Research On Tuna Fishing Forecast Based On EMD-BiLSTM Model

South Pacific bigeye tuna,as an important marine fisheries fishing target,has become one of the important industries in China’s offshore fisheries.Its resources are abundant and widely distributed.Under the background of the rapid development of offshore fisheries,traditional fishing conditions analysis methods It is gradually unable to meet the needs of existing problems,and at the same time,data acquisition capabilities and channels are becoming increasingly diversified.When traditional linear models analyze high-dimensional data,the complexity and complexity of the data often lead to a decrease in the accuracy of the model.Therefore,how to choose a suitable forecast model to deal with the problem of marine fisheries has become a major difficulty.In this paper,in the process of consulting deep neural network data and processing data of South Pacific bigeye tuna fishing grounds,it is found through experiments that Bi-LSTM(bidirectional long short-term memory neural network)has good experimental results in processing fishery data with time series attributes,and at the same time It can effectively integrate the information of the past and the future.Therefore,this paper conducts research on the basis of Bi-LSTM,and proposes a new fishery forecast model based on empirical mode decomposition and two-way long and short-term memory neural network(EMD-BiLSTM),and on this basis The model is further improved and the CEEMD-BiLSTM fishery forecast model is constructed to realize a new production forecast method for fishery applications.The effectiveness of the test is mainly reflected in the following points:(1)A new forecast model for bigeye tuna fishing grounds based on EMD-BiLSTM was constructed.Among them,a new signal processing method is introduced,the empirical mode decomposition mechanism(EMD),which can decompose and extract the sequence under the background of different feature scales of the original data to obtain the eigenmode function of the sequence IMF,and first obtain the high-frequency components of the sequence,and then gradually obtain the low-frequency components,so that the volatility of each component is lower than the original CPUE signal,and the stability is improved.Then the model will skip the original CPUE sequence and directly perform Bi-LSTM predictions for each decomposition component with better stability.While simplifying the difficulty of prediction,it also provides multiple Bi-LSTM neural networks to make the memory function stronger.This network further exerts its advantages in data processing,and finally reconstructs each sequence and obtains the final forecast result.(2)Through experiments under the same conditions with other models,the experimental results show that the new model combined with the EMD mechanism in the prediction visualization graph of the CPUE value is closer to the two curves in the figure than the other comparative models involved.The predicted value is relatively consistent with the actual value,which shows that the new model has a better prediction effect.At the same time,the test uses two evaluation methods of absolute error X＿MAE and root mean square error X＿RMSE to compare and evaluate the results.Among them,the smaller the absolute error,the more accurate the forecast,and the smaller the root mean square error,the more stable the forecast model.According to the test results,among the models used,the EMD-BiLSTM model has the smallest forecast error.The MAE and RMSE are 0.033 and 0.057,respectively.Compared with the random forest model,the MAE and RMSE are reduced by 0.286 and 0.475,respectively,and compared with the BP model.0.05,0.204,compared with the LSTM model,reduced by 0.033,0.085,respectively,compared with the Bi-LSTM model,reduced by 0.018,0.047.It can be seen that the new model has better forecasting performance in terms of accuracy and stability,and has good applicability.(3)The forecast model is further improved,and the complementary ensemble empirical mode decomposition mechanism(CEEMD)is introduced,where CEEMD introduces complementary noises with the same amplitude,independent and identical distribution and negative correlation into the original sequence on the basis of EMD,And then when the signal is integrated,the noise will overwhelm the unstable and irregular signals with higher frequencies,and at the same time extract the low-frequency signals,thereby effectively alleviating the modal aliasing phenomenon of EMD,and further increasing the decomposition of each component.Smoothness.At the same time,because complementary noise is added,the undesirable effect of added noise on the result is avoided after integration.The results show that compared with traditional fishing forecast methods,the CEEMD-BiLSTM model has significantly improved forecast accuracy and stability,which provides a new idea for the forecast of bigeye tuna fishing in the South Pacific.