Improvement Of BP Algorithms And Its Application In Software Cost Estimation

With the continuous development of computer software industry, the size of software systems is increasing and becoming more and more complex, it brings a lot of difficulties to the software project development. Followed by the software crisis has emerged, the most prominent problem that the software cost estimation is not accurate, resulting in the project cannot be completed on schedule or a budget funding shortage, and other problems. Accurately estimate the cost of the software is an important means to control the progress of the project and reduce project costs, to improve the quality of software development also plays an important role.There are many influence factors of software cost, with the continuous development of project development, the influence factors are constantly changing, there are many complex nonlinear relationships, it very difficult to accurately and efficiently estimate software cost. The BP neural network applies to the software cost estimation, and using the advantages of the nonlinear mapping ability, and other advantages, not having to go to look for the relationship between the various cost drive factors and software cost. In this paper, the research content includes the following several aspects: First, studied the software cost estimate foundation, expounds the methods of software cost estimation, introduced in detail the influence software cost estimation cost drive factor and evaluation standards; Second, introduced the related theory of BP algorithm, according to the existing disadvantages of algorithm, on the basis of a large number of data and many scholars improvement, and further improvements. Mainly combine learning rate and add momentum item, giving full play to the advantages of both, to speed up the network convergence speed and reduce network instability oscillation phenomenon. And use the characteristics of strong global search ability of the genetic algorithm to avoid network into a local minimum value, take advantage of the algorithm to avoid the disadvantages of BP algorithm; And then, use the improved algorithm to construct the software cost estimation model, using MATLAB tools for experiments, compare the performance between before and after the improvement of the algorithm, as a result of the software project data collection difficulty, select software project data of the COCOMO model database as the input of the network, the same set of data as input can also convenient and clearly to make comparison with the result of the experiment, and carries on the concrete analysis of the experimental results. Experimental results show that the estimated value and real value error is small, almost the same, which shows that the improved algorithm is suitable for software cost estimation; finally, using the land and sea transport logistics information platform system as an example was validated.In this paper, the BP algorithm was improved and applied to the software cost estimation, and use transport by sea and land transport logistics information platform as an example to verify. The results show that the improved BP algorithm can effectively estimate the software project cost, and the difference between the estimated value and the actual project cost is small, which can improve the accuracy of the estimation. However, there are a lot of problems in the process of algorithm improvement. How to find a kind of cost estimating method is suitable for the higher complexity of software project, get more efficient and accurate estimates, which needs further research.