Prediction Of Large-scale Municipal Solid Waste Incineration Boiler Steam Parameters Based On The Time-span Input Neural Network

With the annual increase of the waste and the requirements of the national environmental protection policies,the waste incinerators in our country are gradually developed towards the large capacity and high parameters.However,the inertia of combustion control is larger,the parameters in the furnace have complex coupling effect,and the main steam parameters fluctuate more frequently in the large scale of furnaces.As the result,it is necessary to carry out the combustion diagnosis and prediction optimization of the large waste incinerators.In this paper,a time-span input neural network model was established to predict the variation trend of the main steam parameters of a 750 t/d municipal solid waste grate furnace and a 750 t/d waste circulating fluidized bed in the next 5 minutes.By means of the data correlation analysis and the data delay analysis,the input parameter tables of the main steam parameters prediction models of the two types of furnace were established,respectively.The dimension of input data of the models was reduced by simplification of the high coupling input variables and elimination of the lag variables.The timespan neural network model has included the characteristic of time-lag between the input and output variables,thus could obtain better prediction performance.The time-span input neural network model could realize 1% of the average prediction error of the variation trend of the main steam parameters in the next five minutes through the sample data,and the time-span input neural network had a higher prediction precision and better generalization ability compared with the traditional neural network model.The average prediction error of the main steam temperature,pressure and quantity of the grate furnace were decreased by 49.6%,19.5%,5.4%,respectively.The average prediction error of the main steam temperature,pressure and quantity of the fluidized bed were decreased by 74.2%,82.5%,65.8%,respectively.At the same time,the time-span input neural network model had the capability of the nearly zero prediction error in the next 1 minute.Finally,the data sensitivity analysis was used to obtain the variables that have the most influence on the main steam parameters of the two types of furnace,and the variables can be seen as the key monitoring objects for the optimization and control of the combustion in the furnace.According to the research,the main steam temperature trend prediction module of the combustion diagnosis and intelligent prediction system for a 750 t/d waste circulating fluidized bed was developed to realize the real-time online prediction of the changing trend of the main steam temperature in the next 5 minutes.The operation results showed that the module had a good prediction performance.The confidence of the prediction error within 2℃ in the next one minute could reach 99.7%,and the confidence of the prediction error within 3℃ in the next two minutes could reach 90%,which met the expected requirements of the engineering.