Skip to main content
Have a personal or library account? Click to login
Multi-Horizon Machine Learning Model for Home Energy Consumption Prediction Considering Appliances Usage Cover

Multi-Horizon Machine Learning Model for Home Energy Consumption Prediction Considering Appliances Usage

International Journal on Smart Sensing and Intelligent Systems
Volume 19 (2026): Issue 1 (January 2026)
By: ,   and    
Open Access
|Jul 2026
Figures & tables

Figures & Tables

Figure 1:

Architecture of Scheme-1. GRU, gated recurrent unit; LGBM, light gradient boosting machine; LSTM, long short-term memory; RF, random forest; RNN, recurrent neural network; XGB, extreme gradient boosting.

Figure 2:

Architecture of Scheme-2. GRU, gated recurrent unit; LGBM, light gradient boosting machine; LSTM, long short-term memory; RF, random forest; RNN, recurrent neural network; XGB, extreme gradient boosting.

Figure 3:

Architecture of Scheme-3. GRU, gated recurrent unit; LGBM, light gradient boosting machine; LSTM, long short-term memory; RF, random forest; RNN, recurrent neural network; XGB, extreme gradient boosting.

Figure 4:

Architecture of Scheme-4. GRU, gated recurrent unit; LGBM, light gradient boosting machine; LSTM, long short-term memory; RF, random forest; RNN, recurrent neural network; XGB, extreme gradient boosting.

Figure 5:

Architecture of Scheme-5. GRU, gated recurrent unit; LGBM, light gradient boosting machine; LSTM, long short-term memory; RF, random forest; RNN, recurrent neural network; XGB, extreme gradient boosting.

Figure 6:

RNN cell. RNN, recurrent neural network.

Figure 7:

LSTM cell. LSTM, long short-term memory.

Figure 8:

GRU cell. GRU, gated recurrent unit.

Table 1:

Tree-based machine learning parameters

RFLGBMXGB
ParameterValueParameterValueParameterValue
No. of estimators200Boosting type“gbdt”Booster“gbtree”
Criterion“Squared error”No. of leaves31Min. loss reduction (gamma)0
Max. depthNoneMax. depth−1Max. depth6
Min. samples split2Learning rate0.05Learning rate0.05
Min. samples leaf1No. of estimators500No. of estimators500
Max. features1.0Objective“Regression”Objective“reg:squarederror”
Max. leaf nodesNoneMin. split gain0Min child weight1
BootstrapTrueSubsample ratio0.8Subsample ratio0.8
Subsample ratio of columns0.8Subsample ratio of columns0.8
Regularization alpha0Regularization alpha0
Regularization lambda0Regularization lambda1

[i] LGBM, light gradient boosting machine; RF, random forest; XGB, extreme gradient boosting.

Table 2:

Deep learning-based machine learning parameters

ParameterValue
RNNLSTMGRU
No. of units646464
Activation“tanh”“tanh”“tanh”
Recurrent activation-“Sigmoid”“Sigmoid”
Use biasTrueTrueTrue
Kernel initializer“glorot uniform”“glorot uniform”“glorot uniform”
Recurrent initializer“Orthogonal”“Orthogonal”“Orthogonal”
Bias initializer“Zeros”“Zeros”“Zeros”
Input shape(WINDOW,1)(WINDOW,1)(WINDOW,1)
Dropout0.20.20.2
Dense units111

[i] GRU, gated recurrent unit; LSTM, long short-term memory; RNN, recurrent neural network.

Figure 9:

NMAE of 1-min ahead individual appliance power prediction. GRU, gated recurrent unit; LGBM, light gradient boosting machine; LSTM, long short-term memory; MO, microwave oven; NMAE, normalized mean absolute error; RF, random forest; RNN, recurrent neural network; WP, Water pump; XGB, extreme gradient boosting; FZ, freezer; LG, lighting; RC, rice cooker; RFG, refrigerator; WH, water heater; WM, washing machine.

Figure 10:

NMAE of 15-min ahead individual appliance power prediction. GRU, gated recurrent unit; LGBM, light gradient boosting machine; LSTM, long short-term memory; MO, microwave oven; NMAE, normalized mean absolute error; RF, random forest; RNN, recurrent neural network; WP, Water pump; XGB, extreme gradient boosting; FZ, freezer; LG, lighting; RC, rice cooker; RFG, refrigerator; WH, water heater; WM, washing machine.

Figure 11:

NMAE of 1-hr ahead individual appliance power prediction. GRU, gated recurrent unit; LGBM, light gradient boosting machine; LSTM, long short-term memory; MO, microwave oven; NMAE, normalized mean absolute error; RF, random forest; RNN, recurrent neural network; WP, Water pump; XGB, extreme gradient boosting; FZ, freezer; LG, lighting; RC, rice cooker; RFG, refrigerator; WH, water heater; WM, washing machine.

