Due to the lack of causality between upstream and downstream industries in the existing load forecasting research, the accuracy of load forecasting between industries is poor. Therefore, this paper proposes an HPO-BiLSTM industry load forecasting method based on causal analysis. Firstly, convergence cross-mapping algorithm is used to identify the causal relationship between ferroalloy industry and related industries, and typical industries with strong causality are selected. Secondly, the original load data is decomposed by an improved adaptive noise complete empirical mode decomposition method. Finally, the hunter and prey algorithm is used to optimize the bidirectional long and short term network, and the decomposed load data is predicted. Finally, the accuracy of the proposed method is compared with that of XGBoost and LSTM prediction models. Experiments show that the proposed method can effectively extract the relevant features of different industries in the load prediction of ferroalloy industry. The prediction accuracy is significantly improved, which provides important support for rational planning and stable operation of urban power grid.