1. Research Background
As the unit capacity of wind turbines continues to increase and wind farms keep expanding, significant multi-timescale and multi-band coupling exists among wind turbines, the farm collector network, energy storage and the external power grid, which readily triggers broadband oscillations and, in severe cases, may cause turbine disconnection, protection tripping and even local system instability. Meanwhile, continuously varying wind speed, power commands and grid strength make system stability distinctly time-varying and scenario-dependent. Existing analysis methods mostly rely on fixed parameters and offline models and can hardly reflect the combined influence of turbine operating states, controller differences and external grid variations on stability; field measurements and simulation models are disconnected, and model parameters cannot be updated in time with the operating state, resulting in deviations between stability assessment results and the actual system. Although purely data-driven methods can extract dynamic features from operating data, they often lack physical constraints and mechanism interpretation, making them unreliable for critical-condition analysis and control decisions. Therefore, digital twin and stability assessment technologies integrating high-fidelity models, real-time operating data and artificial intelligence algorithms are urgently needed.
2. Research Content
To address the above problems, systematic research is carried out in three aspects: refined multi-physics time-frequency-domain digital twins, broadband oscillation stability analysis and active early warning, and turbine-farm-grid coordinated control.
In terms of refined multi-physics time-frequency-domain digital twins, a multi-physics coupling model is established that comprehensively considers wind farm fluid dynamics, aerodynamic loads, blade and tower structural vibration, mechanical transmission, electrical control and grid-connection electromagnetic transients. In the time domain, high-fidelity models are built that can reproduce fault transients, control switching and oscillation instability; in the frequency domain, impedance and modal models are established that account for positive/negative-sequence coupling, control-loop interactions and network distributed parameters, achieving coordinated time-frequency-domain characterization of complex grid-connected dynamics. Multi-source data such as SCADA, PMU, fault recording and equipment condition monitoring are further fused, and parameter identification, state estimation and physics-informed machine learning methods are used to correct model parameters and dynamic characteristics online, keeping the digital model consistent with the physical wind farm in terms of waveforms, spectra and dominant modes.
In terms of broadband oscillation stability analysis and active early warning, impedance measurement and data-driven identification methods applicable to unknown equipment parameters and complex operating conditions are studied to achieve unified acquisition of broadband dynamic characteristics. Combined with methods such as the generalized Nyquist criterion, eigenvalue analysis, oscillation energy analysis and dynamic mode decomposition, the dominant oscillation frequency, oscillation sources, propagation paths and key participating equipment are identified, revealing the broadband coupling mechanism among controllers, the farm collector lines and the external power grid. Methods such as the Koopman operator, graph neural networks and physics-informed neural networks are further used to extract stability evolution features from multi-source time-series data. A dynamic security and stability region is constructed to calculate in real time the safety distance between the operating point and the instability boundary, and risk scenario deduction is carried out based on the digital twin, enabling graded early warning of oscillation risks, identification of weak links and identification of sensitive parameters.
In terms of turbine-farm-grid coordinated control, for oscillations in different frequency bands and from different sources, measures such as control parameter optimization, impedance shaping, supplementary damping control and coordinated active/reactive power allocation are adopted to actively improve the broadband impedance characteristics of turbines and wind farms. At the turbine level, key control parameters such as the phase-locked loop, power loop and current loop are adaptively adjusted according to the turbine operating state; at the farm level, the active and reactive power output of multiple turbines, SVG and energy storage is coordinated to improve the overall damping of the farm and its adaptability to weak grids; at the grid level, the farm power commands and support strategies are optimized in consideration of the stability margin at the point of connection and operating constraints. Based on the online deduction results of the digital twin, model predictive control and safe reinforcement learning are further integrated to form preventive and corrective control strategies under stability-region constraints, shifting wind power grid-connection oscillation treatment from local suppression of a single device to turbine-farm-grid coordinated control, and from passive post-fault handling to proactive pre-risk defense.
3. Achievements
The team has built a refined wind farm digital twin platform and an online grid-connection stability assessment and active early warning platform. The former achieves cross-scale high-fidelity mapping of wind farm fluid dynamics, aerodynamics, structures, mechanics, electrical systems and grid-connection processes, with capabilities for online model correction, reconstruction of complex operating conditions and predictive deduction; the latter achieves online assessment of low-frequency and broadband oscillations, characterization of stability boundaries, risk tracing and active early warning, advancing grid-connection stability analysis from offline post-event handling to online proactive defense. The related achievements are internationally leading and have won the Second Prize of the Shandong Provincial Science and Technology Progress Award and the First Prize of the Science and Technology Award of the China Simulation Federation.
