Soil thickness prediction method and system

Abstract

The invention discloses a soil thickness prediction method and system, and relates to the technical field of transmission tower construction. The method comprises the following steps: acquiring digital elevation model data, multispectral remote sensing image data and geological lithology data, and respectively calculating a terrain humidity index, a lithology weighting coefficient, a terrain roughness index, a slope index and a normalized vegetation index based on the data. According to the method, a digital elevation model, a multispectral remote sensing image and geological lithology data are systematically integrated, a multi-factor evaluation system comprising a terrain humidity index, a lithology weighting coefficient, a terrain roughness index, a slope index and a normalized vegetation index is constructed, and the weight of each factor is scientifically determined by adopting principal component analysis or multiple linear regression; the accumulation potential index comprehensively reflecting the soil is generated through weighted fusion, and the inherent limitation that a traditional Kriging method only depends on the spatial distance is overcome.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power transmission tower construction, in particular to a soil thickness prediction method and system. BACKGROUND

[0002] In the survey and design stage of the pile foundation engineering of the power transmission line tower, accurately obtaining the soil thickness distribution of the construction area is a key prerequisite to ensure that the pile foundation can be seated on the stable bedrock, and this information is directly related to the safety and economy of the project.

[0003] At present, in the engineering practice, the soil thickness prediction of the area between the drilling points mainly relies on the spatial interpolation method based on geostatistics, among which the ordinary Kriging method is the representative. The core idea of this method is "spatial autocorrelation", that is, it is believed that the points adjacent in space have similar attribute values. In operation, a semi-variogram function model is fitted according to the spatial position and soil thickness data of the known drilling points, so as to quantify the law of the change of spatial correlation with distance. Finally, the value of the point to be predicted is obtained by weighted average of the measured values of the surrounding known points, and the weight is completely determined by the spatial distance relationship between the known points and the point to be predicted.

[0004] However, this method has the following key problems in actual application, especially in complex mountainous areas: 1) the prediction result is out of touch with the physical cause. The distribution of soil thickness is not determined by simple geometric distance, but is jointly controlled by many physical and biological factors such as topography, lithology, erosion and accumulation process, vegetation cover, etc. For example, a point located on a ridge and a point located in an adjacent valley, although the spatial distance is very close, due to the erosion of the ridge and the accumulation of the valley, the soil thickness of the two points will be very different. The ordinary Kriging method cannot identify the difference in cause, and only gives high weight because the two points are close in distance, resulting in a large prediction error in these key topographic parts; 2) the precision is insufficient in the key engineering area. The tower site of the power transmission line is often arranged in the ridge, slope and other shallow soil layer areas, which are the areas of spatial variability of soil thickness. The low precision prediction of the traditional prediction method in these areas cannot reliably guide the pile foundation site selection, which may lead to the risk of the pile foundation seated in the soft soil layer or the over-conservative excessive drilling in the subsequent exploration, resulting in unnecessary increase of engineering cost and time.

[0005] Therefore, we provide a soil thickness prediction method and system to solve the above problems. SUMMARY

[0006] The present application aims to provide a soil thickness prediction method and system, which solves the problems of low prediction accuracy and neglecting physical mechanism in complex terrain area by fusing multi-source environmental data to construct accumulation potential index and adaptively modifying the cause-oriented spatial weight of Kriging based on the index.

[0007] To solve the above technical problems, the present application is realized by the following technical solutions: The present application provides a soil thickness prediction method, comprising the following steps: Collecting digital elevation model data, multispectral remote sensing image data and geological lithology data, calculating terrain humidity index , lithology weighting coefficient , terrain roughness index , slope index and normalized difference vegetation index based on these data respectively; Weighted fusion is performed on the terrain humidity index , lithology weighting coefficient , terrain roughness index , slope index and normalized difference vegetation index to calculate the accumulation potential index ; Using the soil thickness data of known drilling points, a semi-variogram function model is established by ordinary Kriging interpolation method to calculate the initial spatial weight of each known point to the interpolation point ; Based on the accumulation potential index , the initial spatial weight is adaptively modified to obtain the cause weight ; Weighted summation is performed by the cause weight to calculate the soil thickness prediction value of the interpolation point, wherein is the soil thickness measurement value of the known point ; The prediction result is verified by the reserved verification method, and the root mean square error and the mean absolute error As an evaluation index, when the prediction error exceeds the threshold value allowed by engineering, the penalty correction training process is started to optimize the model parameters, the influence of multiple factors such as terrain, lithology and vegetation on soil thickness is effectively integrated by comprehensively collecting multi-source data and calculating key environmental indexes, and the limitation of traditional Kriging method relying only on spatial distance is overcome;Through weighted fusion to generate accumulation potential index, and combined with adaptive weight correction, the prediction accuracy in complex topography is significantly improved, and the error caused by the difference between erosion and accumulation process is reduced;At the same time, the verification and penalty correction mechanism is reserved to ensure the reliability and adaptive optimization ability of the model, finally more accurate and economic guidance is provided for the site selection of power transmission tower pile foundation, the exploration cost and engineering risk are reduced, and the pile foundation is ensured to be located on stable bedrock.

