Engine Control Neural Network Structure Against Overlearning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Increasing the number of neurons or layers in neural networks for engine control models leads to overlearning, reducing generalization performance and prediction precision for unknown data.
Innovation Solution
Incorporating a fully connected layer between the input layer and a layer with a return path in the neural network, using a rectified linear unit activation function to remove unnecessary features and compress data, thereby reducing the number of neurons and layers needed, and preventing overlearning.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the number of neurons or layers in the neural network is increased to improve prediction precision, then the model's ability to fit training data improves, but overlearning occurs and generalization performance decreases
Solution Approach 1:
A fully connected layer is introduced as an intermediary component between the input layer and the layer with return path. This intermediate layer processes and compresses input features before they reach the recurrent portion of the network, enabling better generalization without increasing the overall network size. The fully connected layer acts as a feature extractor that reduces dimensionality and removes redundant information.
Solution Approach 2:
The patent extracts and removes unnecessary features from the input data through the fully connected layer with rectified linear unit activation. By selectively removing redundant or irrelevant features, the model avoids overlearning while maintaining prediction precision on relevant patterns.
2Adaptability or versatility
If the number of neurons or layers is increased to capture more complex patterns, then the model's representational capacity increases, but the degree of freedom increases leading to overlearning
Solution Approach 1:
The neural network is segmented into distinct functional portions: an input layer, a fully connected layer for feature extraction, and a layer with return path for temporal processing. This segmentation allows each portion to have a specific function, maintaining representational capacity while controlling overall complexity through modular design.
Solution Approach 2:
The patent transforms the input feature space by adding a new dimensional processing stage through the fully connected layer. This intermediate dimension processes features before they enter the recurrent structure, effectively increasing representational capacity without simply adding more neurons or layers in the traditional sense.
3Adaptability or versatility
If a recurrent neural network structure is used to process time series data, then the model can capture temporal dependencies, but the complexity of the network structure increases
Solution Approach 1:
The recurrent neural network is segmented such that only a specific layer contains the return path for temporal processing, while other layers perform feature extraction. This segmentation concentrates the temporal processing complexity in one portion of the network, making the overall structure more manageable and interpretable.
Data Source
AI summary
An engine control device includes an engine model having a neural network that inputs a manipulated variable of the engine and computes a controlled variable; and a controller that computes the manipulated variable so as to reduce a deviation between the controlled variable and a target controlled variable. The neural network includes an input layer to which the manipulated variable are input; a first hidden layer including a first fully connected layer; a second hidden later including a second fully connected layer that generates a plurality of second output values at a first time and has a return path on which the plurality of second output values at a second time, earlier than the first time, are input into the second fully connected layer; and an output layer from which the plurality of second output values at the first time are output as the controlled variable.