Intelligent design system and method for power electronic model predictive control

Abstract

The invention discloses an intelligent design system and method for power electronic model predictive control, and belongs to the field of industrial artificial intelligence. The system comprises a demand interaction and guidance module, a domain knowledge enhancement module and a core workflow scheduling and execution module. The system supports a full-automatic and autonomous design working mode, in the full-automatic working mode, a large language model intelligent agent analyzes design requirements, and automatically dispatches the processes of parameter calculation, scheme matching, controller design, simulation configuration and code generation, and automatic design of a power electronic model predictive control algorithm is carried out; in the autonomous design working mode, a user manually triggers a process of'performance modeling-parameter optimization-code implementation ', and deep optimization design of a power electronic model control algorithm is carried out. According to the design system and method, full-process automatic and intelligent design and optimization of a power electronic model predictive control algorithm from natural language requirements to deployable codes are achieved, and the design efficiency and quality are remarkably improved.

Description

Technical Field

[0001] This invention relates to the interdisciplinary fields of industrial automation, power electronics, and artificial intelligence, specifically to an intelligent design system and method for a power electronic model predictive control system that integrates knowledge enhancement, intelligent scheduling, hybrid element modeling, multi-objective optimization, and automated code generation. Background Technology

[0002] With the increasing demand for renewable energy grid connection and high-performance motor drives, Model Predictive Control (MPC) algorithms have become the core control strategy for power electronic converters due to their advantages in handling multivariable constraints and dynamic performance optimization. However, designing an MPC control algorithm is a complex systems engineering project involving multiple stages such as topology selection, parameter tuning, algorithm implementation, and simulation verification. It highly relies on the designer's professional experience and suffers from problems such as long design cycles, high trial-and-error costs, and difficulty in obtaining the global optimum.

[0003] In recent years, large language models (LLMs) have shown great application potential in natural language processing and code generation. However, general-purpose LLMs lack in-depth expertise in the field of power electronics MPC. Directly applying them to design tasks can easily lead to "hallucinations" that violate physical laws or engineering constraints. The designed control algorithms have low reliability and cannot autonomously complete the entire process of automated design from requirements analysis to code implementation.

[0004] The above analysis shows that current power electronics MPC algorithm design suffers from low levels of automation and intelligence, heavily relying on human experience. General-purpose LLMs cannot meet the stringent requirements of this vertical field for professionalism, accuracy, and engineering feasibility. Therefore, to address the problems of fragmented, inefficient, and difficult-to-optimize power electronics MPC algorithm design processes, as well as the lack of professionalism in general-purpose AI tools, this invention provides an intelligent design system and method for power electronics model predictive control. This system deeply integrates domain knowledge and a large language model intelligent agent, aiming to achieve automation and intelligence in power electronics model predictive control algorithm design through two collaborative working modes. Summary of the Invention

[0005] To address the problems of fragmented design processes, low efficiency, and optimization difficulties in existing power electronics model predictive control algorithms, as well as the lack of professionalism in general AI tools, this invention provides an intelligent design system and method for power electronics model predictive control.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: An intelligent design system for power electronics model predictive control includes: a requirement interaction and guidance module for receiving and parsing user-input design requirements; this module contains a large language model agent for collecting and structuring complete design requirement information, including the controlled object, algorithm, load, performance target, and constraints, through multi-round guided dialogue in a preset order in a fully automatic working mode; a domain knowledge enhancement module, based on retrieval enhancement generation technology and connected to the requirement interaction and guidance module, for retrieving professional knowledge fragments related to the current task from a pre-built power electronics MPC professional knowledge base to enhance the understanding of requirements and the accuracy and traceability of subsequent processing; and a core workflow scheduling and execution module, connected to the requirement interaction and guidance module, for scheduling and executing the core design workflow according to the current working mode.

[0007] The system supports two working modes: In fully automated operation mode, the large language model agent automatically parses and structures the requirements, and then schedules and executes the corresponding workflows in a fixed order of parameter calculation, scheme matching, controller design, simulation configuration, and code generation.

[0008] In the self-designed working mode, in response to user commands, the corresponding workflow is scheduled and executed in a fixed order of performance modeling, parameter optimization, and code implementation.

