| In the process of coal mining,mine water damage often accompanies the occurrence,rapid analysis of mine water source types,in advance to make mining decisions and response plans,can reduce the risk of water damage and economic losses.Based on this,this paper designs a rapid mine water source identification system based on Raspberry Pi development board and Laser-induced fluorescence(LIF)technology.The system uses the Raspberry Pi development board as the processing core,combined with deep learning models and LIF technology,to realize the overall water source identification system.This paper first introduces the classification and identification index of mine water sources,and then introduces the principle of LIF technology and the construction of water source identification system,uses the spectral acquisition system to obtain the fluorescence spectral data of experimental water samples,uses two preprocessing algorithms of Savitzky-Golay filter(SG)and normalize to reduce noise interference and improve the signal-to-noise ratio of spectral data,and then uses Principal Component Analysis(PCA)and Linear Discriminant Analysis(LDA)reduce the dimensionality of the original spectral data and preprocessed spectral data to 3 dimensions,select three sets of data with better clustering effect,and use them to train the Long Short-Term Memory(LSTM)neural network water source recognition model after dividing the data set.Using Genetic Algorithm(GA)and Mayfly Algorithm(MA)to optimize the water source identification model,a total of 9 water source identification models were built,and the advantages and disadvantages of the models were identified by comparing the prediction situation and model evaluation indicators in the test set,and the comparison of each model was as follows:(1)Test set prediction: the prediction results of the unoptimized LSTM water source identification model are the worst,and the prediction results of the recognition model after MA optimization are close to the real value,and the optimization effect is significantly better than GA;Original-PCA,Normalize-PCA,SG-LDA compared the predicted values of the original LSTM recognition model,and the prediction results of the spectral data after SG-LDA treatment were closer to the real value,which was more suitable for training the water source recognition model.(2)Multi-classification model confusion matrix: the LSTM water source identification model built by SG-LDA has the best classification recognition situation and the highest accuracy,and the three indicators of Precision,Recall,and F1-score are all 1.00;when the LSTM water source identification model built by Normalize-PCA is not optimized by MA,the classification effect is the worst,and there are 2 misjudgments,Precision,Recall,The three indicators of F1-score are all 0.97;the water source identification model built by PCA-Original has a misjudgment when it is not optimized,and the three indicators of Precision,Recall,and F1-score are all 0.99.(3)The accuracy change trend of LSTM: water source identification model after MA optimization reaches 1.00 under shorter iterations,GA has a certain optimization effect on the model;the spectral data after SG-LDA processing is more suitable for building LSTM water source identification model,and after MA optimization,the accuracy of the model reaches the best under the shortest number of iterations.(4)Loss function change trend: The loss function of the water source identification model optimized by MA reaches convergence the fastest and the model tends to be stable,and the LSTM water source identification model built by the spectral data processed by SG-LDA has a faster convergence curve and stability than PCA-Original and Normalize-PCA.Select the SG-LDA-LSTM-MA water source identification model with the best performance,build the Tensor Flow Lite migration environment through the model migration method,migrate the model to the Raspberry Pi 4B development board,connect the communication between the Raspberry Pi and the spectrometer through the host computer,obtain the fluorescence spectrum data of water samples,analyze the embedded water source identification model,and finally display the water sample type and spectral analysis on the LCD to realize the overall construction of the hardware equipment.After actual verification,it can be concluded that the mine water source identification system designed in this paper has certain feasibility.Figure 39 Table 2 Reference 78... |
