# Accurate Deep Model for Electricity Consumption Forecasting Using Multi-Channel and Multi-Scale Feature Fusion CNN–LSTM

^{*}

## Abstract

**:**

## 1. Introduction

- To the best of our understanding, a few types of research focused on using one model for VSTF, STF, MTF, and LTF. This paper addresses this issue with MCSCNN–LSTM.
- The hybrid deep model MCSCNN–LSTM was designed, trained, and validated. The MCSCNN–LSTM obtains the highest performance compared to the current state-of-the-art methods.
- The proposed method can accurately forecast electricity consumption by inputting the self-history data without any additional data and any handcrafted feature selection operation. Therefore, it reduces the cost of data collection while simultaneously keeping high accuracy.
- The feature extraction capacity of each part has been analyzed.
- The excellent transfer learning and multi-step forecasting capacities of the proposed MCSCNN–LSTM have been proven.

## 2. Problem Formulations

- VSTF: Hourly forecasting, power consumption data of previous $H$ hours are employed for next-hour power consumption forecasting.
- STF: Daily forecasting, applying power consumption data of previous $D$ days to get the next day’s power consumption.
- MTF: Weekly forecasting, using power consumption data of previous $W$ weeks to forecast power consumption of the next week.
- LTF: Monthly forecasting, the power consumption data of previous $M$ months are employed to get the next one month.

## 3. Methods

#### 3.1. CNN

#### 3.2. LSTM

#### 3.3. Statistical Components

## 4. Proposed Deep Model

#### 4.1. Input

#### 4.2. CNN Feature Extraction

#### 4.3. LSTM Feature Extraction

#### 4.4. Feature Fusion

#### 4.5. Output

#### 4.6. Updating the Networks

## 5. Experiment Verification

#### 5.1. Dataset Introduction

Algorithm 1: Overlapping sample algorithm |

Input: Hourly electricity consumption historical time series $hourly$ Output: Daily, weekly, and monthly electricity consumption historical time series samples and labels. Define the length of samples $D,\text{}W,\text{}M$ as 24. Step 1: Integrating the original data for different forecasts $daily<-sum\left(hourly,\text{}24\right)$ #adopt the sample rate of 24 h for STF $weekly<-sum\left(hourly,168\right)$ #adopt the sample rate of 168 h for MTF $monthly<-sum\left(hourly,720\right)$ #adopt the sample rate of 720 h for LTF Step 2: Generating the feature and label of each sample corresponding to the (2) and (3) For $i$ in range ($length\left(daily/weekly/monthly\right)$): # different forecasts have different contents $dail{y}_{features}<-daily\left[i:i+D\right]$ $dail{y}_{lables}<-daily\left[i+D+1\right]$ $weekl{y}_{features}<-weekly\left[i:i+W\right]$ $weekl{y}_{labels}<-weekly\left[i+W+1\right]$ $monthl{y}_{features}<-monthly\left[i:i+M\right]$ $monthl{y}_{labels}<-monthly\left[i+M+1\right]$ End for Return $dail{y}_{features}$, $dail{y}_{labels}$, $weekl{y}_{features}$, $weekl{y}_{labels}$, $monthl{y}_{features}$, $monthl{y}_{labels}$ |

