Research On Secondary Voltage Control Based On Adaptive Dynamic Programming Method

Accounting that secondary voltage control(SVC) plays a decisive role in mataining voltage level and enhancing voltage stability, it is becoming indispensable to the auto voltage control. At present SVC in widespread use in practical engineering projects. And the SVC is modelled in the control law by sensitivity, and the principle of the control law is the division of network into zones based on the weak coupling criterion between network nodes.To achieve better balance of reactive power outputs, a part of objective funtion uses the variance of reactive power coordination. Finally the main goal of SVC is to adjust the generators and to maintain the voltage profile inside a zone. There still need some improvements such as: the acuuracy of SVC is dependent on the estimation of appropriate sensitivity coefficients, there is a contradiction between balancing the reactive power output and enhancing the reactive power margin, the reactive power coupling between neighboring zones is control response of neighboring areas is neglected, the effective methods to design discreted SVC controller are less in practical engineering. Adaptive dynamic programming(ADP) is a nonlinear intelligence optimization method, which has a fast calculation speed after the training of ADP controller. So the ADP can be employed for the coordinated secondary voltage control in large scale power system. To improve control accuracy and enhance reactive power margin, as well as to balance the reactive power outputs of the generators and to overcome the pilot voltage oscillation, and to coordinate the capacitor/reactor with discrete character and the generator with continuous character, the research on theory and implementation has been done as follows:1)In order to enhance the voltage control accuracy, an ADP method was proposed to design a secondary voltage controller. The minimum cost are used to enhance the reactive power margin. The ADP controller is based on subarea division. And all the controllers are trained offline individually and then applied on-line. The utility function is the defined on the consideration of the deviation of pilot buses’ voltages with respect to their settings and the increments of generator terminal voltages at current stage. Next the cost-to-go function is constructed by accumulating utility function in future. A Bellman difference equation is derived on Bellman’s principle of optimality. The optimal controls satisfy the Bellman difference equation are solved by an iterative ADP technique. The ADP controller is composed of a critic neural network(CNN) and an action neural network(ANN), while the CNN and ANN are constructed by BP neural networks respectively. A growth of reactive load data during a certain time interval uses as the scenario for off-line training of ADP controller. The Bellman difference equation is served as the target function in training CNN, while the optimal controls derived from the iterative technique are used to train ANN. When the trainings in all sub-areas are finished, the ANN with fixed weights can apply to on-line control.2)In order to balance the reactive power outputs of the controlled generators, a part of utility function related to the balance is proposed, which is the sample variance of the outputs. The momentum factor was introduced to overcome the pilot voltage oscillation. Take account into the interval of ADP controller is short, the mean of the outputs at the previous stages are used to calculate the sample variance at the current stage. The utility function was built to minimize the voltage deviations of the pilot nodes with respect to their settings and balance of the reactive power outputs among the controlled generators, and hence the cost-to-go function was constructed by accumulating utility function in future. The optimal control, derived based on the optimal condition, can be divided into two parts: one part related to the balance can be directed calculated without iteration, another part related to the deviation is solved iteratively. The actual control to the generators is composed of the control at current stage and part of the one at previous stage. The reactive power output oscillation is decreased under the action of a damping effect produced by the momentum factor. As a result the pilot voltage profile becomes smooth.3)A novel mixed integer ADP algorithm with single network is proposed to co-ordinate the capacitor/reactor and the generator. Fist a Bellman equation is established. Then that the mixed integer Bellman equation is converted into the one with only discrete variables attributes to discretization of continuous terminal voltages. Following the convert, a binary particle swarm optimization(BPSO) method is employed to solve the Bellman equation with discrete variables. A technique for accelerating the convergence of BPSO is proposed: first continue the discrete variables, and then derive the optimal controls using conventional ADP method, and find all controls locate in which interval sections. Every section has two discrete points, and those two points constructs a subset. Parts of swarms are initialized using the subsets. Simulations on two real provincial power system illustrate that the voltage porfiles at pilot bus can recovery to their reference by the proposed method and the calculation time is reduced by using the acceleration technique.