Evaluation and prediction of power network system robustness are conducive to system managers' ability to perceive the current status of network system operations and take timely measures to cope with potential risks. Therefore, a robustness prediction model for power dispatching data network is proposed based on an improved whale optimization algorithm. Firstly, an evaluation index system for the robustness of the power dispatching data network is constructed and data dimensionality reduction processing is designed based on methods such as field extraction and formula mapping. Moreover, an improved chaotic mapping and adaptive weight-whale optimization algorithm （CA-WOA-BP） is proposed based on chaotic mapping and adaptive weights （WOA-BP） and a power network robustness prediction method is established. Experimental results show that compared with the WOA-BP algorithm, the proposed improved algorithm speeds up the convergence of the prediction model while overcoming the situation of falling into local optima and reduces the absolute error percentage of the predicted values by 5.3 %, which helps users to discover the robustness of power dispatching data network systems more accurately and timely.