Producing reliable landslide susceptibility maps is crucial for effective landslide prevention. However, previous studies have placed less emphasis on the technique for dividing attribute intervals of evaluation facto...
详细信息
Producing reliable landslide susceptibility maps is crucial for effective landslide prevention. However, previous studies have placed less emphasis on the technique for dividing attribute intervals of evaluation factors when assessing landslide susceptibility. This study aims to compare the influences of the natural breaks method and the frequency ratio method (FR) as techniques for dividing attribute intervals of continuous evaluation factors. Additionally, different susceptibility assessment models, including frequency ratio (FR), frequency ratio coupled with analytic hierarchy process (FR-AHP), frequency ratio coupled with BP neural network (FR-BP), and frequency ratio coupled with mean impact value and BP neural network (FR-miv-BP), were utilized to explore landslide susceptibility in Huichang County and its surrounding areas. Based on field investigations and correlation analysis, ten evaluation factors were selected, comprising land use, lithology, plan curvature, profile curvature, slope, aspect, elevation, distance to fault, distance to road, and normalized vegetation index (NDVI). Finally, the uncertainties were tested using receiver operating characteristic curves (ROC) and the distribution law of susceptibility index. The results indicated that the FR method yielded higher accuracy than the natural breaks method in dividing attribute intervals. The miv algorithm optimized BP neural network demonstrated a 0.7 to 0.8% improvement compared to the traditional BP neural network. The FR-miv-BP models exhibited smaller mean values and larger standard deviations for the landslide susceptibility index, suggesting better alignment with the actual landslide distribution features. This study has important implications for selecting evaluation factor division methods and appropriate models, providing valuable landslide susceptibility mapping for disaster prevention in Huichang County and its surrounding areas.
A computational and test method for calibrating the flight loads carried by aircraft wings is *** wing load is measured in real-time based on the resistance and fiber Bragg grating strain *** linear stepwise regressio...
详细信息
A computational and test method for calibrating the flight loads carried by aircraft wings is *** wing load is measured in real-time based on the resistance and fiber Bragg grating strain *** linear stepwise regression method is used to construct the load *** mean impact value algorithm is employed to select suitable *** the ground calibration experiment,the wing load calculation equations in both forward and reverse installation states are *** correctness of the load equations was verified through equation error and inspection error ***,the actual flight load of the wing was obtained through flight tests.
In order to guarantee the precision of the parameters of the probability integral method (PIM), starting from optimizing input and improving algorithm an algorithm integrating the genetic algorithm (GA) and particle s...
详细信息
In order to guarantee the precision of the parameters of the probability integral method (PIM), starting from optimizing input and improving algorithm an algorithm integrating the genetic algorithm (GA) and particle swarm optimization (PSO) was put forward to optimize the prediction model of BP neural network and the mean impact value algorithm (miv) was applied to optimize the input of BP neural network. The mean impact value algorithm (miv) was applied to optimize the input of BP neural network. The measured data of 50 working faces were chosen as the training and testing sets to build the miv-GP-BP model. The results showed that among the five parameters, the RMSE was between 0.0058 and 1.1575, the MaxRE of q, tan beta, b and theta was less than 5.42%, and the MeaRE was less than 2.81%. The RMSE of s/H did not exceed 0.0058, the MaxRE was less than 9.66% and the MeaRE was less than 4.31% (the parameters themselves were small). The optimized neural network model had higher prediction accuracy and stability.
An online sequence extreme learning machine neural network prediction model based on sample aggregation method (AOSELM) is proposed, which is used to solve the prediction of the charging process of lead-acid batteries...
详细信息
An online sequence extreme learning machine neural network prediction model based on sample aggregation method (AOSELM) is proposed, which is used to solve the prediction of the charging process of lead-acid batteries. The model is based on the online sequence extreme learning machine (OSELM) neural network and finds the similar samples of the test samples based on the mean influence value (MW) algorithm. Similar samples are used for temporary incremental learning of the neural network, then the updated neural network makes predictions on the test samples. Taking the terminal voltage prediction of the battery charging process as an example, the BP neural network, the OSELM neural network and the AOSELM neural network are used to compare the predictions, and the charging process of the battery is predicted based on the AOSELM neural network. The comparison results show that the AOSELM neural network has higher prediction accuracy and adaptive ability than the BP neural network, and reduces the maximum absolute percentage error and the maximum absolute error of the prediction result when compared with the OSELM neural network. The prediction shows that the charging process of lead-acid batteries can be accurately predicted by the AOSELM neural network.
In this paper, we take the average impact value method as the evaluation of neural network variable correlation indicators, analysis the data provided by Professor *** and A. Morais from University of Minho(Portugal) ...
详细信息
ISBN:
(纸本)9781509046584
In this paper, we take the average impact value method as the evaluation of neural network variable correlation indicators, analysis the data provided by Professor *** and A. Morais from University of Minho(Portugal) using the mivBP algorithm to filtrate 13 characterization factors to get 7 characterization parameters affect forest fires, construct the simulation model of the prediction of forest fires based on the support vector machine algorithm, using a test set for testing, the accuracy rate reached 91.89%.
暂无评论