Whale-Optimized Neural Network Curbs Temperature Drift in CO2 Sensors

Non-dispersive infrared (NDIR) carbon dioxide sensors can lose measurement accuracy as ambient temperature changes. A study published in Environmental Monitoring and Assessment on 24 October 2026 reports a temperature compensation method that pairs a backpropagation neural network (BPNN) with the Whale Optimization Algorithm (WOA), which is used to tune the network’s initial weights and biases before training.

The research team, led by Yang He and colleagues, tested the resulting WOA-BPNN model against two benchmarks under the same experimental protocol: a conventional BPNN trained on its own, and a BPNN whose initial parameters were instead optimized with particle swarm optimization (PSO-BPNN). All three models were fitted using the complete 30-point dataset of measured temperature and concentration values, allowing a direct comparison of how well each approach captured the relationship between temperature and sensor output.

The clearest results came from the eight measured conditions at 45 degrees Celsius. On those points, the WOA-BPNN model achieved a maximum absolute error of 27.55 parts per million and a mean absolute percentage error (MAPE) of 3.87 percent. Within the tested conditions, the WOA-optimized network outperformed both the conventional BPNN and the PSO-BPNN, indicating that the choice of optimization algorithm for setting the network’s starting parameters had a measurable effect on compensation accuracy.

The authors describe these findings as demonstrating the WOA-BPNN model’s ability to fit the measured temperature-concentration data and as providing a promising basis for temperature compensation in NDIR CO2 sensors. The study’s scope is limited to the conditions included in the 30-point dataset, and the reported error figures apply to the eight 45-degree-Celsius measurement points; performance outside those tested conditions was not reported in the available results. Even so, the comparison offers a concrete, quantified example of how metaheuristic optimization of neural network initialization can reduce the temperature-induced errors that otherwise degrade NDIR CO2 measurements, and it positions WOA-based tuning as a competitive alternative to particle swarm optimization for this class of sensor compensation problems.

Subject of Research: Research on temperature compensation method of NDIR CO2 sensor based on WOA-BPNN

Article Title: Research on temperature compensation method of NDIR CO2 sensor based on WOA-BPNN

Article References: He, Y., Lin, D., Liu, S., Xu, B., Zhao, L., & Li, H. (2026). Research on temperature compensation method of NDIR CO2 sensor based on WOA-BPNN. Environmental Monitoring and Assessment, 198(11), Article 1249. https://doi.org/10.1007/s10661-026-16101-7

Image Credits: AI Generated

DOI: 10.1007/s10661-026-16101-7

Keywords: Research, temperature, compensation, method, NDIR, sensor, based, WOA-BPNN, scientific research