Complex-Valued Neural Networks for Compact Wind Speed Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing wind speed prediction methods using artificial neural networks face challenges in achieving optimal network structure and parameter design, leading to inefficiencies and inaccuracies due to issues like local minima and slow convergence, particularly in the context of wind energy stability and grid safety.
Innovation Solution
A wind speed prediction method utilizing a feedforward complex-valued neural network is developed, incorporating a Group Lasso regularization term and a specialized complex-valued projected quasi-Newton algorithm to optimize network structure and parameters, effectively deleting redundant neurons and enhancing generalization performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If manual trial and error is used to determine network structure, then appropriate structure can be found, but the process is time-consuming and laborious
Solution Approach 1:
The patent implements automatic network structure determination through a training method that uses a determination unit to automatically select optimal network structures based on training data, eliminating the need for manual trial and error. The system self-adjusts the network structure during the training process, making the structure determination automated and efficient.
2Manufacturing precision
If gradient descent method is used for training, then network weights and biases can be obtained, but the method is prone to local minima and slow convergence
Solution Approach 1:
The patent replaces the traditional gradient descent mechanical optimization method with a neural network training method that uses forward propagation and backpropagation algorithms. This substitution allows the network to learn optimal parameters through data-driven training rather than relying on gradient-based mechanical optimization, thereby avoiding local minima issues and improving convergence reliability.
3Measurement precision
If traditional neural network is used for wind speed prediction, then prediction can be performed, but optimal network structure and parameter design is difficult to achieve
Solution Approach 1:
The patent implements automatic network structure determination through a training method that uses a determination unit to automatically select optimal network structures based on training data, eliminating the need for manual trial and error. The system self-adjusts the network structure during the training process, making the structure determination automated and efficient.
Solution Approach 2:
The patent automatically determines network parameters such as the number of hidden layers and neurons by analyzing training data characteristics and performance metrics. This dynamic parameter adjustment allows the network structure to be optimized for each specific prediction task, achieving high accuracy without manual design complexity.
Data Source
AI summary
The invention discloses a wind speed prediction method based on feedforward complex-valued neural network (FCVNN), including: acquiring a training set and a prediction set for wind speed prediction, and constructing a FCVNN, and initializing a parameter vector; and introducing a Group Lasso regularization term into a target function for training, transferring the training into solving a constrained optimization problem, training the FCVNN by using the training set and a specialized complex-valued projected quasi-newton algorithm, stopping the training until a preset condition is met, obtaining the trained FCVNN, and inputting the prediction set into the established FCVNN to obtain a wind speed prediction result. A Group Lasso regularization term is introduced and a FCVNN is trained by using a specialized complex-valued projected quasi-Newton algorithm to optimize the network structure and parameters, thereby obtaining a compact network structure and high generalization performance and improving the accuracy of wind speed prediction.