The main objective of the automation on the hydro power plants is the control of water turbines. It provides automatic start-up procedure of water turbine usually connected with solid shaft to generator, online monitoring of the operation parameters and different control and safety functions (e.g. emergency shut down). In the doctoral thesis we focus on the predictive based control functions with fuzzy models that provide improved closed-loop response of the frequency and active power control in parallel as well as island regime of operation. The study is based on the 15 years of the author working experience on the field of the automation and measurements on the hydro power plants, results and recommendations of the recent research and standards of the discussed area of development, that continue the guideline of the further development. The control system of water turbine with its guide vane and eventually runner servomotors (in case of doubly regulated), changes angle of the blades and controls the water flow through the turbine. It results in the change of the frequency of generator, connected to the turbine with shaft, as opposed to the change of active power delivered to electrical grid, after the synchronization.
Modelling of the Francis and Kaplan water turbines and identification of the generator’s frequency depended on the guide vane servomotors opening on four different hydro power plants (equipped with the Francis, Kaplan and tubular turbines), reveals a non-minimum phase characteristic of water turbines in case of active power control. Furthermore, the generator’s frequency in the open loop could be unstable at larger changes guide vane’s servomotor opening and in the case of long penstocks could include poorly damped dominant poles that reflect in the underdamped transient responses. For both cases, the nonlinear process with the dead time is exposed, that is changing and consequently affects the robustness of the system. Additional issue is addressed with the transition to the island operation, where the system is subjected to higher frequency deviations and load disturbances.
The scientific literature addresses the considered issues from the aspect of transmission system operators. The main focus is on the load frequency control on large number of generation units and rather simplified models. Model predictive control is usually used in in terms of area control errors with different approaches. The most promising load frequency control cases emphasizes the centralized and decentralized approaches, multi-agents systems for active disturbance networks, online learning stochastic model predictive control for linear time-invariant systems with additive stochastic disturbance described with Dirichlete process mixed model, robust nonlinear model predictive control with the extended Kalman filter, robust model predictive control of load frequency control on multi-area power system considering the uncertainty with polytopic model, internal mode control with two degrees of freedom and two-layer active disturbance rejection controller.
Based on the different scientific approaches, an attempt to eliminate the nonlinear characteristics is applied with fuzzy logic control. With the inference mechanism and centroid method the sharp values of three membership functions of model variables (gain and two time constants) based on turbine power Pm, have been acquired for arbitrary operating point of water turbine. Undesirable effect of non-minimum response of turbine power was minimized with the internal model control and tuning parameter (time constant) that was optimized in three different operating points. The upgrade to the fuzzy IMC control determines tuning parameter in arbitrary operating point in middle sections, contributing to the less aggressive initial response of the actuator (guide vane) and smaller pressure spikes in the penstock in the entire operational range of the water turbine. Affirmative simulation results of the proposed algorithm are presented in the case of non-linear model of Francis turbine 1 with penstock and surge tank on HPP Moste, revealing lower undershoot of turbine power, smaller head changes at turbine admission and damped pressure oscillations in penstock, compared to the classical PI controller, which is often used in practice.
The approach to control the frequency of a generator was based on the parametric identification on turbine 1 and generator 1 on HPP Mavcice with an output-error model derived from autoregressive with external input model. It establishes basis for model prediction calculation and two algorithms of generalized predictive control. The derived control signals are written in discrete difference equations with the ability of straightforward implementation on PLCs. Besides higher computation demand, the algorithm 2 overtakes change of the reference signal to achieve minimization of error in transient response. Consequently, we expect smaller overshoots and shorter settling times compared to the algorithm 1. Derived controllable value is expressed in discrete difference equations, suitable for implementation in the general purpose industrial controllers. Calculated sample of controllable signal depends on two sequential past values and N values of reference signal that is right-shifted for the length of the input-output delay of the discrete model. At generalized predictive control, the length of the reference signal is the same as prediction horizon. In comparison with the fractional order PID and classical PID controller tuned by integral of time multiplied absolute error and integral absolute error, GPC outperforms classical controllers in terms of reference tracking and disturbance rejection. In the proposed approach, the response of the generator’s frequency is depended on the change of the predictive horizon up to N=15.
The research simulation of isolated operation mode, realized in ETAP environment includes complete model with subsystems of HPP Mavcice and generalized predictive control. Algorithm 2 with prediction horizon N=15 and control-weighting factor λ=1.2 was used in three different simulation scenarios. At simulated variable load with the changing active power (first scenario) and variations of the transmission length (second scenario), response on systematically changed disturbance and stability limit of operation was found. Simulation of transition to the real island operation with 25 buses, convincingly presents 50 % shorter settling time of frequency and 51 % shorter settling time of active power of the generator 1 HPP Mavcice as opposed to the PID control, tuned by the ITAE criterion. Proposed methodology enhances quality metrics of the control system in isolated operation mode and decreases probability of the island separation or break down of power system.
|
