Content uploaded by Nadeem Javaid

Author content

All content in this area was uploaded by Nadeem Javaid on Jul 13, 2019

Content may be subject to copyright.

Half Hourly Electricity Load Forecasting

Using Convolutional Neural Network

Abdul Basit Majeed Khan1, Sajjad Khan2, Sayeda Aimal2, Muddassar Khan1,

Bibi Ruqia3, and Nadeem Javaid2(B

)

1Abasyn University Islamabad, Islamabad 44000, Pakistan

2COMSATS University Islamabad, Islamabad 44000, Pakistan

nadeemjavaidqau@gmail.com

http://www.njavaid.com

3Sardar Bhadur Khan Women University Quetta,

Quetta 87300, Pakistan

Abstract. In this paper, enhanced Deep Learning (DL) method is

implemented to resolve the accurate electricity load forecasting prob-

lem. Electricity load is a factor which plays major role in operations of

Smart Grid (SM). For solving this problem, we propose a model which

is based on preprocessing, selection and classiﬁcation of historical data.

Features are selected by Combine Feature Selection (CFS) using Deci-

sion Tree (DT) and Mutual Information (MI) techniques, and then CFS

Convolutional Neural Network (CFSCNN) is used for forecasting of load.

Our proposed scheme is also compared with other benchmark schemes.

Simulation results show better eﬃciency and accuracy of proposed model

for half hourly electricity load forecasting for one day, one week and one

month ahead for the data obtained from ISO NE-CA electricity market.

Keywords: Deep learning ·Classiﬁcation ·Mutual information ·

Smart grid ·Decision tree ·Convolutional neural network

1 Introduction

Smart Grid (SG) is an advanced form of Traditional Grid (TG). TG is the

connection of diﬀerent power systems. They are planted away from the p ower

usage areas. Electricity is transferred by long transmission cables. Energy can

only be provided from the main power plant using traditional power structure.

Traditional power system makes very hard to control the energy because when

electricity leaves the power plant, energy ﬁrms have no more control on dis-

tribution, and this may cause the loss of energy. SG is used for eﬃcient and

reliable distribution of electricity. It is a two way transmission among utility and

consumer [1]. Utilities and consumers, both are able to monitor the activities

of grid system. With the eﬃcient and smart digital structure energy ﬁrms have

better control on power distribution. Power and power consumption is easily

monitored while transferring from source to destination in SG. Through this

c

Springer Nature Switzerland AG 2020

L. Barolli et al. (Eds.): IMIS 2019, AISC 994, pp. 172–184, 2020.

https://doi.org/10.1007/978-3-030-22263-5_17

Half Hourly Electricity Load Forecasting Using Convolutional . . . 173

technology, energy which is produced by renewable resources can also be added

in main power grid. SG also reduced the cost of operations for utilities and low

electricity cost for consumers.

Smart meters are used to record the consumer’s load of electricity consump-

tion and send back to utilities. For the eﬃcient and reliable prediction of future

power consumption, accurate electricity load forecasting is necessary. In this

research, we use diﬀerent Artiﬁcial Neural Network (ANN) and Deep Learning

(DL) based classiﬁers to forecast the better accuracy of load. DL techniques help

us to ﬁnd hidden patterns from the large data accurately. Data pre-processing is

also performed to calculate valuable features from large dataset. These methods

give us accurate prediction of load which beats previous methods. The goal of

this research is to provide eﬃcient method to forecast the electricity load with

higher accuracy.

1.1 Problem Statement

In SG technology, we have a lot of issues regarding energy utilization and

distribution. Both the customer and utility want to get advantages from the

technology. This can be only possible, when the eﬃcient utilization of energy is

occurred. Electricity Load forecasting has a huge impact on reducing electricity

consumption in SG. One of the key problems in power grid system is accurate

forecasting of electricity load. When electricity load is forecasted accurately, it

helps power generators and distributors to modify their power grid operations,

and generate electricity according to the need of consumer. Moreover, when

the generation is reduced, cost is also reduced. To solve the forecasting issues

of diﬀerent models, some advanced and eﬃcient methods are required. We use

enhanced DL strategies to overcome the issues of accurate electricity load fore-

casting and improve the accuracy of forecasting load.