[0009] The functional modules enabling fully automated operation include: a parameter calculation module, used to automatically calculate key electrical and control parameters, including DC bus voltage, sampling time, load parameters, and controller gain, based on the analyzed control algorithm design requirements; a scheme matching module, which embeds a simulation case knowledge base containing various topologies, algorithms, and load combinations, used to match and recommend the most similar reference scheme based on the current design requirements and parameters, providing a complete reference implementation including system configuration, key parameters, and algorithm framework; a controller design module, which stores algorithm framework templates for various control algorithms, load models, and discretization methods, used to automatically generate the core algorithm framework of the predictive controller based on the selected topology, algorithm, and load type; a simulation configuration module, which automatically recommends the simulation step size, solver type, simulation duration, and initial conditions of the algorithm simulation model based on the system type and parameters; and a code generation module, used to generate the MATLAB simulation model code for the designed predictive controller algorithm. These functional modules work collaboratively under the scheduling of the core workflow scheduling and execution module, forming a fully automated design closed loop from design requirements to algorithm simulation model code.

[0010] The functional modules for achieving an autonomous design workflow include: a performance modeling module, employing an adaptive hybrid model modeling method to construct a high-precision performance prediction hybrid model for specific MPC design tasks. This module trains heterogeneous machine learning models in parallel, including at least two of the following: random forest, gradient boosting, neural networks, and support vector machines. Based on a test set, it adaptively selects the optimal prediction model for each performance metric to be predicted, thus constructing the hybrid model. A parameter optimization module, using a multi-objective collaborative optimization engine, uses the constructed hybrid model as a fast performance evaluator to automatically optimize the design parameters of the MPC system under multiple industrial and physical constraints, outputting a Pareto optimal parameter solution set. A code implementation module, employing an end-to-end automatic code generator, integrates design parameters and solutions, and automatically generates compileable and deployable control algorithm code engineering files that meet the requirements of the target hardware platform based on the optimized Pareto optimal parameters. These functional modules work collaboratively under the scheduling of the core workflow scheduling and execution module, forming a closed-loop automated design from design requirement input to final code engineering file delivery.

[0011] Furthermore, the multi-objective collaborative optimization engine is configured to: transform the design task into a multi-objective optimization problem containing at least two conflicting performance indicators, embed electrical performance constraints, parameter coupling physical constraints, and system operation safety constraints into the optimization objectives, and solve the problem using an improved non-dominated sorting genetic algorithm combined with the penalty function method.

[0012] Furthermore, in fully automatic operation mode, the requirement interaction and guidance module maintains a design state machine that records the collected requirement information and determines the next round of guidance questions based on the completeness of the requirements, so as to realize a progressive and structured requirement collection process.

[0013] Furthermore, the large language model agent in the core workflow scheduling and execution module is specifically used to parse the enhanced user input and automatically initiate the sequential scheduling of the workflow sequence of parameter calculation, scheme matching, controller design, simulation configuration, and code generation.

[0014] Furthermore, the system also includes a graphical user interface; in the self-designed working mode, the core workflow scheduling and execution module responds to the user's sequential operation commands on the graphical user interface and initiates the sequential invocation of the workflow sequence of performance modeling, parameter optimization, and code implementation.

[0015] A smart design method for power electronics model predictive control applied to the above system includes the following steps: S1 The system receives design requirements input by the user; S2 Based on retrieval enhancement generation technology, relevant information is retrieved from the power electronics MPC professional knowledge base to enhance the analysis and understanding of the design requirements; S3 The system executes the corresponding core design workflow in sequence according to the analyzed design task and the working mode selected by the user.

[0016] When in fully automatic mode, the following workflow is executed sequentially: S31 Parameter Calculation: The parameter calculation module is invoked to calculate the initial set of electrical and control parameters based on the design requirements; S32 Scheme Matching: The scheme matching module is invoked to match and provide reference implementation schemes from the simulation case knowledge base based on the requirements and the initial parameter set; S33 Controller Design: The controller design module is invoked to generate the core algorithm code framework of the predictive controller according to the requirements, initial parameters, and reference schemes; S34 Simulation Configuration: The simulation configuration module is invoked to generate simulation parameter configuration suggestions for the current design; S35 Code Generation: The code generation module is invoked to integrate the initial parameters, reference schemes, algorithm code framework, and simulation configuration to generate the MATLAB simulation model code of the designed predictive controller algorithm.

[0017] When in autonomous design mode, the following workflow is executed sequentially: S41 Performance Modeling: An adaptive hybrid model modeling method is used to construct a high-precision performance prediction hybrid model based on the input design parameters; S42 Parameter Optimization: A multi-objective collaborative optimization engine is invoked, using the hybrid model obtained in S41 as a fast performance evaluator, to perform multi-objective collaborative optimization of the design parameters under multiple engineering and physical constraints to obtain the Pareto optimal parameter scheme; S43 Code Implementation: An end-to-end automatic code generator is invoked to automatically convert the user-selected optimal parameter scheme into a compileable and deployable control algorithm code project file that meets the requirements of the target hardware platform.

