The information about ships’ fuel consumption is critical for condition monitoring, navigation planning, energy management and intelligent decision-making. Detailed analysis, modelling and optimisation of fuel consumption can provide great support for maritime management and operation and are of significance to water transportation. In this study, the real-time status monitoring data and hydrological data of inland ships are collected by multiple sensors, and a multi-source data processing method and a calculation method for real-time fuel consumption are proposed. Considering the influence of navigational status and environmental factors, including water depth, water speed, wind speed and wind angle, the Long Short-Term Memory (LSTM) neural network is then tailored and implemented to build models for prediction of real-time fuel consumption rate. The validation experiment shows the developed model performs better than some regression models and conventional Recurrent Neural Networks (RNNs). Finally, based on the fuel consumption rate model and the speed over ground model constructed by LSTM, the Reduced Space Searching Algorithm (RSSA) is successfully used to optimise the fuel consumption and the total cost of a whole voyage.

有关船舶油耗的信息对于状态监控，导航计划，能源管理和智能决策至关重要。详细的油耗分析，建模和优化可以为海上管理和运营提供大力支持，对水路运输具有重要意义。本文通过多个传感器采集内河船舶的实时状态监测数据和水文数据，提出了一种多源数据处理方法和实时油耗计算方法。考虑到航行状态和环境因素（包括水深，水速，风速和风向角）的影响，然后对长短期记忆（LSTM）神经网络进行了定制和实施，以建立用于预测实时油耗的模型率。验证实验表明，开发的模型比某些回归模型和常规的递归神经网络（RNN）更好。最后，基于油耗率模型和LSTM建立的地面速度模型，成功地使用了缩减空间搜索算法（RSSA）来优化油耗和整个航程的总成本。