Short Term Load Forecasting Using Evolutionary Optimized Modified Locally Weighted GMDH

Accurate forecasting of electricity load is one of the most important issues in the electricity industry. It is essential part of an efficient power system planning and operation. Due to the limited generic structure of conventional group method of data handling (GMDH) network (quadratic twovariable polynomial), it tends to produce an exceedingly complex network when it comes to highly nonlinear systems. In order to overcome this, the modified locally weighted group method of data handling (M-LWGMDH) based genetic algorithm (GA) is proposed in this paper to solve the short term load forecasting problem. The LWGMDH is derived by combining the GMDH with local regression method and weighted least squares (WLS) regression. In the proposed method, the connectivity configuration is not limited to adjacent layers unlike the conventional GMDH. In addition, GA is used for optimal design of the M-LWGMDH network’s topology where a new encoding scheme is presented. The hourly electricity load and temperature data in New York City are used to evaluate the proposed method. The results show that the proposed model exhibits superior performance to that of other methods. Index Terms Short term load forecasting, locally weighted group method of data handling, genetic algorithm, kernel principal component analysis, state space reconstruction.