| Professional domains Business data such as traffic flow,water consumption,traffic flow,network flow,etc.Are affected to some extent by external event information,especially in certain occasions such as venues,stadiums,office buildings and so on.Event information that is often influenced by these external influences can be collected on the Internet.We call these available priori knowledge of public event data that have a strong impact on predicted professional domains of events.At present,the data prediction tasks are faced with the following problems:(1)Fewer support for multi-source heterogeneous data,modeling more based on internal features,more single,rarely combined with external events(2)Lack of memory for long-term data(3)The result of the prediction task is limited with temporarily information,instead of predicting in a period of time,and does not support the scenes that need sequence prediction.For long-term forecasting tasks,we need to adapt to the change of time series,and we can not predict each time point as discrete and unrelated points.In recent years,many scholars have research work in data tasks related to series.The research methods are mainly divided into methods based on ARIMA model or correlation regression model and methods based on neural network or depth neural network model.Some work also used priori information,but the vast majority used the influence factors which set by experts in advance,but there is seldom a combination of public event information sources as a study of business projections of domain events.The advantages of the model based on ARIMA are that it is simple to construct the network and suitable for forecasting tasks related to time series.However,due to its poor adaptability to irregular events and irregular periodic events,the model does not satisfy the hypothesis of its stability.It is difficult to achieve a better prediction goal.The method based on neural network model has certain effect in exploring nonlinear influence factors,but most of them use artificial neural network.The introduction of deep learning is still under study,and its adaptability to time series is not high.In order to solve the above problems,this article first extracts the parts of the public events that affect the professional domains.The impact of which is quantified into a weight coefficient,and a long short time memory(LSTM)algorithm based on an improved deep neural network is proposed in the paper to solve the problem of the temporal prediction model.Combined with the public event data source as a priori information of the model,The resulting method builds the model.At each step of real-time dynamic forecasting,prior information on the source of public events is added.In this dissertation,LSTM is used to train the model in data forecasting task in the field of public domain.The experimental results show that(1)LSTM can support the data prediction in a sequence-generated manner besides the traditional natural language processing and song generation.(2)The model can correctly predict the trend of data events and has better accuracy after combining the prior information of public event sources.(3)The improved LSTM algorithm can support the Predict when new data is dynamically spliced.And the design and implementation of the decision support system is considered.The knowledge,models,methods and data information of all aspects involved in the overall project and are packaged in a bidirectional decision support system,integrated with the overall system,Data to field data,and domain data to public event data bidirectional decision support system. |
PDF Full Text Request