Integrated optimization method for HIES energy and carbon based on hybrid neural network

• An CGAN-BiLSTM-DQN-SSA network is established to realize parallel training of each task network, and accelerate the model training process. • Introducing BiLSTM for prediction, enabling post agents to make decisions based on deterministic scenarios avoiding the new environment errors. • SSA algorithm is used to pretend the neural network from getting stuck in local optimal and convergence difficulties. Traditional low-carbon operation and maintenance of integrated energy systems(IES) are usually based on automated control systems and manual experience regulation. With the increasing types of loads and energy conversion equipment in the system, traditional operation and maintenance methods are difficult to adjust in a timely manner according to the dynamic changes in the system. Therefore, it is urgent to adopt more advanced and intelligent methods to achieve intelligent low-carbon operation and maintenance of the comprehensive energy system, and to more accurately and timely grasp the system’s operating status, optimization potential, and abnormal situations. Based on this, this article proposes a HIES energy carbon integrated optimization method based on CBAM-BiLSTM-DQN-SSA network, which can simultaneously achieve new energy output prediction, load prediction, scheduling plan generation, and carbon emission optimization in the system. Firstly, based on the HIES structure, a multi-energy load aggregation response model and its carbon trading model are established. Based on the carbon emission flow theory, the carbon emissions in hydrogen integrated energy systems(HIES) are converted from the source to multiple energy load nodes, and a HIES carbon reduction model is established. The HIES source load coordinated carbon reduction is achieved by setting the edge objective function at the load and the cloud objective function. Then, an energy-carbon integrated optimization network based on CBAM-BiLSTM-DQN-SSA network is established. The network uses DSCNN combined with CBAM attention mechanism to learn the original feature data, performs partition weighted output, and strengthens key feature data. The BiLSTM network, which can simultaneously capture forward and backward time dimension features, is used to predict the source load data. The SSA algorithm is combined with DQN to optimize the HIES response plan and solve the initial scheduling plan and carbon reduction plan of the system. Finally, the method proposed in this paper was validated using actual running HIES as an example. The results showed that the proposed method can effectively extract the operating characteristics of equipment within HIES, achieve coordinated optimization of carbon reduction between source and load, and reduce carbon emissions by 8% compared to the initial plan. Compared with other models, the prediction accuracy of our model has been improved by about 11.5% and 22%, the convergence speed of the objective function has been improved by about 67% and 58%, and the optimization degree of the objective function has been improved by about 8.9%, 4.3%, and 6.6%. This provides a feasible solution for the integrated management of energy and carbon in HIES containing EV.