#### 5.2. Evaluation Metrics

#### 5.3. Performance Comparison with Other Excellent Methods

#### 5.4. Feature Extraction Capacity of MCSCNN–LSTM

#### 5.5. Transfer Learning Capacity Test

#### 5.6. Multi-Step Forecasting Capacity Test

## 6. Discussion

## 7. Conclusions

## Author Contributions

## Funding

## Conflicts of Interest

## References

- Ahmad, A.S.; Hassan, M.Y.; Abdullah, M.P.; Rahman, H.A.; Hussin, F.; Abdullah, H.; Saidur, R. A review on applications of ANN and SVM for building electrical energy consumption forecasting. Renew. Sustain. Energy Rev.
**2014**, 33, 102–109. [Google Scholar] [CrossRef] - Ayoub, N.; Musharavati, F.; Pokharel, S.; Gabbar, H.A. ANN Model for Energy Demand and Supply Forecasting in a Hybrid Energy Supply System. In Proceedings of the 2018 IEEE International Conference on Smart Energy Grid Engineering (SEGE), Oshawa, ON, Canada, 12–15 August 2018. [Google Scholar]
- Weng, Y.; Wang, X.; Hua, J.; Wang, H.; Kang, M.; Wang, F.Y. Forecasting Horticultural Products Price Using ARIMA Model and Neural Network Based on a Large-Scale Data Set Collected by Web Crawler. IEEE Trans. Comput. Soc. Syst.
**2019**, 6, 547–553. [Google Scholar] [CrossRef] - Tang, N.; Mao, S.; Wang, Y.; Nelms, R.M. Solar Power Generation Forecasting With a LASSO-Based Approach. IEEE Internet Things J.
**2018**, 5, 1090–1099. [Google Scholar] [CrossRef] - Yu, J.; Lee, H.; Jeong, Y.; Kim, S. Short-term hourly load forecasting using PSO-based AR model. In Proceedings of the 2014 Joint 7th International Conference on Soft Computing and Intelligent Systems (SCIS) and 15th International Symposium on Advanced Intelligent Systems (ISIS), Kitakyushu, Japan, 3–6 December 2014; pp. 1449–1453. [Google Scholar]
- Wu, H.; Wu, H.; Zhu, M.; Chen, W.; Chen, W. A new method of large-scale short-term forecasting of agricultural commodity prices: Illustrated by the case of agricultural markets in Beijing. J. Big Data
**2017**, 4, 1–22. [Google Scholar] [CrossRef][Green Version] - Radziukynas, V.; Klementavicius, A. Short-term wind speed forecasting with ARIMA model. In Proceedings of the 2014 55th International Scientific Conference on Power and Electrical Engineering of Riga Technical University (RTUCON), Riga, Latvia, 14 October 2014; pp. 145–149. [Google Scholar]
- He, H.; Liu, T.; Chen, R.; Xiao, Y.; Yang, J. High frequency short-term demand forecasting model for distribution power grid based on ARIMA. In Proceedings of the 2012 IEEE International Conference on Computer Science and Automation Engineering (CSAE), Zhangjiajie, China, 25–27 May 2012; pp. 293–297. [Google Scholar]
- Mitkov, A.; Noorzad, N.; Gabrovska-Evstatieva, K.; Mihailov, N. Forecasting the energy consumption in Afghanistan with the ARIMA model. In Proceedings of the 2019 16th Conference on Electrical Machines, Drives and Power Systems (ELMA), Varna, Bulgaria, 6–8 June 2019; pp. 1–4. [Google Scholar]
- Jha, G.K.; Sinha, K. Agricultural Price Forecasting Using Neural Network Model: An Innovative Information Delivery System. Agric. Econ. Res. Rev.
**2013**, 26, 229–239. [Google Scholar] [CrossRef][Green Version] - Lecun, Y.; Bengio, Y.; Hinton, G. Deep learning. Nature
**2015**, 521, 436–444. [Google Scholar] [CrossRef] - Li, L.; Ota, K.; Dong, M. Everything is image: CNN-based short-term electrical load forecasting for smart grid. In Proceedings of the 2017 14th International Symposium on Pervasive Systems, Algorithms and Networks & 2017 11th International Conference on Frontier of Computer Science and Technology & 2017 Third International Symposium of Creative Computing (ISPAN-FCST-ISCC), Exeter, UK, 21–23 June 2017; pp. 344–351. [Google Scholar]
- Deng, Z.; Wang, B.; Xu, Y.; Xu, T.; Liu, C.; Zhu, Z. Multi-scale convolutional neural network with time-cognition for multi-step short-Term load forecasting. IEEE Access
**2019**, 7, 88058–88071. [Google Scholar] [CrossRef] - Suresh, V.; Janik, P.; Rezmer, J.; Leonowicz, Z. Forecasting solar PV output using convolutional neural networks with a sliding window algorithm. Energies
**2020**, 13, 723. [Google Scholar] [CrossRef][Green Version] - Kim, D.; Hwang, S.W.; Kim, J. Very Short-Term Photovoltaic Power Generation Forecasting with Convolutional Neural Networks. In Proceedings of the 9th International Conference on Information and Communication Technology Convergence (ICTC), Jeju, Koera, 17–18 October 2018; pp. 1310–1312. [Google Scholar]
- Hochreiter, S.; Urgen Schmidhuber, J. Long Shortterm Memory. Neural Comput.
