Novel Deep Regression and Stump Tree-Based Ensemble Models for Real-Time Load Demand Planning and Management

Abstract

The electrical power companies' load consists of several kinds of customers, for example, agricultural, building, industrial, etc. Agricultural and industrial energy consumption is based on highly inductive loads, and it leads to the electricity shutdown because of highly inductive loads effect large spikes to the power curve. These spikes show their irregular disturbance in the electric power system due to shutdown and start-up of these large loads. It is considered irregular as well as a tough job to forecast the load with the existence of such kinds of spikes. Further, the heterogeneous and multi-source load consumption characteristics necessitate the integration of evolutionary models and the ensemble techniques render better robustness and forecast accuracy. For this purpose, we proposed the novel deep regression and stump tree-based ensembles models (DRTSEM) for accurate load planning and management with the use of highly inductive load data, buildings loads, etc. The hyperparameters e.g., network weight decay, network learning rate, the total number of hidden layers and momentum of the DRTSEM is optimized by using the deep regression and stump tree-based algorithms. For forecasting analysis, two different climate zones and seven model's input model parameter selection (MPS) with new sets of the data samples are used. The forecasting performance is testified with the generalization network error and resubstituting loss, 10fold cross-validated mean square error, four performance evaluation statistics and existing tree bagger (TB) model. The mean absolute percentage error is noted 47%, 64% and 41% better than the TB model at MPS-3 to MPS-5 respectively. The proposed DRTSEM can efficiently tackle the nonlinear characteristics of the heterogeneous and multi-source loads and moreover increase accuracy and reduce the error between actual and forecasted load demand. The modeling results proved that the use of DRTSEM renders more accurate utility energy planning and management compared with the existing ensemble approaches. INDEX TERMS Novel DRTSEM model, building energy yield planning, utility energy management, multiple feature selection, deep regression, medium & long-term forecasting. NOMENCLATURE ANNs Artificial neural networks ARMA Auto regressive moving average model CDD Cooling degree-days CV Coefficient of variation CZ Climate Zone DRTSEM Deep regression and stump tree-based ensembles models DS Decision stumps ENE East-northeast The associate editor coordinating the review of this manuscript and approving it for publication was Mauro Tucci .

Towards developing a systematic knowledge trend for building energy consumption prediction

“…Regression tree [9], [33], [124], [199] Boosted tree [33], [41] [124] Easy to interpret and nonparametric  Tends to over-fit  May get stuck in local minima  Not an online learning Ensemble algorithms: Boosting [4], [33], [38], [41], [48], [90], [125], [139] Bootstrapped aggregation [21], AdaBoost [48], [196], Stacked generalization [41],…”

“…Ensemble methods [28] are combinations of multiple learning algorithms, whereby the individual algorithms can be trained separately and their predictions combined in ways that depict an overall prediction. The main purpose of ensemble algorithms is to determine which of the weaker models can be combined as well as the best means by which such combinations can be performed [47], [48]. Random forest, bootstrapped aggregation, gradient boosting machines, boosting, stacked generalization and adaBoost [49] are some of the most notable examples of ensemble algorithms.…”

Comparison of Tree-Based Ensemble Algorithms for Merging Satellite and Earth-Observed Precipitation Data at the Daily Time Scale

“…Instead, such a selection could rely on attentively designed empirical comparisons. Thus, such comparisons of tree-based ensemble algorithms are conducted with increasing frequency in various scientific fields ( [30][31][32][33][34][35]).…”