Based On Evolutionary Computation Software Process Optimization And Cost Estimation

As the foundation of information society, software industry has become one of the keys in world competition for all countries. While many problems namely software crisis have emerged in the last fifty years during the development of software industry, including overspend, overtime, low productivity and low resource utilization rate, etc. So it is very important to improve software process to alleviate software crisis. Software process is a typical nonlinear complex system with multi-factors in it. It is also a set of ordered steps relating activities, constrains and resources in software life cycle. Presently the main software process models are waterfall model, V-Shape model, incremental model, spiral model, rapid prototype model and parallel model. For its advantages of increasing productivity and resource utilization rate, parallel model has become one of the trends in software process structure development and a popular research field as well. Evolutionary computation is a branch of system science and artificial intelligence research, which has got wide application in engineering for its satisfactory effectivity.Aiming to advance productivity, resource utilization rate and cost estimation accuracy in software process, this thesis discusses problems of cost estimation, activity scheduling and resource configuration with methods of evolutionary computation. The main contents are as follows.The history of software development and the significance of software industry to national economy, society and military are briefly introduced. The crisis and challenges faced by software industry development and the necessity to improve software process are analyzed.The development phases of software process and the advantages and disadvantages of different methodologies adopted in research of software process are summarized. The background, principles, methods and characters of evolutionary computation are presented as well. The difference between software cost and traditional product cost, the necessity andcomplexity of software cost estimation, and estimation technique classification are exposited. The low convergence speed and black-box feature of ANN make it difficult to analyze relationship between cost and its driving factors;while modeling based on evolutionary computation could achieve mathematical formulas, which could be further analyzed with statistic tools. A software cost estimation algorithm combining GAPSO hybrid programming algorithm and CoCoMo cost driving factors is put forward in this part. Making relativity coefficient, PRED and MMRE as evaluation criteria, results of an empirical instance show that the algorithm could increase estimation accuracy.Based on analysis of constraint relations among software process activities, a scheduling method is advanced based on activity priority. Organization and configuration of human resources in software process are researched, and a relative evaluation system is built. The ASRC-GA model employing GA with matrix coding is put forward in this part, which has the steps of task decomposition, interface analysis, circle cutting, and resource evaluation. A project instance is used to verify efficiency of the model in cost and time saving and fitness improvement.