1 Abstract-There are a lot of uncertainties in planning and operation of electric power system, which is a complex, nonlinear, and non-stationary system. Advanced computational methods are required for planning and optimization, fast control, processing of field data, and coordination across the power system for it to achieve the goal to operate as an intelligent smart power grid and maintain its operation under steady state condition without significant deviations. One of the important aspects to operate power system in such manner is accurate and consistent short term load forecasting (STLF). This paper presents a methodology for the STLF using the similar day concept combined with fuzzy logic approach and evolutionary particle swarm optimization (EPSO) technique. A Euclidean distance norm with weight factors considering the weather variables and day type is used for finding the similar days. Fuzzy logic is used to modify the load curves of the selected similar days of the forecast by generating the correction factors for them. The input parameters for the fuzzy system are the average load, average temperature and average humidity differences of the forecasted previous days and their similar days. These correction factors are applied to the similar days of the forecast day. The tuning of the fuzzy input parameters is done using the EPSO technique on the training data set of the considered data and tested. The results of load forecasting shows that the proposed EPSO tunes fuzzy system provides better results than the fuzzy stand alone system (without EPSO). Index Terms--Euclidean norm, Evolutionary particle swarm optimization, Fuzzy logic approach, Particle swarm optimization, Short term load forecasting, Similar day method. I. INTRODUCTION HORT term load forecasting (STLF) is a time series prediction problem that analyzes the patterns of electrical loads. Basic operating functions such as unit commitment, economic dispatch, fuel scheduling and maintenance can be performed efficiently with an accurate load forecast [1]-[3]. STLF is also very important for electricity trading. Therefore, establishing high accuracy models of the STLF is very important and this faces many difficulties. Firstly, because the load series is complex and exhibits several levels of seasonality. Secondly, the load at a given hour is dependent not only on the load at the previous hour, but also on the load at the same hour on the previous day and because there are many important exogenous variables that must be considered, specially the weather-related variables [4]. The load is composed of two components; one is weather dependent, and the other is weather independent. In some traditional methods, each component is modelled separately Dr Amit Jain is Head of Power Systems Research Center at IIIT, Hyderabad, Andhra Pradesh, India (e-mail: amit@iiit.ac.in). M. Babita Jain is PhD student at Power Systems Research Center, IIIT, Hyderabad, Andhra Pradesh, India (e-mail: babita_j2000@yahoo.com). and the sum of these two gives the total load forecast. The behaviour of weather independent load is usually represented by Fourier series or trend profiles in terms of the time functions. The weather sensitive portion of the load is arbitrarily extracted and modelled by a predetermined functional relationship with weather variables. Traditional STLF methods include classical multiply linear regression, automatic regressive moving average (ARMA), data mining models, time-series models and exponential smoothing models [5]-[13]. Similar-day approach and various artificial intelligence (AI) based methods have also been applied [4, 5, 7, 14]. However, these models take long computational time and have some deficiency in the case of the load changed abruptly. Evolutionary and behavioural random search algorithms such as genetic algorithm (GA) [15]-[17], particle swarm optimization (PSO) [18, 19], etc. have been previously implemented for different problems. In spite of its successful implementation, GA does pose some weaknesses such as longer computation time and premature convergence accompanied by a high probability of entrapment into the local optimum [20, 21]. Feed forward neural net structures like multi layer perceptron, functional link, wavelet, recurrent or feedback structures like Hopfield, Elman, Multi Feedback and hybrid structures using fuzzy neural networks have been widely proposed for non-stationary forecasting applications [22]. But in STLF, actual load data put forth many challenges to design a predictive neural network. Prominent of these challenges are, data pre-processing, input parameter selection, type of neural net structure selection, and training algorithm. Computational complexity, which is important for real time implementation of algorithms in power systems, is dependent on the structural complexity and training algorithm. There also exist large forecast errors using ANN method when there are rapid fluctuations in load and temperatures [4, 23]. In such cases, forecasting methods using fuzzy logic approach have been employed. Fuzzy logic allows one to (logically) deduce outputs from fuzzy inputs and in this sense fuzzy logic is one of a technique for mapping inputs to outputs. S. J. Kiartzis et al [24], V. Miranda et al [25], and S. E. Skarman et al [26] described applications of fuzzy logic to electric load forecasting as well as many others [27]-[29]. In this paper, we propose an approach for the short term load forecasting using similarity and the fuzzy parameters tuned by the evolutionary particle swarm optimization (EPSO) algorithm. In this method, the similar days to the forecast day are selected from the set of previous days using a Euclidean norm based on weather variables and day type [30]. There may be a substantial discrepancy between the load on the forecast day and that on similar days, even though the selected days are very similar to the forecast day with regard to weather and day type. To rectify this problem, the load curve Amit Jain, Member, IEEE and M. Babita Jain, Member, IEEE Fuzzy Modeling and Similarity based Short Term Load Forecasting using Evolutionary Particle Swarm Optimization S