Study Of The Load Forecast Based On Artificial Neural Networks And Optimal Operation Of Industrial Boiler Plants Equipped With Thermal Storage

In China, there are a great number of industrial boilers, of which a big part operates at rather low efficiencies. So it is of importance that the efficiency of industrial boiler plants must be improved to contribute to the solution of the energy crisis in our country. Optimization of load assignment in boiler plants can be an effective approach to enhance the energy efficiencies and has a good expectation in its wide use. The existing methods of optimal load assignment for boiler plants are generally based on the famous principle of coordination of incremental fuel costs, and do not work quite well as usual, for actual complexity in real cases. As a rule empirical methods of load assignment have still been used so far, although they may result in waste of energy. Thermal storage is another effective measure of energy saving for boiler plants. However it is a difficult problem in dealing with the optimal design of thermal storage. The smaller the volume of the storage tank, the lower the investment cost of the storage project. On the other hand, it is more difficult to perform the optimal load assignment when the storage volume gets the smallest. The best way to design a thermal storage optimally may be to fix a suitable point of compromise between the smallest volume and the lowest investment cost.In order to solve the current problem of load assignment to boilers in a boiler plant, a new model was proposed in this paper, being referred to as minimal departure model (MDM). The fuel cost curve of the boilers is not required by this model, which depends only on the data of two typical working conditions of boilers however. Unlike certain method, it is not necessary for MDM that the performance of the boilers in a plant be almost the same, which makes MDM applicable to any real cases. Based on the principle of MDM, a model of optimization programming is developed and a suitable algorithm is designed after a series of algorithms have been analyzed. The computer program is completed and applied to an industrial boiler plant. Computation practices show it is easy to find optimal load assignment and the resolutions are obviously better than those by other existing methods.Two optimal design methods of thermal storage for industrial boiler plants have been mentioned before. One is based on minimization of tank volume, with smallest thermal storage but without optimal operation mode of boiler plants. The other is according to the principle of the highest actual efficiencies with which boiler plantsalways work, while the storage volume may increase a lot. The results of both the design methods may not be thought the best. So the third method is proposed in this paper. This method can minimize the life cycle cost (LCC) of boiler plants, considering both the smaller storage volume and the higher actual efficiencies of boiler plants. A comprehensive computer program is developed, embodying the 3 optimal design models. Resolutions of the 3 models can be obtained simultaneously, What is the most preferable design could be determined from the 3 resolutions by the owners who take the charge of decision making.This paper also optimized the segmentation of a load cycle for the industrial boiler plant. In an industrial boiler plant different segmentation pattern means different fuel cost, and the optimal segmentation one provides the most saving of fuel. Boiler plants that work with an optimal segmentation pattern and an optimal load assignment can reach the biggest energy saving.For the sake of optimal operation, forecast of load for boiler plants is a key precondition. In this thesis, artificial neural network (ANN) is adopted to perform the load forecast. After comparing different ANN models, it is found that the cascaded neural network (CNN) is unable to perform well in load forecast for boiler plants. The factors that influence the errors of forecast are analyzed. A structure of BP neural network that yields the most accurate load forecast is constructed after analyses and trials of computation have been made. The final results of...