Aaron A. Jennings

Testing random number generators for microcomputer applications of Monte Carlo simulation

Abstract Monte Carlo simulation is a venerable method of solving probabilistic modeling problems, but has historically been thought of as a mainframe technique because of its computational burden. With the great increases in the speed and power of microcomputers, there is increasing interest in microcomputer-based Monte Carlo techniques applied to probabilistic environmental modeling. Effective Monte Carlo simulation requires fast, accurate random number generators and generator performance is often software and hardware specific. This paper describes a package of diagnostic programs titled RANDALIZE that has been developed to interrogate proposed random number generators to determine if they yield sufficiently random variates with the desired probabilistic properties. The methods used to interrogate a random number sequence are defined and illustrated. Methods for interpreting the results are discussed. Example results are presented to illustrate program functions.

Stability routines for expert system permit application review of hazardous waste surface impoundment dikes

Abstract Expert system compatible diagnostic algorithms are presented for evaluating the slope stability and foundation bearing capacity of proposed hazardous waste surface impoundment dikes. The algorithms capitalize upon conservative assumptions to minimize information requirements and to rapidly provide an easily interpreted measure of design quality. Results are presented to demonstrate that, although these algorithms have been tailored to the specialized requirements automated regulatory review, they perform very well when compared to more conventional and far more extensive analysis software.

Algorithms for automated regulatory review of hazardous waste management facilities

Abstract The regulatory review of proposed hazardous waste facilities can offer considerable technological challenges to federal and state agencies. In addition to numerous procedural details, review can require complex engineering analysis and the judicious application of “expert” judgment and opinion. The Environmental Protection Agency is developing microcomputer-based expert systems to help support its regulatory review activities. These systems will help agency personnel analyze sophisticated design elements and qualitatively interpret the resulting information. The goal of this paper is to discuss the special requirements of engineering diagnosis, expert system design, and artificial intelligence applied to automated regulatory review. Many of the desirable features such as front-end information management, safety factor interpretation, assessment of potential “as-built” perturbation, and dynamic information accumulation have been coded into demonstration modules titles DIKE(1.0) and DIKE(1.1). The algorithmic design and function of these codes will also be discussed.

The microcomputer performance of uniform variate random number generators expressed in FORTRAN

Abstract The microcomputer performance of random number generators is an important component of uncertainty analysis in environmental modeling. The results of a performance study of 110 uniform variate pseudo-random number generators expressed in FORTRAN are presented. The algorithms tested include multiplicative, mixed, additive and quadratic congruential generators plus example combined and generalized feedback shift register generators. Results are presented for generator speed, period, and 17 statistical measures of number randomness quality. These results indicate that, based on both speed and number quality, the multiplicative linear congruential generator can yield superior performance. Specific moduli and multipliers are recommended for this generator. For relatively low period generators, explicit use of the integer modulus function provides superior performance. When high period generators are implemented, they should be based on the Schrage (1979) or Bongiovanni (1987) computational procedure for accomplishing the modulus operation with arithmetic that avoids explicit use of the modulus function.