Hydropower station slope comprehensive health degree prediction method considering ecological environment index data

Abstract

According to the hydropower station slope comprehensive health degree prediction method considering the ecological environment index data, the incidence relation between the ecological environment index and the slope health degree is mined by deeply analyzing the slope health degree influence factors; a hydropower station slope health degree prediction index system including multiple dimensions of landform, meteorological conditions and ecological environment is established, a hydropower station slope comprehensive health degree prediction model based on a random forest algorithm is provided to realize accurate prediction of slope health degree, and the slope health degree is monitored for a long time to realize accurate prediction of the slope health degree. The method provides long-time-scale risk management and disaster prevention and reduction scientific guidance for the hydropower station slope, and shows significant effectiveness and reliability, and the health degree prediction result is highly consistent with the actual situation.

Description

Technical Field

[0001] The invention belongs to the field of geological disaster data analysis, and specifically is a method for predicting the comprehensive health of a hydropower station slope by taking ecological environment index data into consideration. Background Art

[0002] Hydropower projects are often built in high mountain valleys and river areas with complex geological conditions, which inevitably involve slope stability issues. The health of the slope of a hydropower station is directly related to the safe operation of the hydropower station. Slope instability may lead to disasters such as slope landslides and collapses, which in turn affect the normal operation of the hydropower station, cause reservoir dams to burst, damage to hydropower facilities, and even affect the power supply capacity of the power system, causing power outages and even the risk of power system collapse. The instability of the slope of a hydropower station may also affect the safety of the surrounding areas. Slope landslides or collapses may lead to natural disasters such as mudslides and floods, threatening the lives and property of surrounding residents. At the same time, if the instability of the slope of a hydropower station is not promptly addressed, it may lead to long-term safety hazards, thereby affecting the social stability and economic development of the surrounding areas. The unhealthy or unstable slope of a hydropower station may also affect the reliability and stability of the power system. Once a hydropower station is affected by slope instability, it may lead to a decrease in power generation capacity, thereby affecting the power supply capacity and stability of the power system. Especially under extreme weather conditions, slope instability may increase the risks of hydropower stations and have an adverse impact on the operation of the power system.

[0003] Existing slope health prediction methods usually select geological factors such as rock and soil properties and slope shape for evaluation. However, hydropower stations are often located in mountain valleys and flowing water areas, and their environments are more complex and changeable. Existing slope stability assessment methods are difficult to match them. Therefore, it is urgent to conduct in-depth analysis of factors affecting the health of hydropower station slopes, such as topography, ecological environment and meteorological conditions, to mine the evolution trend of landslide disasters in the reservoir area based on big data technology, to introduce evaluation indicators with more dimensions and time scales, to build a more scientific hydropower station slope health prediction quantitative model, to more comprehensively grasp the overall dynamic changes of the slope, and to achieve advanced perception of slope health and precise prevention and control of disasters. Summary of the invention

[0004] In order to solve the above problems, the present invention proposes a method for predicting the comprehensive health of hydropower station slopes taking into account ecological environmental indicator data, provides an evaluation index system for influencing factors with multiple dimensions and time scales including ecological environmental indicators, and uses random forests to construct a more scientific quantitative model for predicting the health of hydropower stations, thereby achieving comprehensive and accurate slope health prediction.

[0005] To achieve the above object, the technical solution adopted by the present invention is:

[0006] The method for predicting the comprehensive health of a hydropower station slope by considering ecological environment index data is characterized by comprising the following steps:

[0007] S1: Factors affecting the health of the excavation slope:

[0008] Select disaster observation point A and disaster observation point B at the landslide site in the reservoir area of ​​the hydropower station, detect and record the topography, meteorological conditions, and ecological environment of the slopes at disaster observation point A and disaster observation point B, and determine the indicators of the health influencing factors of disaster observation point A and disaster observation point B;

[0009] S2: Based on the factors affecting slope health, a multi-dimensional slope health prediction and evaluation index system is constructed:

[0010] The influencing factor indicators of the three types of topography, meteorological conditions, and ecological environment are numbered separately to form a collection, which is the multi-dimensional slope health prediction and evaluation index system;

[0011] S3: According to the multi-dimensional health prediction and evaluation index system, the correlation analysis of the indicators is carried out:

[0012] The Pearson correlation coefficient method is used to measure the correlation between two different evaluation indicators. The calculation formula is as follows:

[0013]

