Application Of Markov Chain In Mid-long Term Load Combination Forecasting

Mid-long term forecasting of power system loading is one of the basic work of power grid planning for cities. It provides the required and basic data for power grid planning, and it's the premise of reliable supplying and economic running of power system. The precision of the forecasting shall directly affect the rationality of investment, network layout and its running.Due to the fact that the mid-long term forecasting is affected by many uncertain factors, no single model can guarantee the satisfaction for the result under any circumstances. How to get the forecast model, how to apply simplicity, practicality, and stability, and how to improve the accuracy of load forecasting has become a research key point. Based on the well comprehension of Markov theory, which is introduced in this paper into mid-long term forecasting, and analyzed and calculated separately form series combination forecast and parallel combination forecast aspect.To begin with, this paper gives a brief introduction of the definition, goal and significance of electricity load forecasting, and makes an analysis of the current conditions and prospects of mid-long term electricity load forecasting both at home and abroad, and discusses the classification of electricity load forecasting and methods for mid-long term electricity load forecasting, and meanwhile, describes the Markov theory, all of which doing foreshadowing for the following discussion. Based on this, consideration of the quality of Markov theory which is applied to combined forecast is developed. For series combination forecast, the features of Markov theory which can reflect the influence on random factors and be extended to the stochastic process which is dynamic and fluctuating is considered, and it is seamless integrated with the GM(1, 1)model. This method makes up the inherent deficiency which showed by the GM(1, 1)model's load forecasting that in precision and dependability aspect. The results of the load forecasting indicated that this method can improve the accuracy. For parallel combination forecast, on the one hand in view of the specific utilization and condition of each single forecasting model, the properties of no aftereffect is implemented to multi-model sifting. The example demonstrated that relative to primeval method, the result which produced by the new method is more ideal. On the other hand, firstly, Markov chain is used to fit the law of status probability distribution of these filtered models, and then the estimating problem of the one-step status probabilities transition matrix is translated into constrained multivariate self-regression analysis model. Secondly, the combination weights of these filtered models are determined through the estimate of the one-step status probabilities transition matrix and the distribution of status probability. Results of calculation examples show that the forecasting results generated by the proposed model is accurate and the proposed method is practicable.