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

Ambient temperature changes can significantly affect the measurement accuracy of non-dispersive infrared (NDIR) carbon dioxide (CO₂) sensors. This study evaluates a temperature compensation method that uses the Whale Optimization Algorithm (WOA) to optimize the initial weights and biases of a backpropagation neural network (BPNN). The resulting WOA-BPNN model is compared with a conventional BPNN and a particle swarm optimization-optimized backpropagation neural network (PSO-BPNN) under the same experimental protocol. Using the complete 30-point dataset for model fitting, the WOA-BPNN achieved a maximum absolute error of 27.55 ppm and a mean absolute percentage error (MAPE) of 3.87% on the eight measured 45 °C conditions, outperforming BPNN and PSO-BPNN within the tested conditions. These results demonstrate the ability of WOA-BPNN to fit the measured temperature-concentration data and provide a promising basis for temperature compensation.