[0014] When the absolute value of the Pearson correlation coefficient is greater than 0.7, it indicates that the selected indicators are closely related; when the absolute value is less than 0.2, it indicates that the selected indicators are not closely related;

[0015] S4: Perform outlier detection on selected indicators:

[0016] Use the local outlier factor to detect outliers and calculate the outlier value of each indicator data. The calculation formula is as follows:

[0017]

[0018] reach_dist(p,q)=max{d(p,q),k_distance(q)}

[0019]

[0020] S5: Select indicator data:

[0021] When the outlier value of the indicator data is significantly greater than other indicators, the data indicator set with the largest outlier value is obtained and removed from the indicator data;

[0022] S6: Use synthetic minority class oversampling technology to balance indicator data:

[0023] x n =x i +Rand(0,1)×(x iz -x i )

[0024] In the formula, x i is a sample in the minority class; x iz According to the sampling rate N, from x i Randomly select the zth sample x from the K nearest neighbors of iz ;x n is a newly generated sample; Rand(0,1) is a random number within (0,1).

[0025] S7: Using random forest to predict the health of hydropower station slopes:

[0026] The random forest algorithm is an ensemble learning algorithm based on decision trees. For classification problems, it is a classifier composed of multiple decision trees. For indicator data, K decision trees are established {l 1 (x),l 2 (x),…,l k (x)}, construct a random forest model based on the obtained decision tree model:

[0027]

[0028] Where L(x) is the random forest model finally constructed; i (x) is a single decision tree model; Y is the target variable; H(.) is the classification function of the decision tree model.

[0029] S8: Use grid search method to optimize the parameters of random forest:

[0030] In the random forest model, the number of decision trees, the maximum number of features, and the depth of the decision tree are key factors affecting the growth of the decision tree, and are also important guarantees for the performance and accuracy of the random forest model. The grid search method is used to find the optimal parameter combination of the model.

[0031] As a further improvement of the present invention, in step S1, the detection of the topography includes the detection of the slope direction, pore water pressure, slope and curvature at the disaster observation point A and the disaster observation point B.

[0032] As a further improvement of the present invention, in step S1, the detection of meteorological conditions includes the detection of rainfall, humidity, and rock and soil moisture content at disaster monitoring point A and disaster observation point B.

[0033] As a further improvement of the present invention, in step S1, the detection of the ecological environment includes the detection of the normalized difference vegetation index and the normalized difference water index at the disaster observation point A and the disaster observation point B.

[0034] As a further improvement of the present invention, in step S1, the normalized difference vegetation index among the ecological environment indicators reflects the growth density and health of plants. The higher the value, the more luxuriant the plant growth and the more developed the root system, which can better fix the soil, enhance the stability of the slope, and make the slope healthier.

[0035] As a further improvement of the present invention, in step S1, the normalized difference water index among the ecological environment indicators reflects the existence of surface water bodies. The higher the value, the greater the risk of erosion and geological disasters on the slope, which will reduce the stability of the slope and the health of the slope.

[0036] As a further improvement of the present invention, in step S4, the distance between the index data and its adjacent points is as shown in formula (1-1):

[0037]

[0038] The distance from the indicator data to the neighboring points is shown in formula (1-2):

[0039] reach_dist(p,q)=max{d(p,q),k_distance(q)}(1-2)

[0040] The local reachable density of the indicator data is shown in formula (1-3):

[0041]

[0042] The outlier calculation of indicator data is shown in formula (1-4):

[0043]

[0044] As a further improvement of the present invention, in step S5, the maximum outlier set to be eliminated is as shown in formula (2-1):

[0045] As shown:

[0046] O={o 1 ,o 2 ,…,o n}(2-1);

[0047] In step S6, the over-synthesized minority class oversampling technique of the index data is shown in formula (3-1):

[0048] x n =x i+Rand(0,1)×(x iz -x i )(3-1)

[0049] In the formula, x i is a sample in the minority class; x iz According to the sampling rate N, from x i Randomly select the zth sample x from the K nearest neighbors of iz ;x n is a newly generated sample; Rand(0,1) is a random number in (0,1);

[0050] As a further improvement of the present invention, in step S7, if there are N samples, the constructed feature vector is as shown in formula (4-1):

[0051] {(x i ,y i )|i=1,2,…,N} (4-1)

[0052] where y i The slope of the hydropower station i The factors affecting the health of the hydropower station slope, namely, the slope aspect, pore water pressure, slope, curvature rainfall, humidity, rock and soil moisture content, normalized difference vegetation index, and normalized difference water index selected in step S1 are used as feature vectors;

