The Study Of Wavelet Neural Networks And Combination Forecast To HABs Intelligent Prediction System

As one of the victims all over the world who have had badly destruction due to the repadily expansion of Harmful Algal Blooms (HABs) which have also casused many negatively environmental and economic consequences around along the coastal areas of China, the annual average direct economic losses amounted to more than 10 billions Yuan in the two recent decades and this trend is intensive year by year. Therefore, study on the key advanced techniques of analysis and forecast for HABs, and development of HABs early warning systems has important practical and strategic significance for promoting sustainable development of marine economy in China.HAB is an anomalous ecological phenomenon caused by many complex factors and is characterized by random, fuzzy, abruptness and nonlinearity, so it is difficult to build up HABs forecast model with high accuracy and stability. Therefore, all of these characteristics of HAB make the research on how to build the effective forecast model one of challenging problems in the frontier field of marine science. Therefore, HABs forecasting system with high accuracy is in great need. This paper gives a preliminary study on forecast modelling of HABs based on Improved Wavelet Neural Network Algorithm, Induced Ordered Weighted Average Operator and Combination Forecast Theory. And then try to design a service oriented intelligent forecast platform aims at improving the level of software design of HABs. Specifically speaking, the paper includes the following aspects:●Building one HABs forecast model uses Improved Wavelet Neural Network Algorithm. First, a new structure of wavelet neural network model is established, in which Morlet wavelet basis function and the sigmoid function are employed as an activation function in the hidden and output layer respectively. Secondly, the entropy error function is used to accelerate the learning speed and the momentum term is aimed at speeding the velocity of network convergence. And then, using the conventional back propagation algorithm studies the weight of wavelet neural network. Finally, the experimental results demonstrated that this IWNN model had excellent functional approximation and generalization abilities. Besides, comparing to a conventional BP neural network model with the same architecture, the convergence velocity and the mean square error of this model also show its superiority.●Even though some presented combination forecasting model performed not bad, there is still an essential need for more accurate forecast models for HABs, which are able to combine effective information of different variables. With this purpose, a new combination forecast model is proposed, composed of Induced Ordered Weighted Average Operator and Improved Wavelet Neural Networks. The results were compared with normal BP models and offered a high accuracy. This combination model proved to be effectual and practicable through comparing experiments using the Red Tide Data collecting from Sishili Bay, Yantai.●To solve the general problems (such as simplified model, coupling of data and algorithm, lack of expandability and reusability, etc.) of traditional HABs forecast software, a solution of HABs Intelligent Forecast Platform is put forward, which is based on service oriented architecture and some new intelligent algorithms. Through the convenient environment provide by the proposed architecture, users could find useful when they would like to construct various combined model, test models, and execute actual business in the future.