Wind Farm Frequency Modulation Modeling for Nonlinear Grid Conditions
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing frequency modulation dynamic modeling methods for wind farms face high difficulty and accuracy challenges due to complex geographical and weather conditions, which are not adequately represented by linear models like transfer functions.
Innovation Solution
A method involving the acquisition of frequency modulation data under various conditions, construction of state space models, measurement of nonlinearity, and training of an LSTM neural network to create a dynamic model that accounts for nonlinearity, improving model accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If linear models like transfer functions are used for frequency modulation dynamic modeling, then the modeling process is simpler, but the model accuracy deteriorates due to inability to capture nonlinearity
Solution Approach 1:
The patent transforms the modeling approach from linear parameters to nonlinear parameters by using LSTM neural networks that can adaptively learn and represent nonlinear relationships in frequency modulation data, thereby capturing complex dynamic behaviors that linear models cannot represent
Solution Approach 2:
The patent replaces traditional mechanical/mathematical linear modeling methods (transfer functions) with an intelligent computational approach (LSTM neural networks), substituting deterministic linear system theory with data-driven nonlinear pattern recognition to achieve higher accuracy
2Measurement precision
If complex modeling methods are used to capture nonlinearity, then the model accuracy improves, but the modeling difficulty increases
Solution Approach 1:
The LSTM neural network performs self-learning and self-adjustment by automatically optimizing its internal parameters through training on frequency modulation data, eliminating the need for manual model structure design and parameter tuning that would otherwise complicate the modeling process
Solution Approach 2:
The patent prepares training data and establishes the neural network framework in advance, then uses automated training processes to populate the model with accurate representations of system behavior, reducing the complexity of real-time model development
3Device complexity
If traditional clustering methods based on wind speed and direction are used, then the characterization of wind farm output is simplified, but the representation of actual wind farm characteristics deteriorates due to ignoring weather factors
Solution Approach 1:
The LSTM neural network serves multiple functions simultaneously: it processes frequency modulation data, captures nonlinear dynamics, adapts to different working conditions, and represents various weather factor influences without requiring separate models for each condition, thereby improving reliability without proportionally increasing complexity
Data Source
AI summary
The present invention provides a frequency modulation dynamic modeling method and device for a wind farm, and an electronic device. The method includes: acquiring first frequency modulation data measured at a grid-connected point of the wind farm under a plurality of preset working conditions; establishing a state space model corresponding to each of the plurality of working conditions according to the first frequency modulation data; measuring the nonlinearity between the state space models corresponding to each two of the plurality of working conditions by using a gap measurement method; combining the first frequency modulation data according to the nonlinearity to obtain second frequency modulation data; and training a preset initial LSTM neural network according to the second frequency modulation data until a preset training requirement is met, and obtaining a trained frequency modulation dynamic model of the wind farm.