Support Vector Machine(SVM) is a new kind of machine learning algorithm, which is based on structural risk minimization criterion to obtain smaller actual risk, effectively improve the generalization ability, has the theory tightly, strong adaptability, the characteristics of global optimization, widely used in pattern recognition and regression problems. The paper is based on the historical load data as input, through the Seeker Optimization Algorithm(SOA) to search algorithm on the parameters of support vector optimization, and get the optimal parameter selection, and then put the optimal parameter generation into the SVM prediction model, get the Seeker Optimization Algorithm of support vector machine (SOA-SVM) model, by using this model for the next 24 hours in one region of the load for short-term prediction. Through the example, the use of SOA-SVM prediction precision accuracy is higher than BP neural network and PSO-SVM,so that use this method to short-term load forecasting is feasible and effective.