Research And Application Of The Extreme Learing Machine Optimized By The Fireworks Algorithm

With the development of science and technology, the algorithm has become the core of intelligent technology. Machine learning algorithms have become the main research field. In the field of the machine learning in recent years, the neural networks, the BP neural networks, the support vector machines(SVM) and the extreme learning machine(ELM) have emerged one after another, and the ELM has become a hot topic in the field of the machine learning in recent years. With the development of science, the original extreme learning machine has been unable to meet its requirements in many fields, so the research personnel at home and abroad have put forward many improved algorithms based on the ELM. For example, the extreme learning machine based on the ant colony and the extreme learning based machine particle swarm, etc. This thesis presents a kind of extreme learning algorithm based on the optimization of fireworks algorithm(FWA). And this algorithm is studied, and the theory and related application of fireworks extreme learning machine are expounded.Firstly, the basic theory and principle of the artificial neural network, the single hidden layer neural network and the extreme learning machine algorithm were studied in this thesis. The formula derivation and realization process of several algorithms were introduced. Based on the study of the principle of the particle swarm extreme learning machine(PSOELM), the fireworks algorithm was used to optimize the input weight of the extreme learning machine. Its principle is as follows: First, the FWA gained the M optimal fireworks through many iterations, and the RMSE of the extreme learning machine's test samples was used as the fitness function in each iteration. Second, the optimization of the input weights and hidden layer deviation matrix of the extreme learning machine was achieved. Finally, the matrix output was obtained based on the generalized inverse.Secondly, based on the completion algorithm, the performance of the algorithm was analyzed by using one dimensional SinC function and one dimensional Gauss function, and compared with the ELM. For example, the effect of the number of the nodes in the hidden layer and the number of the iterations on the performance of the algorithm were analyzed. Finally, the conclusion is that the FWAELM can achieve a higher precision than the ELM with less number of nodes in the hidden layer, which can save the space occupied by the input node in the practical application.Then, the FWAELM was applied to the prediction of the output power of photovoltaic power generation, and a feasible method for the prediction of the output power of photovoltaic power generation was proposed at the same time. In this thesis, the output power of a photovoltaic power generation system in a certain area of Australia and the related meteorological data were used in this thesis. The author firstly carried on the normalization processing to the data, the data processing, the data processing, and then establishes the training sample and the test sample of the forecast model.Finally, the photovoltaic power output model of ELM, the photovoltaic power output model of PSOELM and the photovoltaic output model of FWAELM were established by the author. In the same number of hidden layer nodes under the circumstance, the author compared the test on the three prediction models, the photovoltaic output model of FWAELM can reach higher prediction accuracy.