CN-122021364-A - Intelligent design system and method for predictive control of power electronic model

CN122021364ACN 122021364 ACN122021364 ACN 122021364ACN-122021364-A

Abstract

The invention discloses an intelligent design system and method for predictive control of a power electronic model, and belongs to the field of industrial artificial intelligence. The system comprises a demand interaction and guide module, a domain knowledge enhancement module and a core workflow scheduling and executing module. The system supports full-automatic and autonomous design working modes, in the full-automatic working mode, a large language model intelligent body analyzes the design requirement, and automatically schedules the flow of parameter calculation, scheme matching, controller design, simulation configuration and code generation to conduct automatic design of a power electronic model predictive control algorithm, and in the autonomous design working mode, a user manually triggers the flow of performance modeling, parameter optimization and code implementation to conduct deep optimization design of the power electronic model control algorithm. The designed system and method realize the full-flow automation, intelligent design and optimization of the power electronic model predictive control algorithm from natural language requirements to deployable codes, and remarkably improve the design efficiency and quality.

Inventors

SHEN KUN
CHEN HAOXIN
CHEN RONGBIN

Assignees

湖南师范大学

Dates

Publication Date: 20260512
Application Date: 20260414

Claims (10)

1. The intelligent design system for the predictive control of the power electronic model is characterized by comprising a demand interaction and guide module, a domain knowledge enhancement module and a core workflow scheduling and executing module, wherein the demand interaction and guide module is used for receiving and analyzing user design demands, the module comprises a large language model intelligent body and is used for sequentially collecting structured design demand information through a guided dialogue, the domain knowledge enhancement module is used for retrieving relevant knowledge from a power electronic model predictive control professional knowledge base based on a retrieval enhancement generation technology to enhance demand analysis, and the core workflow scheduling and executing module is connected with the demand interaction and guide module and is used for scheduling and executing core design workflows.
2. The intelligent design system for predictive control of a power electronic model according to claim 1, wherein the system supports two working modes, namely a full-automatic working mode and an autonomous design working mode, wherein in the full-automatic working mode, the large language model intelligent body automatically dispatches the working flows according to the sequence of parameter calculation, scheme matching, controller design, simulation configuration and code generation, and in the autonomous design working mode, the system dispatches the working flows according to the sequence of performance modeling, parameter optimization and code implementation in response to a user design instruction.
3. The intelligent design system for the predictive control of the power electronic model according to claim 1 is characterized in that the functional module for realizing the full-automatic working mode comprises a parameter calculation module, a scheme matching module, a controller design module, a simulation configuration module and a code generation module, wherein the parameter calculation module is used for automatically calculating electric and control parameters according to design requirements, the scheme matching module is used for matching according to the design requirements based on an embedded simulation case knowledge base and recommending a reference scheme, the controller design module is used for generating a core algorithm frame of a model predictive controller according to a preset algorithm frame template, the simulation configuration module is used for generating simulation parameter configuration suggestions, the code generation module is used for generating MATLAB simulation model codes of a designed model predictive controller algorithm, and the functional module is used for cooperatively working under the scheduling of the core workflow scheduling and executing module to form a full-automatic design closed loop from the design requirements to the algorithm simulation model codes.
4. The intelligent design system for power electronic model predictive control according to claim 1 is characterized in that the functional module for realizing an autonomous design working mode comprises a performance modeling module, a parameter optimization module, a code implementation module, an end-to-end code automatic generator and a compiling and deploying function code file, wherein the performance modeling module adopts a self-adaptive mixed meta model modeling method to construct a performance predictive mixed meta model for a specific model predictive control design task, the parameter optimization module adopts a multi-objective collaborative optimization engine, the mixed meta model is used as an evaluator to automatically optimize design parameters in a multi-objective manner, a pareto optimal parameter solution set is output, the code implementation module adopts an end-to-end code automatic generator to integrate design results and generate a compilable and deploying function code file, and the functional module works cooperatively under the scheduling of the core workflow scheduling and executing module to form a design closed loop from design requirements to the engineering code file.
5. The intelligent design system for predictive control of power electronic models of claim 1 wherein said demand interaction and guidance module implements a fully automatic mode of operation via a design state machine that collects demand and determines guidance logic based on demand integrity until complete demand is collected, said core workflow scheduling and execution module's large language model agent that parses user input and automatically initiates in-order scheduling of parameter computation, scheme matching, controller design, simulation configuration, code generation workflow sequences, said system further comprising a graphical user interface, said core workflow scheduling and execution module initiates in-order invocation of performance modeling, parameter optimization, code implementation workflow sequences in response to user sequential operation instructions on said graphical user interface in an autonomous design mode of operation.
6. The intelligent design system for predictive control of a power electronic model according to claim 4, wherein the adaptive hybrid meta-model modeling method trains heterogeneous machine learning models including at least two of random forest, gradient lifting, neural network and support vector machine in parallel, adaptively selects an optimal predictive model for each performance index based on a test set to construct the hybrid meta-model, and the multi-objective collaborative optimization engine is configured to convert a design task into a multi-objective optimization problem including at least two conflicting performance indexes, embed electrical performance constraints, parameter coupling physical constraints and system operation safety constraints in an optimization objective, and solve the problem by adopting an improved non-dominant ordering genetic algorithm combined with a penalty function method.
7. An intelligent design method for power electronic model predictive control applied to the system of any one of claims 1-6 is characterized by comprising the following steps of S1 receiving design requirements input by a user, S2 retrieving relevant information from a power electronic model predictive control expert knowledge base based on a retrieval enhancement generation technology to enhance analysis of the requirements, S3 executing corresponding core design workflow in sequence according to analyzed design tasks and a working mode selected by the user, and executing parameter calculation, scheme matching, controller design, simulation configuration and code generation in sequence when the system is in a full-automatic working mode, and executing performance modeling, parameter optimization and code implementation in sequence when the system is in an autonomous design working mode.
8. The intelligent design method for power electronic model predictive control according to claim 7, wherein in step S1, when entering a full-automatic working mode, the method further comprises the following substeps of S11 initializing a design state machine, S12 providing a guide problem according to the current step of the state machine, collecting requirement information, S13 updating the state machine and judging the requirement integrity, S14 entering the next guide step and repeating S12 and S13 if the requirement is incomplete, and triggering the subsequent workflow in the full-automatic working mode if the requirement is complete.
9. The intelligent design method for predictive control of a power electronic model according to claim 7, wherein in step S3, the workflow in the fully automatic operation mode is automatically parsed and initiated by a large language model agent, and the workflow in the autonomous design operation mode is manually triggered and initiated in sequence by a user through a graphical user interface.
10. A computer readable storage medium, on which a computer program is stored, characterized in that the program, when being executed by a processor, implements the steps of the method according to any of claims 7-9.

