Heterogeneous sensor integration through Co-Evolutionary Genetic Programming with Multiple Individual Representations

Abstract

Efficient integration of heterogeneous sensors is essential for optimizing complex sensor systems, however, variations in sensor types, data formats, and measurement units create significant challenges for seamless data integration and interoperability. Similarity Features (SFs) are commonly used to quantify relationships across diverse sensor data, yet no single SF is universally effective due to the inherent complexity of sensor data. Thus, combining multiple SFs is necessary for accurate integration. Although Genetic Programming (GP) offers a powerful solution for constructing SF combinations, it often encounters difficulties with the complex interactions in heterogeneous data, leading to local optima. To overcome these challenges, this paper introduces Co-Evolutionary Genetic Programming with Multiple Individual Representations (CEGP-MIR), which combines tree-based and linear GP representations to enhance the search and optimization process through dynamic population interaction. This approach includes a novel interaction mechanism for adaptive cooperation between representations and new crossover operators to prevent stagnation in local optima. Experiments use OAEI’s Benchmark, Conference datasets and ten real-world sensor datasets to test the performance of CEGP-MIR. Results demonstrate that the designed CEGP-MIR enhances sensor entity alignment and improves overall efficiency in heterogeneous sensor data integration.