1.2 Contributions

In this paper, accurate load prediction is our primary goal. For this purpose, a

DL based technique is proposed. We proposed a model, which reduces the com-

plexities and gives better results. The contributions of this paper are describes

as follows:

•Selection of best features using Decesion Tree (DT) and Mutual Information

(MI) is made at the start of process.

•Extract the important features by Combine Feature Selection (CFS).

•Design a CFSCNN network, which is the combination of CFS and CNN,

where the best input is provided to the network, which improves performance

of model.

•Our model also improves the accuracy, reduce computational complexity and

respone time.

174 A. B. M. Khan et al.

2 Related Work

Electricity price and load, are the most important factors in electricity market.

To make the market competitive and beneﬁcial, price and load forecasting are

key approaches in SG, to be implemented. Large datasets are diﬃcult to process

with traditional computational and statistical models [1], however, author pro-

posed an eﬀective DL based framework for better forecasting of electricity price

and load. Firstly, data pre-processing is done, then Hybrid Feature Selection and

Extraction (HFSE) is used for prediction of electricity load and price. However,

this model has over-ﬁtting problem. Another method is presented in [2], which

forecasts short term electricity load. In this paper author shows the calendar

eﬀects of intra-day, weekly and seasonal on load forecasting accuracy. The eﬀect

of dataset length on accuracy of diﬀerent neural networks and Support Vector

Regression (SVR) techniques are also studied. Also measures how the forecasting

accuracy eﬀects on the granularity, which makes forecast. Authors also conclude

that computational complexity of Neural Netwok (NN) is greater than LR tech-

niques, and if the historical load is not available, calendar eﬀects become more

important.

In [3] authors present short-term load forecasting of holidays using fuzzy

enhanced same day method. The initial step is to determine the important fea-

tures of the holiday’s proﬁles. For each feature, one similar day is considered.

The important load features of the expected holiday are acquired from these

same days. The holiday’s load is forecasted by combination of these features

and then improved using a fuzzy method. Accuracy of model is needs to be fur-

ther improved. To forecast cooling load, Deep Auto Encoders (DAE) are used in

[4]. It gives accurate results. Eﬃcient energy utilization can reduce the shortage

of energy, minimize the electricity cost [5]. Another method [6] which devel-

ops a technique for Short-Term Load Forecasting (STLF) using Dynamic Mode

Decomposition (DMD) by choosing the meaningful hidden patterns from data.

The proposed model optimize the load series data that is inﬂuenced by diﬀerent

factors consist of day, time, seasons, climate, and socio-economic activities. In

[7], long term load forecasting is done using twelve diﬀerent US Western utlilites.

Peak demand growth and load consumption is considered. The simulation results

showed that complexity and accuracy of diﬀerent forecasting methods is cor-

related. In paper [8], Enhanced Logistic regression (ELR), Classiﬁcation and

Regression Tree (CART), RFE, RF and Grey Wolf Optimization (GWO) tech-

niques are used for forecasting the electricity load and price. This work is working

well in their model.

In [9], a hybrid model for load and price forecasting is presented. The pro-

posed technique uses kalman and wavelet machines. Load and price data is

decomposed into various frequency components and Kalman machine is used to

forecasts each frequency component of load and price data. A novel load forecast-

ing model is developed in [10], which is consist of a feed forward ANN to forecast

demand of hourly load for diﬀerent season of a year. In this method, a Global

Best Particle Swarm Optimization (GPSO) is implemented, which improves the

accuracy and eﬃciency of ANN prediction. To modify network training, ﬁtness

Half Hourly Electricity Load Forecasting Using Convolutional . . . 175

fuction is introduced and weight bias method is also presented. The output of

model shows better results than benchmark techniques. In paper [11], authors

examine the short term price and load forecasting using diﬀerent selection meth-

ods and deep learning techniques. In [12], authors propose seer grid, an alterna-

tive of Smart Grid communication Network (SGN) method, aimed to minimize

the privacy-utility trade-oﬀ. As a result of two-level electricity load forecasting

in seer grid, high relationship occurs among predicted and actual energy con-

sumption values at cluster level, which show excellent utility protection. The

main goal of this study is to ﬁnd substitute practical design for privacy-sensitive

production and sharing of energy utilization data from the SM to the energy

company which allows operation of the energy company in terms of accuracy.

Another approach for distributed system electricity load forecasting is presented

in [13], which gives short term load forecasting with high accuracy using Support

Vector Regression (SVR) and two-step hybrid parameters enhancement method.

