Short-term Load Forecasting Model Of Combined Datadecomposition And ESN

The stand or fall of power load forecasting accuracy has directly affection on the stability of power system and society. At the same time, it can make sure people’s life in order. Short-term power load forecasting has relationship with power system’s scheduling and production planning. Accurate short-term load forecasting can reduce cost, save resource. And it is also an integral part to realize automatic management of electric power system. So short-time power load forecasting model with a high degree of accuracy is of great significance.The characteristics of short-term power load are analyzed in this paper. And short-time power load forecasting model based on the data decomposition and echo state network(ESN) is put forward. According to the characteristics of power load, we use proper decomposition method to decompose load data and dig out the inherent law of power load data. Then this paper respectively set up one echo state network for every decomposition. Train the all ESN networks with the corresponding sample. After training, we can use it to predict respectively. At last, accumulate all components’ forecast value to get the final load prediction. And on this basis, this paper considers how similar day affects short-term load. We select similar days with fuzzy method which can make the training sample more targeted. Decomposing power load data to a different time scale and studying each decomposition can discovery the hiding law of power load data. In the end, a new recursive neural network- ESN network is used to forecast every component which can avoid the complex training of traditional neural network who is easy to fall into local extremum. This paper chooses three kinds of data decomposition method: empirical mode decomposition(EMD), average empirical mode decomposition(EEMD), improvement local mean decomposition(ILMD). And the result of each decomposition method is analyzed and compared here. We do simple explanation for physical meaning of each decomposition. Finally, this paper respectively combines the three kinds of decomposition method and the echo state network. And the predicted results are compared and analyzed.As you can see in simulation results, the forecasting model of combining data decomposition and ESN has better prediction effect than directly predictions without decomposition. Among the forecasting model of combining three kinds of decomposition method and the ESN network, the model of combining ILMD and the ESN network is the best combined forecasting model of prediction accuracy, or predict speed is the most ideal, whether prediction accuracy or predict speed.