RESEARCH ON LOW VISIBILITY FORECAST MODEL OF SEA FOG IN BEIBU GULF BASED ON ATTENTION MECHANISM-EMBEDDED LSTM DEEP LEARNING

： TO ADDRESS THE COMPLEXITIES ASSOCIATED WITH FORECASTING LOW-PROBABILITY, 8 LOW-VISIBILITY FOG EVENTS AND THE UNDERLYING NONLINEAR INTERDEPENDENCIES AMONG 9 VARIOUS INFLUENCING VARIABLES, WE PRESENT AN ATTENTION MECHANISM-EMBEDDED LONG 10 SHORT-TERM MEMORY (ATT-LSTM) DEEP LEARNING MODEL FOR SEA FOG VISIBILITY HAZARD 11 PREDICTION. THIS ARCHITECTURE SEAMLESSLY INCORPORATES AN ATT INTO THE CONVENTIONAL 12 LSTM NEURAL NETWORK FRAMEWORK. SUCH AN INTEGRATION ENABLES THE MODEL TO 13 ADAPTIVELY ASSIGN WEIGHTS TO THE INPUT FEATURES, THEREBY DISTINGUISHING BETWEEN 14 SALIENT AND NONSALIENT VARIABLES. THIS TARGETED ALLOCATION ENHANCES THE 15 CONTRIBUTION OF SIGNIFICANT FACTORS WITHIN THE LSTM FORECASTING ALGORITHM, 16 OPTIMIZES INPUT DATA, AND APPORTIONS VARYING LEVELS OF ATTENTION TO EACH VARIABLE. 17 CONSEQUENTLY, THE MODEL SUBSTANTIALLY MITIGATES PREDICTION ERRORS IN MULTIVARIATE 18 SCENARIOS. AN EMPIRICAL ANALYSIS EMPLOYING AN INDEPENDENT DATASET ENCOMPASSING 303 19 FOGGY DAYS OVER A BIENNIAL PERIOD CORROBORATED THE SUPERIOR PERFORMANCE OF THE 20 PROPOSED ATT-LSTM MODEL. COMPARATIVE EVALUATIONS WITH LSTM, LOGISTIC 21 CLASSIFICATION REGRESSION, AND SUPPORT VECTOR MACHINE (SVM) CLASSIFICATION 22 REGRESSION MODELS REVEALED THAT THE ATT-LSTM MODEL ACHIEVED A RECALL RATE OF 37%, 23 A PRECISION RATE OF 48%, AN ACCURACY RATE OF 91%, AND A THREAT SCORE (TS) OF 0.26. 24 AMONG THE ASSESSED METHODOLOGIES, THE ATT-LSTM MODEL OUTPERFORMED IN TERMS OF 25 RECALL, ACCURACY, AND THREAT SCORE METRICS. THESE FINDINGS SUBSTANTIATE THAT THE