RESUMEN
How to reduce a boiler's NOx emission concentration is an urgent problem for thermal power plants. Therefore, in this paper, we combine an evolution teaching-learning-based optimization algorithm with extreme learning machine to optimize a boiler's combustion parameters for reducing NOx emission concentration. Evolution teaching-learning-based optimization algorithm (ETLBO) is a variant of conventional teaching-learning-based optimization algorithm, which uses a chaotic mapping function to initialize individuals' positions and employs the idea of genetic evolution into the learner phase. To verify the effectiveness of ETLBO, 20 IEEE congress on Evolutionary Computation benchmark test functions are applied to test its convergence speed and convergence accuracy. Experimental results reveal that ETLBO shows the best convergence accuracy on most functions compared to other state-of-the-art optimization algorithms. In addition, the ETLBO is used to reduce boilers' NOx emissions by optimizing combustion parameters, such as coal supply amount and the air valve. Result shows that ETLBO is well-suited to solve the boiler combustion optimization problem.
RESUMEN
The yield of C4 olefin is often low due to the complexity of the associated products. Finding the optimal ethanol reaction conditions requires repeated manual experiments, which results in a large consumption of resources. Therefore, it is challenging to design ethanol reaction conditions to make the highest possible yield of C4 olefin. This paper introduces artificial intelligence technology to the optimization problem of C4 olefin production conditions. A sample incremental eXtreme Gradient Boosting tree based on Gaussian noise (GXGB) is proposed to establish the objective function of the variables to be optimized. The Sparrow Search Algorithm (SSA), which has an improved advantage in the optimization efficiency, is used to combine with GXGB. Therefore, a kind of hybrid model GXGB-SSA that can solve the optimization of complex problems is proposed. The purpose of this model is to find the combination of ethanol reaction conditions that makes the maximum yield of C4 olefin. In addition, due to the insufficient interpretation ability of GXGB on the data, the SHAP (SHapley Additive exPlanations) value method is creatively introduced to investigate the effect of each ethanol reaction condition on the yield of C4 olefin. The constraints of each decision variable for optimization are adjusted according to the analysis results. The experimental results have showed that the proposed GXGB-SSA model obtained the combination of ethanol reaction conditions that maximized the yield of C4 olefin. (i.e., when the Co loading is 1.1248 wt%, the Co/SiO2 and HAP mass ratio is 1.8402, the ethanol concentration is 0.8992 ml/min, the total catalyst mass is 400 mg, and the reaction temperature is 420.37 â, the highest C4 olefin yield is obtained as 5611.46%). It is nearly 25.46 % higher compared to the current highest yield of 4472.81 % obtained from manual experiments.
RESUMEN
Recent advances in communication enable individuals to use phones and computers to access information on the web. E-commerce has seen rapid development, e.g., Alibaba has nearly 12 hundred million customers in China. Click-Through Rate (CTR) forecasting is a primary task in the e-commerce advertisement system. From the traditional Logistic Regression algorithm to the latest popular deep neural network methods that follow a similar embedding and MLP, several algorithms are used to predict CTR. This research proposes a hybrid model combining the Deep Interest Network (DIN) and eXtreme Deep Factorization Machine (xDeepFM) to perform CTR prediction robustly. The cores of DIN and xDeepFM are attention and feature cross, respectively. DIN follows an adaptive local activation unit that incorporates the attention mechanism to adaptively learn user interest from historical behaviors related to specific advertisements. xDeepFM further includes a critical part, a Compressed Interactions Network (CIN), aiming to generate feature interactions at a vectorwise level implicitly. Furthermore, a CIN, plain DNN, and a linear part are combined into one unified model to form xDeepFM. The proposed end-to-end hybrid model is a parallel ensemble of models via multilayer perceptron. CIN and xDeepFM are trained in parallel, and their output is fed into a multilayer perceptron. We used the e-commerce Alibaba dataset with the focal loss as the loss function for experimental evaluation through online complex example mining (OHEM) in the training process. The experimental result indicates that the proposed hybrid model has better performance than other models.
