Research On Generator Emissions Prediction Based On Recursive Neural Network And Generator Scheduling

In recent years, energy data mining, analysis and application has become a new research hotspot both in academia and industry. The problem of generator emission prediction is an important part of the research of energy conservation and emissions reduction. This dissertation put forward a new generator emission prediction model, based on deep learning,and designed a generator scheduling algorithm for energy conservation and emissions reduction.Using traditional statistical methods and machine learning method, researchers have obtained some achievements, but the prediction effect of these methods is dependent on high quality features of data. However,deep learning has the advantage of strong expression capability and not relying on high quality features. So this dissertation adopts deep learning to research generator emissions prediction. Through researching and improving deep neural network, this dissertation designed a precise emissions prediction model. Then based on the predictions of emissions and intelligent optimization algorithm, this dissertation designed an effective scheduling algorithm to reduce energy conservation and emissions.The contribution of this dissertation are:(1)To put forward a generator emissions prediction model based on recurrent neural network (RNN). To solve the problem of low accuracy, this dissertation design two improvement solutions of deep neural network activation function. And through the study of deep neural network details such as Batch Normalization, overfitting,data normalization and choice of time steps, accurate prediction is obtained. Finally through comparison with polynomial fitting based on least square method, support vector machine regression (SVR), hidden markov model (HMM), this dissertation proved that the model based on deep learning is more effective.(2)To put forward a generator scheduling algorithm to reduce energy conservation and emissions,based on the prediction results and intelligent ant colony optimization algorithm. To solve the problem that ant colony algorithm is easy to fall into local optimal solution, this dissertation improved transfer rule, pheromone updating and solution set search to achieve better scheduling results.Through the experiment on the real historical data, compared to previous methods, generators emissions prediction model and scheduling system designed in this dissertation get a significant improvement.