With the development of power grid big data and industrial automation, the number of nonlinear factors influencing load fluctuations is increasing, complicating accurate prediction. This paper addresses predicting short-term electricity load in industrial parks using a Temporal Convolutional Network (TCN) model. In the training phase, Quantum Particle Swarm Optimization (QPSO) tunes hyperparameters to improve accuracy. In quantum space, particles search globally, optimizing predictions. For data preprocessing, artificial features like date and weather that influence fluctuations are incorporated. Additionally, AutoEncoder and Principal Component Analysis (PCA) are used to extract load data features. Finally, feature engineering methods are used to select highly correlated inputs, enhancing model learning and interpretability. Simulation results confirm the proposed method significantly improves prediction accuracy compared to traditional industrial park short-term forecasting methods.