Global efforts to reduce carbon emissions and promote sustainable development have made solar energy integration into contemporary power systems more and more popular. For grid operators and energy planners, however, the intrinsically erratic and weather-dependent character of solar radiation poses a serious problem. Optimizing photovoltaic (PV) power generation, maintaining grid stability, lowering reserve capacity, and enhancing energy management techniques in smart grid frameworks all depend on accurate short-term solar radiation forecasting. For the purpose of short-term solar radiation forecasting using meteorological parameters, this study provides a thorough comparative analysis of several Artificial Intelligence (AI)-based models. These models, which include K-Nearest Neighbors (KNN), Random Forest (RF), Extreme Gradient Boosting (XGBoost), Support Vector Regression (SVR), and Long Short-Term Memory (LSTM) networks, are a combination of both conventional machine learning algorithms and cutting-edge deep learning techniques. A wealth of historical meteorological inputs, including temperature, relative humidity, wind speed, and cloud cover—all of which are important factors in determining the variability of solar irradiance—were used to train these models. To guarantee data quality, extensive preprocessing methods were used, such as temporal alignment, normalization, and handling of missing values. To improve the predictive power of the models, temporal feature engineering was also used to capture seasonal and diurnal variations in solar radiation. To ensure a fair comparison, every AI model was trained and evaluated using the same experimental setup. Standard error metrics, such as Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Coefficient of Determination (R2), were used to assess the model's performance. These metrics shed light on each model's forecasts' dependability and accuracy. The LSTM model consistently outperformed all other models across a variety of forecast horizons and environmental conditions, according to the comparative study's findings, even though more conventional machine learning models like Random Forest and XGBoost did fairly well. The non-linear and time-dependent nature of solar radiation was better modeled by LSTM networks, which are specifically made to capture temporal dependencies in sequential data. The model generated extremely accurate forecasts by skillfully utilizing the temporal structure present in the meteorological data. This emphasizes how crucial it is to apply deep learning architectures to time series forecasting issues in the energy sector. The results of this study highlight how important sophisticated AI models—in particular, deep learning methods like LSTM—are to improving the forecasting accuracy of solar radiation forecasting systems. These developments have the potential to greatly improve the planning and operation of smart grids that integrate renewable energy sources. By comparing the advantages and disadvantages of well-known AI models, the study also establishes a standard for further research in solar energy forecasting. Ultimately, by offering dependable instruments for incorporating renewable energy sources into the electrical grid, this work advances the larger goal of moving toward cleaner energy systems.