Study On Gray Extreme Learning Machine Prediction Algorithm

Prediction is a behavior which describes the development of the future based on the regularity identified from the past data in a certain period of time. In recent years, the prediction is used in a lot of fields , such as electricity price forecasts, stock prices and weather forecasts. However, the traditional forecasting methods due to its accuracy is not high enough or fast enough is unable to meet the needs of today's forecast demand. In this paper, the following work was done:1. Based on the idea of combination forecasting, the study proposes two combination forecasting algorithms using gray prediction model and extreme learning machine by a weighted method. One determines the weights with a covariance. The other determines the weights with a proposed method based on the cumulative function in the reinforcement learning.2. Four datasets, including Microsoft stock price and product transaction price in TAC-SCM game and Mackey-Glass time-series data and Taiwanese color filter manufacturing data, are evaluated in the experiment, and the results show that the two proposed methods not only have good forecasting performance, but also have good forecasting speed.3. Finally, the combination algorithm is applied to TAC-SCM supply chain competition,for the price and demand forecast in the race. Trading agent competition- Supply Chain Management, referred as TAC-SCM, is co-sponsored by Carnegie Mellon University and other schools. TAC is designed to provide a platform to simulate real market environment and participants will applied the current theoretical research of artificial intelligence to implementation the supply chain management process. After the game ,we verify that the predictive accuracy and speed of the Gray extreme learning machine algorithm are suitable for solving the practical problems.