The demand for accurate short-term electricity load forecasting within the power innovation ecosystem has driven the exploration of advanced technologies, with deep learning algorithms emerging as a highly promising solution. In this paper, we discuss the necessity of developing a robust short-term electricity load prediction model tailored to the dynamic nature of the evolving power innovation ecosystem. By integrating deep learning algorithms into the power industry’s innovation ecosystem, we aim to enhance the efficiency, reliability, and sustainability of electricity load forecasting. Leveraging the advantages of convolutional neural networks and recurrent neural networks within deep learning, we propose a Convolutional Neural Network–Bidirectional Gated Recurrent Unit (CNN-BiGRU) hybrid neural network prediction model based on the attention mechanism. Convolutional neural network (CNN) is utilized to extract predictive features and establish the correlation of load data in a high-dimensional space, constructing high-dimensional feature vectors for time-series sequences. These are then input into a Bidirectional Gated Recurrent Unit (BiGRU) network to obtain prediction results. Simultaneously, the Convolutional Block Attention Module (CBAM) attention mechanism is integrated into the CNN network, while a self-attention mechanism is added after the BiGRU network—enhancing the model’s ability to focus on important information within the load features, effectively improving its performance in fitting temporal data. Through the design and implementation of a deep learning-based prediction model, we demonstrate the potential for transformative progress within the power industry. Our research findings underscore the importance of adopting advanced technologies to drive innovation and address the multifaceted challenges within the power innovation ecosystem.