| This thesis is motivated by the recognition that sources of uncertainties in electric power systems are multifold and may have potentially far-reaching effects. In the past, only system load forecast was considered to be the main challenge. More recently, however, the uncertain price of electricity and hard-to-predict power produced by renewable resources, such as wind and solar, are making the operating and planning environment much more challenging.;The near-real-time power imbalances are compensated by means of frequency regulation and generally require fast-responding costly resources. Because of this, a more accurate forecast and look-ahead scheduling would result in a reduced need for expensive power balancing. Similarly, long-term planning and seasonal maintenance need to take into account long-term demand forecast as well as how the short-term generation scheduling is done. The better the demand forecast, the more efficient planning will be as well. Moreover, computer algorithms for scheduling and planning are essential in helping the system operators decide what to schedule and planners what to build. This is needed given the overall complexity created by different abilities to adjust the power output of generation technologies, demand uncertainties and by the network delivery constraints.;Given the growing presence of major uncertainties, it is likely that the main control applications will use more probabilistic approaches. Today's predominantly deterministic methods will be replaced by methods which account for key uncertainties as decisions are made. It is well-understood that although demand and wind power cannot be predicted at very high accuracy, taking into consideration predictions and scheduling in a look-ahead way over several time horizons generally results in more efficient and reliable utilization, than when decisions are made assuming deterministic, often worst-case scenarios. This change is in approach is going to ultimately require new electricity market rules capable of providing the right incentives to manage uncertainties and of differentiating various technologies according to the rate at which they can respond to ever changing conditions.;Given the overall need for modeling uncertainties in electric energy systems, we consider in this thesis the problem of multi-temporal modeling of wind and demand power, in particular. Historic data is used to derive prediction models for several future time horizons. Short-term prediction models derived can be used for look-ahead economic dispatch and unit commitment, while the long-term annual predictive models can be used for investment planning. As expected, the accuracy of such predictive models depends on the time horizons over which the predictions are made, as well as on the nature of uncertain signals. It is shown that predictive models obtained using the same general modeling approaches result in different accuracy for wind than for demand power.;In what follows, we introduce several models which have qualitatively different patterns, ranging from hourly to annual. We first transform historic time-stamped data into the Fourier Transform (Fr) representation. The frequency domain data representation is used to decompose the wind and load power signals and to derive predictive models relevant for short-term and long-term predictions using extracted spectral techniques. The short-term results are interpreted next as a Linear Prediction Coding Model (LPC) and its accuracy is analyzed. Next, a new Markov-Based Sensitivity Model (MBSM) for short term prediction has been proposed and the dispatched costs of uncertainties for different predictive models with comparisons have been developed. Moreover, the Discrete Markov Process (DMP) representation is applied to help assess probabilities of most likely short-, medium- and long-term states and the related multi-temporal risks.;In addition, this thesis discusses operational impacts of wind power integration in different scenario levels by performing more than 9,000 AC Optimal Power Flow runs. The effects of both wind and load variations on system constraints and costs are presented. The limitations of DC Optimal Power Flow (DCOPF) vs. ACOPF are emphasized by means of system convergence problems due to the effect of wind power on changing line flows and net power injections. By studying the effect of having wind power on line flows, we found that the divergence problem applies in areas with high wind and hydro generation capacity share (cheap generations). (Abstract shortened by UMI.). |