Description

Intelligent design system and method for predictive control of power electronic model Technical Field The invention relates to the technical field of industrial automation, power electronics and artificial intelligence intersection, in particular to an intelligent design system and method for a power electronics model predictive control system integrating knowledge enhancement, intelligent scheduling, mixed element modeling, multi-objective optimization and automatic code generation. Background With the increase of renewable energy grid connection and high-performance motor driving demands, a model predictive control (Model Predictive Control, MPC) algorithm has become a core control strategy of a power electronic converter due to the advantages of the algorithm in multivariable constraint processing and dynamic performance optimization. However, the design of the MPC control algorithm is a complex system engineering, and involves multiple links such as topology selection, parameter setting, algorithm implementation, simulation verification and the like, and highly depends on the professional experience of a designer, so that the problems of long design period, high trial-and-error cost, difficulty in obtaining a global optimal solution and the like exist. In recent years, a large language model (Large Language Model, LLM) has a strong application potential in terms of natural language processing and code generation, but general LLM lacks deep expertise in the field of power electronics MPC, and is directly applied to design tasks to easily generate "illusions" (hallucination) against physical laws or engineering constraints, so that the reliability of a designed control algorithm is low, and the full-flow automatic design from demand analysis to code realization cannot be completed autonomously. From the above analysis, the current power electronic MPC algorithm design has low automation and intelligence, and highly depends on manual experience, while the general LLM cannot meet the severe requirements of the vertical field on the professional property, the accuracy and the engineering realizability. In order to solve the problems of the prior art that the design flow of the power electronic MPC algorithm is split, the efficiency is low, the optimization is difficult and the expertise of a general AI tool is insufficient, the invention provides an intelligent design system and method for predictive control of a power electronic model. The system deeply fuses domain knowledge and large language model intelligent body, and aims to realize the automation and the intellectualization of the design of the predictive control algorithm of the power electronic model through two cooperative working modes. Disclosure of Invention The invention provides an intelligent design system and method for predictive control of a power electronic model, which aims to solve the problems of the prior art that the design flow of the predictive control algorithm of the power electronic model is split, the efficiency is low, the optimization is difficult and the expertise of a general AI tool is insufficient. In order to achieve the above purpose, the invention adopts the following technical scheme: An intelligent design system for predictive control of a power electronic model comprises a demand interaction and guide module, a core workflow scheduling and execution module and a core design workflow, wherein the demand interaction and guide module is used for receiving and analyzing design demands input by a user, the module comprises a large language model intelligent body and is used for collecting and structuring complete design demand information comprising control objects, algorithms, loads, performance targets and constraints according to a preset sequence through multi-round guide dialogue in a full-automatic working mode, the field knowledge enhancement module is connected with the demand interaction and guide module based on a retrieval enhancement generation technology and used for retrieving expertise fragments related to a current task from a pre-built power electronic MPC expertise base so as to enhance understanding of demands and accuracy and traceability of subsequent processing, and the core workflow scheduling and execution module is connected with the demand interaction and guide module and is used for scheduling and executing the core design workflow according to the current working mode. The system supports two modes of operation: In the full-automatic working mode, after the large language model intelligent agent automatically analyzes and constructs the requirements, corresponding workflow is scheduled and executed according to the fixed sequence of parameter calculation, scheme matching, controller design, simulation configuration and code generation. In the autonomous design working mode, corresponding workflow is scheduled and executed according to fixed sequences of performance modeling, parameter optimiza