Short-Term Forecasting Study Of Ferry Passenger Flow Based On Time Series Analysis

Roll-on/Roll-off(RoRo)passenger transportation holds a pivotal role in China’s port cities,providing an efficient waterborne conduit for regional economic growth and human mobility.In recent years,due to increased interregional economic activities and population flux in these port cities,travel demand has seen a persistent rise,resulting in a strain between demand and transport supply.To augment passenger satisfaction,ferry corporations have implemented measures such as bolstering capacity,flexible dispatching,reservation systems,and route optimization to augment operational efficiency.In light of this,short-term ferry passenger flow forecasting has emerged as a critical element in improving passenger service competence and rationally deploying transportation resource capacity.Compared to the research on passenger flow prediction in rail transit,the study of short-term passenger flow prediction in ferries still requires deeper investigation in terms of passenger flow characteristics,prediction accuracy,and model selection.Therefore,this thesis conducts statistical analysis on ferry passenger flow based on the basic data related to the Hai’an passenger ferry route,preprocesses passenger flow data,and introduces the characteristics and correlation analysis of passenger flow,exploring the impact of season,weather,suspension of operations,and historical passenger flow on future passenger flows.Considering the characteristics of the long interval between departures and substantial hourly passenger flow fluctuations in RoRopassenger transportation,the prediction time granularity is divided into three periods: morning(8:00-13:00),afternoon(13:00-18:00),and evening(18:00-0:00).Through the correlation analysis of short-term passenger flow influencing factors,it is decided to use the strongly correlated passenger flow from the same period of the previous day as the input variable for the prediction model.In the process of model construction,the bidirectional long short-term memory neural network model(BiLSTM)was used as the core structure.The Sparrow Search Algorithm(SSA)was employed for optimization problems related to six parameters in this model,including learning rate,batch size,and number of iterations.Given the large fluctuations and non-stationary characteristics of ferry passenger flow sequences,the Complete Ensemble Empirical Mode Decomposition with Adaptive Noise(CEEMDAN)was introduced for decomposition.This yielded several intrinsic mode function(IMF)components with less noise,which were clustered and combined into high-frequency,mid-frequency,and low-frequency components.A composite model,CEEMDAN-SSA-BiLSTM,was built as a new short-term ferry passenger flow prediction model.The prediction model’s effectiveness was evaluated using a performance index system constructed based on three error evaluation methods: Root Mean Square Error(RMSE),Mean Absolute Error(MAE),and Mean Square Error(MSE).Compared to the BiLSTM model,the composite model decreased by 27.18%,6.9%,and 9.8% respectively.In comparison with the SSA-BiLSTM model,the composite model decreased by 15.02%,6.3%,and 11.5% respectively.When compared to the CEEMDAN-BiLSTM model,the composite model decreased by 8.3%,16.5%,and 19.2%respectively.This proves that the composite model performs better in the performance index system.The dual optimization of CEEMDAN and SSA algorithms can significantly improve prediction quality and fitting effect.