Explainable district heating load forecasting by means of a reservoir computing deep learning architecture

Energy, Journal Year: 2025, Volume and Issue: unknown, P. 134641 - 134641

Published: Jan. 1, 2025

Language: Английский

---

Operation optimization for a CHP system using an integrated approach of ANN and simulation database

Applied Thermal Engineering, Journal Year: 2025, Volume and Issue: unknown, P. 125771 - 125771

Published: Jan. 1, 2025

Language: Английский

---

Modeling and forecasting electricity consumption amid the COVID-19 pandemic: Machine learning vs. nonlinear econometric time series models

Sustainable Cities and Society, Journal Year: 2023, Volume and Issue: 98, P. 104860 - 104860

Published: Aug. 15, 2023

Accurately modelling and forecasting electricity consumption remains a challenging task due to the large number of statistical properties that characterize this time series such as seasonality, trend, sudden changes, slow decay autocrrelation function, among many others. This study contributes literature by using comparing four advanced econometrics models, machine learning deep models1 analyze forecast during COVID-19 pre-lockdown, lockdown, releasing-lockdown, post-lockdown phases. Monthly data on Qatar's total has been used from January 2010 December 2021. The empirical findings demonstrate both econometric models are able capture most important features characterizing consumption. In particular, it is found climate change based factors, e.g temperature, rainfall, mean sea-level pressure wind speed, key determinants terms forecasting, results indicate autoregressive fractionally integrated moving average three state markov switching with exogenous variables outperform all other models. Policy implications energy-environmental recommendations proposed discussed.

Language: Английский

---

Explainable district heat load forecasting with active deep learning

Applied Energy, Journal Year: 2023, Volume and Issue: 350, P. 121753 - 121753

Published: Aug. 21, 2023

District heat load forecasting is a challenging task that involves predicting future demand based on historical data and various influencing factors. Accurate essential for optimizing energy production distribution in district heating systems. However, most existing models lack transparency interpretability fail to capture the spatial–temporal dependencies data. Moreover, they often require large amount of annotated training, which can be costly time-consuming obtain. In this paper, we present novel approach forecasting, The proposed an Active Graph Recurrent Network (Ac-GRN), leverages strengths active deep learning graph neural networks complex also provides explainability its predictions by using correlation-based attribution methods. component effectively select informative representative samples from pool data, reducing frequency cost collection human effort. network model both linear nonlinear relationships among meters bidirectional recurrent connections, enhancing accuracy robustness predictions. We conduct extensive experiments compare our with eleven state-of-the-art real-world dataset consumption Danish residential buildings. Our results show outperforms other terms accuracy, robustness, reliability, computational efficiency multi-horizon multi-step forecasting. meaningful explanations highlighting influential factors each prediction. This paper makes contribution explainability.

Language: Английский

---

Explainable AI models for predicting drop coalescence in microfluidics device

Chemical Engineering Journal, Journal Year: 2024, Volume and Issue: 481, P. 148465 - 148465

Published: Jan. 2, 2024

In the field of chemical engineering, understanding dynamics and probability drop coalescence is not just an academic pursuit, but a critical requirement for advancing process design by applying energy only where it needed to build necessary interfacial structures, increasing efficiency towards Net Zero manufacture. This research applies machine learning predictive models unravel sophisticated relationships embedded in experimental data on microfluidics device. Through deployment SHapley Additive exPlanations values, features relevant processes are consistently identified. Comprehensive feature ablation tests further delineate robustness susceptibility each model. Furthermore, incorporation Local Interpretable Model-agnostic Explanations local interpretability offers elucidative perspective, clarifying intricate decision-making mechanisms inherent model's predictions. As result, this provides relative importance outcome interactions. It also underscores pivotal role model reinforcing confidence predictions complex physical phenomena that central engineering applications.

Language: Английский

---

Accuracy improvement of the load forecasting in the district heating system by the informer-based framework with the optimal step size selection

Energy, Journal Year: 2024, Volume and Issue: 291, P. 130347 - 130347

Published: Jan. 17, 2024

Language: Английский

---

Predicting hourly heating load in residential buildings using a hybrid SSA–CNN–SVM approach

Case Studies in Thermal Engineering, Journal Year: 2024, Volume and Issue: 59, P. 104516 - 104516

Published: May 8, 2024

This study proposes a hybrid prediction model using sparrow search algorithm (SSA) to optimize the convolutional neural network (CNN) and support vector machine (SVM), in order perform accurate of secondary supply temperature (Ts2). The historical operation data Weifang residential building thermal station was adopted reasonable preprocessing performed suppress interference abnormal data. input variables were screened correlation analysis method, taking influence hysteresis effect into consideration. SSA-CNN-SVM then developed for prediction. performance evaluated by root mean square error, absolute percentage error (MAPE), value relative each time step. results obtained demonstrated that has high accuracy. MAPE values two heat exchange stations between 2.28% 2.4%. indoor significantly affected accuracy Ts2. After introduction temperature, predicted reduced 0.35%. maximum reduction 1.5% compared with other models.

Language: Английский

---

Investigation on the Long Short-term Memory-based Models for Rural Heating Load Prediction in Northeast China

Energy, Journal Year: 2025, Volume and Issue: unknown, P. 134751 - 134751

Published: Jan. 1, 2025

Language: Английский

---

District heating load patterns and short-term forecasting for buildings and city level

Energy, Journal Year: 2023, Volume and Issue: 289, P. 129866 - 129866

Published: Dec. 6, 2023

District heating (DH) load forecasting for buildings and cities is essential DH production planning demand-side management. This study analyzes compares the hourly patterns a city five different types of over an entire year. The various operating modes introduce nonlinear dependencies between outdoor temperature. We compare prediction accuracies multiple linear regression (MLR) artificial neural network (ANN) models. Without dependencies, both ANN MLR provide good, almost identical accuracies. In case superior to MLR. However, novel clustering method eliminates improves accuracy on par with ANN. methods can automatically adapt nonlinearities. advantage combining that it simpler than designing method, although manual work required. addition, more insight into how depends factors compared 'black box' developed methodology be widely applied building- city-level analyses in systems combined or without domestic hot water consumption.

Language: Английский

---

Temporal collaborative attention for wind power forecasting

Applied Energy, Journal Year: 2023, Volume and Issue: 357, P. 122502 - 122502

Published: Dec. 16, 2023

Wind power serves as a clean and sustainable form of energy. However, its generation is fraught with variability uncertainty, owing to the stochastic dynamic characteristics wind. Accurate forecasting wind indispensable for efficient planning, operation, grid integration energy systems. In this paper, we introduce novel method termed Temporal Collaborative Attention (TCOAT). This data-driven approach designed capture both temporal spatial dependencies in data, well discern long-term short-term patterns. Utilizing attention mechanisms, TCOAT dynamically adjusts weights each input variable time step based on their contextual relevance forecasting. Furthermore, employs collaborative units assimilate directional global information from data. It also explicitly models interactions correlations among different variables or steps through use self-attention cross-attention mechanisms. To integrate effectively, incorporates fusion layer that concatenation mapping operations, along hierarchical feature extraction aggregation. We validate efficacy extensive experiments real-world dataset Greece compare performance against twenty-two state-of-the-art methods. Experimental results demonstrate outperforms existing methods terms accuracy robustness Moreover, conduct generality study an additional climate condition characteristics. The show can achieve comparable better than methods, confirming generalization ability TCOAT.

Language: Английский

---