**1997**, 9, 17351780. [Google Scholar] [CrossRef] - Ma, Y.; Zhang, Q.; Ding, J.; Wang, Q.; Ma, J. Short Term Load Forecasting Based on iForest-LSTM. In Proceedings of the 2019 14th IEEE Conference on Industrial Electronics and Applications (ICIEA), Xi’an, China, 19–21 June 2019; pp. 2278–2282. [Google Scholar]
- Xiuyun, G.; Ying, W.; Yang, G.; Chengzhi, S.; Wen, X.; Yimiao, Y. Short-term Load Forecasting Model ofGRU Network Based on Deep Learning Framework. In Proceedings of the 2018 2nd IEEE Conference on Energy Internet and Energy System Integration (EI2), Beijing, China, 20–22 October 2018; pp. 1–4. [Google Scholar]
- Yu, Y.; Cao, J.; Zhu, J. An LSTM Short-Term Solar Irradiance Forecasting under Complicated Weather Conditions. IEEE Access
**2019**, 7, 145651–145666. [Google Scholar] [CrossRef] - Kong, W.; Dong, Z.Y.; Jia, Y.; Hill, D.J.; Xu, Y.; Zhang, Y. Short-Term Residential Load Forecasting Based on LSTM Recurrent Neural Network. IEEE Trans. Smart Grid
**2019**, 10, 841–851. [Google Scholar] [CrossRef] - Agrawal, R.K.; Muchahary, F.; Tripathi, M.M. Long term load forecasting with hourly predictions based on long-short-term-memory networks. In Proceedings of the 2018 IEEE Texas Power and Energy Conference (TPEC), College Station, TX, USA, 8–9 February 2018; pp. 1–6. [Google Scholar]
- Han, L.; Peng, Y.; Li, Y.; Yong, B.; Zhou, Q.; Shu, L. Enhanced deep networks for short-term and medium-term load forecasting. IEEE Access
**2019**, 7, 4045–4055. [Google Scholar] [CrossRef] - Zhou, H.; Zhang, Y.; Yang, L.; Liu, Q.; Yan, K.; Du, Y. Short-Term photovoltaic power forecasting based on long short term memory neural network and attention mechanism. IEEE Access
**2019**, 7, 78063–78074. [Google Scholar] [CrossRef] - Wang, Z.; Lou, Y. Hydrological time series forecast model based on wavelet de-noising and ARIMA-LSTM. In Proceedings of the 2019 IEEE 3rd Information Technology, Networking, Electronic and Automation Control Conference (ITNEC 2019), Chengdu, China, 15–17 March 2019; pp. 1697–1701. [Google Scholar]
- Kim, T.Y.; Cho, S.B. Predicting residential energy consumption using CNN-LSTM neural networks. Energy
**2019**, 182, 72–81. [Google Scholar] [CrossRef] - Hu, P.; Tong, J.; Wang, J.; Yang, Y.; Oliveira Turci, L. De A hybrid model based on CNN and Bi-LSTM for urban water demand prediction. In Proceedings of the 2019 IEEE Congress on Evolutionary Computation (CEC), Wellington, New Zealand, 10–13 June 2019; pp. 1088–1094. [Google Scholar]
- Yan, K.; Wang, X.; Du, Y.; Jin, N.; Huang, H.; Zhou, H. Multi-step short-term power consumption forecasting with a hybrid deep learning strategy. Energies
**2018**, 11, 3089. [Google Scholar] [CrossRef][Green Version] - Yan, K.; Li, W.; Ji, Z.; Qi, M.; Du, Y. A Hybrid LSTM Neural Network for Energy Consumption Forecasting of Individual Households. IEEE Access
**2019**, 7, 157633–157642. [Google Scholar] [CrossRef] - Peng, H.K.; Marculescu, R. Multi-scale compositionality: Identifying the compositional structures of Social dynamics using deep learning. PLoS ONE
**2015**, 10, 1–28. [Google Scholar] [CrossRef] - Chen, Z.; Li, C.; Sanchez, R. Gearbox Fault Identification and Classification with Convolutional Neural Networks. Shock Vib.
**2015**, 2015, 390134. [Google Scholar] [CrossRef][Green Version] - Wang, K.; Qi, X.; Liu, H. A comparison of day-ahead photovoltaic power forecasting models based on deep learning neural network. Appl. Energy
**2019**, 251, 113315. [Google Scholar] [CrossRef] - Gupta, J.N.D.; Sexton, R.S. Comparing backpropagation with a genetic algorithm for neural network training. Omega
**1999**, 27, 679–684. [Google Scholar] [CrossRef] - Kingma, D.P.; Ba, J.L. Adam: A method for stochastic optimization. In Proceedings of the 3rd International Conference on Learning Representations, ICLR 2015—Conference Track Proceedings, San Diego, CA, USA, 7–9 May 2015; pp. 1–15. [Google Scholar]
- Hu, Q.; Zhang, R.; Zhou, Y. Transfer learning for short-term wind speed prediction with deep neural networks. Renew. Energy
**2016**, 85, 83–95. [Google Scholar] [CrossRef] - Liu, S.; Ji, H.; Wang, M.C. Nonpooling Convolutional Neural Network Forecasting for Seasonal Time Series With Trends. IEEE Trans. Neural Netw. Learn. Syst.
**2019**, 1–10. [Google Scholar] [CrossRef] [PubMed]