Residential electricity load forecasting has been playing an important role in

smart grids. A Recurrent Neural Network (RNN) based forecasting with appli-

ance consumption sequences is proposed to optimize such volatile problem. Kong

et al. [14]. Electricity load forecasting is complicated problem because of com-

plex and variable factors. In [15], authors introduce an Internet of Things (IoT)

model to automatically ﬁnds important feature from the obtained data and ulti-

mately provides an accurate estimation of electricity load. A biggest advantage

of this technique is that its two step forecasting method, which clearly increases

the prediction accuracy for daily total utilization. A new eﬀcient hybrid model

for forecasting is proposed in [16], which consists of GELM, IWNNs, wavelet

processing and bootstrapping. Simulation results show the better accuracy and

reliability of model. A method consists of fractal geometry functions to forecast

electricity load density growth in an urban area to better distribute load density

is presented in [17]. Furthermore, the method presents a very low global error

according to the position of loads, when compared with actual data.

The behaviour of energy consumption has been changing over the few

decades, especially due to improvements in the distributed production segment

and technological innovations introduced by SM. Authors in [18]proposeto

build an ANN and fuzzy logic for electricity load forecasting to do an eﬃcient

analysis. This method is able to give calculations of the elasticity of electricity

demand behaviour with satisfactory results. In [19], authors forecast the load

and price based on long short term memory (LSTM). Proposed Deep (DLSTM)

for prediction on ISO NE-CA and NYISO dataset. This work is eﬀective for only

DLSTM model.

3SystemModel

Our methodology is based on the DL techniques. DL techniques help us in the

better forecasting of load in electricity. The method is based on supervised learn-

ing; having input and output variables. Benchmarks are already deﬁned, which

are used later for comparison of predicting and actual results. The structure

176 A. B. M. Khan et al.

of our proposed model is shown in Fig. 1. It consist of three parts, i.e., data

pre-processing, feature selection and classiﬁcation.

Fig. 1. Proposed system model

3.1 Preprocessing Data

The ﬁrst step of our model is to preprocess the data. We used half hourly elec-

tricity load data of ISO-NE CA market. One year half hourly load data of 2017 is

used in this model. Data is divided into two parts i.e. training and testing data.

Train test splitt() is used for splitting the data into; feature train, feature test,

Labels test, Labels train, and these values are further usead as input. Testing is

done on 25% of data and remainig part is used for training. For the ﬁx number

of inputs, random state is also deﬁned. Data is also normalized at this stage.

3.2 CFS Feature Selection

In this section, we describe the method of feature selection. DT and MI tech-

niques, calculate important features from data. We also drop the features, which

have low importance. After combining the results of these two techniques we

select the features by deﬁning a thresh hold value, which drops the unimportant

features. DT importances are shown in Fig. 2while MI importances are shown

in Fig. 3. We found the importances in vector form.

Half Hourly Electricity Load Forecasting Using Convolutional . . . 177

Fig. 2. DT importances

Fig. 3. MI importances

178 A. B. M. Khan et al.

3.3 Load Forecasting

In our proposed model, electricity load forecasting is done with CFSCNN, which

is described as.

CFSCNN In machine learning, CNN is a class of deep neural network. It consist

of one or more convolution layers, and then followed one by one fully connected

layer as in NN. CNN contains input layer, multiple hidden layers and output

layer. Hidden layers consist of, convolutional layer, dense layer, droupout layer,

ﬂatten layer and pooling layer. The convolution layer calculates the output of

neurons that are associated with local boundary or receptive ﬁelds in the input,

each simulates a dot product with their weights and a receptive ﬁeld by which

they are connected to the input data. FFNN trains the network and also classify

the data.

In our model, we use the output of CFS as input for network and the net-

work is named as CFSCNN. Due to CFSCNN, accuracy of forecasting load is

increased. In proposed model, two convolutional layers are used. The ﬁrst layer

consist of, 96 ﬁlters and 2 kernels and the second layer have 32 ﬁlters and 3 ker-

nels. In addition, one max pooling layer with pool size 2 is used in our network.

Pooling layer sums the output of large data into neuron and passes that input

to the next layer. Dropout layer is added to avoid overﬁtting.

4 Simulation and Discussion

In this section, we describe simulation results of our proposed technique in detail,

for showing accuracy of electricity load forecasting. Our model results are sum-

marized as follows.