[0053] The established single decision tree randomly selects m feature variables for splitting, and i and health i , search for split variables and split values, and divide the entire vector space into m partitions, as shown in formula (4-2):

[0054] {R 1 ,R 2 ,…,R m}(4-3)

[0055] For any partition, the partition is classified into two parts, as shown in formulas (4-4) and (4-5):

[0056] R 1 (j,s)={I|I j ≤s}(4-4)

[0057] R 2 (j,s)={I|I j >s}(4-5)

[0058] In the formula: j represents the variable type, and s represents the value at the time of splitting.

[0059] The objective function for searching for split variables and split values ​​in vector space is shown in formula (4-6):

[0060]

[0061] in:

[0062] c 1 =ave(y i |x i ∈R 1 (j,s))(4-7)

[0063] c 2 =ave(y i |x i ∈R 2 (j,s))(4-8)

[0064] A random forest is constructed based on the constructed decision tree, as shown in formula (4-9):

[0065]

[0066] As a further improvement of the present invention, in step S8, a grid search method is used to seek the best parameter combination with a sample training set ratio of 0.8, a number of decision trees of 50, a maximum number of features of 4, and a maximum depth of 17.

[0067] Compared with the prior art, the present invention has the following beneficial effects:

[0068] From the perspective of the long-term development and changes of the health of the slopes of hydropower stations, the present invention deeply explores the factors affecting the health of slopes, establishes a comprehensive prediction index system for the health of slopes including topography, meteorological data, ecological environment and other dimensions, and proposes a quantitative model for slope health assessment based on random forests, which can achieve accurate prediction of slopes with different degrees of health. By predicting and evaluating the health of slopes over a long period of time, it provides long-term multi-dimensional health prediction and slope disaster prevention and mitigation guidance for the slopes of hydropower stations.

[0069] Through correlation analysis, the indicators with strong generalization, large information content and great influence on slope health were determined as the key indicators for slope health prediction. Taking into account the characteristics of the hydropower station slope system, such as changeable characteristics, complexity and uncertainty, random forest was selected as the health prediction model, and the optimal parameter combination of random forest was selected through the grid search method, which further improved the accuracy of slope health prediction.

[0070] Through outlier detection and sample balancing, the data sets with too large outliers in the indicator data are eliminated, the proportion of disaster points in the indicator data is increased, and the performance, accuracy and anti-overfitting ability of the model are further improved.

[0071] The comprehensive health prediction model of hydropower station slopes proposed in the present invention, which takes into account ecological environmental indicator data, shows remarkable effectiveness and reliability. Its health prediction results are highly consistent with the actual situation, providing health development and change trends for the safe management of hydropower station slopes and providing timely disaster prevention and mitigation guidance for power stations. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 is a schematic diagram of the geographical location of the research area in the embodiment;

[0073] Figure 2 It is a schematic diagram of the installation of the example point;

[0074] Figure 3 It is a schematic diagram of the indicator impact indicator system;

[0075] Figure 4 is a correlation heat map;

[0076] Figure 5 is a schematic diagram of outlier detection;

[0077] Figure 6 It is the structure diagram of random forest model and grid search method;

[0078] Figure 7 This is a table showing the training effect of the model with different sample ratios;

[0079] Figure 8 It is a graph showing the performance of the model changing with the number of decision trees;

[0080] Fig. 9 It is a graph of the model performance changing with the maximum number of features in the model;

[0081] Fig.10 It is a graph showing the variation of model performance with maximum depth. DETAILED DESCRIPTION

[0082] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments:

[0083] like Figure 1-10 As shown, the present invention proposes a method for predicting the comprehensive health of a hydropower station slope by considering ecological environment index data, comprising the following steps:

[0084] S1: Factors affecting the health of the excavation slope:

[0085] Select disaster observation point A and disaster observation point B at the landslide site in the reservoir area of ​​the hydropower station, detect and record the topography, meteorological conditions, and ecological environment of the slopes at disaster observation point A and disaster observation point B, and determine the indicators of the health influencing factors of disaster observation point A and disaster observation point B;

[0086] S2: Based on the factors affecting slope health, a multi-dimensional slope health prediction and evaluation index system is constructed:

[0087] The influencing factor indicators of the three types of topography, meteorological conditions, and ecological environment are numbered separately to form a collection, which is the multi-dimensional slope health prediction and evaluation index system;