**Figure 2.**The architecture of the proposed multi-channels and scales convolutional neural networks (MCSCNN)–LSTM at three levels.

**Figure 3.**The electricity consumption at different durations. (

**a**) Hourly electricity consumption for VSTF. (

**b**) Daily electricity consumption for STF. (

**c**) Weekly electricity consumption for MTF. (

**d**) Monthly electricity consumption for LTF.

**Figure 4.**The averaged MAPE improvement results for each duration forecast compared to other excellent deep learning-based methods on three data sets. Each data set contributes the same to the final outputs.

**Figure 5.**The comparative forecasting results using two different deep learning-based methods. The x-axis is the time stamp at different duration; the y-axis is electricity consumption. (

**a**) VSTF (hourly) forecasting results. (

**b**) STF (daily) forecasting results. (

**c**) MTF (weekly) forecasting results. (

**d**) LTF (monthly) forecasting results.

**Figure 6.**The averaged improvement of the proposed deep model for each duration forecast based on MSCNN on three data sets.

**Figure 7.**Feature maps visualization. Each part of the proposed deep model extracted different features. (

**a**) The raw sample channel. (

**b**) The statistic components channel. (

**c**) CNN-learned feature map, which almost has no changes around 0. (

**d**) LSTM-learned feature map, which ranges from –0.10 to 0.075. (

**e**) Statistic components feature map of a reshaped tensor. The raw statistic components channel was reshaped into [1,6], which ranges from 0.00 to 1.75. (

**f**) Reshaped comprehensive feature map. The shape of the obtained feature is 1 by 66, and we reshaped it into 11 by 6 to clearly see and analyze.

**Figure 8.**The average MAPE of different methods to validate the transfer learning capacity of the proposed MCSCNN–LSTM for different forecasts. (

**a**) Transfer learning capacity test for VSTF. (

**b**) Transfer learning capacity test for STF. (

**c**) Transfer learning capacity test for MTF. (

**d**) Transfer learning capacity test for LTF.

**Figure 9.**A comparison of the results of the five-step forecasting using the proposed method and CNN–LSTM [27]. The results indicate the proposed method has an absolute advantage in terms of STF, MTF, and LTF. For VSTF, CNN–LSTM performs a little better than proposed MCSCNN–LSTM. (

