| Vigorously promoting the development of new energy and building a new multienergy coupling energy network system is one of the important measures to solve the problems of energy shortage and environmental pollution.However,the renewable energy sources mainly wind power and photovoltaic are unevenly distributed and have different capacities.The power connected to the main network is characterized by strong uncertainty.Secondly,the coupling network of multi-energy flow system becomes complex and diverse,the multi-energy transmission characteristics are variable,and the system control constraints and uncertainties are greatly increased,which brings certain challenges to the system operation optimization.Accordingly,this thesis supported by the Sichuan Provincial Key R&D Project "Research and Application of Intelligent Joint Dispatch Key Technology for 100 MW-class Photovoltaic Storage Power Plant under Extreme Conditions"(2023YFG0108)addresses the above-mentioned problems on the Integrated electrical and heating system(IEHS)and conducts a relevant study as follows.1)A Gaussian mixture model(GMM)is constructed to address the problem that load uncertainty is difficult to characterize based on the expectation maximization algorithm.Firstly,a deep neural network is used for load day-ahead forecasting.Then an opinion probability distribution is constructed according to the discrepancy between the forecast and the actual value.Finally,the Gaussian mixture model is used to capture the uncertainty of the load by matching the error probability distribution.The foundation for the subsequent system optimization study is established.2)The IEHS energy management model based on deep reinforcement learning(DRL)assisted by load prediction is constructed for the multi-objective integrated energy system optimization problem of load uncertainty and high proportion of hybrid renewable energy injected into the grid.DRL maximizes the reward value by continuous learning to obtain the optimal control action strategy.Empirical results prove that the optimal control strategy based on load prediction-aided DRL has better performance and load uncertainty handling capability.3)The IEHS operation optimization model based on the surrogate model-TD3 algorithm is constructed to address the problem that the topological parameters of the physical model of the grid and heat network are difficult to obtain.Firstly,two surrogate models are trained based on the sparse Gaussian process method to simulate the electric power tide and thermal network tide calculation process.Then the two agent models are embedded in the DRL and they are used to estimate the rewards of DRL intelligence training.The experimental results show that the proposed surrogate model in this paper has better accuracy compared with other methods.The proposed surrogate model-DRL’s IEHS operational optimization model has the ability to achieve better management policies comparing to other algorithms. |