[0088] S3: According to the multi-dimensional health prediction and evaluation index system, the correlation analysis of the indicators is carried out:

[0089] The Pearson correlation coefficient method is used to measure the correlation between two different evaluation indicators. The calculation formula is as follows:

[0090]

[0091] When the absolute value of the Pearson correlation coefficient is greater than 0.7, it indicates that the selected indicators are closely related; when the absolute value is less than 0.2, it indicates that the selected indicators are not closely related;

[0092] S4: Perform outlier detection on selected indicators:

[0093] Use the local outlier factor to detect outliers and calculate the outlier value of each indicator data. The calculation formula is as follows:

[0094]

[0095] reach_dist(p,q)=max{d(p,q),k_distance(q)}

[0096]

[0097] S5: Select indicator data:

[0098] When the outlier value of the indicator data is significantly greater than other indicators, the data indicator set with the largest outlier value is obtained and removed from the indicator data;

[0099] S6: Use synthetic minority class oversampling technology to balance indicator data:

[0100] x n =x i +Rand(0,1)×(x iz -x i )

[0101] In the formula, x i is a sample in the minority class; x iz According to the sampling rate N, from x i Randomly select the zth sample x from the K nearest neighbors ofiz ;x n is a newly generated sample; Rand(0,1) is a random number within (0,1).

[0102] S7: Using random forest to predict the health of hydropower station slopes:

[0103] The random forest algorithm is an ensemble learning algorithm based on decision trees. For classification problems, it is a classifier composed of multiple decision trees. For indicator data, K decision trees are established {l 1 (x),l 2 (x),…,l k (x)}, construct a random forest model based on the obtained decision tree model:

[0104]

[0105] Where L(x) is the random forest model finally constructed; i (x) is a single decision tree model; Y is the target variable; H(.) is the classification function of the decision tree model.

[0106] S8: Use grid search method to optimize the parameters of random forest:

[0107] In the random forest model, the number of decision trees, the maximum number of features, and the depth of the decision tree are key factors affecting the growth of the decision tree, and are also important guarantees for the performance and accuracy of the random forest model. The grid search method is used to find the optimal parameter combination of the model.

[0108] In step S1, the detection of topography includes the detection of slope direction, pore water pressure, slope, and curvature at disaster observation point A and disaster observation point B; the detection of meteorological conditions includes the detection of rainfall, humidity, and rock and soil moisture content at disaster monitoring point A and disaster observation point B; and the detection of ecological environment includes the detection of normalized difference vegetation index and normalized difference water index at disaster observation point A and disaster observation point B.

[0109] In step S1, the normalized difference vegetation index (NDVI) among the ecological environment indicators reflects the growth density and health of plants. The higher the value, the more luxuriant the plant growth and the more developed the root system, which can better fix the soil, enhance the stability of the slope, and make the slope healthier.

[0110] In step S1, the normalized difference water index among the ecological environment indicators reflects the existence of surface water bodies. The higher the value, the greater the risk of erosion and geological disasters on the slope, which will reduce the stability of the slope and the health of the slope.

[0111] Topography, meteorological conditions, and ecological environment are key factors that must be considered in predicting the health of hydropower station slopes.

[0112] Among them, topography is the basic condition for landslides, which determines the stability and bearing capacity of the slope. It includes slope aspect, pore water pressure, slope, and curvature. The slope aspect refers to the direction of the slope surface, usually expressed in geographic azimuth. Different slope aspects will affect the light, temperature, and water evaporation of the slope. On sunny slopes, water evaporates quickly and the soil is dry, which may lead to sparse vegetation and increase the risk of erosion. On shady slopes, due to insufficient sunlight, water retention time is longer, which may lead to good plant growth, thereby enhancing soil stability. Therefore, the slope aspect will directly affect the health of the slope, especially during drought and rainy seasons. Pore water pressure refers to the pressure state of water in the soil, which is directly related to the effective stress of the soil. Excessive pore water pressure may weaken the strength of the soil and increase the risk of landslides on the slope. Especially after rainfall, the increase in pore water pressure will further increase the fluidity of the slope, which may cause the slope to lose stability. Therefore, monitoring the changes in pore water pressure is an important indicator for assessing the health of the slope. Slope is the angle of inclination of the slope. The steeper the slope, the stronger the effect of gravity on the soil, and the risk of slope instability increases. Generally speaking, when the slope exceeds a certain critical value (such as 30°-40°), the stability of the slope decreases significantly. In addition, the slope also affects soil erosion. Steep slopes easily cause rainwater to flow in a concentrated manner, aggravating erosion and further affecting the health of the slope. Curvature describes the shape characteristics of the slope surface, which is usually divided into concave curvature and convex curvature. Slopes with concave curvature can gather rainwater more effectively and help replenish groundwater, but they may also increase soil moisture and increase the risk of landslides. On the contrary, slopes with convex curvature may cause water flow to accelerate, thereby increasing the degree of soil erosion. Therefore, curvature has a significant impact on factors such as soil moisture distribution, erosion, and plant growth, and indirectly affects the health of the slope.