4.1 Data Description and Simulation Setup

In this paper, ISO New England Control Area (ISO NE-CA), market data from

January 2017 to December 2017 is used. We consider the half hourly data of

each day. Our dataset is consist of 15 columns and 8,760 instances. Each day

consist of 48 instances. For this purpose we use a simulator, which consist of

Python framework with Intel Core i3, 4GB RAM, and 500GB hard disk. Before

moving to next step we also normalize the data.

4.2 Simulation Results

After preprocess the data, we apply our techniques to get result. Results are

described in following sections.

Half Hourly Electricity Load Forecasting Using Convolutional . . . 179

CFS Feature Selection Selection of best features helps in accurate prediction

of load. Two techniques, i.e., DT and MI are used to ﬁnd the importances of

features. Figure 2shows importances of DT and Fig. 3shows the importances of

MI for diﬀerent features. As we see in graphs that MI shows more appropriate

feature than DT. After ﬁnding the importances, features which have low impor-

tances are dropped by denining a thresh hold value. Features are selected by

combining the importances of both DT and MI. The output is then named as

CFS.

Daily, Weekly and Monthly Comparison with Actual Load In our model,

half hourly load data is used for comparisons. In this section, daily, weekly and

monthly forecasting results of CFSCNN are compared with actual values. One

day prediction is consists of 48 values of day, one week prediction consist of

330 values of a week, and one month prediction consists of 1440 values of a

month. Figures 4,5and 6show these comparisons. Solid line shows the actual

load and dashed line shows the predicted load. The proposed model predicts the

load with very small error, which makes the model accurate. This comparison

shows that, weekly and monthly load forecasting results are better than daily

load forecasting result, which shows the better accuracy of our model with larger

input.

Fig. 4. Comparison graph of CFSCNN with actual load for one day

180 A. B. M. Khan et al.

Fig. 5. Comparison graph of CFSCNN with actual load for one week

Fig. 6. Comparison graph of CFSCNN with actual load for one month

Daily, Weekly and Monthly Comparison with Existing Classiﬁers In

this section, we compare the daily, weekly and monthly results of our proposed

model with existing classiﬁers. Accuracy of our model is better than existing

techniques, which is shown in Figs. 7,8and 9. Similarly, forecasting for larger

inputs is much better than lesser inputs. The model, reduces the computational

complexity of existing classiﬁers. Adding more layers in model helps in better

training of model. However, the training of NN model is very diﬃcult, because

some times it takes random values, which overﬁtt the data and gives bad accu-

racy.

4.3 Performance Metrics

For calculate the performance and accuracy of model, Two evaluators i.e. Mean

Square Error (MSE) and Mean Absolute Error (MAE), are assumed. Error com-

parison of daily, weekly and monthly load forecasting are shown in Figs.10 and

11. These graphs show that CFSCNN has low error than other techniques, which

Shows better accuracy of proposed model.

Half Hourly Electricity Load Forecasting Using Convolutional . . . 181

Fig. 7. Comparison graph of CFSCNN with diﬀerent classiﬁers for one day

Fig. 8. Comparison graph of CFSCNN with diﬀerent classiﬁers for one week

Fig. 9. Comparison graph of CFSCNN with diﬀerent classiﬁers for one month

182 A. B. M. Khan et al.

Fig. 10. Comparison of MSE score

Fig. 11. Comparison of MAE score

5 Conclusion

A half hourly load forecasting model by combining a feature selection model

and DL based model is presented in this study. The combine feature selection

is used as input for the DL model, which avoids the model from unimportant

calculations and improve accuracy. Important features are examined, and from

among them suitable features are selected by CFS. The experimental results of

diﬀerent forecasting models showed that the proposed model CFSCNN reduced

computational complexity and increased accuracy eﬃciently.