Figure 12:

NMAE of 1-day ahead individual appliance power prediction. GRU, gated recurrent unit; LGBM, light gradient boosting machine; LSTM, long short-term memory; MO, microwave oven; NMAE, normalized mean absolute error; RF, random forest; RNN, recurrent neural network; WP, Water pump; XGB, extreme gradient boosting; FZ, freezer; LG, lighting; RC, rice cooker; RFG, refrigerator; WH, water heater; WM, washing machine.

Table 3:

NMAE comparison of individual appliance power prediction

Time horizonLowest NMAE (%) (appliance, method)
Reschedulable loadNon-reschedulable load
Tree-based algorithmDeep-learning algorithmTree-based algorithmDeep-learning algorithm
1-min0.14 (MO, RF)0.10 (MO, RNN)0.26 (WP, RF)0.21 (WP, LSTM)
15-min0.16 (MO, RF)0.09 (MO, RNN)1.77 (FZ, XGB)1.39 (WP, RNN)
1-hr0.18 (MO, XGB)0.18 (MO, GRU)2.00 (FZ, XGB)1.81 (WP, GRU)
1-day0.20 (MO, RF/LGBM)0.26 (MO, LSTM)0.35 (LG, LGBM)1.80 (WP, GRU)

[i] GRU, gated recurrent unit; LGBM, light gradient boosting machine; LSTM, long short-term memory; MO, microwave oven; NMAE, normalized mean absolute error; RF, random forest; RNN, recurrent neural network; XGB, extreme gradient boosting; FZ, freezer; LG, lighting; WP, water pump.

Figure 13:

NRMSE of 1-min ahead individual appliance power prediction. GRU, gated recurrent unit; LGBM, light gradient boosting machine; LSTM, long short-term memory; MO, microwave oven; NRMSE, normalized root mean square error; RF, random forest; RNN, recurrent neural network; WP, Water pump; XGB, extreme gradient boosting; FZ, freezer; LG, lighting; RC, rice cooker; RFG, refrigerator; WH, water heater; WM, washing machine.

Figure 14:

NRMSE of 15-min ahead prediction. GRU, gated recurrent unit; LGBM, light gradient boosting machine; LSTM, long short-term memory; MO, microwave oven; NRMSE, normalized root mean square error; RF, random forest; RNN, recurrent neural network; WP, Water pump; XGB, extreme gradient boosting; FZ, freezer; LG, lighting; RC, rice cooker; RFG, refrigerator; WH, water heater; WM, washing machine.

Figure 15:

NRMSE of 1-hr ahead individual appliance power prediction. GRU, gated recurrent unit; LGBM, light gradient boosting machine; LSTM, long short-term memory; MO, microwave oven; NRMSE, normalized root mean square error; RF, random forest; RNN, recurrent neural network; WP, Water pump; XGB, extreme gradient boosting; FZ, freezer; LG, lighting; RC, rice cooker; RFG, refrigerator; WH, water heater; WM, washing machine.

Figure 16:

NRMSE of 1-day ahead individual appliance power prediction. GRU, gated recurrent unit; LGBM, light gradient boosting machine; LSTM, long short-term memory; MO, microwave oven; NRMSE, normalized root mean square error; RF, random forest; RNN, recurrent neural network; WP, Water pump; XGB, extreme gradient boosting; FZ, freezer; LG, lighting; RC, rice cooker; RFG, refrigerator; WH, water heater; WM, washing machine.

Table 4:

NRMSE comparison of individual appliance power prediction

Time horizonLowest NRMSE (%) (appliance, method)
Reschedulable loadNon-reschedulable load
Tree-based algorithmDeep-learning algorithmTree-based algorithmDeep-learning algorithm
1-min2.63 (MO, LGBM)2.45 (MO, RNN)3.33 (WP, LGBM)3.14 (RFG, GRU)
15-min2.55 (MO, LGBM)2.52 (MO, RNN/LSTM)4.13 (FZ, LGBM/XGB)4.37 (FZ, LSTM)
1-hr2.56 (MO, LGBM/XGB)2.56 (MO, GRU)4.81 (RFG, LGBM)5.34 (RFG/FZ, GRU)
1-day2.65 (MO, LGBM)2.66 (MO, GRU)5.30 (LG, RF)5.90 (RFG, LSTM/GRU)

[i] GRU, gated recurrent unit; LGBM, light gradient boosting machine; LSTM, long short-term memory; MO, microwave oven; NRMSE, normalized root mean square error; RF, random forest; RNN, recurrent neural network; XGB, extreme gradient boosting; FZ, freezer; LG, lighting; RFG, refrigerator; WP, water pump.

Figure 17:

NMAE of multi-horizon average appliance power prediction. GRU, gated recurrent unit; LGBM, light gradient boosting machine; LSTM, long short-term memory; NMAE, normalized mean absolute error; RF, random forest; RNN, recurrent neural network; XGB, extreme gradient boosting.

