In view of an accurate prediction of the impact of power load on optimizing power generation and scheduling plans, as well as an improvement of economic efficiency, so as to ensure safe operation of the power grid, a short-term power load forecasting model has thus been proposed based on perceived temperature and improved Fick’s law algorithm (IFLA) optimized CNN BiLSTM. Logistic mapping, Cauchy Gaussian mutation strategy, spiral wave search, and other techniques are used to improve FLA. Firstly, the features of meteorological data are amplified by adopting the somatosensory temperature formula. Secondly, the CNN BiLSTM network is subjected to hyperparameter optimization using IFLA. Finally, the CNN BiLSTM performs feature extraction on the data and outputs prediction results. On the basis of simulation experiments on the residential load dataset of a certain location in Hunan Province in March 2022, the experimental results show that the IFLA-CNN BiLSTM prediction model outputs root mean square error, average absolute error, average absolute percentage error, and coefficient of determination of 1.305, 0.882, 2.558%, and 0.989, respectively, verifying the generalization and reliability of the IFLA-CNN-BiLSTM model in practical environmental applications.