[0018] Furthermore, in step S1, when the user enters the fully automatic working mode, the system initiates the requirement guidance process. Step S1 also includes: S11 initializing the design state machine and setting the current stage as the initial phase of requirement collection; S12 posing guiding questions to the user to collect requirement information in specific dimensions; S13 updating the design state machine, recording the collected requirements, and determining whether the requirement information is complete; S14 if the requirements are incomplete, proceeding to the next guidance step and repeating S12-S13; if the requirements are complete, triggering the core design workflow of step S3.

[0019] Furthermore, in step S3, the workflow in the fully automated working mode is automatically parsed and initiated by the large language model agent.

[0020] Furthermore, in step S3, the workflow in the self-designed work mode is initiated by the user through the interactive interface in a manual sequence.

[0021] Compared with the prior art, the present invention has the following significant advantages: 1. Full-process automation and high scenario adaptability: The system innovatively defines two collaborative working modes: fully automated and autonomous design. The fully automated mode, guided by an LLM agent, achieves end-to-end one-click design from fuzzy natural language requirements to simulation model code, greatly improving the efficiency of power electronic model predictive control algorithm prototype development. The autonomous design mode focuses on a deep optimization loop of "performance modeling - parameter optimization - code implementation," supporting multi-objective global optimization of existing power electronic model predictive control schemes. Both modes seamlessly cover the complete design lifecycle from rapid prototyping of control algorithms to performance optimization.

[0022] 2. Deep Domain Knowledge Fusion and Highly Reliable Output: By using retrieval-enhanced generation technology for power electronics MPC, the generation and reasoning process of large language models are anchored on a structured domain knowledge base, effectively alleviating the "illusion" problem and ensuring the professionalism and engineering feasibility of the entire chain of output from requirement analysis and parameter calculation to code generation.

[0023] 3. Agent-driven dynamic planning and scheduling: Innovatively, a large language model agent is used as the central scheduler of the system, enabling it not only to understand user intent, but also to dynamically plan and sequentially trigger the invocation of a series of professional tools (modules) based on built-in logic and domain knowledge, thereby realizing the intelligent decomposition and execution of complex design tasks.

[0024] 4. A Leap in Design Quality and Efficiency: Adaptive hybrid modeling provides high-precision and rapid performance prediction for control algorithm optimization strategies; the multi-objective collaborative optimization engine can automatically search for Pareto optimal solutions under strict constraints, providing a global optimization perspective that traditional trial-and-error methods cannot obtain. The combination of these two technologies significantly improves the design quality of power electronic model predictive control algorithms while shortening the traditional design-optimization cycle of several days or even weeks to minutes or hours. Attached Figure Description

[0025] Figure 1 The overall system architecture and dual-mode workflow diagram of this invention.

[0026] Figure 2 The present invention designs a state transition diagram for a state machine.

[0027] Figure 3 This is a diagram illustrating the interaction process in Embodiment 1 of the present invention.

[0028] Figure 4This is a diagram of the MATLAB model code generated in Example 1 of this invention.

[0029] Figure 5 This is a flowchart of adaptive hybrid element modeling in Embodiment 2 of the present invention.

[0030] Figure 6 This is a diagram of the Pareto optimal solution set generated in Example 2 of this invention. Detailed Implementation

[0031] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention.

[0032] The present invention discloses an intelligent design system for power electronic model predictive control, the overall system architecture and dual-mode workflow diagram of which are shown below. Figure 1 As shown, it includes the following parts: ① The requirement interaction and guidance module is used to receive and parse the design requirements input by the user. This module contains a Large Language Model (LLM) agent, which is used to collect and structure complete design requirement information, including control objects, algorithms, loads, performance goals and constraints, in a preset order through multi-round guided dialogue in fully automatic working mode.

[0033] ② Domain knowledge enhancement module, based on Retrieval-augmented Generation (RAG) technology, is connected to the requirement interaction and guidance module. It is used to retrieve professional knowledge fragments related to the current task from the pre-built power electronics MPC professional knowledge base to enhance the understanding of requirements and the accuracy and traceability of subsequent processing.

[0034] ③ The core workflow scheduling and execution module is used to schedule and execute the core design workflow according to the current work mode.

[0035] ④ Graphical User Interface (GUI) is used to respond to the user's sequential operation commands on the GUI in the self-design working mode, and to initiate the sequential invocation of the workflow sequence of performance modeling, parameter optimization, and code implementation.

