|
This article proposes a research framework based on neural network methods for predicting and decision support the effectiveness of soil pollution remediation in mines. To address the complexity, nonlinearity, and uncertainty of soil pollution in mines, neural network models in deep learning are utilized to train and predict data from multiple sources, including soil pollutant types, contents, pollution ranges, remediation techniques, meteorological conditions, etc. By constructing multi-layer perceptron models or convolutional neural network models and utilizing existing mining soil data to predict the effectiveness of future mining soil remediation, it is possible to simulate and reveal the inherent laws in the process of mining soil pollution remediation, achieve quantitative prediction of remediation effects, and provide scientific decision support for the selection and optimization of pollution remediation strategies based on the monitoring data of current mining soil, We constructed backpropagation neural network (BPNN) and convolutional neural network (CNN) prediction models for soil remediation performance indicators in mines, and compared and analyzed the prediction results of basic soil physical and chemical properties index. The data showed that the predicted values of the BPNN model were consistent with the actual soil conditions, but overfitting occurred during the fitting process. We improved the parameter selection method of the BPNN model using particle swarm optimization (PSO) algorithm, this phenomenon can be avoided, and CNN models have a more complex structure and more scientific fitting methods, thereby avoiding the model from falling into local extremes during the fitting process. At the same time, the use of deconvolution feature fusion optimizes the network feature extraction ability, improving the accuracy of model prediction results. Using the root mean square error (RME), coefficient of determination (R2), and mean absolute error (MAE) of different models as evaluation parameters, compared with traditional BPNN and PSO-BP models, the improved model has a more stable fitting effect, which is of great value for guiding practical repair engineering practices.
|
