With the rapid development of the Energy Internet, higher accuracy is demanded in the load forecasting and classification of power systems. This study aims to explore load classification and forecasting methods based on neural network models to improve the efficiency and accuracy of load management in the Energy Internet. By deeply analyzing the load characteristics of the Energy Internet, this paper constructs a neural network model containing multilayer perceptrons, specifically targeting the classification and prediction of different types of load data. This study first establishes a comprehensive dataset by collecting and processing large-scale historical load data from the Energy Internet, including residential, industrial, and commercial electricity usage data. Then, we use these data to train the neural network model, enabling it to identify and predict consumption patterns of different load types. Furthermore, this paper also explores the impact of different network structures, activation functions, and training methods on the model's performance to determine the optimal model configuration. Experimental results show that our neural network model demonstrates high accuracy and reliability in load classification and prediction. Particularly in peak load forecasting and abnormal load identification, the model performs better than traditional statistical methods. This research not only provides an effective tool for load management in the Energy Internet but also lays a foundation for future research in smart grids and sustainable energy systems.