[0008] The application is further provided with a terrain humidity index The calculation formula is

[0009] Among them The specific catchment area obtained by hydrological analysis of digital elevation model is obtained by calculating the upstream catchment area flowing to the point, The slope angle calculated by digital elevation model, 0.001 The small constant added to prevent the denominator from being zero, by accurately calculating the terrain humidity index, the influence of terrain on soil humidity and accumulation potential can be quantified, based on digital elevation model hydrological analysis to obtain catchment area and slope angle, the sensitivity of the model to topographic changes is enhanced, so that the difference in soil thickness between humid and dry areas can be more accurately identified, the physical cause correlation of the prediction is improved, and the error in key topographic parts is reduced, providing more reliable topographic basis for pile foundation site selection.

[0010] The application is further provided with a lithology weighting coefficient The calculation formula is: Among them The weight of the first Class of lithology is determined according to the anti-weathering ability of lithology by expert experience method or analytic hierarchy process, the hard and difficult weathering lithology is given lower weight, and the soft and easy weathering lithology is given higher weight, The lithology indicator variable is 1 when the point belongs to the first Class of lithology, otherwise 0, by introducing the lithology weighting coefficient, the influence of the anti-weathering ability of lithology on soil formation is fully considered, the hard lithology and soft lithology are distinguished, the adaptability of the model to geological background is enhanced, so that the soil thickness distribution in different lithology areas can be more accurately predicted, the prediction deviation caused by uneven lithology is reduced, and the reliability under complex geological conditions is improved, providing scientific support for the positioning of power transmission tower pile foundation in bedrock stable area.

[0011] The present invention is further configured such that the terrain roughness index The calculation formula is:

[0012] in The elevation value of the center point. For its surroundings The elevation values ​​of the neighboring points are adopted. When performing calculations by moving the window The calculation is completed by traversing all pixels in the study area. The terrain roughness index can quantify the degree of surface undulation and reflect the intensity of erosion and deposition processes. The moving window traversal analysis ensures the comprehensiveness of the data, thereby more accurately identifying the difference in soil thickness between rough and smooth terrain, enhancing the model's ability to capture micro-topographic changes, improving the spatial resolution of predictions, reducing errors in areas with large undulations, and providing a more detailed terrain risk assessment for pile foundation site selection.

[0013] The present invention is further configured such that the slope index The calculation formula is: in For a specific catchment area, The slope angle is given by the index 1.2, which is an empirical value based on the law of water flow power; the normalized vegetation index is... The calculation formula is:

[0014] in and These are the reflectance values ​​in the near-infrared and red bands of remote sensing images, respectively, extracted from the corresponding band data of satellite remote sensing images. By combining the slope index and the normalized vegetation index, they comprehensively reflect the water erosion capacity and the regulatory effect of vegetation cover on soil thickness. The slope index, based on the water flow power law, enhances the model's simulation accuracy of the erosion process, while the normalized vegetation index introduces biological factors and can indirectly indicate soil organic matter and stability, thereby comprehensively improving the prediction's comprehensiveness, reducing errors in densely vegetated or steep slope areas, and providing more comprehensive environmental adaptability and safety assurance for pile foundation engineering.

[0015] The present invention further includes the method comprising: standardizing the obtained indices, wherein the standardization of the indices is performed using the maximum-minimum standardization method or... Standardization method, the formula for standardizing the maximum and minimum values ​​is:

[0016] Standardized formula ,in The original value, and min and max respectively, average, standard deviation, by standardizing each index to eliminate the influence of data dimension and distribution difference, ensure the fairness and comparability of weighted fusion, the maximum and minimum standardization method is suitable for the case where the data range is clear, and the standardization method is more suitable for normal distribution data, thereby enhancing the stability and generalization ability of the model, reducing the prediction fluctuation caused by uneven data, improving the applicability in different geographical regions, and providing more consistent and reliable soil thickness prediction results for power transmission tower engineering.

[0017] The application further provides that the method further comprises verifying the prediction results by a reserved verification method, and using root mean square error and mean absolute error as evaluation indexes, when the prediction error exceeds the threshold value allowed by the project, a penalty correction training process is started to optimize the model parameters, and when the prediction error is lower than the predetermined threshold value, the generated soil thickness prediction map is directly used to guide the targeted drilling of the power transmission tower pile foundation, so as to determine the optimal position of the pile foundation in the stable bedrock.

[0018] The application further provides that the reserved verification method comprises: reserving 10%-20% of the known drilling data as a verification set before modeling, calculating the root mean square error and mean absolute error between the predicted soil thickness value and the actual drilling measured value, wherein is the number of verification points, is the predicted value of the verification point, is the measured value of the verification point, and the calculation result is compared with the error threshold value and allowed by the project, the reserved verification method is used to realize objective evaluation of the prediction model, the root mean square error and the mean absolute error are used as quantitative indexes, the prediction accuracy and deviation can be comprehensively reflected, the result is ensured to meet the actual demand by comparing with the engineering threshold value, thereby enhancing the credibility and practicability of the model, reducing the risk of overfitting or underfitting, providing data-driven decision support for the design of the power transmission tower pile foundation, and reducing the engineering rework or safety accidents caused by inaccurate prediction.

