As the integrated energy system continues to deepen towards multi-energy complementarity and collaborative optimization, the coupling and interaction among its internal electric, thermal, gas, storage and other multi-energy networks have become increasingly complex. The emergent behaviors triggered by the nonlinear interactions among intelligent agents have further increased the uncertainty of system state prediction. In response to the shortcomings of traditional prediction models, such as insufficient generalization ability, difficulty in adapting to cross-scenario dynamic changes, and ignoring the influence of emergent behaviors, this paper proposes a meta-adaptive learning framework that integrates the perception of emergent behaviors of intelligent agents, aiming to build a high-precision state prediction model for the integrated energy system. Firstly, a multi-agent interaction structure and emergent behavior modeling structure for the integrated energy system are designed to quantify the emergent features generated by the coupling and interaction among intelligent agents; Secondly, a meta-adaptive learning core structure is constructed, where the meta-learner extracts cross-scenario general knowledge and combines the adaptive modulation mechanism of the basic learner to achieve dynamic scene adaptation; Finally, a state prediction execution structure is designed to complete feature fusion and precise prediction. Experimental results show that the proposed model achieves an average absolute error of 0.023 on three typical scenario datasets, a root mean square error of 0.031, and an average absolute percentage error within 1.8%. Compared with traditional LSTM, Transformer, and ordinary meta-learning models, the prediction accuracy is improved by 25% to 42%; in cross-scenario transfer tasks, the adaptation time of the model is shortened by 68%, and the computing efficiency is increased by more than 35%; meanwhile, in the scenario of sudden load fluctuations, the model still maintains prediction stability, with the average absolute percentage error fluctuation not exceeding 0.3%, providing efficient and reliable technical support for real-time scheduling and safe operation of the integrated energy system, and applicable to the state prediction scenarios of complex multi-energy coupling and dynamic scene switching of the integrated energy system.

The integrated energy system in the park faces challenges in producing and consuming renewable energy on a large scale as well as in achieving equilibrium between supply and demand for energy, making it a novel form in the study of integrated energy systems. The study takes the integrated energy system of the park as an example, and constructs a real-time rolling regulation model of two-layer optimal dispatch with multiple time scales. The model includes an upper-layer rolling economic optimization scheduling model and a lower-layer dynamic performance optimization control model, which takes economy and real-time as the objectives and realizes dynamic rolling optimization through model predictive control theory. The electric chillers are producing power to give cold energy during the whole dispatching cycle, while the absorption chillers produce power to supply cold energy only during the peak cold load period. The cold storage tank lowers the system’s operational costs by storing cold energy during low hours and releasing it during portions of the system’s high cold load hours. For the park's integrated energy system's primary energy exchange nodes 1 and 2, the micro gas turbine, and the gas boiler. The dynamic response process of the output power of the equipment takes a long time in model 2, with a value of about 10 min, while the time for the output value to reach the desired value is greatly reduced in model 1, with a value of about 4 min, and at the same time, it can foresee the change of the output power in advance, and make adjustments accordingly. The model constructed in the study has a more rapid calculation process and higher calculation accuracy in a short period of time, which has obvious advantages in online real-time prediction operation.

Most of the renewable energy sources have unstable supply and high volatility. With the growing share of renewable energy in the integrated energy system, it is more and more difficult to execute the energy pre-dispatch regulation decision of the integrated energy system. To address the problem of increased volatility of the system, the study proposes to optimize the pre-dispatch decision-making of the system’s control center by analyzing the difference between the day-ahead market clearing price and the declared price of electricity supply. The results show that the node declared power is much higher than the ground's actual clearing power during the period from 1:00 am to 4:00 am. During this period the declared power of the nodes is at 2500kW and the actual clearing power of the nodes is around 1500kW. The outgoing power of the integrated energy system electrical load can be reduced in advance during the period from 10 am to 3 pm. The proposed pre-dispatch decision of the integrated energy system on the basis of the difference between the day-ahead clearing price and the node declared price can ensure the stability of the system operation while reducing the operating cost of the integrated energy system.