The Research on Neural Network Computing Model and Method of Interval Arithmetic
|Course||Computer Software and Theory|
|Keywords||neural network interval algorithm iterative computation model roots neurons non-linear equations connection weights activation function|
With the development of the artificial neural network technology, its use is gradually widespread; the application field also is developing, to broad already in every project field unceasingly. But in the science computation and the project technology application field, many questions finally come down to the non-linear algebraic equation solution problem. This article is mainly about using the artificial neural network technology for the traditional numerical calculation. And combining interval algorithm concept, proposed a neural network computing model and method based on interval iterative algorithm, in order to develop traditional interval computational method, expands its applying range.At present, the people use the traditional numerical calculation method, successive iterative algorithm is one of people’s commonly used methods. Find the non-linear equation numerical root not to be exceptional, the traditional iterative algorithm is from some or several initial values start, leaves a series of approximate solutions according to the specific iterative process structure. But it has two significant weaknesses. First, it is very strict to the starting value selection request, requesting initialing value fully approaches the equation thezero spot x*. Second, error is cannot give x（k）(x（k） is iteration number of times) out at thesame time to the iterative process in every step. Combining interval iterative algorithm with neural network, this paper a neural network computing model and method based on interval iterative algorithm is proposed to solve these problems. This method successive iteration each step all can iterate contains the root interval, exactly can make up for these two aspects of their weaknesses, thus has overcome the traditional iteration method weaknesses. These proposed five neural network computing models and method are: （l）Bisect interval iteration method, （2）N-sect interval iteration method, （3） Chord or secant interval iteration method, （4）Newton interval iteration method, （5） Secant interval iteration method. Finally, through to specifically calculate the example simulation experiment, confirmation this paper proposes neural network computing model and method based on interval iterative algorithm is effective, feasible. At the same time, has developed under the traditional value significance "the root value" the concept, the root value usually is "the spot" expands the traditional significance under to satisfies the certain precision "the interval ", in this interval willfully a value all is asks the question the approximate solution, this solves the project application problem for the people to bring conveniently, the people may according to the question actual need, the selection suit the satisfactory solution which the question needs.