Research On Energy Consumption Prediction And Units Scheduling

Energy contributes greatly to the rapid development of national economy. At the same time, energy consumption situation is the important factor restricting the economic revenue. Reasonable and economical utilization of energyis one of the efficient ways to increase the national revenue and reduce production costs.This research is implemented from the angles of energy generation and consumption. For the energy consumption problem in Iron and Steel industry, efficient predictive algorithms are designed to obtain accurate estimation of energy consumption in future horizon to as to reduce the energy cost, finally reduce the production cost. For the energy generation problem stemed from practical production, the unit commitment problem with pressure consideration is studied. Lagrangian relaxation algorithm is adopted for solving. Reasonable scheduling of units can reduce production cost, save maintanence cost of units, and increase the lifespan of units.This paper’s work is summarized as follows:1) The energy consumption in Iron and Steel enterprise is analyzed, and the modeling method to solve this kind of problem is discussed, finally the support vector machine(SVM) algorithms is designed for prediction considering the characteristics of the addressed problem.Due to the inefficiency of SVM in industrial application, the least squares SVM algorithm is adopted for prediction in this paper.2) The improving stategy for least square SVM is proposed to overcome its disadvantages such as the veak learning ability and the serious influences of parameter selection to the prediction results. To deal with the parameter selection, the the PSO algorithm is used to optimize the parameters. At the same time, the standard PSO is also improved to have global searching ability since it is easily trapped in local optimal solution. To improve the learning ability of the least square SVM, the paper introduces the concept of reinforcement learning and uses Q-Learning method.3) The unit commitment problem is derived from the practicle production and added with pressure characteristics. Based on the tradictional model of unit commitments, the pressure constraints are considered, and units are scheduled by being divided into batch according to its geographical positions. Because of the consideration of the pressure constraints, the problem is more difficult tha before. The models of unit commitment problems with/without climbing constraints are established respectively.4) Lagrangian relaxation method is adopted to solve the unit commitment problem. Two types of dynamic programming methods are designed in the Lagrangian relaxation algorithm for solving the problem without climbing constraints. The comparison results are presented. To construct feasible solution, the pressure constraints are considered. A two-tuple dynamic programming method is proposed to deal with the unit commitment problem with climbing constraints, and the results are satisfying. Then a two-stage heuristic algorithm is designed to obtain feasible solution. The first stage of the heuristic algorithm is the same as the one used for no climbing contrains case, and the second stage of the heuristic algorithm considers primarily the climbing constraints. Finally, the proposed algorithms are implemented using c++language and the results are compared with the Cplex optimization software. The comparison results indicate that good solutions can be obtained within a relatively short period of time, and that the proposed method is very effective.