The disclosed is a method for forecasting a demand load during navigation of a ship based on working condition classification. The method of the disclosure comprises: training historical data values by employing a least squares support vector machine to obtain a classification plane, classifying working conditions during the navigation into a fast-changing working condition and a stable working condition, and determining in real time a working condition type of the ship in an online stage. A method for forecasting a demand load by means of a Markov chain is employed for the stable working condition, and a method for forecasting a demand load by means of a radial basis function neural network optimized by a genetic algorithm is employed for the fast-changing working condition. A desirable forecasting effect can be achieved by selecting forecast models suitable for either type of working condition.