The growing interest in saving building energy has increasingly motivated studies on model predictive control (MPC), where system operation proceeds according to a planned operation strategy. Data-driven models that perform learning using past operation data of buildings are favorable for MPC applications owing to their fast computation speed. However, it is difficult to apply MPC to buildings with insufficient operation data, as the prediction accuracy varies depending on the data used for learning. To address this, we propose a method that involves generating data through a detailed building energy model and utilizing a long short-term memory (LSTM) network that performs learning using the data as an MPC model. The model was verified through a comparison with the reference model using the same optimization algorithm. In the MPC of the objective function, which is to reduce electrical energy expenditure by optimizing the indoor temperature of the target building, approximately 35% grid energy consumption was reduced compared to a reference case, by increasing self-consumption of photovoltaic (PV) energy and avoiding PV curtailment. Further, the required computation time was reduced to approximately 30%, even including the data generation time for daily learning, thereby confirming the feasibility of the MPC model that employs LSTM.

Solar irradiance prediction is significant for maximizing energy-saving effects in the predictive control of buildings. Several models for solar irradiance prediction have been developed; however, they require the collection of weather data over a long period in the predicted target region or evaluation of various weather data in real time. In this study, a long short-term memory algorithm–based model is proposed using limited input data and data from other regions. The proposed model can predict solar irradiance using next-day weather forecasts by the Korea Meteorological Administration and daily solar irradiance, and it is possible to build a model with one-time learning using national and international data. The model developed in this study showed excellent predictive performance with a coefficient of variation of the root mean square error of 12% per year even if the learning and forecast regions were different, assuming that the weather forecast was correct.

The aim of this study is to develop a model that can accurately calculate building loads and demand for predictive control. Thus, the building energy model needs to be combined with weather prediction models operated by a model predictive controller to forecast indoor temperatures for specified rates of supplied energy. In this study, a resistance–capacitance (RC) building model is proposed where the parameters of the models are determined by learning. Particle swarm optimization is used as a learning scheme to search for the optimal parameters. Weather prediction models are proposed that use a limited amount of forecasting information fed by local meteorological centers. Assuming that weather forecasting was perfect, hourly outdoor temperatures were accurately predicted; meanwhile, differences were observed in the predicted solar irradiances values. In investigations to verify the proposed method, a seven-resistance, five-capacitance (7R5C) model was tested against a reference model in EnergyPlus using the predicted weather data. The root-mean-square errors of the 7R5C model in the prediction of indoor temperatures on all the specified days were within 0.5 °C when learning was performed using reference data obtained from the previous five days and weather prediction was included. This level of deviation in predictive control is acceptable considering the magnitudes of the loads and demand of the tested building.

To save energy consumed by a building, utilizing optimal predictive control with model predictive control (MPC) makes the most of energy storage systems (ESSs) to reduce the electrical energy consumption of peak and heavy loads. This study evaluated MPC applicability in a multi-zone commercial building using the EnergyPlus model and conducted multi-objective optimization of thermal comfort and energy savings. As a result of the simulation, optimal ESS charging scenarios responded to the fluctuating electricity pricing system, and changing the peak load time reduced the electricity bill of the grid by 55% compared to the existing operating method. At the same time, room temperatures stayed within the thermal comfort range, and the Pareto curve showed a proper balance between energy saving and thermal comfort. Especially, the proposed method with a white model is applicable for MPC applications in commercial buildings, as it gave optimal solutions within the target time interval.

Prediction of solar irradiance that influences the building load and production of new and renewable energy is crucial in the predictive control of buildings. However, most solar irradiance prediction models are developed using weather data measured over a long term in the target region. This study proposed a solar irradiance prediction model based on a hybrid layer that combined a long short-term memory layer and a bi-directional long short-term memory layer. Next, the solar irradiance of the following day was predicted using only the hourly irradiance measured on site one day before the prediction day using a model trained with the weather data of another region. The proposed solar irradiance prediction model showed an RMSE of 69.5 W/m2, a level suitable for the predictive control of buildings. Moreover, although training data from a different region was used, the performance of the proposed model was similar to that of a model trained with long-term data measured in the target region.

Problem: Most published decomposition‐hybrid energy forecasters—a variational mode decomposition (VMD) or empirical‐mode‐family front end followed by a learned predictor—are evaluated by decomposing the entire series before the train/test split. Because these decompositions are global, noncausal transforms, the features describing any test‐period time step then encode that step’s own future, and the reported accuracy is inflated by an unknown amount. Objective: We measure the size of this evaluation artifact under a controlled protocol and attribute it to its cause. Methods: Feature regimes that differ only in the provenance of the decomposition—no decomposition; full‐series decomposition (the prevailing protocol); split‐then‐decompose; rolling causal windows of 30–365 days; and an expanding causal window that keeps the full history but is denied postorigin samples—predict the same targets with the same models on four hourly series (German load, solar, and wind generation, plus the ETTh1 benchmark), using three model families (ridge, LightGBM, and a multilayer perceptron), 10 random seeds, Diebold–Mariano (DM), Wilcoxon, and Friedman significance tests, and dimensionality‐ and capacity‐matched controls. Results: Full‐series decomposition appears to reduce day‐ahead mean absolute error (MAE) by up to 94%, yet it understates the error of the identical causal pipeline by 30%–93% in the 11 of 12 dataset–model combinations where the leak takes hold (mean 60% across all 12; significant in 11 of 12). The expanding‐window regime retains on average 91% of that gap (88%–104% in 11 of 12 combinations), so the artifact is attributable to look‐ahead itself, not to the longer history the full‐series protocol enjoys; the popular split‐then‐decompose correction leaves 77%–101% of the artifact in place; no causal window length from 30 days to the full record recovers the leaky advantage; and the dimensionality‐ and capacity‐matched controls show none of these conclusions is driven by feature count or network size. Honestly decomposed pipelines are no better than the undecomposed baseline in 10 of 12 combinations. The leaky advantage persists essentially undiminished across the full 24‐h horizon, giving a one‐figure diagnostic for detecting look‐ahead, and the same pattern reappears at smaller magnitude when empirical mode decomposition (EMD) replaces VMD. Conclusion: The accuracy advantage that the prevailing protocol reports is, to first order, an artifact of seeing the future; a five‐point checklist for leak‐free evaluation of decomposition‐based forecasters is provided.