The economy of any country primarily depends on the availability of electricity at all the times to all of its consumers. There is continuous increase in the number of consumers to meet their basic livelihood which as a result increases the power demand. There is around 5% increase of load every year in India. This is because of increase in population and industrial expansion etc. Different power utilities of Indian regions have to maintain sufficient balance between the power generated and power consumed at all the times to operate the power system stability and reliability. As the storage of electricity is linked with high outlays, it is highly impossible to store bulk amount of power. So, the balance between the power generated and consumed can be obtained by predicting or forecasting the future load at high accuracy. Estimating the future load or power demand can be termed as load forecasting. When the estimated load almost matches with the actual load it becomes easy to plan the future requirements effectively. Load forecasting plays an important part in the following aspects:-

•  Electric energy price setting
•  Exchanging power with neighboring regions
•  Preventing over loads and decrease occurrences of equipment failures
• Regulating the yield of generators
• Scheduling the spinning reserves

The Indian economy greatly depends on the availability of electricity. Since the storage of electricity is limited and it involves high costs, there should be a balance between the production and consumption of electricity at all the times. In order to attain it, it is very important to forecast the energy demand precisely. So, estimating the future energy demand or power load is termed as energy demand or load forecasting. The main objective of energy demand forecasting is to estimate the energy demand with minimum error. energy Demand forecasting plays key role in power system planning, power market design, etc., which can lead to higher reward. Estimation of the energy demand gets influenced or affected by many factors like temperature, humidity, day of the week (whether holiday or working-day), tariff structure etc., which will influence the daily load curve. The primary goal of the proposed project is to develop improved energy demand forecasting models for load dispatch centre (LDCs) of Indian states at long term i.e. yearly. Different Artificial Intelligence (AI) techniques like Artificial Neural Network (ANN), Support Vector Machines (SVM), Decision Tree (DT), Ensemble classifiers and fuzzy logic etc. will be used to develop intelligent energy demand forecasting schemes and the performance of these schemes will be compared. Among the AI strategies, Artificial neural systems (ANNs) have got a few consideration from a part of analysts in this region due to its adaptability in information modeling. Our study proposes new model for electric load forecasting is proposed.

• Develop a graphical user interface (GUI) based load forecasting models.

• Load forecasting for long term for LDCs of India

• Improve the accuracy of the load forecasting using different AI techniques

Scope

Despite the significant developments in computational optimization algorithms, fast and stable real-time applications in power distribution systems have yet to be achieved by traditional mathematical methods. To escape computational difficulties like bad-conditioning and cohesion difficulties, significant efforts are needed. The present study will focus on the power output forecasting in a new method and a more efficient way and improve the input data with small deviations, which can make the simulation results more accurate with the help of artificial neural network (ANN).

Methodology

The methodology used for this work is forecasting a load pattern of 5 Regional Load Despatch Centres (RLDCs) and the National Load Despatch Centre (NLDC) using ANN algorithms that has a place to the method of Information Revelation and Data Mining. The stages included in carrying out the assignment incorporate the taking after:

Data gathering:

To  gather/collect  the  preliminary data of load and weather from the RLDCs and NLDC for    practical implementation. The parameters found in this datasets are as follows: Date, Time (hourly record), Temperature for 24 hours daily, Input voltage and Output voltage. The input voltage is in coming voltage or load into the Transmission Company which is high voltage of 132kv and before it is stepped down into the lower voltage levels of 33kv/11kv/0.415kv.

Pre-processing:

This stage involves datasets preparation before applying data mining techniques. At this stage, data cleaning, data transformation and data partitioning were employed as data pre-processing methods. Modelling accurate load forecasting model for load dispatch centres (LDC) of Indian region is the basic objective of the proposed project. Load forecasting is a burning issue in the power industry where the existing or currently used high error yielding solutions require replacement. In this project detailed analysis of all the factors like temperature, humidity, day of the week (whether holiday or working-day), tariff structure etc will be taken into consideration. Different Artificial intelligence techniques like Artificial Neural Network (ANN), Support Vector Machines (SVM), Decision Tree (DT), fuzzy logic etc. will be explored to develop intelligent load forecasting schemes and the performance of these schemes will be compared.

Data Mining:

 At this stage, data mining algorithms were applied in order to forecast electric load. In doing this, two ANN algorithms (MLP and RBF) and SMO algorithm are employed and compared.

Interpretation:

 At this stage, the results of models obtained were analyzed to determine the patterns in the load-forecasting model.

The main steps include:

• Develop long term load forecasting models to estimate yearly loads for Indian states/regions using artificial intelligent techniques.
• Gathering the historical load data daily/weekly/monthly/yearly loads from SLDC Gathering the historical weather data such as temperature/humidity/rainfall from weather forecasting department.
• Gathering data related to the day of the week whether it is holiday or working-day etc. Minimize the existing margin of error in the existing methodologies like regression, similar day etc., by adopting new Artificial intelligence based solutions like Artificial Neural Network (ANN), Support Vector Machines (SVM), Decision Tree (DT), fuzzy logic etc.
• Develop a module that should be integrated to RLDC scheduling modules and also NLDC report module to automatically fetch the data in order to keep the algorithm updated according to the daily changes that are happening in the power system.

Outcome

A load forecasting model is proposed by combining 24 regression models, with six of the most effective models being evaluated further in the study. A nonparametric kernel-based probabilistic model such as Gaussian Process Regression (GPR) can be useful for forecasting load demand, according to the study.Unlike other models with functional form constraints, GPR is able to provide information about consumption trends and do statistical interpolation by combining the parameters of all admissible functions. Due to the fact that exponential GPR algorithms are computationally inexpensive, generate patterns based on the average and standard deviation of a value, and are computationally inexpensive, the study recommends using them for optimal load forecasting efficiency. The study evaluates the model's accuracy using mean absolute percentage errors (MAPE) and R-squared validation techniques. The study found that the GPR model was able to accurately forecast load with a relatively low MAPE and R-squared value. Therefore, the study suggests that GPR algorithms can be an effective tool for load forecasting.

Gochhait, Saikat, and Deepak Sharma. 2023. “Regression Model-Based Short-Term Load Forecasting for Load Despatch Centre”. Journal of Applied Engineering and Technological Science (JAETS) 4 (2):693-710. https://doi.org/10.37385/jaets.v4i2.1682.