Abstract: The surface movement vector is used to solve the problems of instability, initial value dependence, difficult selection of optimization index, non-rectangular panel and difficult invesion of multi-panel for parameter calculation of probability integral method (PIM). The model uses the minimum squared error of surface movement vectors as the index and the genetic algorithm (GA) as the core optimization inversion algorithm to calculate PIM parameters. Using the surface movement vector can solve the problem of different results obtained when using the mining subsidence and horizontal movement separately, and assess the accuracy of the results. GA can solve the parameter inversion problem for non-rectangular working panel and multi-panel of surface movement observation station. The vector inversion model has no particularly stringent requirements for observation station setting and it reduces the calculation error caused by improper design observation station. To solve the problem of different results for using GA repeatedly, a combination forecasting method of weighted average results is established, which can obtain only one result at the same time so as to improve the reliability and stability of the model. The surface vector GA inversion model avoids a number of shortcomings of the traditional methods, and it is computationally efficient and easy to combine with the existing mining subsidence prediction program. The research results provide a new way to solve the engineering problem of parameter inversion for mining subsidence observation station influenced by non-rectangular multi-panel.