With the widespread application of deep learning in power load forecasting, its nonlinear modeling capability has significantly improved prediction accuracy. However, while pursuing high precision, deep learning techniques face the problem of a large number of model parameters, leading to low computational efficiency and difficulty in meeting the deployment requirements of high real-time scenarios and resource-constrained environments. To address these issues, this paper proposes a lightweight power load forecasting model based on adaptive knowledge distillation, aiming to achieve a superior balance between prediction accuracy and computational efficiency. First, in the teacher model construction phase, a deep information extraction module is designed to fully explore the deep-level features of load data, and a high-precision forecasting model is built by combining it with a long short-term memory (LSTM) network. On this basis, a more compact student model is further designed to perform load forecasting tasks with a small number of parameters. To enhance the knowledge transfer effect, a learning-based knowledge distillation method is proposed to achieve efficient knowledge transfer from the teacher model to the student model. Additionally, an adaptive knowledge correction strategy is introduced during the distillation process to further improve the learning capability of the student model. Simulation results demonstrate that the proposed method not only further improves prediction accuracy but also significantly reduces the number of model parameters (to only 1.36% of that of the original model), effectively achieving model lightweighting and overall performance enhancement, which validates the effectiveness and superiority of the proposed method in load forecasting tasks.