Keywords

1 Introduction

Transmission of reliable power to customers is the prime duty of any utility. Consumer load demand in electric distribution systems varies because of person-to-person activities that follow daily cycles. During daylight, especially in the afternoon, the load demand is generally high due to industries and lighting. On the contrary, the demand is lesser in the evening and early morning. Load forecasting (LF) is like weather forecasting, as weather experts tell us about the weather conditions in the upcoming future [1]. Load forecasting allows the future estimation of load conditions, and power system planning can be done in the future. The energy requirement of the clients varies each minute because of human activities that are not the same all day [2]. Thus, a significant priority for estimation of load demand is required in the future, and LF, therefore, plays a vital part in scheduling, operation and control of electrical power systems [3]. It approximates the variable(s) of the forecasted value at a future point in time. This paper mainly focuses on FL-based short term load forecasting (STLF) [4].

1.1 Literature Review

Balancing a power system is mostly done a day ahead of forecasted values given on the side of demand [5]. In nature, several types of loads make the performance highly nonlinear, which is quite dissimilar from the nonlinearity of the power system; then comes the fault due to which the system increases the price [6]. The methods used are based on regression analysis, time analysis, artificial neural network (ANN), similar day method, fuzzy logic, etc. [7]. Neutral Networks is a prolonged convergence period and a poor capability to process a large amount of variable quantity at a time; on the other side, Fuzzy logic provides a medium for representing and processing data in linguistic standings that makes the systems easily understandable [8,9,10]. The fuzzy forecasting method has widely been considered for forecasting dynamic and nonlinear type data in the last decade [11, 12]. In [13]. use Mamdani implication to attempt fuzzy logic for STLF. The basic rule of Fuzzy is prepared based on time, temperature and similar previous day load. Forecasts focus on minimization of the error among the actual and forecasted values [14]. A fuzzy-based STLF method that uses Gaussian membership function, If–Then rules and fuzzy logic process and further reduces the forecasted error and the processing time [15]. A methodology on STLF contains functioning mechanisms under the different regimes of power segment in India [16]. It also demonstrates the effects of load and temperature on generation, transmission and distribution. Here load forecasting is processed by using the function of triangular membership. Also, weather parameters such as temperature can be used to predict the consumption of energy based on week/special days with a support vector machine and generic neural regression network [17].

From the above literature survey, it is concluded that LF plays a significant role in the economic operation of a power system. The conventional methods used for STLF are Multiple Linear Regression (MLR) and Time Series (TS). The significant disadvantages of the MLR method are that this model is susceptible to temperature fluctuations. It needs a very accurate temperature forecast, as a minute variation in temperature leads to a significant change in load prediction and that of the TS method is it is difficult to use and is time-consuming. This encourages the author to implement an intelligent scheme like FL, which is more effective and faster than the abovementioned schemes.

2 Short-Term Load Forecasting (STLF)

The STLF is an integrated planning sector. Nowadays, demand-scheduling for several energy divisions like generation, transmission and distribution is required for designing a time-ahead power system, and STLF plays a significant role in that. It helps operators know about power systems with several decisions, including planning, generation reserve, system security, scheduling, demand side management, financial planning, and so forth. While it is essential for the time-ahead power system operation, inaccurate demand forecasting will cost the effectiveness of a spectacular financial loss.

2.1 Block Diagram and Flow Chart of the Proposed Scheme

For STLF, mathematical modelling is used for FL. The FL is multi-valued logic where the estimation of data is done according to Boolean logic (yes or no and true or false). The FL approach is one of the generalizations of Boolean logic. This logic is implemented for load forecasting weather data like temperature and humidity as the input and load as the output data. Figure 1 represents the configuration of the FL, which is implemented here. The block diagram in Fig. 2 presents an overview of FL methodology. Figure 3 presents the proposed intelligent scheme algorithm which depicts the stepwise implementation of the proposed work.

Fig. 1
F L configuration diagram. Crisp output goes to fuzzifier, fuzzy input, goes to inference engine with rules base, fuzzy output, goes to defuzzifier, comes out as crisp input.

Fuzzy logic configuration

Full size image
Fig. 2
A diagram of fuzzy logic methodology. Time and load undergo fuzzification, go to interference system with the rule base applied, go to defuzzification, yield to forecasted load.

Block diagram of fuzzy logic

Full size image
Fig. 3
A flowchart has five steps and a decision box that asks, Is actual load forecasted load. If yes, the process stops, otherwise return to step 3.

Flow chart of the proposed intelligent scheme

Full size image

3 Model Description of Fuzzy Logic

3.1 Base Design

Make a note of the variable quantity of input and output by using arithmetical analysis with another user. This paper consists of 2 input and 1 output variable. Each variable is taken by resolving for membership value [0, 1]. After determining each variable, the input and output variable, the shape of fuzzy membership is selected and its shape can be triangular, trapezoidal, Gaussian or ballcapped. For all single variables, fuzzy membership functions are well-defined. The provided variables should not remain equal, but it is ensured that the variable must equal the number of functions. Next, the data needs to be trained in a fuzzy model. Couples of data is referred to as training sample and it is further said that if the time is midnight and the load is average, then the load is high.

