A Study on Project Cost Forecasting Based on LSTM and Multi-Source Data Fusion

Project cost forecasting constitutes a core element of engineering management, with its accuracy directly impacting project decision-making and resource optimisation. Addressing limitations inherent in traditional forecasting methods—such as reliance on single data sources, loss of temporal information due to static modelling, and poor adaptability—this study proposes a project cost prediction model integrating LSTM (Long Short-Term Memory) with multi-source data fusion. First, multi-source heterogeneous data relevant to project costs (resource pricing data, schedule progress data, market environment data, policy regulation data) are integrated. Feature-layer fusion strategies achieve data dimensionality optimisation and information complementarity. Second, the gating mechanism of LSTM neural networks captures long- and short-term temporal dependencies in cost data, resolving the vanishing gradient problem inherent in traditional neural networks. Finally, newly generated data throughout the project lifecycle are incorporated in real-time to enable cost forecasting. Simulation experiments employed 50 large-scale construction projects as research samples, selecting GM (1,1) grey prediction, BP neural networks, and single-data-source LSTM as comparative algorithms. Evaluation metrics comprised Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE). Results indicate that the proposed model reduces MAE, RMSE, and MAPE to ¥23,700, ¥31,200, and 1.89% respectively, representing reductions of 62.8%, 58.7%, and 65.3% respectively compared to the traditional GM(1,1) model, and reductions of 28.5%, 25.3%, and 30.1% respectively compared to the single-source LSTM. This validates the model's significant advantage in predictive accuracy. This research provides a novel technical pathway for precise project cost forecasting, holding substantial theoretical and practical value for enhancing the intelligence of engineering management.