Algorithm Of Render Time Prediction Based On Neural Network Combined With Particle Swarm Optimization

In recent years because of the great improvement on computer production process, computer performance is greatly improved, the applications technology of compute science has great development and has been used into various industrial fields. computer graphics has deeply changed the film and production process as an important branch of computer applications technology. People trend to pursue more high quality and more realistic animation, realistic film special effects. Therefore more and more technology companies, film production companies devote themselves to high quality image rendering to improve their products. Because the rendering of a high quality picture requires a very long time, so submit tasks to the cluster parallel rendering is very important, we come up an idea that predicting the rending time as a basis of the scheduling system before a model submitted to the render system in order to improve the utilization efficiency of the cluster.For the prediction of the time of rendering task, many algorithms may be able to achieve this goal, each algorithm focuses on different aspects and has various application conditions. Most of the time can be predicted through analysis of its code, and can also be estimated by analyzing historical data, such as we can use machine learning, artificial intelligence and other means to achieve time prediction purposes.Neural network is a common machine learning models, and is widely used especially with the rapid development of the computer, the computing speed and the rapid increase in storage capacity. Neural network simulates the human’s thinking mode and related research content is very widespread. It’s a interdisciplinary technology and an interesting classification algorithm, giving correct analysis of some complicated problem from a relatively simple way as well as easy to understand. It establish human neural network structure based on the historical data analysis and a series of simulation. There are several different models of neural network, we select neural network base on the current context is. BP neural network is a neural network, which has been widely used in various fields to model different predictive model; due to the further research of BP neural network and applied to many fields, some shortcomings of its own in training process has also been found. Slow convergence speed and local minimization problem among these shortcomings in general to solve practical problems have big impact, and the causes of these problems are mainly due to BP neural take a gradient descent algorithm as network training algorithm, the algorithm updates to change the network through the gradient error. This paper proposed to improve the disadvantage of using particle swarm optimization algorithm for neural network learning algorithm considering the problem in a comprehensive research project and according to the objective reality.This paper based on historical data supported by Bleman rendering system which is based on RenderMan. First, using principal component analysis to normalize input historical data to obtain the sample, and give a neural network render time prediction model trained by particle swarm optimization algorithm through the analysis of the use of BP neural network prediction model. We found that the subsequent time estimates obtained by the new render time prediction model has a better effect through the compare between prediction time and real render time.