[0113] Meteorological conditions are also important factors affecting slope health, including rainfall, humidity, and rock and soil moisture content. Rainfall is the main meteorological factor affecting slope stability. Heavy rainfall will rapidly increase the moisture content in the soil, increase pore water pressure, and cause slope sliding. At the same time, rainfall will also increase soil erosion, loosen the soil structure, and then reduce the bearing capacity of the slope. Therefore, timely monitoring of rainfall changes is crucial for slope health assessment. Soil moisture reflects the moisture content in the soil and is one of the important indicators for judging slope health. Excessive soil moisture may lead to increased pore water pressure, reduce the bearing capacity of the soil, and increase the risk of slope landslides. In addition, humidity affects plant growth. Areas with moderate humidity are conducive to vegetation coverage and improve soil stability. In drought conditions, soil drying and vegetation death increase the fragility of the slope. Rock and soil moisture content is directly related to the physical and mechanical properties of the soil. When the soil moisture content is too high, the soil strength decreases, and landslides or collapses are prone to occur. In addition, soil with moderate water content is conducive to plant growth, thereby strengthening the root system's consolidation of the soil and improving the stability of the slope. Therefore, changes in rock and soil water content have a significant impact on the health of the slope.

[0114] The ecological environment of the hydropower station slope is also an important comprehensive indicator for measuring the health of the slope, including the normalized difference vegetation index and the normalized difference water index. The normalized difference vegetation index is an indicator for measuring the growth of vegetation, reflecting the biomass and health of plants. A high normalized difference vegetation index value usually indicates lush vegetation, which helps to stabilize the soil and retain water. Vegetation absorbs soil through its roots, enhances the cohesion of the soil, and reduces the risk of landslides. Therefore, the normalized difference vegetation index value is closely related to the health of the slope, and monitoring its changes can detect the potential instability risk of the slope at an early stage. The normalized difference water index is mainly used to monitor the distribution of water bodies and changes in moisture, reflecting the soil moisture and moisture conditions. Through the normalized difference water index, the changes in water bodies near the slope and their impact on the health of the slope can be evaluated. Changes in water bodies may directly affect soil moisture, thereby affecting plant growth, erosion, and landslide risk. Therefore, the normalized difference water index also plays an important role in the prediction of slope health.

[0115] In summary, the present invention analyzes the factors affecting slope health by different time scales and geographic space dimensions, and finally determines Figure 3 The nine influencing factors presented are used as an indicator system for slope health prediction.

[0116] Figure 3The constructed large-scale mountain hydropower station slope health prediction index system covers multiple types of influencing factors in multiple dimensions of topography, meteorological conditions, and environmental factors. The selected influencing factors have rich data types and strong logical correlations. By collecting data on these indicators, it is possible to fully grasp the changes in slope health from multiple perspectives in space and time. It can more scientifically and accurately determine the risks of geological disasters such as landslides and slope instability caused by the decline in slope health, increase the early warning processing time of power stations for geological disasters, and provide scientific guidance for the long-term stability and safety of hydropower station slopes.

