Abstract: The power grid dispatcher often uses the coal consumption reference value when making the research plan of the graded early warning system for coal-fired supply or the economic load distribution schedule for each power plant. It plays an important role in both monitoring the coal stockpiles in terms of days of burn in the power plant and formulating a reasonable power generation dispatch schedule. However, there are only few methods applicable so far, most of which simply process the historical data into the mathematical calculation as the forecast value for future coal consumption benchmark and hence may introduce considerable computation errors. By taking account of the actual demand, a method using the rolling calculation of unit coal consumption benchmark value is proposed on the basis of particle swarm optimization support vector machine for power grid scheduling.Three typical power plants in the power grid are selected in the case studies. Through detailed data analysis and comparison testing, from the results the support vector machine model based on particle swarm optimization has demonstrated satisfactory performance in rolling test and forecast update on coal consumption reference value, which means it can even provide further data support for the grid dispatch department to estimate the number of days of coal storage available, establish a coal storage early warning mechanism and formulate a power generation dispatch plan.

Key words: power grid dispatch, coal consumption benchmark, support vector machine, particle swarm, rolling calculation