Effects of meteorological variables on electricity consumption

Today, energy consumption is one of the most important indicators of countries' development levels. Energy, the most important input of social and economic development, is a necessity in order to increase the standard of living and sustainable development. Electricity is one of the most preferred and consumed energy types because of easy use and clean energy. Due to industrialization, population growth and rising living standards in our country, electricity consumption is increasing rapidly. Technological tools that make human life easier have become indispensable elements of life. Until the technological devices are designed and are consumed by the users, energy is the most important factor. The healthy progress of production, social and economic activities, which are obligations of human life, depends on the uninterrupted supply of energy. Quality and reliable electric energy can reach the consumers and the electricity generation to be needed in the future can be planned with the correct consumption estimation. Electricity consumption depends on various social and economic variables such as population, economic growth, cost and gross domestic product, as well as on climatic variables such as temperature, precipitation and humidity. The electricity used for the heating and cooling needs is bigger in the electricity consumption series. The effect is the highest meteorological variable temperature, while weather conditions cause an increase and a decrease in consumption. Humidity, wind and precipitation are other meteorological variables that cause an increase and decrease in electricity demand. While the effect of these variables varies on a regional basis, factors such as population or industry change which air condition affects consumption more intensively in that region. In regions with high population density, especially in summer, temperature, humidity and temperature are influential in the change of consumption due to the need for cooling, while rainfall consumption is the most influential variable in agriculture areas in summer. The energy used for irrigation causes the delayed effect of rainy days and irrigation to decrease and consequently the electricity consumption to decrease. The researches on the literature were mentioned in the first part of the thesis. In the second part, the structure of the electricity markets and the historical progression process in our country are explained. It was explained in which processes the electricity consumption was evaluated in the Turkish electricity markets and how the trade was done. By exploring the variables affecting electricity consumption, it was explained how meteorological variables affect consumption. In the third part, artificial neural networks, which are model structures used in working, are explained. The network and structure details of the model used are given. In the fourth part, it is the part where the actual work is told. Electricity consumption profiles in Turkey are announced on provincial and regional basis. Information about the selected data and the data analysis process are explained. Then the details of the model and the results are presented in this section. In this thesis, it is examined how much the electricity demand of Turkey is affected by the temperature changes. In monthly and seasonal time periods, it was researched how and how much the consumption was changed due to the temperature and it was aimed to make more consistent consumption estimation by adding the temperature as input to the consumption model. Within the scope of the study, short-circuited electricity consumption estimations were made by using the Levenberg-Marquardt back propagation data group which is modeled by YSA method. The study period is January 2012-November 2016. Turkey's total consumption data and temperature data for these years are used. The temperature data was produced with a population-weighted average. The temperatures of 11 provinces with the largest share of consumption in Turkey are weighted according to their consumption rates. In the model, 38 intermediate layer artificial neural networks were found to give the best results. When estimates are established with this structure, the consumption of the hottest and coldest August and January months for 2016 is estimated according to the monthly average. Day effect was used as an input to model in addition temperature and electricity consumption data. Using the same network structure and algorithm two electricity consumptions were forecasted. One of these X1 included the temperature data in addition to historical consumption and day effect data, other model X2 included just historical consumption and day effect data. Results of the study, MAPE of August was 1.62% for hourly and 1.04% for daily of X1 model. MAPE of August was 2.67% for hourly and 2.08% for daily of X2 model. Moreover, MAPE results of January was 1.93% for hourly and 1.34% for daily of X1 model. MAPE of January was 2.20% for hourly and 1.81% for daily of X2 model. The model X1 which include temperature data, was more successful with higher performance. Based on these results, it is predicted that the effect of temperature on electricity consumption will be higher in summer. The use of air conditioning for summer cooling needs directly leads to an increase in electricity consumption. However, as winter heating needs to be covered by natural gas and coal in winter, the use of electric heaters and air conditioners is low for heating needs. Thus, the direct effect of temperature on electricity consumption is decreasing. In the transition seasons, it is thought that the model will work close to performance at this period because the temperatures are close to the comfort temperature. In the model, the day type was used as input. The results were analyzed on a weekly basis. According to month of August result; on Monday, the models performed very close to each other while the X2 performed better than X1 at 0.04%. Looking at the other days, it is seen that X1 has lower MAPE value per day. In the chart, the X2 model predicts Tuesday that the MAPE average is very high. The reason for this is that on August 30, 2016, it will coincide with Tuesday. In the model, August 30 has a separate day type as a public holiday. But the artificial neural network is limited in the number of data to be learned because there are few official holidays in hand when making weight assignments. It is much easier to assign the right weight to today as it is on Tuesday with data on it. This result shows us that the model has a higher error rate on special days. The X1 performs better on August 30 but still has the error above average. According to month of August result; On Thursday the average MAPE X2 model is less than X1. For other days, the X1 model was more successful. The day with the highest error rate on both models has been Saturday. The highest error rates were observed on Friday, January 1 and Saturday, January 2, when looking at daily faults. Parallel to August results, on weekdays the errors were less and the errors of the weekend models more. It was taken as a public holiday on January 1, but it is not enough that consumption is lower than on a typical Friday, and that there are only five January 1 data in model training. Because of the special day electricity consumption, the highest error rate during the month was realized on that day. In the model weighting, the percentage of effect 1 day before is more than 1 week before and 1 week before is more than 1 year before. For this reason, deviations in the recent history more influence the model. The study presents a model for forecasting electricity demand, which is very important for today's electricity markets and which is very effective in terms of economic and financial performance of market participants. When model results are evaluated, error rates are considered reasonable. It is planned to establish the model on a regional basis for future work. When estimating regional electricity demand, a model can be developed by using different meteorological variables. It is predicted that rainfall data will increase the performance of estimates of the inclusion of temperature data in the Bosporus Region, which has a high population density and a high level of residential consumption.