[0036] The system supports two working modes: ① In fully automatic working mode, the user inputs initial requirements via natural language description. The large language model agent, aided by the domain knowledge enhancement module, then processes the input through methods such as... Figure 2 The design state machine shown interacts with the user, gradually transforming vague requirements into a structured and quantifiable design task book. Then, it automatically schedules the complete workflow of "parameter calculation → scheme matching → controller design → simulation configuration → code generation" to deliver the basic design scheme.

[0037] ② The user-designed workflow allows for clear design parameters and optimization goals. Users manually trigger the "performance modeling → parameter optimization → code implementation" workflow through a graphical interface. The system first constructs a high-precision hybrid element model, then performs multi-objective parameter optimization, and finally generates high-performance code based on the user-selected optimal solution. Example 1

[0038] This embodiment demonstrates the complete workflow of the system in fully automatic operation mode, such as... Figure 3 As shown.

[0039] 1. Requirement Input: The user enters the fully automatic working mode and inputs: "Design a three-level inverter for a 30kW permanent magnet synchronous motor, using FCS-MPC".

[0040] 2. Knowledge Enhancement and Guided Parsing: The domain knowledge enhancement module retrieves relevant professional knowledge. The LLM agent initiates a state machine-guided dialogue, parsing the key parameters input by the user: Vdc=540V, fs=30kHz, rated current 8A, speed 3000rpm, prediction steps N=1, speed loop Kp=5.2, Ki=0.12, constraint switching frequency ≤10kHz, current ripple ≤3.5%, weight λi=1.2, λs=0.01.

[0041] 3. Automated workflow execution: The intelligent agent automatically schedules subsequent modules: ① Parameter calculation: The calculated parameters are Ts=33.333µs, system bandwidth≈79.58Hz, current loop gain Kp≈100.0, Ki≈500.0, etc.

[0042] ②Solution matching and recommendation: The solution matching module outputs four candidate solutions, and the user chooses solution 1 (standard version) based on the trade-offs.

[0043] ③ Controller design: Generate a 27-switch state enumeration for a three-level NPC topology, a dq-axis PMSM current prediction model, and a cost function code framework with integrated specified weights.

[0044] ④ Simulation configuration: Recommended simulation step size 1e-5s, solver ode3.

[0045] ⑤ Code generation: Integrate all results and package them into a directly runnable MATLAB / Simulink project, such as... Figure 4 As shown.

[0046] 4. Output: Deliver a structured design report and a complete source code project package, enabling rapid delivery from a single requirement to a verifiable solution. Example 2

[0047] This embodiment demonstrates in-depth parameter optimization of the FCS-MPC control algorithm with integrated error compensation in the self-designed working mode.

[0048] 1. Problem Definition: The user enters the self-design working mode and, for a system with an existing error compensation module, the goal is to minimize the current THD value and system switching losses under the conditions of Vdc=500V and PF=0.95.

[0049] 2. Parameter configuration: The user fixes Vdc and PF, and sets the filter inductor Lf and capacitor Cf as optimization variables (Lf: 0.8-4.4mH, Cf: 15-70µF).

[0050] 3. Performance Modeling: The user triggers "Performance Modeling". This is achieved using methods such as... Figure 5 The adaptive hybrid element modeling method shown constructs a high-precision hybrid element model for "current THD" and "switching loss".

[0051] 4. Parameter Optimization: User-triggered "Parameter Optimization". The multi-objective collaborative optimization engine uses a hybrid meta-model as the evaluator and employs an improved non-dominated sorting genetic algorithm (NSGA-II) for optimization, outputting as follows: Figure 6 The Pareto optimal solution set is shown. The system automatically sorts and recommends the optimal solution, for example: [Rank 1] Parameters: Lf=2.3mH, Cf=65µF; Performance: Voltage THD=2.444%, etc.

[0052] 5. Validation and Decision: The optimized parameters are substituted into the detailed model for validation. The performance is in high agreement with the prediction and is significantly better than the empirical parameters. The user selects this solution.

[0053] 6. Code Implementation: When the user triggers "Code Implementation", industrial-grade C code for the TI C2000 DSP with integrated and optimized parameters is generated.

[0054] 7. Output: Deliver an optimization parameter table, a verification report, and deployable code, completing the paradigm shift from "manual trial and error" to "data-driven automatic optimization".

[0055] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An intelligent design system for predictive control of power electronic models, characterized in that, include: The requirement interaction and guidance module is used to receive and parse user design requirements. This module includes a large language model intelligent agent, which is used to collect structured design requirement information in sequence through guided dialogue; a domain knowledge enhancement module, which retrieves relevant knowledge from the power electronics model predictive control professional knowledge base based on retrieval enhancement generation technology to enhance requirement parsing; and a core workflow scheduling and execution module, which is connected to the requirement interaction and guidance module, is used to schedule and execute the core design workflow.

