人工智能在新型电力系统优化调度与控制中的应用

Chen Shi College of Electrical Engineering, Sichuan UniversityJiilin, ChinaDec. 2025 Content I、 Background ISteady State Optimal Dispatch Based on Model-Free DRL TII.Another Technical Route-Model-Based Approach IV,Transition Process - Wide Area Damping Control Method Not Enough Training Data? Few-Shot Learning Characteristic of New Power System > Compared with the traditional power system, the coupling relationship between components in new power systemis more complex, and it has changed into a mode of " Double-sided Uncertainty of Source and Load", and multi-agent interaction between Source-Network-Load-Storage" Complex and yaried loads Multi-form power grid Multi energy complementarityWind and photovoltaic powergeneration have become the mainforce of power generation.Operationofnewpowersystemmustcoordinatethegoalsof lowcarbon, safety and economy. Multiple types of storage The load characteristics of thencw power system arc morccomplex, including the rapiddevelopment of new formatssuch as electric vehicles, whichincreases the bidirectionalityandcomplexityof loadregulation. Multi-form power grids such asthe main distribution and microdistribution makes the systemoperation and control morecomplex,andthe adaptability ofthe network architecture at thesending and receiving end shouldbetakenintoaccount. Energy storage has become animportant part of the new powersystem, and a variety oftechnical routes are parallel, soit is necessary to consider itsadaptabilityanddynamiccoordination. Challenges in New Power System Modeling, Operation and Control Due to the complexity of the new power system, the modeling, operation and control of the system are bothfacingwithenormouschallenges The uncertainty and coupling relationship of the system make precise physical modeling become more difficult. Operation strategies should be devised based on factors such as cost, resource availability, and efficiency. At the control level, the intermittent and fluctuating characteristics of renewable energy generation requireadvanced control solutions to maintain the stability and safety of the power grid Advantages of Deep Neural Networks in Power System Modeling In DNNs, data moves from input through hidden layers, each output feeding the next layer and undergoingnonlinear transformation, reaching the final output layer. Advantage: Apply: DPowerful Nonlinear Modeling Capability②AutomaticFeatureExtractionAdaptabilityto MassiveData@Real-Time AdaptabilityinDynamic ConditionsAbility to Handle Complex Interdependencies DLoadForecastingand Demand Prediction?Fault Detection and DiagnosisCondition Monitoring and Predictive Maintenance@Cybersecurity in Power Systems Application of Deep Reinforcement Learning(DRL)in Power System Regulation and Control DRL uses deep neural networks to simulate complex states and optimize decisions through reward mechanisms toadapt to the dynamics of the environment. Apply: Advantage: Adaptation to Complex and Dynamic EnvironmentsAdaptive Optimization of Scheduling Strategies? Effective Handling of Large-Scale System ProblemsEnhancing System Flexibility and Robustness @ Real-Time Energy Management and DispatchVolt-VAR Control and Voltage RegulationFrequency Regulation and Grid Stability@ Power Market Participation and Trading Challenge of Data-Driven Approach in Power System Application Deep Reinforcement Learning (DRL) optimizes power system decisions through reward-based interactions.adapting strategies to dynamic changes.Transfer Learning [a,., &] [0, , ., 0.8, , , 0.3, , . , 0.5, ](feature representation) Challenge: Measure: Data acquisition difficulty?Uneven sample distributionWeak feature and invalid feature data@ Data management and privacy security protection @ Few-Shot Learning? Transfer Learning? Unsupervised /Self-Supervised LearningMeta-Learning Content I、 Background Steady State - Optimal Dispatch Based on Model-Free DRL Another Technical Route-Model-Based Approach IV, Transition Process Wide Area Damping Control Method Not Enough Training Data? Few-Shot Learning Optimal Dispatch Based on Model-Free DRL According to the Chinese government's plan, a total of 455Gw of wind and solarpower generation capacity will be built by 2030 So huge amount of electricity produced can be transmitted through ultra-high voltageline, but it is constrained by factors such as high cost and insufficient channels Local consumption is an effective and feasible alternative Long-distanceTransmission?? List of Renewable EnergyGenerationProiects Optimal Dispatch Based on Model-Free DRL We take a real isolated power grid in the remote desert of Inner Mongolia as anexample to study how to promote the consumption of new energy through optimizeddispatching. Optimal Dispatch Based on Model-Free DRL We emphasize new energy consumption at the operational level, try to address the combination of day-aheadfrequency capacity reserve and real-time frequency regulation. Therefore, two t