Search

Both technically and economically favorable operating states of voltage and reactive power are of interest from the point of view of efficient and reliable operation of high voltage grid systems. A special multicriterial optimization method is being offered which compromises various target functions. It facilitates a situation-dependent and weighted combination of the target functions and evaluates the influence of potential control variables on optimization targets basing on sensitivity coefficients. Furthermore the sensitivities are appropriate to preselect the effective actuating variables of the optimization run in advance. This enables in combination with a post filtering of small changes of actuating variables an essential reduction of the. control actions over the day without a substantial loss of performance.

The investigation of voltage/reactive power phenomena in large-scale power systems and the development of improved voltage control strategies require rather sophisticated simulation models, so that margins of power transmission and system limitations of real systems are correctly reproduced. For a large system this modeling has to be done as a compromise between the necesserary aggregation of parts of the complete system model and the model accuracy. An engineering-based methodology for such a modeling of the voltage/reactive power behaviour of large power systems is described. The resulting simulation model allows the realization of the usual voltage commands and allows detailed operating studies of voltage problems.

For a given power system a method is presented which for each week of a year or a subperiod of it allows to determine firstly the power reserve necessary for a definite reliability index, secondly the weeks with a deficiency of power reserve, and thirdly new maintenance schedules which guarantee to avoid or at least to decrease the deficiencies of power reserve. For rescheduling of maintenance two ways are derived applicable for short respectively large deficiencies of power reserve.

A hybrid pattern algorithm is presented combining statistical and neural methods to forecast hourly load of an electrical power supplying system. Compared with ordinary neural techniques which require a large stationary data-set for the parametrization of the huge number of net-weights, the algorithm yields to a sufficient prediction even with a small reference data-set and is especially suited for power utilities with instationary load patterns. In this sense the choice of appropriate model structures and parsimonious parametrization are considered in particular. The presented modular design yields to a high transparency of the entire prediction algorithm. Furthermore a clear assessable performance measure of the prediction accuracy of the four individual steps of the forecasting algorithm is presented.

In large interconnected electric power systems the bus voltages of the feeding nodes are to be maintained at local optimum values. To avoid large information transmission networks this control task must be carried out by a decentralized control. bach control station has to ensure asymptotic regulation for the corresponding feeding node. Moreover, the controller parameters should be chosen so that the closedloop system remains stable even if several control stations are disconnected. By me ans of a necessary and sufficient condition for the existence of such controllers and the data of a real-existing power system with 20 feeding no des it is proved that vOltage control can actually be carried out in a decentralized way. As a byproduct, it is shown that this result holds for all systems with a ‘near-ideal’ voltage-reactive power behaviour.

The maintanance of an optimal voltage profile in an interconnected power system is necessary for minimizing the transfer losses and for the system security. It can be done by a decentralized control using local bus voltage controllers. Necessary and sufficient conditions for the existence of such a control will be given and proved for a larg5 real power system. Therefore a state space model of the power system will be developed by suitable aggregation and decomposition steps. This model will also be used for the design of robust local voltage controllers by the Generalized Direct Nyquist-Array Method. For the proof of the robustness a specially developed method will be applied. The result is a decentralized control with a good performance in all normal operation conditions of the power system.

Identifiability of closed loop subsystems in large scale systems is considered. Here in contrast to separate closed loop systems all outer signals influencing the closed loop may be correlated. For the general case that all outer signals are correlated the corresponding solutions for the determination of the process dynamics according to the direct method, the indirect method and the correlation method are derived. It is shown that only some definite correlations of the outer signals must be known a priori. Moreover, it is shown that the solutions obtained for uncorrelated outer signals can also be applied to the identification of closed loop subsystems successfully if in addition definite conditions on some of the correlated outer signals are fulfilled. The results obtained are of practical relevance for an optimal design of identification experiments in large scale systems.

For linear time invariant MIMO closed loop systems a derivation of the correlation method in the frequency domain is given first. It is shown that in contrast to the indirect method and to the direct method the correlation method permits as well the determination of the unknown frequency response matrices of the plant and the regulator as the determination of the unknown spectral density matrices of the noise signals influencing the loop. This is denoted by the term “complete system-and signal identification”. Then for the indirect method, the direct method and the joint process method the solutions for the frequency response matrices and the spectral density matrices which can be determined from the measured signals only are summarized. Finally the interactive identification package IMSCLF containing the four different identification methods is described in a short manner.

For linear time-invariant MIMO closed loop systems two ways for identifying a frequency response model are compared. First the determination of a parametric model in form of a vector difference equation by the method of correlation and mean square estimation and the transformation of this model in a corresponding frequency response model is given. Then the determination of such a frequency response model by the direct way of the correlation method in the frequency domain is shown. Subsequently the simulation results obtained for the same data are discussed and the merits of both the ways are listed. It is shown that the two ways give comparable accuracy results sufficient for the application of the modern design approaches in the frequency domain.