[0019] The application further provides that the penalty correction training process comprises: firstly calculating the penalty factor

[0020] of each point in the verification set, wherein is a penalty intensity coefficient, and the value is is a hyperbolic tangent function; then, based on the penalty factor of all validation points, the similarity bandwidth parameter is adjusted by gradient descent method , and the specific updating formula is

[0021] wherein is a learning rate, is a partial derivative of the root mean square error to the bandwidth parameter, an adaptive optimization mechanism is introduced by a penalty correction training process, the penalty factor is dynamically adjusted based on the error size, the hyperbolic tangent function ensures that the correction is smooth and stable, and the bandwidth parameter is automatically optimized by combining the gradient descent method, so as to quickly converge to the best model state, reduce the prediction error and improve the model robustness, especially suitable for soil thickness prediction under complex terrain, and provides continuous improved prediction ability for power transmission tower engineering, reduces long-term operation and maintenance cost and improves engineering efficiency.

[0022] The application further sets that the calculation process of the partial derivative involves layer-by-layer derivation of the root mean square error , the soil thickness prediction value , the normalized cause weight , the cause weight , finally obtains a specific expression containing the initial space weight , the accumulation potential index difference and the current bandwidth parameter , and the layer-by-layer derivation of the partial derivative by the chain rule ensures the mathematical rigor and calculation efficiency of the penalty correction training, can accurately quantify the influence of the model parameters on the error, so as to realize fast parameter optimization, enhance the self-learning ability and adaptability of the model, reduce the need for manual intervention, improve the prediction stability in a variable environment, and provide a more intelligent and reliable soil thickness prediction solution for power transmission tower pile foundation engineering.

[0023] The application further sets that when the prediction error is lower than a predetermined threshold, the generated soil thickness prediction map is directly used to guide the targeted intensive drilling of the power transmission tower pile foundation, the best position of the pile foundation in the stable bedrock is determined, the efficient transformation of engineering practice is realized by directly applying the high-precision prediction map to guide the intensive drilling, the key area can be focused on, unnecessary exploration work is reduced, time and cost are saved, and at the same time, the pile foundation is ensured to be located on the stable bedrock, the engineering safety and economy are improved, a visual decision tool is provided for power transmission tower construction, the environmental disturbance and engineering risk are reduced, and the sustainability and reliability of the whole project are enhanced.

[0024] In a second aspect, the embodiments of the present application also provide a system for predicting soil thickness, which is deployed in a computer or server with data processing capability and comprises the following modules: a data acquisition module comprising a digital elevation model reading unit, a multispectral remote sensing image analysis unit and a geological lithology database interface unit, configured to obtain terrain, remote sensing and lithology raw data from external data sources; an index calculation module comprising a plurality of sub-calculation units, each configured to calculate a plurality of cause indexes based on terrain and lithology, wherein the cause indexes at least include a terrain humidity index based on catchment area and slope angle, a lithology erosion resistance index based on lithology classification and weight table, a terrain roughness index calculated by an elevation neighborhood window, a slope slope index calculated by combining catchment area and slope power, and a vegetation index calculated by remote sensing image infrared and red light band reflectivity; a standardization processing module configured to perform standardization processing on each cause index by using the maximum and minimum value method to unify the data scale; a accumulation potential calculation module configured to perform weighted fusion on the plurality of standardized cause indexes to generate a comprehensive accumulation potential index; an initial weight calculation module configured to construct a semi-variogram function model by using ordinary Kriging interpolation method based on soil thickness data of known drilling points, and calculate initial spatial weights of each known point to a target point; a weight correction module configured to adaptively correct the initial spatial weights by using a preset rule according to the accumulation potential index to obtain a cause weight, and perform normalization processing on the cause weight; a soil thickness prediction module configured to perform weighted summation on soil thickness values of known points by using the normalized cause weight to output a predicted soil thickness of a point to be interpolated; a verification module configured to calculate a prediction error by using reserved verification set drilling data, and compare the prediction error with a preset error threshold; a parameter optimization module configured to start an optimization process by adjusting a bandwidth parameter in the model by using a gradient descent method to optimize model accuracy when the prediction error exceeds the preset threshold; an output and application module configured to generate a soil thickness spatial distribution map when the prediction error meets the requirements to guide the intensive drilling and bedrock positioning work of the power transmission tower pile foundation.

[0025] Optionally, the index calculation module further comprises a terrain humidity index calculation unit for calculating an index reflecting the terrain catchment capacity based on the catchment area and slope angle parsed from the digital elevation model, and the adaptive correction process in the initial weight calculation module is as follows: a correction factor inversely proportional to the difference between the accumulation potential index at the known point and the accumulation potential index at the target point is constructed, and the initial spatial weight obtained by the ordinary Kriging interpolation method is dynamically adjusted using the correction factor to obtain the genetic weight which can better reflect the genetic relationship.

[0026] The present application has the following beneficial effects: 1. The present application integrates digital elevation model, multispectral remote sensing image and geological lithology data to construct a multi-factor evaluation system including terrain humidity index, lithology weighted coefficient, terrain roughness index, slope index and normalized vegetation index, and scientifically determines the weight of each factor by principal component analysis or multivariate linear regression to generate the accumulation potential index of soil by weighted fusion, which overcomes the inherent limitation of traditional Kriging method which only relies on spatial distance, makes the prediction model fully integrate into the physical mechanism of soil formation environment, and improves the identification accuracy and genetic rationality of the spatial distribution of soil thickness in complex topography conditions such as mountainous areas.

