Abstract:To improve the accuracy of predicting anomalies in optical cable systems based on optical power data, this paper employs the simulated annealing long short-term memory (SA-LSTM) algorithm to forecast optical power fluctuations. First, the raw optical power data is normalized. Subsequently, the simulated annealing (SA) algorithm is utilized to perform global optimization of the long short-term memory (LSTM) algorithm’s key parameters, including the number of iterations, hidden layer units, and learning rate, across a wide range. Using the optimized parameters, an LSTM model is constructed to predict future trends in optical power. The experimental results demonstrate that the proposed method achieves a root mean square error of 2.83×10-3 dB for optical power prediction under optimal parameters, which is only 0.14% of the mean value. Compared with the autoregressive moving average radial basis function (ARMA-RBF) algorithm, the autoregressive integrated moving average support vector machine (ARIMA-SVM) algorithm, the autoregressive moving average long short-term memory (ARMA-LSTM) algorithm, and the autoregressive integrated moving average long short-term memory (ARIMA-LSTM) algorithm, the proposed method improves prediction accuracy by 91.7%, 91.0%, 96.4%, and 96.3%, respectively. Thus, the proposed method can more accurately predict optical power trends in power cable systems, offering significant potential for engineering applications.