**a**) Five-step electricity forecasting results for VSTF. (

**b**) Five-step electricity forecasting results for STF. (

**c**) Five-step electricity forecasting results for MTF. (

**d**) Five-step electricity forecasting results for LTF.

Forecasts Types | Length of Input History Data | Outputs |
---|---|---|

VSTF (hourly) | Previous $H$ hours | Next one hour |

STF (daily) | Previous $D$ days | Next one day |

MTF (weekly) | Previous $W$ weeks | Next one week |

LTF (monthly) | Previous $M$ months | Next one month |

Layer | Output Shape | Connected To | Parameters |
---|---|---|---|

Input1 (Raw) | (None, 24, 1) | − | 0 |

Input2 (Statistic) | (None, 6, 1) | − | 0 |

Conv1_1 | (None, 12, 16) | Input1 | 48 |

Conv1_2 | (None, 8, 16) | Input1 | 64 |

Conv1_3 | (None, 6, 16) | Input1 | 80 |

Conv2_1 | (None, 12, 16) | Conv1_1 | 528 |

Conv2_2 | (None, 8, 16) | Conv1_2 | 528 |

Conv2_3 | (None, 6, 16) | Conv1_3 | 528 |

Concatenate_1 | (None, 26, 16) | Conv2_1, Conv2_2, Conv2_3 | 0 |

Static_Conv | (None, 11, 10) | Concatenate_1 | 2570 |

Global_Maxpooling | (None, 5, 10) | Static_Conv | 0 |

Flatten_1 | (None, 50) | Global_maxpooling | 0 |

LSTM_1 | (None, 24, 20) | Input2 | 1760 |

LSTM_2 | (None, 10) | LSTM_1 | 1240 |

Flatten_2 | (None, 6) | Input2 | 0 |

Concatenate_2 | (None, 66) | LSTM_2 Flatten_1 Flatten_2 | 0 |

Dense (Output) | (None, 1) | Concatenate_2 | 67 |

Dataset | Start Date | End Date | Length |
---|---|---|---|

AEP | 2004-12-31 01:00:00 | 2018-01-02 00:00:00 | 121,273 |

COMED | 2011-12-31 01:00:00 | 2018-01-02 00:00:00 | 66,497 |

DAYTON | 2004-12-31 01:00:00 | 2018-01-02 00:00:00 | 121,275 |

Forecasts | Dataset | Samples |
---|---|---|

VSTF (Hourly) | AEP | 121,249 |

COMED | 66,473 | |

DAYTON | 121,251 | |

STF (Daily) | AEP | 121,225 |

COMED | 66,449 | |

DAYTON | 121,227 | |

MTF (Weekly) | AEP | 121,081 |

COMED | 66,305 | |

DAYTON | 121,083 | |

LTF (Monthly) | AEP | 120,529 |

COMED | 65,753 | |

DAYTON | 120,531 |

Method | Structure# Layer (Neurons) | Activation Function |
---|---|---|

[34] DNN | Input-Dense(24)-Dense(10)-Flatten-Output | Sigmoid |

[35] NPCNN | Input-Conv1D(5)-Max1D(2)-Flatten(Dense(1))-Dense(10)-Output | ReLu |

[20] LSTM | Input–LSTM(20)–LSTM(20)-Output | ReLu |

[25] CNN–LSTM | Input-Conv1D(64)-Max1D(2)-Conv1D(2)-Flatten(Max1D(2))–LSTM(64)-Dense(32)-Output | ReLu |

Dataset | Method | VSTF | STF | MTF | LTF |
---|---|---|---|---|---|

AEP | [34] DNN | None | None | None | None |

[35] NPCNN | 2 | 3 | 2 | 2 | |

[20] LSTM | 5 | 4 | 5 | 2 | |

[25] CNN–LSTM | 2 | 2 | 3 | 2 | |

Proposed | None | None | None | None | |

COMED | [34] DNN | None | None | None | None |

[35] NPCNN | 3 | 3 | 2 | 2 | |

[20] LSTM | 5 | 4 | 5 | 4 | |

[25] CNN–LSTM | 2 | 3 | 4 | 2 | |

Proposed | None | None | None | None | |

DAYTON | [34] DNN | None | None | None | None |

[35] NPCNN | 2 | 2 | 3 | 2 | |

[20] LSTM | 4 | 4 | 2 | 3 | |

[25] CNN–LSTM | 2 | 2 | 1 | 2 | |

Proposed | None | None | None | None |

Dataset | Method | RMSE (VSTF) | RMSE (STF) | RMSE (MTF) | RMSE(LTF) |
---|---|---|---|---|---|

