols. Conclusions: The review underscores the fragmented yet growing body of literature addressing the multidimensional nature of surge preparedness in healthcare. While algorithmic forecasting and capacity modelling are advancing, gaps remain in standardising metrics for organisational readiness and incorporating ethical considerations in surge planning. Limitations of this review include potential selection bias and the subjective categorisation of articles. Future research should aim to develop integrative frameworks that couple technical, operational, and ethical readiness for patient surge scenarios.

Abstract - The traditional power system's structure is rapidly evolving due to the shift towards electric vehicles (EVs) and renewable energy sources. With more electric vehicles on the road and solar and wind power plants joining the grid, it is getting increasingly challenging to predict the level of electrical usage due to the unpredictability of these technologies. Keep in mind that precise electric demand forecasting is essential for energy dispatch scheduling, grid stability, and the development of an efficient smart grid and EV model. The chaotic behaviour of today's electricity system and its non-linearity are beyond the scope of traditional statistical models and approaches. The ability to work with large volumes of data, exploit hidden data, and generate precise short-term (from minutes to weeks) and long-term (from months to years) load estimates has always been a strength of machine learning (ML) models. This research introduces a machine-learning technique to load forecasting in power systems combining renewable energy and EV. The approach has included a feature engineering process, a preparation phase, and the use of many machine learning algorithms, such as Random Forest, Support Vector Regression, and LSTM. According to the study, machine learning techniques improve forecasting accuracy when compared to conventional techniques. The study concludes by outlining the difficulties, restrictions, and potential avenues for further research in intelligent energy management. KeyWords: Machine learning, Load forecasting, Electric Vehicles, Renewable Energy, Smart Grid.

The increasing demand for flexible and sustainable mobility services has intensified interest in ridesharing and mobility-on-demand systems. However, current solutions remain limited, as they often neglect critical factors such as demand forecasting, vehicle repositioning, workload balance, and real-time adaptability. This study proposes an integrated data-driven framework for ridesharing and dynamic vehicle routing, composed of a periodic and dynamic components. By synchronising these stages, the system captures interdependencies between demand prediction and routing decisions, reducing inefficiencies caused by treating them separetely. The approach was validated on over 300,000 historical trips and tested with 9,016 real-time demands. Results show that the framework achieves a demand cancellation rate below 9%, balances driver workloads, and responds to new requests within 4 s. Sensitivity analyses reveal that omitting clustering or forecasting raises costs, while the full integration yields 6–10% savings. The proposed system provides a holistic, scalable, and sustainable real time optimisation model for future urban mobility.

Accurately forecasting the demand for air passengers is vital for the aviation industry to formulate appropriate management strategies. Decomposition ensemble learning has attracted much attention from researchers of this problem because it is an effective way to improve forecasting accuracy. In contrast to common ways of generating ensemble forecasts, such as artificial intelligence and linear addition, our study employs the Choquet fuzzy integral. The Choquet integral is effective regardless of the training sample size and it uses a nonadditive fuzzy measure to explain the influence of the inputs on air passenger demand. Data on monthly air passenger flows from major airports in Taiwan were used to assess the effectiveness of the proposed decomposition ensemble models using the Choquet fuzzy integral to generate ensemble forecasts. The results in terms of level and directional forecasting accuracy showed that the proposed models— especially those that integrated smoothing (LOESS) (STL) and radial basis function network with the Choquet integral—significantly outperformed single (non-ensemble) forecasting models and the benchmark models considered.