The calculation of compaction degree for earth-rock dams based on non-embedded methods can assess compaction levels. However, due to the variability in the source, filling processes, and environmental conditions of earth-rock dams, the spatial distribution of fill materials is often uneven, particularly in mixed fill, leading to discrepancies in compaction degree and complicating assessment efforts. To enhance the accuracy of compaction degree evaluation, the latest deep learning EvoLearn model is utilized to correct compaction parameters. This method employs a weight optimization strategy that combines Genetic Algorithm (GA) with Backpropagation to optimize neural network weights, thereby improving model robustness and performance. Through selection, crossover, mutation, and other operations, the mixed error between the sample set and the test set is optimized, so as to improve the generalization performance of the compression prediction model and reduce overfitting. The experimental results show that the new algorithm proposed in this paper can effectively improve the accuracy of predicting the compaction degree of earth-rock dams with intelligent sensing technology.