Heterogeneous Data Fusion for Customized Product Performance Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for predicting the performance of customized products in the design stage are either costly and time-consuming or provide low reliability due to high errors in calculation simulations.
Innovation Solution
A customized product performance prediction method based on heterogeneous data difference compensation fusion, using a BP neural network model combined with a neighborhood association method, similarity difference compensation method, and depth auto-encoder for data encoding and difference compensation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If historical actual measurement performance data is used for deduction and prediction, then the credibility of performance prediction is improved, but the cost increases, the cycle becomes long, and the response becomes slow
Solution Approach 1:
The patent merges two data sources: high-fidelity historical actual measurement data and low-fidelity calculation simulation data. By combining these heterogeneous data sources through a unified prediction model, the system achieves both high credibility (from actual measurement data) and fast response (from simulation data), resolving the contradiction between reliability and productivity.
Solution Approach 2:
The prediction model is designed to handle multiple types of data (actual measurement data and simulation data) with different fidelities. This multi-functional capability allows the system to leverage both data sources simultaneously, achieving reliable predictions while maintaining fast response times through the complementary strengths of each data type.
2Productivity
If calculation simulation is used for performance prediction, then the response is efficient and fast, but the calculation simulation has a relatively large error, such that the reliability of performance prediction is low
Solution Approach 1:
The patent introduces a data fusion mechanism as an intermediary between calculation simulation data and final predictions. This intermediary process compensates for errors in simulation data by integrating it with actual measurement data, thereby maintaining the fast response advantage of simulation while improving the reliability of predictions.
Solution Approach 2:
The system dynamically adjusts the weighting and fusion parameters of different data sources based on their respective fidelities and uncertainties. By changing these parameters adaptively, the model optimizes the contribution of simulation data (for speed) and actual measurement data (for accuracy), resolving the contradiction between response efficiency and prediction reliability.
3Reliability
If traditional product performance prediction methods are used, then either high credibility or fast response can be achieved, but both efficient and credible prediction cannot be achieved simultaneously
Solution Approach 1:
The patent employs adaptive parameter adjustment in the data fusion process, where the model dynamically optimizes weighting parameters based on data characteristics. This parameter optimization approach enables the system to achieve both credibility and efficiency without requiring overly complex multi-model architectures, as the same unified model adapts its parameters to handle different data scenarios.
Data Source
AI summary
Disclosed is a customized product performance prediction method based on heterogeneous data difference compensation fusion. The method includes: on the basis of a depth auto-encoder, a neighborhood association method and a similarity difference compensation method, performing difference compensation correction on a calculation simulation data set by using a historical actual measurement data set; and training a BP neural network model by using the calculation simulation data set after the difference compensation correction to serve as a performance prediction method of a customized product. According to the method of the present application, by combining the depth auto-encoder, and by utilizing the neighborhood association method and the similarity difference compensation method, low-fidelity calculation simulation data is associated with high-fidelity historical actual measurement data, such that the difference compensation correction of the low-fidelity calculation simulation data is realized by using the high-fidelity historical actual measurement data.