Aiming at the problem of low accuracy of regional level total electricity load curve forecasting under the “dual carbon” goals, a forecasting framework that incorporates industry-specific electricity consumption characteristics and multi-factor influences is proposed. First, industry load curve clustering analysis and the construction of comprehensive electricity consumption evaluation indicators are used to qualitatively and quantitatively analyze the electricity consumption characteristics of different industries, demonstrating the necessity of refined, industry-level load profiling. Furthermore, the impact of various external factors, such as seasonal temperatures and day types, on industry-specific electricity usage patterns is quantified using nonlinear correlation coefficients and violin plot visualizations, laying a data foundation for fine-grained forecasting. Finally, a hybrid forecasting model is constructed by extracting electricity usage features via convolutional neural network (CNN), integrating them with the bi-directional long short-term memory (BiLSTM) network and the attention mechanism. This ensemble model is used to predict load curves for each industry, and an indirect forecasting approach is applied to reconstruct the total electricity load curve of the entire society. Using load data from eleven industries and residential electricity users in a region of East China, ten groups of comparative experiments are conducted. Results show that the proposed forecasting framework significantly reduces the prediction errors compared to traditional methods.