2. The intelligent design system for predictive control of power electronic models according to claim 1, characterized in that, The system supports two working modes: fully automatic working mode and autonomous design working mode. In fully automatic working mode, the large language model agent automatically schedules the workflow in the order of parameter calculation, scheme matching, controller design, simulation configuration, and code generation. In the autonomous design working mode, the system responds to user design instructions and schedules the workflow in the order of performance modeling, parameter optimization, and code implementation.

3. The intelligent design system for predictive control of power electronic models according to claim 1, characterized in that, The functional modules that enable fully automated operation include: a parameter calculation module, used to automatically calculate electrical and control parameters according to design requirements; a scheme matching module, which matches schemes based on the embedded simulation case knowledge base and recommends reference schemes according to design requirements; a controller design module, which generates the core algorithm framework of the model predictive controller based on a preset algorithm framework template; a simulation configuration module, used to generate simulation parameter configuration suggestions; and a code generation module, used to generate the MATLAB simulation model code of the designed model predictive controller algorithm. The above functional modules work together under the scheduling of the core workflow scheduling and execution module to form a fully automated design closed loop from design requirements to algorithm simulation model code.

4. The intelligent design system for predictive control of power electronic models according to claim 1, characterized in that, The functional modules for realizing the autonomous design working mode include: a performance modeling module, which uses an adaptive hybrid model modeling method to construct a performance prediction hybrid model for a specific model predictive control design task; a parameter optimization module, which uses a multi-objective collaborative optimization engine to automatically optimize design parameters based on the hybrid model as an evaluator, and outputs a Pareto optimal parameter solution set; and a code implementation module, which uses an end-to-end code generator to integrate design results and generate compileable and deployable engineering code files. The above functional modules work collaboratively under the scheduling of the core workflow scheduling and execution module to form a design closed loop from design requirements to engineering code files.

5. The intelligent design system for predictive control of power electronic models according to claim 1, characterized in that, The requirement interaction and guidance module implements a fully automatic working mode through a design state machine. This design state machine is used to collect requirements and determine the guidance logic based on the completeness of the requirements until complete requirements are collected. The large language model agent in the core workflow scheduling and execution module is used to parse user input and automatically initiate the sequential scheduling of workflow sequences for parameter calculation, scheme matching, controller design, simulation configuration, and code generation. The system also includes a graphical user interface. In the self-design working mode, the core workflow scheduling and execution module responds to the user's sequential operation commands on the graphical user interface and initiates the sequential invocation of workflow sequences for performance modeling, parameter optimization, and code implementation.

6. The intelligent design system for predictive control of power electronic models according to claim 4, characterized in that, The adaptive hybrid model modeling method trains at least two heterogeneous machine learning models, including random forest, gradient boosting, neural network, and support vector machine, in parallel, and adaptively selects the optimal prediction model for each performance index based on the test set to construct the hybrid model; the multi-objective collaborative optimization engine is configured to: transform the design task into a multi-objective optimization problem containing at least two conflicting performance indices, embed electrical performance constraints, parameter coupling physical constraints, and system operation safety constraints in the optimization objectives, and solve the problem using an improved non-dominated sorting genetic algorithm combined with the penalty function method.

7. An intelligent design method for power electronic model predictive control applied to the system described in any one of claims 1-6, characterized in that, Includes the following steps: S1 receives the design requirements input by the user; S2 retrieves relevant information from the power electronics model predictive control professional knowledge base based on retrieval enhancement generation technology to enhance the analysis of requirements; S3 executes the corresponding core design workflow in sequence according to the analyzed design task and the working mode selected by the user. When in fully automatic working mode, the following steps are executed in sequence: parameter calculation, scheme matching, controller design, simulation configuration, and code generation. When in autonomous design mode, the following steps are executed in sequence: performance modeling, parameter optimization, and code implementation.

8. The intelligent design method for power electronic model predictive control according to claim 7, characterized in that, In step S1, when entering the fully automatic working mode, the following sub-steps are also included: S11 Initialize the design state machine; S12 Propose guiding questions based on the current step of the state machine and collect requirement information; S13 Update the state machine and determine the completeness of the requirements; S14 If the requirements are incomplete, proceed to the next guiding step and repeat S12 and S13. If complete, it will trigger subsequent workflows in fully automated work mode.

9. The intelligent design method for predictive control of power electronic models according to claim 7, characterized in that, In step S3, the workflow in the fully automatic working mode is automatically parsed and initiated by the large language model intelligent agent; the workflow in the self-designed working mode is manually triggered and initiated by the user through the graphical user interface.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the method as described in any one of claims 7-9.

Images