Long-Term Load Forecasting Using Neural Network Technique for Distribution Substation System Expansion Planning (Case Study: Debre Work Town)

Abstract

Load forecasting is the prediction of demand for the next period and is an essential requirement for new planning and expansion planning of power systems. Knowing the future demand in concise way has been still a big problem in Ethiopia to meet the power demand for different load types like industrial, commercial, city municipal services or public services, agricultural and domestic services. In this thesis, load forecast for Debre Work town and its surrounding has been obtained for the period of 2023 to 2037 and distribution substation system expansion planning model has been planned with technical feasibility study, which has covered load flow analysis, contingency analysis and short circuit analysis to meet the forecasted load . The following independent variables have considered, namely - GDP, number of populations, GDP/CAP, weather conditions like temperature and electricity consumption by the people (EP) as the factors that affecting the future load growth in Debre Work town. MATLAB software has been used for training and testing for the general neural network techniques such as LSTM-RNN and GRU-RNN. LSTM-RNN result was better than GRU-RNN in good fitting from assessment measurement parameters results. From training time view point, GRU-RNN was better than LSTM-RNN. However, LSTM-RNN has selected for this thesis because the training time difference was not that much big. PSSE (power system simulator for engineers) software has been used for distribution substation system expansion planning model to simulate and test the network feasibility. After forecasting the future load demand of Debre Work town, new distribution substation system expansion planning model has been applied to meet the forecasted power demand and testing of network technical feasibility and these analyses has meet the requirements of technical feasibility analysis limit of EEA/2022 standards. The thesis has also studied the cost of the substation and transmission line from LILO point to Debre Work substation which was 528,653,214 Ethiopian Birr currently. Generally, load forecasting for Debre Work area has been done and comparison of forecasting techniques such as LSTM and GRU has been conducted and even their result has shown almost similar, LSTM result was good, modeling of distribution substation has been done and finally substation cost estimation was done.