The paper represents an algorithm for the estimation of multivariate ARMA models. A orthogonalization procedure proposed by Phadke enables a decomposition of the extremely high dimensional estimation problem in subproblems of lower dimensions. Some problems of this method resulting in difficulties in its practical use are discussed. By means of a further decomposition in subproblems with maximal two inputs and one output and by an appropriate prefiltering the whole estimation problems splits in relatively simple, more practicable parts. The resulting successive determination of the submodels is especially effective in a direct dialog with the computer.

For hierarchically structured systems consisting of linear subsystems an approach for the determination of strategies for the subsystem and noise signal identification is given. This approach is based on the solvability of the matrix equation system formed by the spectral density matrices of the input and output signals. It is applied to a hierarchically structured system consisting of three levels. First it is shown that for the identification of closed loop subsystems only the direct method and the correlation method should be used. Then for some selected subsystems and noise signals unknown in the system under consideration corresponding strategies are derived. The results obtained for different signals measured and a priori information known are discussed under the point of view of an optimal design of the identification experiments.

In the paper two different identification situations being of practical interest are described first. Then some open problems related to it are summarized. Subsequently for hierarchically structured systems consisting of linear subsystems an approach is given for the determination of identifiability conditions guaranteeing a successful identification of the subsystems and the disturbances. These conditions are only based on the signal vectors which must be measured and on the a priori information given. They allow an optimal experiment design. As a further consequence of the identifiability results favourable identification strategies can be derived. Finally this is shown for a special class of large scale systems

For two and three state Markov models the outage probabilities P(t) depending on the transition rates are given first. Then the maximal deviations in the outage probabilities caused by the partial and postponable outages and an upper bound by which those outages can be neglected are derived. Subsequently for a power system consisting of 52 generating units of different capacities the effect of the partial and postponable outages on the power reserve necessary for a reliability index given is calculated. The results obtained for various reliability indices and transition rates show that the postponable outages are negligible and that the partial outages should be taken into consideration.

A suitable measure of error is described for the determination of the necessary accuracy of analysis of non-parametric mathematical models of multivariable plants. For this measure of error, the relationship between the magnitude of error and effect upon the closed loop is established. These relations enable the computation of the required accuracy of analysis to be made for the purpose of experimental analysis or mathematical modeling of the plant concerned.

The effioienoy of the method of correlation and least squares estimation with the identification of the frequencyresponse matrix of the plant in multivariable olosed loop systems is investigated. Assuming a suitable oanonical model of the veotor difference equation approximating the behaviour of the original oontinuous plant, the coeffioients of this veotor difference equation and a parameter minimal structure of the model can be found. By simulation of different plants, the quality of estimation was compared depending on the effeotive value and frequency speotra of the disturbance signal, the implemented regulators and the sampling rate in measurement. In result of the investigations the method has been proved suitable for identification in closed loop systems, giving satisfactory estimation results.

The dispatcher control of large energy systems today needs a decision support by advisory systems. In the paper the development of an advisory system tor the control of active power and frequency in a national power pool will be described. A hierarchically structured process model of the energy system is formed by different steps of decomposition and aggregation to simulate the static and dynamic behaviour of the system. It is realized as a real time interactive computer simulation. Furthermore an advisory system was developed for supporting the dispatcher in the national dispatching centre in controlling the national power pool. Its principle is a modified predictive controller described in this paper. The advisory system has been used for the training of the dispatcher staff of the national system and will be introduced in the daily system operation control.

In many cases an effective computer-aided daily load forecast for electric power systems exists already. Therefore the short-term load prediction can be reduced to the prediction of the rest component representing the deviation of the daily forecast from the real load. By the analysis of the data is obviously that already the use of the momentary value of the rest component decreases the prediction error substantially in comparison to the daily prediction. The paper deals with the question wether or not the accuracy of the prediction may be improved essentially by means of AR-models. Consequences of the practical realization of the short-term prediction are discussed.

In many cases an effective computer-aided daily load forecast for electric power systems exists already. Therefore the short-term load prediction can be reduced to the prediction of the rest component representing the deviation of the daily forecast from the real load. By the analysis of the data is obviously that already the use of the momentary value of the rest component decreases the prediction error substantially in comparison to the daily prediction. The paper deals with the question wether or not the accuracy of the prediction may be improved essentially by means of AR-models. Consequences of the practical realization of the short-term prediction are discussed.

For linear time invariant MIMO closed loop systems a derivation of the correlation method in the frequency domain is given first. It is shown that in contrast to the indirect method and to the direct method the correlation method permits as well the determination of the unknown frequency response matrices of the plant and the regulator as the determination of the unknown spectral density matrices of the noise signals influencing the loop. This is denoted by the term "complete system- and signalidentification". Then for the indirect method, the direct method and the joint process method the solutions for the frequency response matrices and the spectral density matrices which can be determined from the measured signals only are summarized. Finally the interactive identification package IMSCLF containing the four different identification methods is described in a short manner.