Page 217 - 2024-Vol20-Issue2

213 | Gochhait, Sharma & Bachute

TABLE III. TABLE V.
INTERPRETATION OF CORRELATION FOR THE TABLE OF LAYERS AND OPTIONS FOR ODISHA STATE

PARAMETERS OF DELHI STATE Layer Options
sequenceInputLayer ”Name”,”input”
Parameter Correlation Coefficient convolution1dLayer 11,96,’Padding’,’same’
T2MWET 0.686474222 batchNormalizationLayer
0.077530395 -
RH2M 0.588813736 reluLayer -
QV2M 0.022154842 convolution1dLayer 20,180,’Padding’,’same’
WS10M 0.659413161 batchNormalizationLayer -
T2M 0.569944321 -
T2MDEW reluLayer 30,300,’Padding’,’same’
convolution1dLayer -
Effective data pre-processing is crucial for accurate load fore- batchNormalizationLayer -
casting in power system management. Several steps are in- 32,320,’Padding’,’same’
volved in this process: Data Cleaning: This step involves reluLayer -
filling in missing values, removing noise, and detecting and re- convolution1dLayer -
solving outliers and discrepancies within the dataset. An auto- batchNormalizationLayer 0.2
fill feature can be utilized to fill partial missing weather data 100,’OutputMode’,’sequence’
using patterns or data from other cells. Data Transformation: reluLayer 0.1
Multiple files are integrated into a single usable format,and dropoutLayer 105,’OutputMode’,’sequence’
attributes are scaled based on specific properties. After clean- bilstmLayer 0.2
ing the data and identifying correlations between datasets, the dropoutLayer 110,’OutputMode’,’sequence’
final predictor’s dataset is created. Data Reduction: This step bilstmLayer 0.2
aims to reduce the number of attributes or sample data while dropoutLayer 1
capturing most of their properties. It helps prevent overfit- bilstmLayer -
ting and improves the efficiency of load forecasting. Overall, dropoutLayer
proper data pre-processing ensures reliable data and enhances fullyConnectedLayer
the performance of regressions, ultimately improving power regressionLayer
system management.
loss onthe validation set is calculated using the current model
2) Data Set Training and Testing status. It’s important to notethat the validation set does not
In the research, the focus lies on constructing and training a impact the training process, enabling us to assess themodel’s
robust prediction model.This process includes training with robustness on independent data. If the validation loss fails to
validation, which is crucial for ensuring a robust model. The converge whilethe training loss does, it indicates the model
data frame obtained through feature engineering is divided may suffer from overfitting or under fitting.The final step
into three sets:training, validation, and testing. Typically, the involves prediction and evaluation, where we predict time
data is split with 70% for training, 20%for validation, and series data and evaluate the results. The predictions are made
10% for testing. During the training phase, the prediction using the previously developed model.

TABLE IV. In certain cases, the predicted features can be utilized to
COMPARISON OF WEATHER PARAMETER CORRELATIONS predict other features, forming the final target feature set.
To achieve this, the dataset is updated with the predicted
BETWEEN ODISHA AND DELHI values, and a new model is created for further predictions. One
commonly used method for validation is holdout validation,
Parameter Correlation Odisha Correlation Delhi which is the sim- plest form of crossvalidation. In this method,
T2M 0.1537 0.6594 the data is randomly divided into two sets: the training set and
0.0889 0.5699 the test/validation set (hold-out data). The model is trained on
T2MDEW 0.1305 0.6865 the training dataset and evaluated on the test/validation dataset.
T2MWET -0.0614 0.0775 To assess the error on the validation dataset, various model
0.0948 0.5888 evaluation techniques can be employed, such as mean squared
RH2M 0.0981 0.0222 error (MSE) for regression problems or metrics indicating the
QV2M misclassification rate for classification problems. Typically,
WS10M the training process utilizes a larger dataset than the hold-out