[0027] 2. The present application introduces a spatial weight adaptive correction mechanism based on the accumulation potential index, uses the index similarity to optimize the initial weight of the traditional Kriging, and combines the reserved verification method and the punishment correction training process to realize the dynamic adjustment of model parameters and the autonomous control of prediction error by using the hyperbolic tangent punishment factor and the gradient descent algorithm, thereby forming an intelligent prediction system with self-optimization ability, effectively solving the problem of large prediction deviation of traditional methods in key engineering positions, providing reliable technical support for the accurate positioning and exploration layout optimization of the pile foundation of the transmission tower, and greatly reducing the safety risk of the pile foundation located in soft soil layer and the economic cost caused by excessive exploration due to the misjudgment of soil thickness. BRIEF DESCRIPTION OF DRAWINGS

[0028] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed for the embodiment description will be briefly introduced as follows.

[0029] Figure 1 It is a flowchart of a soil thickness prediction method; Figure 2 It is a flowchart of multi-source data acquisition and environmental index calculation in a soil thickness prediction method; Figure 3 It is a flowchart of accumulation potential index synthesis in a soil thickness prediction method; Figure 4 It is a flowchart of genetic weight calculation in a soil thickness prediction method; Figure 5 It is a schematic diagram of a verification and self-optimization process for a soil thickness prediction method. DETAILED DESCRIPTION

[0030] The technical solutions in the embodiments of the present application will be described below with reference to the drawings in the embodiments of the present application. The described embodiments are only some of the embodiments of the present application, not all the embodiments.

[0031] Embodiment 1 Please refer to Figure 1 The present application is a soil thickness prediction method, comprising the following steps: Collecting digital elevation model data, multispectral remote sensing image data and geological lithology data, calculating terrain humidity index , lithology weighted coefficient , terrain roughness index , slope index and normalized vegetation index based on these data respectively; The indexes obtained are standardized, and then the weighted fusion formula

[0032] Calculate the accumulation potential index , wherein is a positive weight coefficient is a negative weight coefficient, and its value is determined by principal component analysis or multiple linear regression fitting with known soil thickness data; Using the soil thickness data of known drilling points, a semi-variogram function model is established by ordinary kriging interpolation method, and the initial spatial weight of each known point to the interpolation point is calculated i ; Based on the accumulation potential index , the formula

[0033] The initial spatial weight is adaptively corrected to obtain the genetic weight , wherein and are the accumulation potential index of the interpolation point and the th known point respectively, is a bandwidth parameter for controlling the similarity decay rate; The obtained genetic weight is normalized to obtain the normalized genetic weight , and then the normalized genetic weight is used to obtain the weighted sum formula

[0034] Calculate the predicted soil thickness at the interpolation point, where For known points The measured value of soil thickness; The prediction results were validated using the reserved validation method, with the root mean square error employed. and mean absolute error As an evaluation metric, when the prediction error exceeds the engineering-allowed threshold, a penalty correction training process is initiated to optimize the model parameters, including the topographic humidity index. The calculation formula is

[0035] in To obtain a specific catchment area through hydrological analysis using a digital elevation model, the area of ​​the upstream catchment area flowing towards that point is calculated. The slope angle is calculated using a digital elevation model. 0.001 The lithology weighting coefficient is a small constant added to prevent the denominator from being zero. The calculation formula is: ,in For the first The weight of lithology is determined based on the lithology's resistance to weathering through expert experience or analytic hierarchy process. Hard, difficult-to-weather lithologies are assigned lower weights, while soft, easily weathered lithologies are assigned higher weights. As a lithological indicator variable, when this point belongs to the first... The value is 1 for lithological types and 0 otherwise; the topographic roughness index. The calculation formula is:

[0036] in The elevation value of the center point. For its surroundings The elevation values ​​of the neighboring points are adopted. When performing calculations by moving the window The slope index is calculated by traversing all pixels within the study area. The calculation formula is: in For a specific catchment area, The slope angle is given by the index 1.2, which is an empirical value based on the law of water flow power; the normalized vegetation index is... The calculation formula is:

[0037] in and Respective reflectance values of near-infrared band and red band in remote sensing image, obtained by extracting corresponding band data of satellite remote sensing image, the method further comprising: standardizing each index obtained, the standardization of each index adopts maximum minimum value standardization method or standardization method, the maximum minimum value standardization formula is

[0038] standardization formula , wherein is the original value, and are minimum value and maximum value respectively, is the average value, is the standard deviation, the method further comprising verifying the prediction result by the reserved verification method, adopting root mean square error and mean absolute error as evaluation indexes, when the prediction error exceeds the threshold value allowed by the project, starting the punishment correction training process to optimize the model parameters, when the prediction error is lower than the predetermined threshold value, directly using the generated soil thickness prediction map to guide the targeted drilling of the power transmission tower pile foundation, determining the best position of the pile foundation in the stable bedrock, the reserved verification method comprising: reserving 10%-20% of the known drilling data as a verification set before modeling, calculating the root mean square error and mean absolute error between the predicted soil thickness value and the actual drilling measured value, wherein is the number of verification points, is the predicted value of the verification point, is the measured value of the verification point, comparing the calculation results with the error threshold value and allowed by the project, the punishment correction training process comprising: firstly calculating the punishment factor of each point in the verification set

[0039] , wherein is the punishment intensity coefficient, taking the value of is the hyperbolic tangent function; then based on the punishment factors of all verification points, adjusting the similarity bandwidth parameter by gradient descent method, the specific update formula is

[0040] , wherein is the learning rate, is the partial derivative of the root mean square error to the bandwidth parameter, the calculation process of the partial derivative involves the root mean square error , soil thickness prediction value , normalized genetic weight , genetic weight , the final expression containing the initial space weight , the difference value of the accumulation potential index and the current bandwidth parameter , when the prediction error is lower than the predetermined threshold value, the generated soil thickness prediction map is directly used to guide the targeted drilling of the power tower pile foundation, and the final position of the pile foundation in the stable bedrock is determined.

