Prediction And Optimization Of Municipal Solid Waste Production And Heating Values Based On Machine Learning

Municipal solid waste production has been surging as the economy has grown rapidly and urbanization has been promoted.Exploring Municipal Solid waste reduction,recycling and harmless disposal is an important way to solve the dilemma of "besieged by waste".In recent years,incineration has become a mainstream method of waste disposal and has been heavily promoted.The heating value of the waste limits whether it can be incinerated and affects the incineration process.It is important to predict the amount of municipal solid waste production in advance and accurately predict the heating value of municipal solid waste after the event for urban pollution prevention,low-carbon economy,energy utilization and economic benefits.Using Xinjiang as an example,this dissertation first analyzes the volume of municipal solid waste production and disposal methods in Xinjiang,and then collects data on the impact factors such as population size,economic development level,industrial structure and living standards of Xinjiang residents in recent years.Six variables,including population,GDP,total retail sales of consumer goods,total investment in fixed assets,GDP of tertiary industries,and consumption expenditure per capita,were determined as input parameters for the model through data preprocessing.The prediction model was established by Multilayer Perceptron neural network(MLP)algorithm.Forecast of municipal solid waste production in Xinjiang.The average annual growth rate of municipal solid waste in Xinjiang from 2009 to 2020 is 1.94 percent,but 85.4 percent of municipal solid waste in Xinjiang is still disposed of in landfills,which is relatively backward compared to other Chinese cities.Compared with 2020,the average growth rate of Xinjiang municipal solid waste production in 2021-2025 is 0.3 percent,which is slower.The mean absolute error and mean relative error between the predicted and true values are 4.94 and 1.39 percent,respectively,and the correlation coefficient is 0.963,indicating good prediction accuracy.The predicted results can provide some references for urban planning and construction,source control and terminal disposal of production in Xinjiang.Measures should be taken to control municipal solid waste generation in advance,and resources should be properly allocated for the terminal disposal of municipal solid waste.However,due to the small amount of training model data and the many factors that affect the amount of municipal solid waste production,the model proposed in this dissertation still has certain limitations,and the amount of municipal solid waste production in the next few years still needs to be modified based on the actual situation of the current year.The heating value of municipal solid waste is a key factor in determining whether municipal solid waste can be incinerated.This dissertation collected the real operation data of a municipal solid waste incineration power plant,conducted data preprocessing and feature selection on 50 characteristic variables in the original data under the advice of experts,and finally selected 16 characteristic variables as input parameters for model training,and took "steam flow at the outlet of the furnace" as the index to measure heating value of municipal solid waste.Three different machine learning algorithms,namely decision regression tree,support vector regression machine,and multi-layer perceptron neural network,are used to build the prediction model.The parameters of the three models are optimized using a mesh search method to find the combination of parameters that minimizes the mean squared error.After optimization,the above three models are able to predict the municipal solid waste heating value to a certain extent,and the mean squared error of the three models in the training set is 0.079,0.025,and 0.018,respectively.The predicted times are 0.244 s,277.750 s,and 10.235 s,respectively.On the validation set,the mean squared error of the predictions is 0.091,0.027,and 0.020,respectively.The predicted times were 0.004 seconds,17.046 seconds,and 0.073 seconds.The prediction errors of the three models are not very different on the training and validation sets,and the models generalize well.Upon comprehensive evaluation,the multi-layer perceptron neural network has the highest prediction accuracy among the three models and is short-lived.Therefore,a multi-layer perceptron neural network is used as the optimal model to predict the final heating value of municipal solid waste.The prediction time of the model on the test set is 0.087 seconds with a mean squared error of 0.019.In this dissertation,we explore previously used or unused machine learning algorithms for building online predictive models of municipal solid waste heating values based on real incineration operation data and larger sample sizes.The prediction accuracy is high,the prediction speed is fast,and the prediction accuracy and time resolution of the municipal solid waste heating values are further improved.The predicted results can provide some guidance on the parameters of the equipment operation and the operation management of the waste incineration power plant.