[0117] When predicting slope health, it is crucial to analyze the correlation between various indicators. These indicators include slope aspect, pore water pressure, slope, curvature, rainfall, humidity, rock water content, normalized difference vegetation index and normalized difference water index, which interact with each other to affect slope stability. First, slope and curvature directly affect the gravity effect and water flow distribution of the slope. The steeper the slope, the stronger the gravity effect, which is prone to landslides, while curvature affects the concentration and dispersion of water flow, and concave slopes are prone to water accumulation, increasing the risk of landslides. Therefore, the combination of slope and curvature is the basis for evaluating slope stability. Secondly, rainfall is closely related to pore water pressure, humidity and rock water content. When precipitation increases, the moisture content of the soil will rise rapidly, resulting in an increase in pore water pressure and humidity, reducing the effective stress of the soil and promoting slope instability. Excessive rainfall will cause saturation of soil in the basin, further increasing the risk of potential landslides. In addition, vegetation index plays an important role in soil stability. Vegetation interacts with soil moisture. Good vegetation coverage can reduce soil erosion and enhance soil water holding capacity, thereby improving slope stability. The distribution of water bodies near the slope has a significant impact on soil moisture and vegetation growth, and its changes are directly related to the health of the slope. In summary, various indicators act together on the health of the slope by affecting water dynamics, soil properties and vegetation conditions. By comprehensively analyzing the correlation between these indicators, potential landslide risks can be effectively identified, and targeted monitoring and control strategies can be formulated to ensure the safety and stability of the slope.

[0118] In summary, slope health prediction is a very complex nonlinear process. Due to the large number of selected indicators, in order to understand the mutual influence and action mechanism between the indicators and the internal principle of the development and change of slope health, the correlation analysis of the indicators is carried out. If the correlation analysis is not carried out, first of all, the interaction between the indicators may be ignored. For example, the increase in rainfall may lead to an increase in pore water pressure, thereby reducing the effective stress of the soil. If the correlation analysis is not carried out, these interactions may be ignored, leading to a one-sided understanding of the slope health; secondly, it is easy to make wrong decisions. Evaluation based on a single indicator is easy to lead to wrong decisions. If only the slope is focused on and the rainfall and vegetation conditions are ignored, the risk of slope instability under extreme weather conditions may be underestimated, resulting in omissions in the formulation of prevention and control measures; next, it is easy to cause a waste of resources. Correlation analysis can help identify the most important influencing factors, concentrate resources on monitoring and governance, and avoid unnecessary expenses. If no analysis is carried out, monitoring and management may be scattered on multiple unimportant indicators, wasting funds and human resources; secondly, it may cause defects in the early warning mechanism. By analyzing the correlation between indicators, the relationship between the changes in certain key indicators and the slope health can be identified, thereby establishing an effective early warning mechanism. Without correlation analysis, potential risks of the slope may not be detected in time, delaying the timing of emergency response and increasing safety hazards; finally, the internal mechanism of slope health changes may not be scientifically explained. Correlation analysis provides data support for the formulation of scientific slope management and protection strategies. Without analysis, decisions may lack basis, thereby reducing the effectiveness of policies and the feasibility of implementation.

[0119] The present invention uses the Pearson correlation coefficient method to measure the degree of correlation between two different indicators. The calculation formula is as follows:

[0120]

[0121] According to the correlation judgment rule, when the absolute value of the correlation coefficient is less than 0.2, the relationship between the indicators is distant, and when the absolute value is greater than 0.7, it indicates that the relationship between them is close. The final correlation coefficient of the health prediction index system is calculated as follows: Figure 4 As shown, the darker the color, the stronger the correlation.

[0122] like Figure 4As shown in the figure, the correlation coefficient between topographic and geomorphic indicators and ecological environment is low. The correlation coefficients between the normalized difference vegetation index and the normalized difference water index and slope aspect are 0.16 and 0.17, both less than 0.2; but the correlation coefficient between the normalized difference vegetation index and the normalized difference water index is 0.94, which has a strong correlation; the correlation coefficient between rainfall and rock and soil moisture content is 0.43, which has a certain correlation; the correlation coefficient between rainfall and humidity is 0.43, which has a certain correlation.

[0123] Local Outlier Factor Algorithm The local outlier factor (LOF, also called local anomaly factor) algorithm is a density-based local outlier detection algorithm proposed by Breunig in 2000. This method is suitable for data with different density dispersion of different clusters. According to the density of the data, it calculates the local outlier factor of each data object to identify the degree of outliers of the data, and selects the top (n) points with the largest degree of outliers as outliers. The method is simple and is not affected by data distribution.

[0124] The outliers in the data are identified by calculating the outlier values ​​of the indicator data. First, the distance from the data point p to the kth neighboring point is calculated, that is, the k distance:

[0125]

[0126] After obtaining the k-distance of the data, the set of all data points within the k-distance is calculated, that is, the k-distance neighborhood, denoted as knn(p). The distance from data point q to p in the set is denoted as d(p,q).