Figure 18:

NRMSE of multi-horizon average appliance power prediction. GRU, gated recurrent unit; LGBM, light gradient boosting machine; LSTM, long short-term memory; NRMSE, normalized root mean square error; RF, random forest; RNN, recurrent neural network; XGB, extreme gradient boosting.

Figure 19:

NMAE of 1-min ahead total power prediction. GRU, gated recurrent unit; LGBM, light gradient boosting machine; LSTM, long short-term memory; NMAE, normalized mean absolute error; RF, random forest; RNN, recurrent neural network; XGB, extreme gradient boosting.

Figure 20:

NMAE of 15-min ahead total power prediction. GRU, gated recurrent unit; LGBM, light gradient boosting machine; LSTM, long short-term memory; NMAE, normalized mean absolute error; RF, random forest; RNN, recurrent neural network; XGB, extreme gradient boosting.

Figure 21:

NMAE of 1-hr ahead total power prediction. GRU, gated recurrent unit; LGBM, light gradient boosting machine; LSTM, long short-term memory; NMAE, normalized mean absolute error; RF, random forest; RNN, recurrent neural network; XGB, extreme gradient boosting.

Figure 22:

NMAE of 1-day ahead total power prediction. GRU, gated recurrent unit; LGBM, light gradient boosting machine; LSTM, long short-term memory; NMAE, normalized mean absolute error; RF, random forest; RNN, recurrent neural network; XGB, extreme gradient boosting.

Table 5:

NMAE comparison of total power prediction

Time horizonLowest NMAE (%) (feature, method)
Tree-based algorithmDeep learning algorithm
1-min2.03 (C, LGBM/XGB)2.53 (C, LSTM/GRU)
15-min5.02 (C, LGBM)5.00 (B, GRU)
1-hr5.77 (C, XGB)6.09 (C, LSTM)
1-day7.26 (C, RF)8.08 (A, LSTM)

[i] GRU, gated recurrent unit; LGBM, light gradient boosting machine; LSTM, long short-term memory; NMAE, normalized mean absolute error; RF, random forest; RNN, recurrent neural network; XGB, extreme gradient boosting.

Figure 23:

NRMSE of 1-min ahead total power prediction. GRU, gated recurrent unit; LGBM, light gradient boosting machine; LSTM, long short-term memory; NRMSE, normalized root mean square error; RF, random forest; RNN, recurrent neural network; XGB, extreme gradient boosting.

Figure 24:

NRMSE of 15-min ahead total power prediction. GRU, gated recurrent unit; LGBM, light gradient boosting machine; LSTM, long short-term memory; NRMSE, normalized root mean square error; RF, random forest; RNN, recurrent neural network; XGB, extreme gradient boosting.

Figure 25:

NRMSE of 1-hr ahead total power prediction. GRU, gated recurrent unit; LGBM, light gradient boosting machine; LSTM, long short-term memory; NRMSE, normalized root mean square error; RF, random forest; RNN, recurrent neural network; XGB, extreme gradient boosting.

Figure 26:

NRMSE of 1-day ahead total power prediction. GRU, gated recurrent unit; LGBM, light gradient boosting machine; LSTM, long short-term memory; NRMSE, normalized root mean square error; RF, random forest; RNN, recurrent neural network; XGB, extreme gradient boosting.

Table 6:

NRMSE comparison of total power prediction

Time horizonLowest NRMSE (%) (feature, method)
Tree-based algorithmDeep-learning algorithm
1-min4.61 (C, LGBM)4.91 (C, GRU)
15-min8.15 (C, LGBM)8.27 (C, LSTM)
1-hr9.44 (C, XGB)9.75 (B, RNN)
1-day10.66 (B/D, RF)11.75 (A, RNN)

[i] GRU, gated recurrent unit; LGBM, light gradient boosting machine; LSTM, long short-term memory; NRMSE, normalized root mean square error; RF, random forest; RNN, recurrent neural network; XGB, extreme gradient boosting.

Figure 27:

NMAE of multi-horizon total power prediction. NMAE, normalized mean absolute error.

Figure 28:

NRMSE of multi-horizon total power prediction. NRMSE, normalized root mean square error.

DOI: https://doi.org/10.2478/ijssis-2026-0038 | Journal eISSN: 1178-5608
Journal RSS Feed
Language: English
Submitted on: Dec 6, 2025
Published on: Jul 11, 2026
Published by: International Journal on Smart Sensing and Intelligent Systems
In partnership with: Paradigm Publishing Services
Publication frequency: 1 issue per year
Keywords:
multi-horizon,
machine learning,
energy consumption,
prediction,
home appliance
Related subjects:
Engineering,
Introductions and overviews,
Engineering, other

© 2026 Irrine Budi Sulistiawati, Aryuanto Soetedjo, Irmalia Suryani Faradisa, published by International Journal on Smart Sensing and Intelligent Systems
This work is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License.

Volume 19 (2026): Issue 1 (January 2026)
Previous article
Next article