Page 220 - 2024-Vol20-Issue2
P. 220
216 | Gochhait, Sharma & Bachute
Fig. 4. Hybrid model of CNN + LSTM.
model demonstrates accurate load forecasting for Delhi as the average RMSE was 0.864 over the same duration. These
well. RMSE values were lower compared to other machine learning
models and empirical methods examined in previous stud-
IV. RESULT ANALYSIS ies. The utilization of the 1D CNN-BI LSTM architecture
contributed to the improved performance of the model. Fig.
The comparative long-term electricity forecasting analysis of 2 and Fig. 3 present a graphical comparison between the
the states of Orissa and Delhi was conducted using a hybrid actual load values and the predicted load values for the fi-
1D CNN-BI LSTM model implemented in MATLAB R2023a. nal quarter of the datasets obtainedfrom Odisha and Delhi.
The model architecture consisted of a sequence input layer, These figures depict the accuracy of load forecasting using
four convolutional layers, two bi-LSTM layers, two dropout the 1D CNN-BI LSTM model, visually demonstrating its per-
layers, and a fully connected layer with a regression layer. formance. The results clearly show that the proposed hybrid
The training process utilized Adam optimization with a gra- model surpasses LSTM time series models and traditional
dient threshold of 1, an initial learning rate of 0.001, and a STLF models such as ARIMA and ANN-based methods, as
maximum of 10 epochs. The accuracy of the model’s load evident fromthe lower RMSE values. Furthermore, to vali-
forecasting was evaluated using the root mean squareerror date the effectiveness of the CNN-BI LSTM hybrid model,
(RMSE). a comprehensive comparison was conducted with other fore-
The results demonstrated that the hybrid 1D CNN-BI LSTM casting methods under various scenarios. Consistently, the
model achieved accurate load forecasting for both Orissa and proposed hybrid model consistently achieved significantly
Delhi. The model exhibited a relatively low RMSE, indicating lower RMSE values, indicating its superior forecasting capa-
a good fit to the data. In the case of Orissa, the average RM- bility in different situations. This comparison solidifies the
SEover a six-month period was 0.952. Similarly, for Delhi, model’s reliability and confirms its potential for accurate load
