The Research Of Algal Bloom Predicting Model Based On Time Series Neural Network

In recent years, with the rapid development of economy and society, a large amount of industrial waste water and sewage rich in nitrogen, phosphorus and other nutrients speeds up the eutrophication of the water. The seasonal outbreak of algae blooms due to the proliferation of aquatic organisms, poses a serious threat to drinking water safety, which cause concern around the world. In order to minimize the losses caused by algae blooms, real-time monitoring, and accurate forecast are very important.As the generation process of algae blooms is complex, nonlinear and time-variant, its accurate prediction remains a world-wide an problems. Though research about the eutrophication and its influence factor analyses and its current state evaluation has been widely reported, the prediction research is rare. Thus it is extremely important to study algae variation rules and to find efficient prediction methods. With the practical application background of the topics, this paper tries to establish a practical and effective early warning non-mechanistic model based on the water pollution control. The main research contents of this paper are as follows:(1) Algae bloom generation mechanism is studied. From the algae physical features and external conditions, the reasons of the outbreak of algae bloom are explored and this provides theory basis for establishing the prediction model. Besides, negative consequences caused by algae bloom are listed and the importance of this research is pointed.(2) An algae bloom prediction model is established. Through the comparison of algae bloom forecast method, the advantages and disadvantages of all kinds of models are found out. On the basis of combining time series method and neural network's advantages, this paper establishes a suitable subject of practical application prediction model.(3) Simulation of the algae bloom prediction model are analyzed. With a certain test pool as the research object, using the collected historical data, the univariate and multivariate models qualitied in chlorophyll-a are respectively established, and take Yuqiao reservoir as an actual application verification. The simulation results show that the model has a strong flexibility and practicability. (4) The algae bloom prediction model is optimized. Aiming at the problem that BP neural network weights and threshold randomness limits the model prediction accuracy problem, a genetic algorithm is adopted, which further improves the network generalization ability.(5) The number of the training samples of the algae bloom prediction model is studied. A method based on a self-optimizing RBF neural network is proposed to identify the minimal number of training sample, which can guarantee the model's prediction performance requirements. It is of strong practical significance.