Study On Dynamic Economic Dispatch Considering Stochastic Wind Power Integration

As a representative of the suatainable energy, the large-scale development and utilization of wind power makes the structure, operation and contorl of power systems changing fundamentally and brings new challenges. It prominently presents that the randomness of wind power, low level of short-term wind forecasting and its negative influence to power system’s dispatching, and so on. Traditional dispatch methods are not applicable for systems with large-scale wind integration. Therefore, studies on models and algorithms of stochastic optimal dispatch become imperative. This paper summarizes three main aspects: uncertainty analysis of wind power, modeling of dynamic economic dispatch with wind power integration and algorithms to solve these models. Details are given as follows:1) We present a multi-objective stochastic economic dispatch with variable wind generation using scenario-based decomposition and asynchronous block iteration. Multi-objective stochastic economic dispatch problem could be transformed into a series of single-objective optimization problems, which can be solved using the normal boundary intersection method and interior-point method. During the iterations while solving each single-objective optimization problem, the coefficient matrix of the correction equation could be rearranged in the block bordered diagonal form according to the sequence of the forecast scenario and sampling scenarios. Thus, this correction equation can be decomposed further into a number of low-dimensional equations corresponding to the forecast scenario and sampling scenarios, respectively. To enhance the computational ability to adapt large-scale systems with massive scenarios, the asynchronous block iteration method was used when decomposing the reduced correction equation. With the increase of scenarios in our model, power generation output is more and more adaptable to the uncertainty brought about by the integration of wind power. When this model was used on a real provincial power system, 1,000 sampling scenarios were generated by Monte Carlo sampling, the number of variables reached 16,432,417, the number of inequality constraints reached 115,319,756, and the number of equality constraints was exactly 99,101. Even though the loop of “solving model → verifying all branches’ limits → adding overloaded branches’ limits to model”(so-called “while-loop detection method”) was used to handle the network constraints, the number of inequality constraints was still 49,686,237. Nevertheless, the size of this non-linear programme was very large. We performed a scenario-based decomposition to solve this single-objective problem with high performance computing clusters. The computation time reduced to 1~3.7h. Similarly, different Pareto optimal solutions can also be parallel computed. A double-layer parallel computational framework is built on high-performance clusters to make obvious enhancements in computational speed.2) We present a correlation analysis of wind power based on mixed copula model and its application into stochastic dispatch. Outputs of multiple wind farms have temporal and spatial correlations. Different Copula function can capture characteristics of symmetry, asymmetry and tail dependence from different data. Therefore, convenience can be brought to grid’s stochastic dispatch when we use Copula functions to analysis these correlation information and generate wind power scenarios of multiple wind farms. An error scenario generation method is proposed using mixed-Copula function model. It is verified that the mixed-Copula function has better performance than the single Copula function in fitting accuracy when dependent parameters are selected reasonablely. The new method is beneficial to absorb more wind energy through dispatching than the traditional method when the power system is integrated multiple wind farms. The traditional method take forecast error of wind power as normal distribution.3) Comparison study on 6 typical modeling methods of stochastic dynamic economic dispatch with wind power integration, and analysis on computation size, speed and optimal performance are used to them. These models are deterministic model considering the up/down spinning reserve constraints, stochastic model with chance-constrained programming, stochastic model considering probabilistic optimal power flow, scenario-based deterministic model, deterministic model considering extreme scenario constraints and deterministic model with sample average approximation method. The simulation results show that a fast solution to deterministic model considering the up/down spinning reserve constraints, stochastic model with chance-constrained programming, stochastic model with probabilistic power flow or deterministic model considering extreme scenario constraints is enough when the dispatch center has low-level compiling environments.; An accurate solution with high-performance computing clusters can be performed through scenario-based deterministic model or deterministic model with sample average approximation method.