| The power flow as the basic analysis tool plays an important role in a smart grid. In order to combine the power flow with economy, safety and power quality precisely, the optimal power flow calculation has been proposed. Optimal power flow is preferred by people who plan, operate and schedule the power system. Thus it plays a very important role in power system planning, operation, analysis and control.Firstly, to deal with the sensitivity problem of initial guess in power flow based on Newton method, the seletion method about optimal initial value is proposed in this thesis. This method solves the divergence problem of power flow due to the improper selection of initial value. Secondly, considering the problems of sensitive to initial value and large amount of calculation by using the interior-point method, simplified gradient method and other methods, a generalized gradient projection algorithm based on the fisher function, which is used tosolve the optimal power flow, is proposed in this thesis. Finally, in order toincrease convergence rate and deal with discrete variables more easily, an optimal power flow methods based on double Hopfield neural network is proposed. The main works of this thesis are listed as follows:(1) A convergence criterion of NL power flow is proposed to judge whether the convergent solution of the power flow can be obtained through selecting the initial value properly. If the initial value is feasible, then according to the proposed maximum iterations estimation criterion, iterations of NL power flow can be estimated preliminary. According to the two abovecriterions, then it can be judged that whether initial value can make the power flow equation convergence before the calculation. The redundant computing of power flow can be avoided. A practical adjustment method based on GA and the proposed criterions is improved for selecting optimal initial value to overcome the guess problem. Aiming at the sensitivity problem about initial value of Newton method, the optimal initial value selction method is proposed combining two criterions by GA. The proposed criterions and GANL methods The analysis of IEEE node systems and the Tongliao power grid can be used to proved the effectiveness of the proposed method.(2) A generalized gradient projection algorithm based on the fisher function for solving the optimal power flow is proposed in this thesis. Compared with simplified gradient method, the power flow equations are not calculated in each iteration and the amount of calculation is greatly reduced. Avoiding the morbid condition number due to the penalty function improper selection, the dynamic selection of penalty function technique is used in this thesis. Compared with the interior point method, the range of initial value selection is enlarged. The oscillation process of convergence and slow process of optimizing can be avoided. The effetiveness of the proposed algorithm is verfied through the IEEE standard node system. The interior-point method, simplified gradient method and proposed method are applied to the Tongliao power grid respectively, together with their comparative analysis. The simulation results verify the quick convergence and stability of the proposed method based on a generalized gradient projection algorithm.(3) Solving optimal power flow calculation by using double Hopfield neural network double Hopfield neural network is divided into two networks:one network is used to minimize the cost function, finding the solution towards the feasible descent direction of the objective function. The other network is used to satisfy the constraints, making the solution in the feasible solution subspace. The two networks are independent and operated alternately. Double Hopfield neural network avoids contradiction between the satisfication of the constraint conditions and high quality solutions. The algorithm greatly improves the calculation rate. The simulation results of the IEEE standard node systems by double Hopfield neural network has been achieved better optimization results compared with Hopfield neural network. |
PDF Full Text Request