Algorithm For Unit Commitment Based On Artificial Neural Network And Dynamic Programming

With the penetration of our country's reform and the continuous development of the economy, the load demand of the power system is increasing rapidly, so lots of generating units must be extended and the new generating units shall be built in the power system to accommodate the increasing load demand. In such situations, the dimensions of the interconnected operation generating units are expanding. But to the extent of its efficiency, the coal consumption rates of the units are different. Some of its rates are very high, others are low. In addition, the load is changing at any time, and exits peak and valley load in different period of time. So it's necessary to consider the unit commitment in drawing the short term generating scheduling according to the load and unit characteristics. From above, we can conclude that the unit commitment is to study how to commit the unit in the period of the time, and how to distribute the load in the units which is committed in order to minimize the coal consumption. This thesis makes systematic and thorough studies on the problem of unit commitment, tries to improve the method of the BP neural network and dynamic programming to meet the requirements of the unit commitment, reduces the condition number so that avoid the curse of dimensionality which caused by the dynamic programming.Based on comprehensive references, this thesis analyses the combination of units and the consumption rates based on the operating characteristics of the units in the power system, summarizes methods for unit commitment and the latest research results. Compared with the traditional model, this thesis looks the hydro power plant as a hydro unit, and takes into account the hydro unit and its cost in the optimization object, employs the low startup costs and flexibility of the hydro units to coordinate the hydro and thermal units. In this thesis, the unit commitment problem is divided into two steps according to the neural network and dynamic programming. The first step is to create the pre-schedule of the unit commitment based on the BP neural network, the second is to deal with the pre-schedule further based on the dynamic programming, find the ultimate unit commitment schedule. The Levenberg-Marquardt method is used to train the BP neural network .since its large scale memory usage, a revised method is presented based on it. Compared with the other training methods, it can be convergence in a few iterations and reduce the memory usages. Dynamic programming is revised in the selection and construction of the element which comprised in the window. Verified...