[0041] Specifically, by comprehensively collecting multi-source data and calculating key environmental indexes, the influence of multiple factors such as terrain, lithology, and vegetation on soil thickness is effectively integrated, overcoming the limitation of traditional Kriging method which only relies on spatial distance; by weighted fusion to generate accumulation potential index and combined with adaptive weight correction, the prediction accuracy in complex topography is significantly improved, and the error caused by differences in erosion-accumulation process is reduced; at the same time, the reserved verification and penalty correction mechanism ensures the reliability and adaptive optimization ability of the model, finally providing more accurate and economic guidance for transmission tower pile foundation site selection, reducing exploration cost and engineering risk, and ensuring that the pile foundation is located on stable bedrock, through accurate calculation of terrain humidity index, the influence of terrain on soil humidity and accumulation potential can be quantified, based on digital elevation model hydrological analysis to obtain catchment area and slope angle, the sensitivity of the model to topographic changes is enhanced, so that the difference in soil thickness between wet and dry areas is more accurately identified, the physical cause correlation of the prediction is improved, the error in key topographic parts is reduced, and more reliable topographic basis is provided for pile foundation site selection, by introducing lithology weighting coefficient, the influence of lithology weathering resistance on soil formation is fully considered, hard and soft lithology are distinguished, the adaptability of the model to geological background is enhanced, so that the soil thickness distribution in different lithology areas is more accurately predicted, the prediction deviation caused by uneven lithology is reduced, the reliability in complex geological conditions is improved, scientific support is provided for the positioning of transmission tower pile foundation in bedrock stable area, by calculating terrain roughness index, the degree of surface undulation can be quantified, reflecting the intensity of erosion and accumulation process, using moving window traversal analysis to ensure data comprehensiveness, so that the difference in soil thickness between rough and smooth terrain is more accurately identified, the model's ability to capture micro-topographic changes is enhanced, the spatial resolution of the prediction is improved, the error in areas with large undulations is reduced, more detailed topographic risk assessment is provided for pile foundation site selection, by combining slope index and normalized vegetation index to comprehensively reflect the adjustment effect of water flow erosion capacity and vegetation cover on soil thickness, slope index is based on water flow power law to enhance the simulation accuracy of the model to the erosion process, while normalized vegetation index introduces biological factors, which can indirectly indicate soil organic matter and stability, thereby improving the comprehensiveness of the prediction, reducing the error in areas with dense vegetation or steep slopes, providing more comprehensive environmental adaptability and safety protection for pile foundation engineering, by standardizing each index to eliminate the influence of data dimension and distribution difference, ensuring the fairness and comparability of weighted fusion, the maximum and minimum value standardization method is suitable for cases with clear data range, while the Z-score standardization method is suitable for cases with large data range, and the final prediction result is more accurate and reliable, providing more scientific and reliable guidance for transmission tower pile foundation site selection, reducing exploration cost and engineering risk, and ensuring that the pile foundation is located on stable bedrock. Standardization rules are better suited to normally distributed data, thereby enhancing the model's stability and generalization ability, reducing prediction fluctuations caused by data heterogeneity, and improving applicability across different geographical regions. This provides more consistent and reliable soil thickness prediction results for power transmission tower projects. The reserved validation method enables objective evaluation of the prediction model. Using root mean square error and mean absolute error as quantitative indicators comprehensively reflects prediction accuracy and bias. Comparison with engineering thresholds ensures the results meet actual needs, thus enhancing the model's credibility and practicality, reducing the risk of overfitting or underfitting, and providing data-driven decision support for power transmission tower foundation design. This reduces engineering rework or safety accidents caused by inaccurate predictions. An adaptive optimization mechanism is introduced through a penalty correction training process. The penalty factor is dynamically adjusted based on the error magnitude, and the hyperbolic tangent function ensures smooth and stable correction. Combined with gradient descent, the bandwidth parameter is automatically optimized, thereby quickly converging to the optimal model state, reducing prediction errors and improving model performance. This model exhibits robustness, making it particularly suitable for predicting soil thickness in complex terrains. It provides continuously improving predictive capabilities for power transmission tower projects, reducing long-term operation and maintenance costs and increasing project efficiency. Through a chain rule-based approach, it calculates partial derivatives layer by layer, ensuring mathematical rigor and computational efficiency in penalty correction training. It can accurately quantify the impact of model parameters on errors, enabling rapid parameter optimization, enhancing the model's self-learning ability and adaptability, reducing the need for manual intervention, and improving predictive stability in variable environments. It provides a more intelligent and reliable soil thickness prediction solution for power transmission tower foundation engineering. By directly applying high-precision prediction maps to guide intensified drilling, it achieves efficient transformation of engineering practice, enabling targeted focus on key areas, reducing unnecessary exploration work, saving time and costs, while ensuring the foundation is situated on stable bedrock, improving project safety and economy. It provides a visual decision-making tool for power transmission tower construction, reducing environmental disturbances and engineering risks, and enhancing the overall project's sustainability and reliability.

