Load forecasting and decision support system for electric vehicles use

There is a dynamic process affecting many sectors in the world within the framework of global production and consumption understanding coordinated with technological innovations. This situation, which fundamentally affects the social order and the increasing environmental awareness, has revealed the need for the energy sector to be structured with different dynamisms. Many operational and administrative processes are carried out in the dynamic infrastructures of the energy sector and market. In these processes carried out with the aim of maintaining the supply-demand balance of energy, consumers, distribution companies and other market stakeholders interact directly or indirectly. Today, depending on the strategic goals established for the change and development of the global energy sector, governments have certain commitments towards consumers (customers/users) and market stakeholders. These commitments include the production of digital transformation-supported smart grid solutions for the grid infrastructures of power systems and the achievement of net zero CO2 emission targets. The integration of electric vehicles (EV), distributed generation resources of different scales and energy storage systems into medium/low voltage networks and the establishment of smart embedded systems are of critical importance in achieving these goals. With the smart grid infrastructure, automation and digitalization in information, communication, monitoring, control and management systems will be increased, and accurate and reliable guidance mechanisms will be created for decision makers. With these systematic and technological infrastructures created for companies operating in the energy sector to develop strategies suitable for their needs analysis, decision makers can be provided with broad and different perspectives. With the real-time operation of forecasting, planning, control and traceability in the management of networks, crises that may be experienced can be instantly turned into opportunities and the support of decision-making systems can be increased. Thus, time and resources can be saved by providing an analysis, evaluation, guidance, interaction, balancing and reconciliation environment with decision support systems that assist the decision-making process of the relevant parties. In this thesis study, the conceptual framework of the strategies to be applied to minimize the impact of peak load on the distribution network and the variable factors determined accordingly, which are considered based on the general approach in the energy sector, is created. In this conceptual framework, where the impact of EV usage on the distribution network is addressed within the scope of the cause-effect relationship; there are six different variable categories as dependent and independent variables and the confounding, moderating, mediating and control variables that affect them. In the case of intensive use of charging stations by EV users, many elements should be considered and monitored in terms of both the user and local distributed generation resources. In the continuation of the study, a conceptual Decision Support System was developed, which includes the elements affecting the operational and managerial decision-making process of organizations from the perspective of electric vehicle users and public service providers, considering the factors determined in the conceptual framework. In order to meet the intense electricity consumption demands of consumers, energy conversions are provided from various renewable energy sources. The development of new energy demand created by the increase in the use of electric vehicles; causes some changes and transformations in the behaviors of electric vehicle users in the consumer role and in the renewable energy sources that feed the production. Therefore, it is important to understand the new demand and consumption profile that will occur on the distribution network with the intense use of electric vehicles. Different solution strategies need to be produced in order to cope with the increase in the load demand that will occur on the electricity network and the overload problem during the peak time period of this demand. The management of daily energy demand created by electrical energy consumption and the load forecast plans that determine the hourly demand amount for the next day are the determining and guiding elements for the energy sector. With the spread of electric vehicle charging stations, which are a component of distributed production (or distributed energy) where the energy consumption data of EVs are obtained, energy demand forecasts need to be made in more detail, including advanced network analyses instead of regional focus. The accuracy of energy demand forecast is of great importance in order to be able to effectively make load forecast plans related to demand side management and load management in the energy sector. The subject and scope of the study; analysis of new energy demand and periodic peak power (or peak load) demand on the distribution network and load forecasting algorithms to be applied for short-term load forecasting are discussed. In the study, it is aimed to obtain the most realistic, consistent and effective results for the forecasting of EV energy consumption or EV charging load that constitutes the future new demand by utilizing past data. The forecasting methods used in the study are considered as time series forecasting in the literature, and deep learning models were used in the development process of forecasting algorithms. For this purpose, the Stacked Deep Learning Ensemble Model (SDLEM) was proposed in the study to increase the short-term charging load forecasting accuracy of electric vehicles. In the proposed model, CNN, LSTM, GRU, BiLSTM and BiGRU models were considered as base learners, while LightGBM algorithm was used as meta learner. For the implementation of the proposed model, real-time charging session events collected from 54 public charging stations located in the Caltech campus garage of a research university located in Pasadena, California, USA were used. The data belonging to these real-time charging session events; EV charging connection time, EV charging disconnection time and energy consumption attributes were obtained from the open data platform called ACN-DATA. The effectiveness of the developed Stacked Deep Learning Ensemble Model was evaluated by comparing it with the performances of deep learning based single models using MAE, RMSE, NRMSE and R² performance metrics. As a result of the evaluation, the MAE, RMSE, NRMSE and R² performance metric values of the proposed model were obtained as 3.5894, 4.9935, 0.0551 and 0.9444, respectively, and it was seen that it gave the best performance compared to the compared single models. In order to manage the demand for electric vehicles and optimize charging control in the capacity planning of the EV charging infrastructure of distribution networks, problems such as demand uncertainty, energy uncertainty, EV mobility, network uncertainty, and price uncertainty should be addressed. In order to manage these uncertainties with the integration of information and communication technologies into the power system, optimality, improvement, and sustainability should be ensured with artificial intelligence applications in the activities of the predictive dynamic processes of energy. In summary, the study; a conceptual framework for the impact analysis of EV charging load demand on distribution networks, a conceptual Decision Support System and a methodology for EV charging load forecasting have been created, and load forecasting models based on deep learning algorithms have been developed. In this context, it is aimed to increase the flexibility of the power system by supporting the construction of smart charging control and decision-making mechanisms based on data obtained and interpreted by utilizing information and communication technologies. It is thought that this study will also guide other smart grid technology development studies.