AEP | [34] DNN | 389.79 | 756.15 | 2864.03 | 15,387.72 |

[35] NPCNN | 476.38 | 1866.71 | 4220.96 | 40,393.06 | |

[20] LSTM | 298.28 | 124.19 | 757.13 | 4876.94 | |

[25] CNN–LSTM | 374.39 | 711.02 | 2408.09 | 20,060.97 | |

Proposed | 294.03 | 424.14 | 665.29 | 3385.70 | |

COMED | [34] DNN | 310.69 | 765.16 | 3908.04 | 10,934.82 |

[35] NPCNN | 439.07 | 1090.17 | 6,274.38 | 14,900.91 | |

[20] LSTM | 251.47 | 426.46 | 2925.53 | 30,407.07 | |

[25] CNN–LSTM | 272.18 | 501.70 | 3082.33 | 4654.41 | |

Proposed | 240.51 | 377.74 | 520.02 | 3122.94 | |

DAYTON | [34] DNN | 61.49 | 112.43 | 311.37 | 1299.99 |

[35] NPCNN | 71.16 | 183.90 | 390.88 | 1399.25 | |

[20] LSTM | 43.84 | 142.49 | 107.87 | 444.95 | |

[25] CNN–LSTM | 47.08 | 109.42 | 175.67 | 502.58 | |

Proposed | 43.68 | 65.68 | 95.84 | 270.40 |

Dataset | Method | MAE (VSTF) | MAE (STF) | MAE (MTF) | MAE (LTF) |
---|---|---|---|---|---|

AEP | [34] DNN | 246.41 | 583.44 | 2052.89 | 10,384.48 |

[35] NPCNN | 332.21 | 1682.52 | 2506.76 | 16,803.83 | |

[20] LSTM | 198.67 | 995.27 | 613.45 | 3732.41 | |

[25] CNN–LSTM | 248.65 | 508.70 | 1705.03 | 11,723.84 | |

Proposed | 180.94 | 250.15 | 494.04 | 2788.86 | |

COMED | [34] DNN | 198.20 | 611.87 | 2951.25 | 8023.53 |

[35] NPCNN | 333.18 | 813.53 | 5427.71 | 11,953.35 | |

[20] LSTM | 156.24 | 316.02 | 2831.15 | 30,082.99 | |

[25] CNN–LSTM | 179.20 | 405.19 | 1181.74 | 3274.47 | |

Proposed | 142.60 | 244.50 | 345.84 | 2434.41 | |

DAYTON | [34] DNN | 39.93 | 88.92 | 244.97 | 1145.00 |

[35] NPCNN | 49.81 | 135.24 | 312.21 | 1247.53 | |

[20] LSTM | 29.10 | 116.53 | 613.45 | 372.93 | |

[25] CNN–LSTM | 28.82 | 79.36 | 131.67 | 392.52 | |

Proposed | 27.12 | 38.68 | 70.04 | 212.64 |

Dataset | Method | MAPE (VSTF) | MAPE (STF) | MAPE (MTF) | MAPE (LTF) |
---|---|---|---|---|---|

