To ensure system stability, the fixed-PID (F-PID) controller with small parameters is usually adopted in hydropower stations. This involves a slow setting speed and it is difficult to realize optimal control for full working conditions. To address the problem, this paper designs a variable-PID (V-PID) controller for a hydraulic turbine regulation system (HTRS) based on the improved grey wolf optimizer (INGWO) and back propagation neural networks (BPNN). These can achieve excellent regulation under full working conditions. First, the nonlinear HTRS model containing the nonlinear hydro-turbine model is constructed and the stable domain is obtained using Hopf bifurcation theory to determine the available range of PID parameters. The optimal PID parameters in typical working conditions are then calculated by the INGWO, and the optimal PID parameters are generalized through training the V-PID neural networks which take the optimal PID parameters as sample data. The V-PID neural networks with different structures are compared to determine the optimal structure of the variable-PID controller model. The V-PID controller-based nonlinear HTRS model shows that the PID parameters can be automatically adjusted online according to the working condition changes, realizing optimal control of hydropower units in full working conditions.