Fast Loss Allocation Using Shapley Values and Neural Networks

• Main Authors: Cheng-Fu Chang, 張丞甫
• Other Authors: Shih-Chieh Hsieh
• Format: Others
• Language: zh-TW
• Published: 2007
• Online Access: http://ndltd.ncl.edu.tw/handle/68336026585651063463

碩士 === 義守大學 === 電機工程學系碩士班 === 95 === In a competitive power market, the main difficulty of allocating a component of system losses to an individual bilateral transaction or power provider is due to the highly nonlinear and non-separable properties of the loss function. The methodologies presented in the literature are complicated and time-consuming, and usually need some assumptions or approximations to get an approximated solution. Thus, a fast loss allocation method without any assumption or approximation is necessary for a system operator. However, the market participants, especially the electricity consumers, can not easily understand the details of computing process, but would like to get an effective and reliable predicted loss allocation for doing some market decision making. Hence, a simple and fast loss allocation tool or predictor, without complicated power system data, is appealing to the market players or consumers in a competitive electricity market. This thesis will utilize induction to derive a simple formulation based on Shapley-Value value to speedup the traditional method for both pool-based and contract-based power markets. In a pool-based market, we use an equivalent current injection for a generation bus and an equivalent constant-impedance model for a load bus; then find the voltage contribution to each bus and current contribution to each branch by each generation bus. In a contractual market, we use an equivalent generation-load current injection pair for each power transaction, and find its voltage contribution to each bus and current contribution to each branch. We then use the proposed formulation to compute the normalized Shapley values and line loss allocations of each transaction. A 6-Bus, a modified IEEE 14-Bus, and a modified IEEE 118-Bus power systems will be tested for both pool-based and contractual markets to show the applicability and feasibility of the proposed formula. Finally, comparision among the results of the proposed formula, the original Shapley-Value method, and simplified Shapley-Value method will also be made. Besides, for the market participants or electricity consumers in a pool-based market, we use back-propagation methodology of neural networks to train a set of line loss allocation calculator. Players can use the neural calculator to predict their loss allocation for decision making in the power market.