To address the challenge of inaccurate electricity load forecasting in start-up enterprises caused by insufficient historical data, highly fluctuating production plans, and unstable electricity consumption patterns, an electricity load forecasting method for start-up enterprises based on conditional generative adversarial networks (CGAN) and bidirectional gated recurrent units (BiGRU) is proposed. Firstly, on the basis of an analysis of key internal and external factors affecting enterprise electricity consumption, a structured influencing-factor system is established, and a multivariate feature set including power distribution data, equipment energy consumption, production planning information, and meteorological conditions is constructed. Subsequently, multivariate features are fused with electricity load time-series data to form original samples, and data augmentation is implemented using CGAN. Finally, the augmented samples are fed into a BiGRU network for training, enabling the capture of complex bidirectional temporal dependencies and the realization of electricity load forecasting. Experimental results based on a case study of an aircraft engine maintenance start-up enterprise demonstrate that high prediction accuracy can be maintained under data-scarce conditions, with prediction errors significantly reduced compared to baseline models, thereby effectively validating the superiority of the proposed data and feature enhancement strategy. Theoretical and practical support is provided for electricity load forecasting in data-limited scenarios.