Vehicle Drive Control Using Design-Value Neural Network Layers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing neural network models fail to effectively use design values of hardware as input parameters, limiting their applicability to hardware with different design values, and thus struggle to maintain prediction precision for output parameters.
Innovation Solution
A control device for vehicle drive devices that employs a neural network with separate input layers for design values and other input parameters, where the second hidden layer is configured to be closer to the output layer than the first hidden layer, allowing design values to correct feature values generated from other input parameters, thereby improving prediction precision.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If design values of hardware are used as input parameters of the neural network, then prediction precision of output parameter is improved, but device complexity increases
Solution Approach 1:
The neural network is segmented into multiple input layers: a first input layer for receiving input parameters other than design values, and a second input layer for receiving design values. This segmentation allows the network to process different types of inputs separately while maintaining improved prediction precision without excessive complexity increase.
Solution Approach 2:
The patent adds a new dimension to the neural network architecture by introducing a second input layer dedicated to design values. This dimensional expansion enables the network to incorporate hardware design parameters as additional input features, thereby improving prediction precision while maintaining a structured and manageable network complexity through organized layer configuration.
2Adaptability or versatility
If a model using a neural network is applied to a plurality of hardware with different design values, then adaptability is improved, but prediction precision deteriorates
Solution Approach 1:
The neural network is designed with universal applicability to multiple hardware configurations by incorporating design values as explicit input parameters through the second input layer. This universal structure allows the same model to be applied across different hardware while maintaining prediction precision by accounting for design value variations in the computation process.
Solution Approach 2:
The patent addresses adaptability to different hardware by allowing parameter changes through the inclusion of design values as input parameters. The neural network processes varying design values (such as compression ratio, number of cylinders, cylinder array) as inputs, enabling the model to adapt to different hardware configurations while maintaining accurate predictions through the configured multi-layer architecture.
3Device complexity
If design values are input together with other input parameters to the same input layer, then device complexity is reduced, but prediction precision deteriorates
Solution Approach 1:
Rather than mixing design values with other input parameters in a single input layer, the patent segments them into separate input layers. The first input layer receives input parameters other than design values, while the second input layer receives design values. This segmentation improves prediction precision by allowing the network to process different parameter types through appropriately configured layers, with the second hidden layer positioned closer to the output layer to give design values appropriate influence.
Data Source
Figure 1
Figure 2~3
Figure 4
AI summary
A control device 60 of a vehicle drive device comprises a processing part 81 configured to use a trained model using a neural network to calculate an output parameter of a vehicle, and a control part 82 configured to control the vehicle drive device based on the output parameter. The neural network includes a first input layer to which input parameters of the vehicle other than a design value are input, a second input layer to which the design values are input, a first hidden layer to which outputs of the first input layer are input, a second hidden layer to which outputs of the second input layer are input, and an output layer outputting the output parameter, and is configured so that the second hidden layer becomes closer to the output layer than the first hidden layer.