Improvements in pipeline modeling have greatly outpaced improvements in market load modeling, yet the effectiveness of any pipeline model in short-range operations planning is critically dependent upon the latter. A generalized load forecasting model for any weather-sensitive market id herein derived from a consideration of human behavior and comfort, and from the physics of heat loss. Typical needs and behaviors are then generalized by statistical methods.The results of using equations thus developed are then compared to historical data for several cases. This method differs from its predecessors in being a derivation from first principles rather than an empirical fit of lines to historical data. Time and cost problems are encountered in applying non-linear multivariate analysis to specific solutions are discussed, as are possibilities for further refinements such as time gradients, solar heating, and problems of non-weather-sensitive, and cooling loads. By addressing the why of variations, this method has proved superior to both linear and non-linear empirical methods in providing both accuracy and insight.
Pipeline modeling programs have achieved an astounding level of sophistication, accuracy, and universal application. The applicable equations of transient fluid dynamics have been successfully reduced to manageable proportions. In other words, the mechanical portion of the modeling problem has essentially been solved. On the other hand, when it comes to predicting the market portion of a pipeline system, there is no general agreement as to the number or form of the equations, or the number or types of variables involved. This is unfortunate, since the usefulness of a modeling system in short-range forecasting is critically dependent upon the accuracy and reliability of the market load model. Part of the reason for this situation is that market determination had its origins in accounting functions and not in scientific analysis of physical first principles. As a result, most market models are strictly empirical in nature. They are either straight line approximations which make bookkeeping comparisons easier (a degree-day method) or they are some polynomial expression where, if the Nth degree equation fits the data reasonably well, there is not much effort put into understanding the why of the behavior. In these approaches correlation is sufficient proof of cause and effect. In a 1975 paper [l] given before the Society for Industrial and Applied Mathematics, W. G. Michaelson of Public Service Electric & Gas defined six essential qualities of a good forecast method.