Half Hourly Electricity Load Forecasting Using Convolutional . . . 183

References

1. Wang, K., Xu, C., Zhang, Y., Guo, S., Zomaya, A.: Robustbig data analytics for

electricity price forecasting in the smartgrid. IEEE Trans. Big Data 5(1), 34–45

(2017)

2. Lusis, P., Khalilpour, K.R., Andrew, L., Liebman, A.: Short-term residential load

forecasting: impact of calendar eﬀects and forecast granularity. Appl. Energy 205,

654–669 (2017)

3. Ebrahimi, A., Moshari, A.: Holidays short-term load forecasting using fuzzy

improved similar day method. Int. Trans. Electr. Energy Syst. 23(8), 1254–1271

(2013)

4. Cheng, F., Xiao, F., Zhao, Y.: A short-term building cooling load prediction

method using deep learning algorithms. Appl. Energy 195, 222–233 (2017)

5. Zafar, I., Javaid, N., Iqbal, S., Aslam, S., Khan, A.Z., Abdul, W., Almogren, A.,

Alamri, A.: A domestic microgrid with optimized home energy management sys-

tem. Energies 11(4), 1002 (2018)

6. Mohan, N., Soman, K.P., Kumar, S.S.: A data-driven strategy for short-term elec-

tric load forecasting using dynamic mode decomposition model. Appl. Energy 232,

229–244 (2018)

7. Carvallo, J.P., Larsen, P.H., Sanstad, A.H., Goldman, C.A.: Long term load fore-

casting accuracy in electric utility integrated resource planning. Energy Policy 119,

410–422 (2018)

8. Naz, A., Javed, M.U., Javaid, N., Saba, T., Alhussein, M., Aurangzeb, K.: Short-

term electric load and price forecasting using enhanced extreme learning machine

optimization in smart grids. Energies 12(5), 866 (2019)

9. Nazar, M.S., Fard, A.E., Heidari, A., Shaﬁe-khah, M., Catal˜ao, J.P.: Hybrid model

using three-stage algorithm for simultaneous load and price forecasting. Electr.

Power Syst. Res. 165, 214–228 (2018)

10. Raza, M.Q., Nadarajah, M., Hung, D.Q., Baharudin, Z.: An intelligent hybrid

short-term load forecasting model for smart power grids. Sustain. Cities Soc. 31,

264–275 (2017)

11. Zahid, M., Ahmed, F., Javaid, N., Abbasi, R.A., Kazmi, Z., Syeda, H., Ilahi, M.:

Electricity price and load forecasting using enhanced convolutional neural network

and enhanced support vector regression in smart grids. Electronics 8(2), 122 (2019)

12. Boustani, A., Maiti, A., Jazi, S.Y., Jadliwala, M., Namboodiri, V.: Seer grid: pri-

vacy and utility implications of two-level load prediction in smart grids. IEEE

Trans. Parallel Distrib. Syst. 28(2), 546–557 (2017)

13. Jiang, H., Zhang, Y., Muljadi, E., Zhang, J.J., Gao, D.W.: A short-term and

high-resolution distribution system load forecasting approach using support vector

regression with hybrid parameters optimization. IEEE Trans. Smart Grid 9(4),

3341–3350 (2018)

14. Kong, W., Dong, Z.Y., Hill, D.J., Luo, F., Xu, Y.: Short-term residential load

forecasting based on resident behaviour learning. IEEE Trans. Power Syst. 33(1),

1087–1088 (2018)

15. Li, L., Ota, K., Dong, M.: When weather matters: IoT-based electrical load fore-

casting for smart grid. IEEE Commun. Mag. 55(10), 46–51 (2017)

16. Raﬁei, M., Niknam, T., Aghaei, J., Shaﬁe-Khah, M., Catal˜ao, J.P.: Probabilistic

load forecasting using an improved wavelet neural network trained by generalized

extreme learning machine. IEEE Trans. on Smart Grid 9(6), 6961–6971 (2018)

184 A. B. M. Khan et al.

17. Melo, J.D., Carreno, E.M., Padilha-Feltrin, A., Minussi, C.R.: Grid-based simula-

tion method for spatial electric load forecasting using power-law distribution with

fractal exponent. Int. Trans. Electr. Energy Syst. 26(6), 1339–1357 (2016)

18. Tondolo de Miranda, S., Abaide, A., Sperandio, M., Santos, M.M., Zanghi, E.:

Application of artiﬁcial neural networks and fuzzy logic to long-term load forecast

considering the price elasticity of electricity demand. Int. Trans. Electr. Energy

Syst. 28(10), e2606 (2018)

19. Mujeeb, S., Javaid, N., Ilahi, M., Wadud, Z., Ishmanov, F., Afzal, M.K.: Deep long

short-term memory: a new price and load forecasting scheme for big Data in smart

cities. Sustainability 11(4), 987 (2019)