[0042] Example 2 Please see Figure 2 , Figure 3 and Figure 4 Based on Specific Implementation Example 1, in this scenario, on a sloping area with sparse vegetation cover and severe surface erosion... and The index has become a key factor controlling soil thickness distribution. In the calculation... When exponentiating, assign and (corresponding weight) and ) higher negative weights accurately reflect the characteristics of strong erosion area. The adaptive weight correction process effectively reduces the weight interference between the erosion area and the adjacent relatively stable area. In the verification stage, the punishment correction mechanism further enhances the model's ability to identify the abrupt boundary between erosion and accumulation processes, enabling the final generated prediction map to clearly delineate the potential risk area of bedrock exposure, directly guiding the construction of tower foundation by avoiding these areas.

[0043] Embodiment 3 Please refer to Figure 2 and Figure 5 , on the basis of specific embodiment one, in the area from hilly to plain, the lithology changes complex, the soil thickness is controlled by microtopography and lithology. In implementation, the focus is to use geological map to accurately calculate , and give different weights to different lithology through analytic hierarchy process. After constructing index and carrying out adaptive kriging interpolation, it is found that the prediction result has deviation in the lithology contact zone. Through the reserved verification method, the problem is identified and the punishment correction training is started. Using the same partial derivative calculation logic as in the previous embodiment, the bandwidth parameter is dynamically adjusted to make the model more sensitive to lithology changes. The final model's prediction result in this transition area is in good agreement with the actual situation, effectively guiding the precise location of the transmission tower foundation and avoiding engineering risks caused by misjudgment of lithology.

[0044] Embodiment 4 Please refer to Figure 2 and Figure 5 , on the basis of specific embodiment one, in the multi-stage terrace of river, the terrain is relatively flat but the soil thickness is significantly affected by ancient river and current hydrological conditions. In the method implementation, special emphasis is placed on and index calculation to accurately capture the influence of hydrology and erosion dynamics on material accumulation. In the initial prediction, due to insufficient differentiation of the accumulation environment of different terraces, the prediction error of some areas is large. Through the closed-loop verification process, the punishment correction is triggered. The model focuses on learning the subtle differences in values between the edge and center areas of the terrace, and the automatically optimized model successfully distinguishes the soil thickness distribution of different terrace levels , providing high-precision spatial data support for selecting stable pile foundation locations in river terrain areas.

[0045] Embodiment 5 Please refer to Figure 5 , on the basis of specific embodiment one, the mountainous area with dramatic terrain fluctuations is selected as the study area. In this area, based on , remote sensing images and geological data, the and and the weight of each factor is determined by fitting with a small amount of known borehole data Exponent; then the initial spatial weight is obtained by using ordinary Kriging method, and the adaptive correction formula

[0046] Weight correction and normalization are performed to generate the soil thickness prediction map; in the verification stage, it is found that the bandwidth parameter set initially fails to fully reflect the differences in the formation of ridges and valleys, resulting in a large error in the reserved verification points, so the punishment correction training is started, and the gradient descent method is used to optimize The core is to calculate the objective function The partial derivative of The specific expression of the partial derivative is derived by chain rule

[0047] Among them

[0048] And Further, The above components are substituted layer by layer to complete the analytical calculation of And used for iterative update of After this correction, the adaptability of the model to the complex formation environment in mountainous areas is significantly enhanced, and the prediction accuracy meets the engineering requirements, effectively guiding subsequent drilling.

[0049] Example 6 Please refer to Figure 2 , Figure 3 and Figure 5 On the basis of the specific example one, in the area with dense forest coverage, dense vegetation has an important holding and improvement effect on soil, in this scenario The calculation and contribution of Exponent are mainly considered, which is given a higher positive weight in Synthesis (w The initial spatial interpolation underestimates the positive effect of vegetation coverage on soil preservation, and the prediction is thinner in some forest areas. Through the reserved verification and subsequent punishment correction training, the model adaptively reduces the weight reference between points with large differences in vegetation coverage. The corrected prediction result is more in line with the actual situation that the soil thickness in the forest area is generally thicker, successfully guiding the exploration personnel to find the pile foundation landing point in the limited open area under the forest.

[0050] ​This invention also provides a soil thickness prediction system for implementing a soil thickness prediction method. The system can be deployed on a computer or server with data processing capabilities and includes the following modules: a data acquisition module comprising a digital elevation model reading unit, a multispectral remote sensing image analysis unit, and a geological lithology database interface unit, used to acquire raw data from external data sources or databases; and an index calculation module comprising multiple sub-calculation units, each used to calculate: Calculated based on catchment area and slope angle; Weighted summation based on lithological classification and weight table; : Calculate terrain roughness using an elevation neighborhood window; Calculated by combining the catchment area and the power of the slope; : Calculation based on infrared and red band reflectance of remote sensing images; the standardization processing module uses the maximum-minimum method or The method standardizes each index to ensure data scale consistency; the accumulation potential calculation module performs weighted fusion of the standardized indices to generate the accumulation potential index. The initial weight calculation module, based on the soil thickness data of known borehole points, uses ordinary kriging interpolation to construct a semi-variogram model and calculates the initial spatial weights of each known point relative to the target point. According to the accumulation potential index The initial weights are adaptively adjusted using a preset formula to obtain the causal weights. The soil thickness prediction module uses normalized causal weights, and performs normalization processing. The soil thickness values ​​at known points are weighted and summed to output the predicted soil thickness at the points to be interpolated; the validation module reserves a portion of borehole data as a validation set for calculation. and The error is compared with a preset error threshold; when the error exceeds the limit, the parameter optimization module initiates a penalty correction training process to calculate the penalty factor. Adjusting the bandwidth parameter using gradient descent The model accuracy is optimized; when the prediction error meets the engineering requirements, the output and application module generates a spatial distribution map of soil thickness, which is used to guide the intensified drilling and bedrock positioning of the transmission tower pile foundation.

