• Energy Research

This paper proposes a novel online and self-learning algorithm to the identification of fuzzy neural networks, which not only learns the structure and parameters online but also learns the threshold parameters by itself and automatically. For structure learning, a self-constructing approach including adding neurons and merging highly similar fuzzy rules is proposed based on the criteria of the system error between actual and model output and the maximum firing strength of neurons. In order to achieve the efficient merging computing, a new calculation method of similarity degree between fuzzy rules is developed. Further and more importantly, the varying width of Gaussian membership functions can be learned by itself according to the underfitting and overfitting criteria. Similarly, different from the existing constant threshold of similarity degree for merging, the varying threshold of similarity degree can be self-learned according to the real-time accuracy of model. The proposed self-learning mechanism significantly improves the model accuracy and greatly enhances the easy usability. Several benchmark examples are implemented to illustrate the effectiveness and feasible of the proposed approach.

Deep belief network (DBN) is an efficient learning model for unknown data representation, especially nonlinear systems. However, it is extremely hard to design a satisfactory DBN with a robust structure because of traditional dense representation. In addition, backpropagation algorithm-based fine-tuning tends to yield poor performance since its ease of being trapped into local optima. In this article, we propose a novel DBN model based on adaptive sparse restricted Boltzmann machines (AS-RBM) and partial least square (PLS) regression fine-tuning, abbreviated as ARP-DBN, to obtain a more robust and accurate model than the existing ones. First, the adaptive learning step size is designed to accelerate an RBM training process, and two regularization terms are introduced into such a process to realize sparse representation. Second, initial weight derived from AS-RBM is further optimized via layer-by-layer PLS modeling starting from the output layer to input one. Third, we present the convergence and stability analysis of the proposed method. Finally, our approach is tested on Mackey-Glass time-series prediction, 2-D function approximation, and unknown system identification. Simulation results demonstrate that it has higher learning accuracy and faster learning speed. It can be used to build a more robust model than the existing ones.

The accurate and real-time measurement of oxygen content in flue gas is the cornerstone of high incineration efficiency and economic benefits for municipal solid waste incineration (MSWI) plants. However, conventional hardware oxygen analyzers are difficult to obtain the oxygen content in flue gas timely and precisely. In this article, a weighted principal component analysis (WPCA) algorithm combined with improved long short-term memory (ILSTM) network is proposed for oxygen content prediction. First, to reduce the model complexity, a WPCA is developed to calculate mutual information correlation coefficients between principal components and the quality variable. Second, the LSTM network is exploited to establish a prediction model, and its hyperparameters are obtained with the particle swarm optimization (PSO) algorithm to improve the generalization ability of the prediction model. Finally, the effectiveness of the proposed prediction method is validated by a benchmark simulation and the real industrial data. And the comparison results with other methodologies demonstrate the considerable prediction performance of the proposed WPCA-ILSTM model.