AEP | [34] DNN | 1.68 | 0.16 | 0.08 | 0.10 |

[35] NPCNN | 2.32 | 0.46 | 0.10 | 0.17 | |

[20] LSTM | 1.65 | 0.27 | 0.03 | 0.03 | |

[25] CNN–LSTM | 1.70 | 0.15 | 0.07 | 0.11 | |

Proposed | 1.23 | 0.06 | 0.02 | 0.03 | |

COMED | [34] DNN | 1.79 | 0.23 | 0.16 | 0.10 |

[35] NPCNN | 3.02 | 0.30 | 0.29 | 0.15 | |

[20] LSTM | 1.41 | 0.12 | 0.15 | 0.38 | |

[25] CNN–LSTM | 1.68 | 0.15 | 0.07 | 0.04 | |

Proposed | 1.30 | 0.09 | 0.02 | 0.03 | |

DAYTON | [34] DNN | 1.99 | 0.19 | 0.07 | 0.08 |

[35] NPCNN | 2.58 | 0.28 | 0.09 | 0.08 | |

[20] LSTM | 1.36 | 0.23 | 0.03 | 0.03 | |

[25] CNN–LSTM | 1.49 | 0.16 | 0.04 | 0.03 | |

Proposed | 1.38 | 0.08 | 0.02 | 0.01 |

Dataset | Method | VSTF | STF | MTF | LTF |
---|---|---|---|---|---|

AEP | MSCNN | 1.91 | 1.17 | 0.85 | 1.26 |

MCSCNN | 2.28 | 0.93 | 0.79 | 1.19 | |

SCNN–LSTM | 1.84 | 0.64 | 0.57 | 0.75 | |

Proposed | 1.23 | 0.06 | 0.02 | 0.03 | |

COMED | MSCNN | 2.36 | 1.02 | 0.79 | 0.86 |

MCSCNN | 2.31 | 0.87 | 0.89 | 0.58 | |

SCNN–LSTM | 1.93 | 0.80 | 1.08 | 0.43 | |

Proposed | 1.30 | 0.09 | 0.02 | 0.03 | |

DAYTON | MSCNN | 2.28 | 1.14 | 0.77 | 0.91 |

MCSCNN | 2.61 | 0.81 | 0.70 | 0.83 | |

SCNN–LSTM | 2.08 | 0.63 | 0.45 | 0.45 | |

Proposed | 1.38 | 0.08 | 0.02 | 0.01 |

Forecasts | Training Sets | Testing Sets |
---|---|---|

VSTF | AEP | COMED, DAYTON |

STF | COMED | AEP, DAYTON |

MTF | DAYTON | AEP, COMED |

LTF | DAYTON | AEP, COMED |

Forecasts | Transfer vs. DNN [34] | Transfer vs. Proposed |
---|---|---|

VSTF | 0.0380 | 0.4650 |

STF | 0.4800 | 0.3600 |

MTF | 0.3120 | 0.1840 |

LTF | 0.1300 | 0.4230 |

Metrics | RMSE | MAE | MAPE | |||
---|---|---|---|---|---|---|

Forecasts (Data) | Proposed | CNN–LSTM [27] | Proposed | CNN–LSTM [27] | Proposed | CNN–LSTM [27] |

VSTF(AEP) | 508.33 | 477.08 | 354.52 | 324.73 | 2.43 | 2.22 |

STF(COMED) | 1.8311 $\times $ 10^{3} | 1.8320 $\times $ 10^{3} | 1.3592 $\times $ 10^{3} | 1.3612 $\times $ 10^{3} | 0.50 | 0.51 |

MTF(DAYTON) | 4.6342 $\times $ 10^{2} | 1.04$18\times $10^{3} | 3.0828 $\times $ 10^{2} | $8.2458\times {10}^{2}$ | 0.11 | 0.23 |

LTF(AEP) | 1.1251 $\times $ 10^{4} | 3.0284$\times $10^{4} | $6.9459\times $ 10^{3} | 2.35$23\times $ 10^{4} | 0.07 | 0.22 |

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## Share and Cite

**MDPI and ACS Style**

Shao, X.; Kim, C.-S.; Sontakke, P.
Accurate Deep Model for Electricity Consumption Forecasting Using Multi-Channel and Multi-Scale Feature Fusion CNN–LSTM. *Energies* **2020**, *13*, 1881.
https://doi.org/10.3390/en13081881

**AMA Style**

Shao X, Kim C-S, Sontakke P.
Accurate Deep Model for Electricity Consumption Forecasting Using Multi-Channel and Multi-Scale Feature Fusion CNN–LSTM. *Energies*. 2020; 13(8):1881.
https://doi.org/10.3390/en13081881

**Chicago/Turabian Style**

Shao, Xiaorui, Chang-Soo Kim, and Palash Sontakke.
2020. "Accurate Deep Model for Electricity Consumption Forecasting Using Multi-Channel and Multi-Scale Feature Fusion CNN–LSTM" *Energies* 13, no. 8: 1881.
https://doi.org/10.3390/en13081881