[0127] Calculate the reachability distance of data in a collection:

[0128] reach_dist(p,q)=max{d(p,q),k_distance(q)}

[0129] Then calculate the local reachability density and local outliers of the data points:

[0130]

[0131] The outlier density diagram of all indicator data of the present invention is as follows: Figure 5 As shown in the figure, the indicator data with outliers greater than 8 are removed.

[0132] The synthetic minority oversampling technique, or SMOTE for short, was proposed by Nitesh Chawla et al. in 2002. First, a line is drawn between samples in the feature space that are close to each other, and then new samples are drawn at points along the line; then a sample is randomly selected from the minority class, and the k nearest neighbors of the sample are found. Finally, neighbors are randomly selected and synthetic samples are created at randomly selected points between two samples in the feature space. The calculation formula is as follows:

[0133] x n =x i +Rand(0,1)×(x iz -x i )

[0134] In the formula, x i is a sample in the minority class; x iz According to the sampling rate N, from x i Randomly select the zth sample x from the K nearest neighbors of iz ;x n is a newly generated sample; Rand(0,1) is a random number within (0,1).

[0135] Random forest is an integrated learning algorithm based on decision trees, that is, it randomly uses variables and data to generate many classification trees, and then summarizes the results of the classification trees. Random forest improves the accuracy of prediction without significantly increasing the amount of calculation. At the same time, random forest is insensitive to multicollinearity, is more robust in predicting missing data and unbalanced data, and can handle large amounts of multi-dimensional complex data with high accuracy. It is known as one of the best algorithms currently. Based on the idea of ​​ensemble learning, for regression problems, the output of random forest is the average of the output of decision trees; for classification problems, the output of random forest is the optimal classification result determined by the decision tree through a simple majority voting method.

[0136] First, construct the feature vector of the model, as shown in the formula:

[0137] {(x i ,y i )|i=1,2,…,N}

[0138] where y i is the health of the hydropower station slope i, and the factors affecting the health of the hydropower station slope, namely, the slope aspect, pore water pressure, slope, curvature rainfall, humidity, rock and soil moisture content, normalized difference vegetation index, and normalized difference water index selected in step S1 are used as feature vectors.

[0139] Then a single decision tree is established, m feature variables are randomly selected for splitting, and the feature vector x of the sample data isi and health i , search for split variables and split values, and divide the entire vector space into m partitions. As shown in the formula:

[0140] {R 1 ,R 2 ,…,R m}

[0141] For any partition, the partition is classified into two parts as shown below:

[0142] R 1 (j,s)={I|I j ≤s}

[0143] R 2 (j,s)={I|I j >s}

[0144] In the formula: j represents the variable type, and s represents the value at the time of splitting.

[0145] The objective function for searching for vector space splitting variables and splitting values ​​is shown in the formula:

[0146]

[0147] in:

[0148] c 1 =ave(y i |x i ∈R 1 (j,s))

[0149] c 2 =ave(y i |x i ∈R 2 (j,s))

[0150] Finally, a random forest is constructed based on the constructed decision tree, as shown in the following formula:

[0151]

[0152] like Figure 6 As shown, the training of the present invention uses the F1 score and AUC value to measure the performance of the slope health prediction model.

[0153] In the random forest model, the number of decision trees (ntrees), the maximum number of features (max_features), and the depth of the decision tree (depths) are key factors affecting the growth of the decision tree, and are also important guarantees for the performance and accuracy of the random forest model. This paper uses the grid search method to find the optimal parameter combination of the model.

[0154] In order to find the best parameter combination of the random forest model, the present invention uses a grid search method to find the optimal parameter combination of the model training set ratio, the number of decision trees, the maximum number of features, and the depth of the decision tree.

[0155] according to Figure 7 The data shows that when the proportion of the test set begins to change, the accuracy and F1 score of the model are on an upward trend. When the test set proportion is 0.2, the scores of the two indicators reach a peak, with the model accuracy of 99.74% and the F1 score of 0.9732, and then they continue to decline. Therefore, when training the model, the test set proportion is selected to be 0.2, and the training set proportion is selected to be 0.8.

[0156] The number of decision trees is the number of iterations of the random forest model. The more luxuriant the trees grow, the better the classification performance of the random forest. By constructing random models with different numbers of decision trees, the relationship between the F1 score and AUC value and the number of decision trees is obtained as follows: Figure 8 When the number of decision trees is greater than 50, the F1 score and AUC value are relatively stable. When the number of decision trees is further increased, the model performance cannot be significantly improved and the computational burden will increase. Therefore, the number of decision trees is determined to be 50.

