Impact of Exogenous Variables on Short-Term Electricity Price Forecasting Accuracy: A Deep Learning Approach Using CNN-LSTM Networks in the Indian Energy Exchange

In deregulated electricity markets such as the Indian Energy Exchange (IEX), Day-Ahead Market (DAM) prices are driven by a complex interplay of internal market dynamics and external (exogenous) factors. While deep learning models have improved Short-Term Electricity Price Forecasting (STEPF), the selection of input variables remains a critical challenge. This paper investigates the impact of various exogenous variables— namely system load, ambient temperature, and renewable generation—on forecasting accuracy. A hybrid deep learning framework combining 1-Dimensional Convolutional Neural Networks (1D-CNN) and Long Short-Term Memory (LSTM) networks is employed to conduct an ablation study. Using two years of hourly data from the IEX, four input combinations are tested. Results indicate that a univariate model (historical price only) yields a Mean Absolute Percentage Error (MAPE) of 10.5%. The sequential integration of load, temperature, and solar generation data reduces the MAPE to 8.2%, 6.8%, and ultimately 6.1%, respectively. The study concludes that incorporating localized weather and renewable penetration data is vital for mitigating the impact of sudden price spikes in the Indian context.