3.2 Compute the Forecast Value

A nonlinear mapping is done by fuzzy if–Then rule implements a fuzzy inference system starting from input to output space. The centroid of the area of a numerical forecast is expressed as

$$ {\text{Centroid of area, M}}_{{{\text{coa}}}} { = }\frac{{\int {z\mu_{A} (z){\text{dz}}} }}{{\int {z\mu_{A} (z){\text{zdz}}} }} $$
(1)

whereas membership function ‘M’ that defines the fuzzy set A is µ, on the universe Z.

3.3 Test Performance

For obtaining the accurateness of the STLF, it is verified by using old data. If it becomes improper, the shape of the membership function of fuzzy is changed, and a new rule base is formed. The testing system and membership function shapes are tested for several periods until the correct rule base system is developed. In the “train and set” method, the lowest error is measured for forecasting, which is very useful for large data sets.

3.4 Error Analysis

To decrease the error in the system, STLF is done.

$$ {\text{APE = }}\left( {\frac{{\text{actual load - forecasted load}}}{{\text{forecasted load}}}} \right)*100 $$
(2)
$$ {\text{MAPE = }}\frac{1}{n}\sum\limits_{i = 1}^{n} {\left( {\frac{{{\text{actual load(}}i{\text{) - forecasted load(}}i{)}}}{{{\text{forecasted load(}}i{)}}}} \right)} $$
(3)

4 Modelling of FL for STLF

4.1 Fuzzification

Fuzzification can be defined as a procedure in which crisp values change the degree of membership related to the fuzzy set shown in Figs. 4, 5 and 6. The time which is divided into 6 sets of fuzzy are Midnight, Morning, Forenoon, Afternoon, Evening and Night.

Fig. 4
A window depicts variable time as the membership function. It illustrates six sets of fuzzy, Midnight, Morning, Noon, Afternoon, Evening, and Night.

Time membership function

Full size image
Fig. 5
A window depicts variable load as the membership function. It illustrates four points in a plot, Very low, Low, Average, and High.

Load membership functions

Full size image
Fig. 6
A window depicts variable forecast load as a membership function. It illustrates four points in a plot, Very low, Low, Average, and High.

Membership functions of forecasted load

Full size image

4.2 Rule Base

The fuzzy rule base is considered an essential part of the fuzzy system. In this, If–then rule is used for getting the load forecasted output, some of which are mentioned below and shown in Figs. 7 and 8 as.

Fig. 7
A window depicts a dropdown box for the rule, if time is blank and load is blank, then forecast load is blank. There are several preset combinations.

Rule base of fuzzy

Full size image
Fig. 8
A window depicts a figure on a three-dimensional surface. X is the time, Y is the load, and Z is the forecasted load.

Three-dimensional surface

Full size image

5 Results and Discussion

Fuzzy logic simulated for STLF is shown in Fig. 9. MATLAB software is used for simulation work. From the workshop file, actual load data and input data are taken and then simulated. The simulated diagram is shown in Fig. 9. The input data is given to the controller of fuzzy logic, block. “fis” of the fuzzy interference system is put in the fuzzy logic controller block. The controller block of fuzzy logic is prepared with the fuzzy rules, which gives the forecasted output as shown in Fig. 10.

Fig. 9
A diagram of fuzzy logic simulated for S T L F. The time input and load input are put in the fuzzy logic controller and yield to forecasted output.

Simulated FL-based STLF model

Full size image
Fig. 10
A window with three variables, time, load, and forecast load, with several plots for each. Time has a value of 12.5, load 36.5, and forecast load 38.

Rule viewer of one sample data

Full size image

It can be observed from Fig. 11 that the actual and forecasted load is almost identical, and significantly with less error.

Fig. 11
A graph of hours in a day versus load in k w. Actual load and forecasted load are plotted, have close values with each other, and both fluctuate.

Comparison of the actual and forecasted load of a day for FL

Full size image

From Table 1, it is concluded that the mean error found from the result obtained from the fuzzy logic is 1.60% and the actual error range is between −2.63% to + 3.11%.

Table 1 Forecasted load and percentage error for 24 h
Full size table

It is observed from Table 2 that the proposed method gives much less error as compared to other researchers’ work. So, FL-based STLF is recommended.

Table 2 Comparison of different schemes
Full size table

6 Conclusion

In this report, the STLF methodology is discussed by using FL. For safety analysis of generation, STLF is a very useful tool. It has been observed that by using input data and by making a rule base, the load forecasting is predicted precisely and its margin error is between + 3.11% and −2.63%, and MAPE is 1.60%, which is relatively minor. Further, during simulation, it has been observed that the fuzzy logic toolbox is easy to understand because it works on the If–Then rule and the error can be reduced if the data is large and by using many memberships function.