1. Research Background
Broadband oscillations and complex transient processes in renewable power stations seriously threaten the safe and stable operation of the power grid. Refined simulation that simultaneously considers mechanical vibration and the electromagnetic transient characteristics of the electrical network can finely characterize the coupling between the fast dynamics of power electronic equipment and the low-frequency vibration of structures, and is a key fundamental tool for revealing oscillation mechanisms, reproducing transient evolution processes, verifying control strategies and building digital twin systems. However, large-scale renewable power stations have a large number of devices, large-scale network nodes and simulation steps as low as below ten microseconds, making traditional simulation methods extremely slow. For example, refined simulation of a wind farm model containing 40 turbines requires several hours to simulate 2 s. Although partitioned parallel simulation can significantly improve efficiency, existing methods are highly prone to numerical instability and accuracy degradation, leaving large-scale renewable power stations facing the difficulty of being "unable to simulate, unable to simulate accurately and unable to simulate fast".
2. Research Content
To address the above problems, the team carries out systematic research on multi-timescale refined modeling and partitioned parallel simulation methods with strong numerical stability.
In terms of refined modeling, a refined wind farm model has been established that comprehensively considers aerodynamic loads, structural vibration, mechanical transmission and the electromagnetic transients of the electrical network, and a high-precision multi-timescale interface modeling method based on frequency response characteristic analysis has been proposed. This method can accurately transfer multi-frequency dynamic information between the slow mechanical subsystem and the fast electrical subsystem, overcoming the difficulty of traditional zero-order-hold methods in reflecting rapid and large state variations during fault transients and effectively improving the simulation accuracy of the whole "wind farm fluid - mechanical vibration - electromagnetic transient" coupling process.
In terms of parallel decoupling, based on central-difference integration and explicit-implicit hybrid integration methods, a large-scale wind farm is partitioned into dozens of sub-models that can be solved in parallel, and parallel computing resources such as multi-core CPUs, GPUs and FPGAs are combined to execute model computation tasks concurrently, greatly improving the speed of farm-level electromagnetic transient simulation. To address the numerical instability caused by partitioned decoupling, a numerical stability and decoupling-parameter sensitivity analysis method has been established, which can quantitatively evaluate the influence of different decoupling locations and parameters on the numerical stability of the system and accurately locate the key decoupling points that cause instability. A numerical stability enhancement method based on small-step synthesis is further proposed, which restores stability to originally unstable partitioned models by reconstructing the numerical structure of the decoupling branches.
3. Achievements
Based on the above theories and methods, the team has developed refined simulation software and a real-time simulation platform for renewable power stations. The refined simulation software is independently developed in C++, making full use of the large-scale parallel computing capability of multi-core CPUs and GPUs to achieve faster-than-real-time simulation of a farm-level model with 40 wind turbines, supporting large-scale scenario construction, broadband oscillation mechanism analysis and control parameter optimization. The real-time simulation platform adopts a CPU-GPU-FPGA heterogeneous architecture, with a wind farm fluid simulation module developed on GPU-based large-scale parallel computing and an FPGA electromagnetic transient simulation solver based on deep pipelining and parallel matrix operations, breaking through key technologies such as multi-FPGA collaborative computing, task scheduling and low-latency inter-chip data exchange, and meeting the requirements of controller hardware-in-the-loop and power equipment hardware-in-the-loop testing. Together, the software and hardware platforms support large-scale renewable power stations in being "able to simulate, able to simulate fast and able to simulate accurately", providing technical support for broadband oscillation analysis, verification of coordinated control of multiple devices and the construction of farm digital twin systems.
1. Research Background
Driven by the "dual carbon" strategy and the construction of new-type power systems, industrial parks, as concentrated areas of energy consumption and carbon emissions, have their green and low-carbon transition become an important part of the national energy strategy. In 2024, China's total energy consumption reached 158.9 million megajoules, of which the industrial sector accounted for more than 65%, while total carbon emissions reached 12,533.4 million tonnes over the same period. Energy-intensive industries such as metallurgy, chemicals and machinery manufacturing rely heavily on multiple energy sources including electricity, heat, cooling and gas, but their energy subsystems are relatively independent and the level of cascade energy utilization is low, making it urgent to build integrated energy systems in industrial parks with coordinated "electricity-heat-cooling-gas" multi-energy operation.
