The development of a real-time wave energy device control algorithm based on artificial neural network

Student thesis: Doctoral Thesis

Abstract

This thesis is aimed at proposing an artificial intelligence controller to maximize the energy absorption of wave energy converter (WEC) in practical application. The controller maximizes the energy absorption by locking and releasing the WEC alternately, and such control is known as the latching control. The model predictive control strategy is used to implement real-time control. Since the control inputs are future wave forces, the controller is a non-causal system. An artificial neural network is developed and trained by the machine learning algorithm to predict the short-term wave forces.Firstly, the state-space dynamic model is developed to simulate the kinetic motion of the WEC in sea waves. Compared with traditional convolution dynamic model, the state-space representation extremely enhances the computation efficiency. Moreover, the state-space model is represented by a differential formula so that it is more convenient to implement the control algorithm. The state-space model is acquired based on the so-called system identification, which transforms the frequency-domain hydrodynamic coefficients of the WEC into a set of state vectors. A Rankine source boundary element method code is developed to calculate the hydrodynamic coefficients. By separating the entire fluid boundary into a set of elements and distributing the Rankine source across the elements uniformly, the velocity potential of the fluid is determined.The latching control, a kind of phase control, maximizes the energy absorption by tuning the velocity phase and making it in phase with the wave excitation forces. The tuning of velocity phase is fulfilled by locking the WEC at some time instants following the control command. The Pontryagin'™s maximum principle is applied to derive the control command, which maximizes the energy absorption over a time horizon [ti, ti+p]. By receding the time horizon forward step by step, the control command is updated at each time instant and thereby the control is implemented in a real-time manner. It is the basic idea of the model predictive control. Since the derived control command just maximizes the energy absorption over the time horizon [ti,ti+p] rather than the entire interval [0, T], the model predictive control is a sub-optimal control strategy.The inputs to a non-causal wave energy controller are future wave forces, and it is what prevents the practical application of wave energy control. An artificial neural network is developed to predict the short-term wave forces. Unlike traditional deterministic forecasting approach, which is based on a specific model defined by the users, the neural network forecasts the future wave forces by learning from examples itself since even the users don'™t know anything about the prediction model. In another word, the neural network is a data-driven prediction approach. The developed neural network is trained by a set of examples using the machine learning algorithm.With the artificial neural network, the real-time smart controller is incorporated into a heaving-point absorber to maximize the energy absorption. It is shown that the energy absorption is indeed increased substantially even if the wave forces are predicted. It indicates that the real-time controller is applicable in the real practice. Besides, the effects of prediction duration and prediction deviation on the control performance are also investigated. Long prediction duration enhances the control efficiency whereas the energy absorption is reduced with the presence of prediction deviation. Since the deviation accumulates with the prediction duration, a moderate prediction duration is thus recommended.The multi-stable mechanism of the WEC is also investigated in this thesis. A non-linear PTO (power take-off) system with two oblique springs is proposed. The PTO system is either monostable or bistable depending on the initia
Date of Award1 Feb 2018
LanguageEnglish
Awarding Institution
  • University Of Strathclyde
SponsorsUniversity of Strathclyde
SupervisorZhiming Yuan (Supervisor) & Alexander Day (Supervisor)

Cite this