You are using an outdated browser Internet Explorer. It does not support some functions of the site.

Recommend that you install one of the following browsers: Firefox, Opera or Chrome.


+7 961 270-60-01

Neural network approach to forecasting energy consumption in an urban environment


Neural network approach to forecasting energy consumption in an urban environment

Gorbunova E.B., Sinutin E.S., Sinutin S.A.

Incoming article date: 14.11.2018

Environmentally friendly approach to energy consumption implies planning the resources' amount which could be delivered and utilized by the end user. For the resource provider this means being able to get remote data from a huge number of users, with further data analysis being performed to predict future consumption. The main objective of this work was a comparison of classical regression analysis methods with neural network analysis for solving a household energy consumption prediction problem. Research was based on an array of energy consumption data collected from 47 households situated in the Rostov region (Russia) over a period of 730 days. We investigated the forecasting models based on statistical regressing of 1-st, 2-d and 3-d order and a recurrent neural network (the best result we achieved had been given by a neural network with one hidden layer of 10 neurons). Forecasting time-frame was 20% of the data-set (nearly three months) and 80% we used for training. For our data the best result was achieved by a neural network where the ratio error of the forecast practically didn't exceed 5% (mean ratio error was 0.22 and standard deviation of 3.7). For the regression models these terms were 0.37 and 6.24 for the 1-st order, 3.31 and 6.45 for the 2-d order, 0.29 and 6.7 for the 3-d order model respectively. As a result of this work we determined, that rather simple recurrent neural network provides a better result in terms of energy consumption prediction in comparison with classical regression analysis methods, though more complex networks, such as LSTM, need a wider data array to learn.

Keywords: energy resources, metering device, housing and communal services, energy consumption forecasting, data analysis, regression, time series, recurrent neural network, machine learning, “smart city”