At present, integrated energy systems in industrial parks still face the "three-many and one-high" challenge: first, multiple energy flows are tightly coupled, but the fusion theory of "energy flow-carbon flow" is missing, so low-carbon operation lacks theoretical support; second, there are many types of equipment and complex operating conditions, with multi-source heterogeneous data that is underutilized, so effective data support is lacking; third, the integration of a high proportion of renewable energy and diverse loads intensifies supply-demand fluctuations and safety risks, and distributed measurement and control terminals are insufficiently coordinated, so collaborative decision support is lacking.
2. Research Content
To address the above problems, the team carries out systematic research in four aspects: unified multi-energy modeling, all-condition state sensing, dynamic carbon accounting and low-carbon dispatch.
In terms of unified multi-energy modeling, the team takes entropy flow as an extensive quantity and builds a unified model of multi-energy flow networks that considers transport characteristics. By representing multiple energy forms such as electricity, heat and gas uniformly within the entropy flow framework, it overcomes the energy loss calculation error caused by interface mismatch in traditional joint simulation of multi-energy flow systems. In terms of solution algorithms, the holomorphic embedding semi-analytical method is used to replace traditional numerical integration methods, and by combining analytical solution with numerical computation, the convergence and computational efficiency of the model are significantly improved.
In terms of all-condition state sensing, a mechanism model characterizing the potential spatio-temporal correlation of multi-source heterogeneous data is first constructed and embedded with an adversarial neural network to form a spatio-temporal generative adversarial imputation network based on a prompt-mask fusion mechanism, comprehensively improving the quality and completeness of measurement data. A two-layer "teaching and mutual learning" machine learning framework is then created, and a non-intrusive and accurate identification method for the operating states of multi-energy equipment at system node ports is developed, achieving "zero-blind-spot" identification of all equipment in the integrated energy system and providing data support for subsequent carbon accounting and optimal regulation.
In terms of dynamic carbon accounting, by analyzing the transmission characteristics of carbon emissions and the analogy with "energy entropy", the team has built a carbon flow transmission network model for integrated energy systems that accounts for transmission losses and dynamic characteristics. By quantifying the carbon flow transmission relationship among nodes, a time-varying weight mechanism is established to dynamically reflect the carbon flow transmission characteristics of lines and pipelines; carbon flow distribution based on power flow tracing is used to calculate the carbon flow carrying ratio of each line; and an adaptive correction mechanism for dynamic energy-carbon parameters based on a meta-learning framework is established, which uses small-sample training to quickly adapt to scenarios such as renewable energy fluctuations and load variations, autonomously calibrating the carbon factor calculation model and accurately characterizing the spatio-temporal distribution of carbon emissions.
In terms of low-carbon dispatch, the team proposes cloud-edge collaborative optimal dispatch technology. On the cloud side, a day-ahead dispatch model based on source-load probabilistic fuzzy sets and dynamic carbon accounting is established, and distributed robust chance-constrained programming is used to handle the uncertainty of renewable energy forecasting. Intraday, a two-layer nested rolling correction mechanism is established, in which the upper layer optimizes the output of coupled equipment and the lower layer rolls to optimize system base values and feeds them back to the intraday upper layer. On the edge side, a distributed consensus control method based on singular perturbation is proposed, enabling dispersed edge devices to autonomously approach the global optimum or a consistent state and solving the consistency convergence problem of output optimization for massive edge devices. In addition, an information-physics fusion control technology based on fast-slow timescale separation is invented, which hierarchically decouples system dynamic behavior by timescale to improve the response speed and coordination of real-time system control.
3. Achievements
Based on the above theories and methods, the team has developed integrated energy system energy-flow and carbon-flow real-time simulation and analysis software and an integrated energy intelligent management and control system. As a core tool supporting the modeling, operation and optimization of complex energy systems, the energy-flow and carbon-flow real-time simulation and analysis software provides comprehensive functions covering modeling, simulation, analysis and decision-making. Its core lies in multi-energy flow coupled dynamic modeling of renewable energy, energy storage and various conversion devices (such as combined heat and power units, heat pumps and power-to-gas devices), and it supports unified or hybrid simulation across multiple timescales from electromagnetic transients to medium- and long-term dynamics. The integrated energy intelligent management and control system is an integrated management and control platform for the coordinated operation of sources, grids, loads and storage in industrial parks. It provides core functions such as unified modeling of electricity, heat, cooling and gas, all-condition state sensing of equipment, lines and pipelines, dynamic carbon accounting with adaptively updated energy-carbon parameters, and cloud-edge collaborative optimal dispatch, enabling real-time monitoring, analysis and optimal control of energy flows, carbon flows and equipment operating states, driving park energy management from passive response to active optimization and providing a replicable and scalable digital solution for the green and low-carbon transition of industrial parks.
