Dynamic System Simulation Methods And Technology Research Based On PNN

The computer simulation of dynamic system (CSDS) is a new edge-subject based on multiple subjects including computer science, random network theory, statistics and analysis of time series etc. The main processing object of CSDS is engineering system and various social economic systems, and the main research tools of CSDS are mathematical model and digital computer. Through observation and statistics to the running process of simulation model in dynamic system, CSDS can acquire the system simulation output and grasp the basic characteristics of the model, then infer the true parameters (or design the best parameters) of the objects being simulated, expecting to acquire the evaluation or prediction of the actual performance of simulation object, and then realize the improvement or optimization to the design and structure of the true system. CSDS is a kind of technology means used in analyzing and evaluating system running status or optimizing designing the new system according to the given performance and function request. CSDS has been become the important tool of analyzing and researching various systems, especially the complex system.The academic methods and realizing technologies of dynamic system simulation based on PNN have been researched in this paper. The theory part mainly researched the modeling methods of dynamic system simulation and establishing the PNN simulation models of various dynamic systems. The learning algorithm part mainly researched the learning algorithms with high efficiency and stability aiming at the structure characteristic and mapping mechanism of various PNN simulation models. The part of learning algorithm is an important issue needed to be solved when applying PNN simulation models to resolve the practical problems of dynamic system simulation.There are three parts in this paper, including modeling theory and method of system simulation based on PNN, learning algorithm design and practical application. There are four simulation models have been established in modeling theory and method part, including continuous system simulation model based on PNN with two hidden layers, discrete system simulation model based on PNN with two hidden layers, mix system simulation model based on PNN, and fuzzy dynamic system simulation model based on FPNN. And the optimization methods of the structure of dynamic system simulation model have also been researched in this part. Aiming at the information transform mechanism and input and output data type of PNN simulation model, the learning algorithm part have researched the learning algorithms fit for the adaptive mechanism of neural network simulation model, mainly including the PNN learning algorithm based on genetic-simulated annealing algorithm and PNN learning algorithm based on particle swarm-simulated annealing algorithm, and have also made correlative discussion about the influence of algorithm parameters to optimize performance. The forming process of underground oil reservoir was affected by the complex geological sedimentary environment and the constraints of physical and chemical conditions, and the relation between various influence factors was very complex. So the oil field developing is a complex nonlinear dynamic system. Aiming at this problem, the practical application part of this paper has presented the concrete application and realization methods in some practical domains of oil field developing based on PNN simulation models, mainly including the application in dynamic index prediction of developing process, the application in diagnosis of indicator diagram of pumping well, the application in physical parameters calculation, as well as the application in other domains such as sunspots prediction problem, and have obtained good application results.