[0051] The index calculation module also includes a topographic humidity index calculation unit, which is used to calculate an index reflecting the topographic water catchment capacity based on the catchment area and slope angle analyzed by the digital elevation model. The adaptive correction process in the initial weight calculation module is as follows: based on the difference between the deposition potential index at the known point and the deposition potential index at the target point, a correction factor that is inversely proportional to the difference is constructed, and the initial spatial weights obtained by ordinary Kriging interpolation are dynamically adjusted using the correction factor to obtain causal weights that better reflect the causal relationship.

[0052] The index calculation module includes a terrain humidity index calculation unit for calculating a terrain humidity index reflecting the terrain water collection capacity based on the catchment area and the slope angle derived from the digital elevation model, wherein the calculation process includes a small constant to prevent the denominator from being zero; the index calculation module includes a lithology weighting coefficient calculation unit for assigning weights to different lithology categories according to the weathering resistance of the lithology by expert experience method or analytic hierarchy process, and calculating a comprehensive lithology weighting coefficient based on the lithology indicator variable, wherein soft and easily weathered lithology is given a higher weight; the index calculation module includes a terrain roughness index calculation unit for traversing all pixels in the study area by moving window, and calculating a terrain roughness index reflecting the degree of surface relief based on the elevation values of the center point and its surrounding neighborhood points; the index calculation module includes a slope index calculation unit and a normalized vegetation index calculation unit; the slope index calculation unit is used to calculate the slope index based on the empirical value of the water flow power law in combination with the specific catchment area and the slope angle; the normalized vegetation index calculation unit is used to calculate the normalized vegetation index based on the reflectivity values of the near-infrared band and the red band in the remote sensing image; the standardization processing module is used to standardize each factor index by maximum and minimum value standardization method or standardization method to unify the data scale of each index; the verification module is used to calculate the prediction error by the reserved validation set drilling data, and when the prediction error exceeds the engineering allowable threshold, the parameter optimization module is started to optimize the model parameters; when the prediction error is lower than the predetermined threshold, the soil thickness prediction map is directly output to guide the intensive drilling and bedrock positioning of the power transmission tower pile foundation; the verification module randomly reserves a certain proportion of known drilling data as the validation set before modeling, and calculates the root mean square error and the mean absolute error between the predicted value and the actual measured value, and compares the calculation results with the engineering allowable error threshold; when the optimization process is started, the parameter optimization module first calculates the penalty factor based on the prediction error of the validation points, and then adjusts the similarity bandwidth parameter in the model by gradient descent method; when adjusting the bandwidth parameter, the calculation process of the partial derivative is realized by chain rule, and the root mean square error, the soil thickness prediction value and the weight are derived layer by layer.

[0053] The preferred embodiments disclosed above are only used to help explain the application, and the preferred embodiments do not describe all the details and limit the application to the specific embodiments described. The embodiments are selected and described in the specification in order to better explain the principles and practical applications of the application, so that those skilled in the art can well understand and utilize the application.

Claims

1. A method for predicting soil thickness, characterized in that: Includes the following steps: Collect digital elevation model data, multispectral remote sensing image data, and geological lithology data, and calculate the topographic humidity index based on these data. Lithology weighting coefficient Topographic roughness index Slope index and Normalized Difference Vegetation Index ; By analyzing the terrain humidity index Lithology weighting coefficient Topographic roughness index Slope index and Normalized Difference Vegetation Index Weighted fusion is performed to calculate the accumulation potential index. ; Using soil thickness data from known borehole points, a semi-variogram model was established using ordinary kriging interpolation to calculate the initial spatial weights of each known point relative to the interpolation point. ; Based on the accumulation potential index The initial spatial weights are adaptively corrected to obtain the causal weights. ; Through the causal weights Perform weighted summation to calculate the predicted soil thickness at the interpolation points, where For known points The soil thickness measurement value.

2. The method for predicting soil thickness according to claim 1, characterized in that: The terrain humidity index The calculation formula is: in To obtain a specific catchment area through hydrological analysis using a digital elevation model, the area of ​​the upstream catchment area flowing towards that point is calculated. The slope angle is calculated using a digital elevation model. 0.001 A tiny constant added to prevent the denominator from being zero.

3. The method for predicting soil thickness according to claim 1, characterized in that: The lithology weighting coefficient The calculation formula is: ,in For the first The weight of lithology is determined based on the lithology's resistance to weathering through expert experience or analytic hierarchy process. Hard, difficult-to-weather lithologies are assigned lower weights, while soft, easily weathered lithologies are assigned higher weights. As a lithological indicator variable, when this point belongs to the first... The value is 1 for lithological types, and 0 otherwise.

