Research On Short-Term Load Forecasting And Optimization Of Economic Dispatch Decision-Making Of Electric Power System

As an important component of the urban lifeline engineering, electric power system plays a significant role in maintaining urban operational functions. Especially, with the rapid modernizing process of the society, the economic construction and national life are becoming more and more dependent on the electric energy, the electricity demand is growing very quickly, and the power customers'demand for the power supply quality and reliability is becoming higher and higher. As a result, the realization of the intelligent control of power system which can ensure the security stable and economic operation of electric power system, does not only help cities keep working orderly, but also help allocate resources properly and relieve the growing energy pressure. It is of great theoretical and engineering significance.This dissertation aims to realize the intelligent control of power system. Firs of all, it focuses on the improvement of the short-term load forecasting accuracy. For the factors which can cause forecasting errors, such as the systematic errors of models, the weights of the input indexes, the chaotic characteristics of the historical data, and the weights of the sample data, several forecasting methods are developed from different angles to improve the forecasting accuracy. The main research results are summarized as follows:(1) The short-term load forecasting method based on neural network with random weights and kernels and error correction is proposed. In order to reduce the systematic errors and improve the forecasting accuracy, for the short-term load forecasting task, an input index system is built firstly. Then the short-term load is forecasted based on the model of neural network with random weights and kernels. At last, the method of error correction is applied to correct and revise forecasting results. This proposed method can overcome shortcomings of traditional feedback forward neural networks based on gradient training, and reduce the redundant information in the forecasting errors based on traditional single forecasting models. It can improve the forecasting accuracy, training speed and generalization performance effectively.(2) The short-term load forecasting methods based on multivariate chaotic time series and weighted neural network with random weights and kernels is proposed. At first, taking into account different contributions of different inputs weights to forecasting results, the mutual information weighting algorithm is used to allocate different weights to the inputs. As a consequence, the short-term load forecasting method based on improved weighted neural network with random weights and kernels is presented. It can avoid the forecasting errors led by ignoring inputs weights based on traditional methods. Then, considering chaotic time series characteristics of historical data and different effects of different samples weights on forecasting results, relevant time series are reconstructed to the multivariate phase space, and samples are weighted by the exponential weighting method. Consequently, the short-term load forecasting method based on multivariate chaotic time series and weighted neural network with random weights and kernels is developed. It can reduce forecasting errors led by the chaos dynamic behavior within the system and samples weights. Both of these two methods above can improve the forecasting accuracy, generalization performance, and training efficiency effectively.Accurate load forecasting can provide a more scientific and effective basis for the economic dispatch decision-making. Then, based on achievements above, this dissertation makes a further study in the field of optimization of the short-term economic dispatch decision-making. The main research results are summarized as follows:(3) The robust security constrained economic dispatch considering uncertain load forecasting method is proposed. Aiming at the uncertainty of load forecasting, the security of networks, and the economy of system operation, the uncertainty model for load forecasting is established, the robust optimal theory is applied, and the security constrains are introduced into the unit commitment model. As a result, the robust security constrained economic dispatch considering uncertain load forecasting method is developed. In the simulation experiment, load parameters are obtained based on the proposed load forecasting method, and are set as the boundary condition of the economic dispatch model. And eventually the optimal economic dispatch strategy is solved. The simulation results shows that the proposed method is able to ensure the security stable and economic operation of electric power system, and the improvement of the economic benefit of day-ahead short-term economic dispatch.