Elsevier

Energy

Volume 154, 1 July 2018, Pages 7-16
Energy

Artificial neural networks and adaptive neuro-fuzzy inference systems approaches to forecast the meteorological data for HVAC: The case of cities for Turkey

https://doi.org/10.1016/j.energy.2018.04.069Get rights and content

Highlights

  • Humidity, temperature, solar radiation was forecasted for 50 cities of Turkey.

  • Surfer and ArcGIS were used to create humidity, temperature, solar radiation maps.

  • Artificial neural networks and adaptive neuro-fuzzy inference systems were modelled.

  • The results of the approaches are compatible with the result of the Turkey.

  • The policy maker from General Directorate of Meteorology can use these approaches.

Abstract

Limited energy resources and increasing need for energy due to population growth seem to lead researchers to focus on these issues. Forecast of meteorological data has significant importance in design of thermal systems. In this study, forecasting of meteorological data used in thermal system design was performed for fifty cities to represent the entire Turkey. Data obtained from General Directorate of Meteorology (MGM) were modelled by artificial neural networks and adaptive-network based fuzzy inference systems. Matlab software was used for modeling and forecasting of prospective data with high sensitivity in thermal systems. Surfer and ArcGIS software were used to create humidity, temperature, solar radiation maps for Turkey. Root mean square error (RMSE), mean absolute error and (MAE), coefficient of variation (COV) and the coefficient of determination (R2) were used to validate the result of the proposed approaches. The results were satisfactory with respect to RMSE, MAE, COV and R2 to forecast the meteorological data. Annual solar power potential maps for Turkey were also proposed and compared with MGM results. The results of the proposed approaches are compatible with the result of the MGM. The Turkey policy maker(s) from MGM can easily use these approaches if a software is constructed.

Introduction

Among renewable energy resources, solar energy technology is the fastest improving and growing. Due to continuously advancing technology and decreasing costs, solar energy is deemed as one of the most important energy resources of the future [1]. Quantitative information regarding global radiation is necessary in calculation of evaporation and soil moisture, hydrological studies, agricultural studies, climatology, building energy analysis and many more fields. Today, solar radiation values have a significant importance for many engineering designs such as design and optimization of solar powered systems and architecture. Therefore, information about change in solar radiation is necessary [2] in many research fields such as agriculture, hydrology, and meteorology [3]. It is not possible to infer directly the radiation over large areas because of measuring at only a small number of stations [4].

One of the fields which require solar radiation values and other meteorological data is climate control and planning of comfort elements in modern buildings. HVAC (Heating, Ventilating and Air Conditioning) systems regulate and control the climate, temperature and air flow in buildings and help ensure a comfortable environment. HVAC systems are important for health of those who live in that environment as well. Because, systems with well-regulated climatic conditions and appropriate fixed values keep hazardous organisms such as mold away from the environment. Besides, the most significant benefit of automated HVAC systems is low energy consumption and therefore energy saving. Any improvements in energy efficiency of HVAC systems could be instrumental for avoiding further dependency on fossil fuels [5].

Input/output implementation, input devices, controlled devices, and all human factor subcategories have significant effects on energy use with respect to control experts [6]. Different inputs are used for evaluating the HVAC systems in literature [7]. The software input/output implementation, input devices, controlled devices, and all human factor subcategories to have significant impacts on energy use [6]. In this study, the considered input variables are months, number of sky clear days, cloudiness, average air pressure, ground surface minimum temperature, mean solar time, and evaporation. In this paper, Sugeno-type ANFIS also used for forecasting values for temperature, solar radiation, relative humidity. ANFIS is a kind of adaptive neuro-fuzzy inference system [8] which connects fuzzy logic system with neural network [9]. The flowchart of the study is presented in detail in Fig. 1.

This study presents a forecast of solar radiation, temperature and relative humidity based on other meteorological parameters using ANN and ANFIS for fifty cities in different regions of Turkey in a way that it will represent the entire country. Values obtained because of this forecast are compared with meteorological data and TS-825. Results are presented in charts and tables. The rest of paper is organized as follows: The proposed approaches are presented in Section 3 and Section 4. The application is detailed in Section 4. Finally, the conclusion is presented in last section.

Section snippets

Background

Behrang et al. [10] estimated daily global solar radiation on a horizontal surface due to using ANN. The solar irradiance values of Dezful, Iran were estimated using the daily mean air temperature, relative humidity, sunshine duration, evaporation and wind speed Between 2002 and 2006. Tymvios et al. [11] applied the Angström linear approach and ANN for forecasting of the solar radiation on a horizontal surface. Mellit et al. [12] developed a hybrid model based on ANN to estimate daily global

Artificial neural networks

The general structure of ANN (Fig. 2) consists of at least 3 layers. The first layer is the input layer and the last layer is the output layer. The other layer is the hidden layer and this layer may consist of multiple layers depending on the property of the problem in hand. Each layer has artificial nerve cells and artificial nerve cells in a layer relate to artificial nerve cells of the next layer through weight coefficients, except for the output layer. It is obvious that the number of nerve

Adaptive network-based fuzzy inference systems

Fuzzy inference systems and multi-layer perception are special examples of adaptive networks used for very general calculation studies [37]. Both examples make use of back-propagation learning skills of the adaptive network. ANFIS is a class of adaptive networks which is the equivalent of fuzzy inference system in terms of function [38]. ANFIS stands either for Adaptive Network-based Fuzzy Inference System or Adaptive Neuro Fuzzy Inference System. ANFIS is mentioned in some sources as neural

The performance criteria of ANN

Performances of ANN and ANFIS models created in this study were compared based on three different statistical criteria: These criteria include statistical parameters such as root mean square error (RMSE), mean absolute error and (MAE), coefficient of variation (COV) and the coefficient of determination (R2). RMSE is used to determine the degree of error between measured values and model forecasts. The closer the RMSE value to zero, the better the forecast capability of the model. RMSE is

Conclusion

A back-propagation neural network perceptron model with seven inputs, three outputs and multi layers was used to forecast temperature, solar radiation and relative humidity values, which are meteorological data. The performance of the ANN model was ensured through trial and error by changing parameters such as the number of neurons in the intermediate layer, the number of inputs and the learning coefficient.

Similarly, an ANFIS model was used as well to forecast meteorological data. Performances

References (40)

Cited by (60)

View all citing articles on Scopus
View full text