Research On Grey Prediction Models And Their Applications In Medium-and Long-Term Power Load Forecasting

The electric power sector is an important basic industry and a key public utility, which is crucial to the country’s development and people’s livelihood. The medium-and long-term power system load forecasting, as the basis of planning, investment, production, dispatching, trading and etc., is playing a significant role in the safe and economic operation of power sector. The country’s medium-and long-term power load is featured with both a certainty of annual increment and an uncertainty of random variation. Thus, it can be seen as a typical grey system and shall be suitable for grey prediction modeling. However, with the increasing complexity and improving marketization of power sector, the traditional grey prediction method gradually cannot meet the requirement of power load forecasting, and needs to be enriched and improved. This dissertation studies on various models and methods of grey prediction and their applications in medium-and long-term power load prediction, and mainly focus on optimal GM(1,1) model, multi-variable prediction model, interval grey number prediction model, variable weights buffer grey model, and etc.The dissertation studies the modeling mechanism of GM(1,1) model, systematically analyzes its inherent flaw, including issues like boundary value, background value, least squares parameter identification, and etc. According to the time response expression of GM(1,1) model, the dissertation proposes a optimized method to directly identify the boundary value x(0)(1) of GM(1,1) model, developing coefficient a and grey coefficient b by using ant colony algorithm, so that it establishes a optimized GM(1,1) prediction model based on ant colony algorithm. This optimized model can reduce the impact of boundary value, and also avoid the errors brought by background value structure and least squares parameter estimation. And the effectiveness of the optimized model has been proved by the load data simulation.The medium-and long-term power load does not exist in isolation. In contrast, it is closely related to economic and social development. Since power load and multiple factors are correlated to and checking each other, it shall be of more practical meaning to establish a multi-variable prediction model rather than a single-variable one. In terms of multi-variable grey modeling prediction issue, this dissertation elaborates the role of mean-value generation in reducing the disturbance of random noise, analyzes the system error brought by inverse mean-value generation, and proposes the multi-variable grey MGMmv(1,N) model and its error modified EMGMmv(1,N) model. And the example analysis and practical application demonstrate that this model can overcome the external disturbance and avoid the system deviation, and can be applied to the practical medium-and long-term load forecasting.The existing grey power load prediction methods are mostly point predictions, and hardly focus on interval predictions. The dissertation raises the concept of grey degree of compound grey number based on the "field" and elaborates its characteristics, and proposes its linear convergence characteristics for the first time. Based on these, it analyzes the deficiencies of interval grey number prediction model, which is based on kernel and grey degree. Furthermore, a prediction model for grey degree is built to replace the identification method of grey degree prediction value so as to improve the original interval grey number prediction model. The improved model can explore both potential information and developing trend of interval grey number series from the perspectives of both "kernel" and "grey degree", thus overcomes the deficiency of the original model, and also supports error analysis and precision test. The effectiveness and availability of the improved model have been proved by the practical application in peak load forecasting.In regard to the application research of buffer operator, firstly, this dissertation proposes three kinds of harmonic mean weakening buffer operators based on harmonic mean number concept and the three axioms of buffer operators, and applies them to the practical medium-and long-term load prediction; secondly, it also introduces the variable-weights buffer operator, illustrates its characteristics and functions in dynamically adjusting buffer amplitude and increasing the smoothness of modeling data series, and proposes the variable-weight buffer grey model and its parameters’ optimizing method. This method, based on grey correlation analysis and particle swarm optimization (PSO), takes the maximum grey correlation degree between predicted value and actual value as objective function to select the optimal buffer factor. It can effectively increase the prediction precision and the stability of fitting and prediction. At last, the application results in practical national electricity consumption have been given to prove the effectiveness and availability of the proposed medium-and long-term load prediction model.The study above shall have a practical significance in both improving grey prediction theory and enhancing the medium-and long-term prediction methods.