Dynamic analysis for employment substitution effect of robots fused with deep learning models and artificial intelligence

The purpose of this study is to address the problem of separation between temporal dynamics and network structure in analyzing the employment substitution effect of robots. The study proposes a dynamically coupled model integrating Graph Convolutional Network (GCN) and Bidirectional Long Short-Term Memory (Bi-LSTM). In the temporal dimension, the framework adopts Bi-LSTM to capture the forward and backward dependencies of technological penetration. Meanwhile, in the structural dimension, this study introduces GCN to model the skill correlation network between jobs. The core innovation lies in designing a gated fusion mechanism, which dynamically integrates temporal features and network embeddings at each time step to achieve a refined simulation of the interaction process. The experiment is based on panel data of Chinese manufacturing enterprises from 2018 to 2022, covering 327 samples. Results show that the proposed model achieves an accuracy of 0.912 and an Area Under the Curve value of 0.928 on the test set, significantly outperforming single Bi-LSTM, GCN, and linear regression models. A key finding is that the model can accurately identify pre-substitution states with a prediction accuracy exceeding 89%, providing a 6-12-month early warning window for policy interventions. Gating analysis reveals that the contribution rates of temporal dynamics and skill networks are 48.3% and 42.7%, respectively, verifying the existence of a dual-drive mechanism. The study confirms that AI substitution is not an immediate event but a gradual process involving multiple stages. This model provides a new paradigm for understanding the interaction between technological change and the labor market, shifting policy formulation from passive response to proactive guidance.