4. The method for predicting soil thickness according to claim 1, characterized in that: The terrain roughness index The calculation formula is in The elevation value of the center point. For its surroundings The elevation values ​​of the neighboring points are adopted. When performing calculations by moving the window The calculation is completed by traversing all pixels within the study area.

5. The method for predicting soil thickness according to claim 1, characterized in that: The slope index The calculation formula is: in For a specific catchment area, The slope angle is given by the index 1.2, which is an empirical value based on the law of water flow power; the normalized vegetation index is... The calculation formula is: in and These are the reflectance values ​​in the near-infrared and red bands of the remote sensing image, respectively, obtained by extracting the corresponding band data from the satellite remote sensing image.

6. The method for predicting soil thickness according to claim 1, characterized in that: The method also includes: standardizing the obtained indices, wherein the standardization of each index is performed using the maximum-minimum standardization method or... Standardization method, the formula for standardizing the maximum and minimum values ​​is: Standardized formula ,in The original value, and These are the minimum and maximum values, respectively. This is the average value. The standard deviation is denoted as .

7. The method for predicting soil thickness according to claim 1, characterized in that: The method also includes validating the prediction results using a reserved validation method, employing the root mean square error. and mean absolute error As an evaluation indicator, when the prediction error exceeds the threshold allowed by the project, a penalty correction training process is initiated to optimize the model parameters. When the prediction error is lower than the predetermined threshold, the generated soil thickness prediction map is directly used to guide the targeted intensified drilling of the transmission tower pile foundation to determine the optimal location where the pile foundation finally sits in stable bedrock.

8. The method for predicting soil thickness according to claim 7, characterized in that: The reserved verification method includes: randomly reserving 10%-20% of known borehole data as a verification set before modeling, and calculating the root mean square error between the predicted soil thickness value and the actual drilling measurement value. and mean absolute error in Number of verification points Predicted values ​​for verification points To verify the measured values ​​at the points, the calculated results are compared with the allowable error threshold for the project. and Compare them.

9. The method for predicting soil thickness according to claim 7, characterized in that: The penalty correction training process includes: first, calculating the penalty factor for each point in the validation set. in The penalty intensity coefficient has a value of [value missing]. The similarity bandwidth parameter is then adjusted using gradient descent based on the penalty factor for all validation points, given the hyperbolic tangent function. The specific update formula is as follows: in For learning rate, This is the partial derivative of the root mean square error with respect to the bandwidth parameter.

10. The method for predicting soil thickness according to claim 9, characterized in that: The partial derivatives The calculation process involves applying the chain rule to the root mean square error. Predicted soil thickness Normalized causal weights Causal weights By performing layer-by-layer differentiation, the final result includes the initial spatial weights. Difference in accumulation potential index and current bandwidth parameters The specific expression.

11. A system for predicting soil thickness, characterized in that, The system is deployed on a computer or server with data processing capabilities and includes the following modules: a data acquisition module, including a digital elevation model reading unit, a multispectral remote sensing image analysis unit, and a geological lithology database interface unit, used to acquire raw topographic, remote sensing, and lithological data from external data sources; The index calculation module includes multiple sub-calculation units, which are used to calculate various genetic indices based on topography and lithology. The genetic indices include at least the topographic humidity index based on catchment area and slope angle, the lithological erosion resistance index based on lithology classification and weight table, the topographic roughness index calculated through the elevation neighborhood window, the slope index calculated by combining catchment area and slope power, and the vegetation index calculated through infrared and red band reflectance of remote sensing images. The standardization module is used to standardize each causal index using the maximum-minimum method to unify the data scale. The accumulation potential calculation module is used to perform weighted fusion of multiple standardized causal indices to generate a comprehensive accumulation potential index. The initial weight calculation module is used to construct a semi-variogram model based on the soil thickness data of known borehole points using ordinary kriging interpolation, and to calculate the initial spatial weight of each known point to the target point. The weight correction module is used to adaptively correct the initial spatial weights according to the stacking potential index using preset rules to obtain causal weights, and to normalize the causal weights. The soil thickness prediction module is used to perform a weighted summation of the soil thickness values ​​of known points using normalized causal weights, and output the predicted soil thickness of the points to be interpolated. The verification module is used to calculate the prediction error using reserved verification set borehole data and compare the prediction error with a preset error threshold. The parameter optimization module is used to initiate an optimization process when the prediction error exceeds a preset threshold, and adjust the bandwidth parameter in the model using the gradient descent method to optimize the model accuracy. The output and application module is used to generate a spatial distribution map of soil thickness when the prediction error meets the requirements, so as to guide the densification drilling and bedrock positioning of transmission tower pile foundations.

12. The system for predicting soil thickness according to claim 11, characterized in that: The index calculation module also includes a topographic humidity index calculation unit, which is used to calculate an index reflecting the topographic water catchment capacity based on the catchment area and slope angle analyzed by the digital elevation model. The adaptive correction process in the initial weight calculation module is as follows: based on the difference between the deposition potential index at the known point and the deposition potential index at the target point, a correction factor that is inversely proportional to the difference is constructed, and the initial spatial weights obtained by ordinary Kriging interpolation are dynamically adjusted using the correction factor to obtain causal weights that better reflect the causal relationship.

Images