Model Predictive Control for Modular Multilevel Converter Harmonic Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional control methods for modular multilevel converters, such as Statcoms, face challenges in rapidly reducing harmonics in electrical grids, compensating for reactive power, minimizing voltage ripple on module capacitors, and maintaining capacitor voltages and branch currents within bounds while dealing with dynamic load transients and faults.
Innovation Solution
A method employing model predictive control (MPC) for a modular multilevel converter that predicts future load currents and converter states, optimizing switching sequences to minimize harmonics, reactive power compensation, and voltage ripple, by solving an optimization problem that includes constraints on capacitor voltages and branch currents, and applying the first element of the optimized sequence to control the converter in a receding horizon approach.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If conventional PI-based control methods are used, then the control structure is simple and easy to implement, but the response speed to transients and faults is slow and harmonics reduction is inefficient
Solution Approach 1:
The patent replaces the conventional PI controller (analog control mechanism) with a model predictive control algorithm that uses digital optimization. The MPC solver computes optimal switching sequences by solving a constrained optimization problem at each control interval, substituting the continuous PI control law with discrete optimization-based control. This enables faster response to transients and faults while maintaining systematic control structure.
Solution Approach 2:
The patent employs a prediction model that forecasts future converter states and load currents over a prediction horizon. By predicting future system behavior and pre-computing optimal switching sequences, the controller proactively compensates for upcoming transients and disturbances before they fully manifest, achieving faster effective response time compared to reactive PI control.
2Manufacturing precision
If conventional control methods are used, then the implementation is straightforward, but the ability to minimize voltage ripple on module capacitors and maintain capacitor voltages within bounds is insufficient
Solution Approach 1:
The patent implements a closed-loop model predictive control where the prediction model uses actual measured converter states (capacitor voltages, branch currents) and load currents as feedback. The controller continuously updates predictions based on real-time system state, and the optimization algorithm adjusts switching sequences to maintain capacitor voltages within specified bounds while minimizing voltage ripple. This feedback mechanism provides precise control that conventional open-loop or simple PI control cannot achieve.
Solution Approach 2:
The patent dynamically adjusts control parameters including the prediction horizon length, optimization weights for different objectives (voltage ripple minimization, capacitor voltage balancing), and switching frequency. By changing these parameters adaptively, the controller optimizes voltage ripple minimization and capacitor voltage maintenance for different operating conditions, achieving high precision control that is impossible with fixed-parameter conventional control methods.
3Productivity
If conventional control methods are used, then the system is easier to operate, but the effectiveness in active power filtering and reactive power compensation is reduced
Solution Approach 1:
The patent implements a unified model predictive control framework that simultaneously performs multiple functions: active power filtering, reactive power compensation, capacitor voltage balancing, and harmonic mitigation. The single optimization problem handles all these objectives concurrently through a multi-objective cost function, achieving superior overall performance compared to separate conventional control loops for each function. This multi-functionality is achieved through systematic MPC formulation rather than multiple independent control systems.
4Reliability
If conventional control methods are used, then the control logic is simpler, but the response to dynamic load transients and faults is too slow
Solution Approach 1:
The patent employs a dynamic prediction model that adapts to changing system conditions in real-time. The model predicts future converter states and load currents over a moving prediction horizon, allowing the controller to respond dynamically to transient conditions and faults. The receding horizon approach continuously updates predictions based on actual measured states, enabling the system to track and respond to dynamic load changes and fault conditions with high speed and accuracy, unlike static conventional control methods.
Data Source
AI summary
A method for controlling a modular converter connected to an electrical grid for active power filtering the electrical grid to compensate for a load connected to the electrical grid, comprises: receiving an actual load current and an actual converter state of the modular converter; determining, from the actual load current and a history of previous load currents, a sequence of future load currents over a prediction horizon; predicting a sequence of future converter states of the modular converter and a sequence of manipulated variables for the modular converter over the prediction horizon by solving an optimization problem based on the actual converter state and the future load currents by minimizing an objective function mapping control objectives to a scalar performance index subject to the dynamical evolution of a prediction model of the modular converter and subject to constraints; and applying a next switching state, which is determined from a first element of the sequence of manipulated variables, to the modular converter.