Research On Combined Model For Short-Term Load Forecasting Based On Crisscross Optimization Algorithm

Power load forecasting has been a hot topic in the research field of power system, and it is also an important task of power dispatching department. Load forecasting, predict the future load based on historical data, is an important basis for power system dispatching operation and making production planning, and the premise of the dynamic balance of electric energy production and consumption. The accuracy of load forecasting is related to the security, economy and stability of power system.In order to improve the accuracy of short-term load forecasting, this paper presented a variable-weight combined forecasting model based on the theoretical analysis by using the Crisscross Optimization Algorithm. In order to effectively capture the load fluctuation, Firstly, the original load data is decomposed to three layers using wavelet packet transform, then sub sequences are inputted to combined model to predict and finally superimpose on the final results. The combined model consists of error feedback weighted time series model, grey model and BP neural network model. Different from the traditional fixed-weight combined models, weight of the three individual model are optimized and dynamically determined by using a new swarm intelligence algorithm--Crisscross Optimization Algorithm. This algorithm consists of alternating horizontal and vertical crossover operation, which would effectively avoiding the shortcoming that easily trap in partial optimum while using Current swarm intelligence algorithms. It has significant advantages in dealing with complex optimization problems in large scale and multi dimension.In order to accurately grasp the load variation of Shaoguan City, the daily and monthly load characteristics of the area are analyzed in this paper, the main factors that affect the load in Shaoguan are summarized. In model parameter identification, the key parameters are determined by several experiments, including layers of wavelet packet decomposition, vertical cross probability and neuron number in hidden layer, which will reduce the influence of human factors on the accuracy of the prediction. In the simulation study, the typical work day, weekend days and holidays were repeatedly 96 point prediction, the prediction results and five kinds of reference model prediction results in-depth comparison, and analysis the dynamic weight coefficient to improve forecasting precision of an important role. The simulation results show that the variable weight combination model can adapt to the load change law of Shaoguan region, and the prediction accuracy of different days is better than that of the five models. Finally, in August 2014 loading on the Shaoguan Power Grid for the continuous prediction, further validation of the proposed based on criss cross algorithm of variable weight combination prediction model has higher prediction accuracy and strong generalization capability and stability.