With the advancement of new-type power systems and the implementation of the "cost reduction and efficiency improvement" strategy, accurate prediction of total investment in engineering projects has become an essential approach to enhancing enterprise management efficiency. Targeting the high-dimensional, nonlinear, and multivariable characteristics in substation project cost forecasting, this study develops an intelligent prediction model that integrates Grey Relational Analysis (GRA) with a Sparrow Search Algorithm (SSA)-optimized Backpropagation Neural Network (BP), referred to as the SSA-BP model. Using fishbone diagram analysis and GRA, 20 key technical indicators with correlation values greater than 0.8—such as single-unit capacity, number of low-voltage circuit breakers, quantity and unit price of control cables—were selected as model inputs. The model was trained and tested using five years of substation engineering data from a province. Results show that the model achieves an average relative error of 4.85%, with a maximum error of 14.12% and a minimum error of 0.11% across 30 test samples. Compared to conventional BPNN, PSO-BP, and WOA-BP models, the SSA-BP model exhibits improved prediction performance, with RMSE, MAE, and MAPE reduced to 545.78, 367.18, and 6.56%, respectively, demonstrating strong practical value in supporting investment decision-making for engineering projects.