In order to effectively optimize the load distribution between power sources during the navigation of hybrid ships, a method for predicting ship load demand based on real-time classification according to different working conditions is proposed. The k-means clustering algorithm is used to quantify the voyage history data to classify the ship’s navigation conditions into fast-changing conditions and slow-changing conditions. Some characteristic parameters related to working conditions are selected as input. Then, input and the category of working conditions are put into least squares support vector machine to learn and train to get an online working condition classifier. The genetic algorithm is used to optimize the radial-based neural network to predict the load demand under fast-changing conditions, use the Markov chain model to predict the load demand under slow-changing conditions, so as to obtain the most accurate future load demand of the ship. The simulation results show that the proposed prediction models under different conditions have higher precision, which is an effective means of predicting the load demand for hybrid power ships.

Abstract. In the view of the anti-peaking characteristics of wind power, the electric power companies and the users were selected as the object to study and formulate reasonable demand side hourly price mechanism and the demand response model was established based on dynamic game time-of-use price. Proposed model combined with the scheduling optimization of wind power, thermal power and demand response of the three kinds of power generation resources to improve absorptive capacity of wind power and reduce the cost of the system scheduling, and the model mentioned above was solved by an improved particle swarm algorithm in which the simulated annealing theory was introduced. Simulation results showed that compared with the scheduling optimization model under fixed time-of-use electricity price, the model adopted in this paper could effectively reduce the amount of abandoned wind, enhance the running level of the unit and the economy of the system, and prove the effectiveness of the model.