[0157] The maximum number of features is the number of randomly extracted feature factors when each decision tree grows. The worse the correlation between each tree, or the independence between trees, the lower the impact of the extracted features on the results, and the better the classification performance of the random forest. When the number of features is reduced, the correlation and classification ability of the decision tree will also decrease accordingly. By constructing random forest models with different numbers of randomly selected features, the relationship between the F1 score and AUC value and the number of decision trees is obtained as follows: Fig. 9 When the number of randomly selected features is 4, the model score is the highest and the model performance is the best, so the maximum number of features is determined to be 4.

[0158] The decision tree depth limits the maximum depth of each decision tree to prevent the model from becoming too complex. If the depth is too large, it is easy to cause overfitting; if the depth is too small, the model may not fit the data well. By constructing random forest models with different maximum depths, the relationship between the F1 score and AUC value and the number of decision trees is obtained as follows Fig.10 As shown in the figure, when the maximum depth of the tree is 17, the model score is the highest and the performance of the model is the best, so the maximum depth is determined to be 17.

[0159] The specific application of the present invention is implemented as follows:

[0160] like Figure 1 As shown in the figure, the disaster observation point A and the disaster observation point B were selected as two geological disaster risk points, namely the Longtan slope of the Shaping Second-level Hydropower Station and the Zhangcungou slope of the Zhentouba First-level Hydropower Station in the Dadu River Basin, to verify the slope health prediction example.

[0161] For each selected index, slope aspect, pore water pressure, slope, curvature, rainfall, humidity, rock and soil moisture content, normalized difference vegetation index, and normalized difference water index, the slopes of Zhangcungou and Longtan are consistent, defined as "1", with slopes of 18.9% and 22.3%, and curvatures of 0.0024rad / m and 0.0029rad / m, respectively.

[0162] For other indicators with higher real-time requirements, such as pore water pressure, rainfall, humidity, rock and soil moisture content, normalized difference vegetation index, and normalized difference water index, monitoring points are deployed on the slopes, and sensors are installed at different monitoring points and at different depths to conduct real-time monitoring of the slopes and collect data on health prediction indicators.

[0163] In order to realize the periodic real-time collection of slope data, a measuring point was set up at Zhangcungou and Longtan slopes respectively, equipped with a variety of sensors, such as Figure 2 These sensors can monitor the changes of various slope parameters in real time and provide important data support for slope health prediction.

[0164] The experimental results prove that the slope health prediction model of the hydropower station shows remarkable effectiveness and reliability. Its health prediction results are highly consistent with the actual situation of the slope. It provides the long-term gradual change trend of geological disasters for the safe management of the slope in the reservoir area of ​​the hydropower station and provides timely disaster prevention and mitigation guidance for the power station.

[0165] The above description is only a preferred embodiment of the present invention and does not constitute any other form of limitation to the present invention. Any modification or equivalent change made based on the technical essence of the present invention still falls within the scope of protection required by the present invention.

Claims

1. A method for predicting the comprehensive health of a hydropower station slope considering ecological and environmental index data, characterized in that: The following steps are involved: S1: Factors affecting the health of the excavation slope: Select disaster observation point A and disaster observation point B at the landslide site in the reservoir area of ​​the hydropower station, detect and record the topography, meteorological conditions, and ecological environment of the slopes at disaster observation point A and disaster observation point B, and determine the indicators of the health influencing factors of disaster observation point A and disaster observation point B; S2: Based on the factors affecting slope health, a multi-dimensional slope health prediction and evaluation index system is constructed: The influencing factor indicators of the three types of topography, meteorological conditions, and ecological environment are numbered separately to form a collection, which is the multi-dimensional slope health prediction and evaluation index system; S3: According to the multi-dimensional health prediction and evaluation index system, the correlation analysis of the indicators is carried out: The Pearson correlation coefficient method is used to measure the correlation between two different evaluation indicators. The calculation formula is as follows: When the absolute value of the Pearson correlation coefficient is greater than 0.7, it indicates that the selected indicators are closely related; when the absolute value is less than 0.2, it indicates that the selected indicators are not closely related; S4: Perform outlier detection on selected indicators: Use the local outlier factor to detect outliers and calculate the outlier value of each indicator data. The calculation formula is as follows: reach_dist(p,q)=max{d(p,q),k_distance(q)} S5: Select indicator data: When the outlier value of the indicator data is significantly greater than other indicators, the data indicator set with the largest outlier value is obtained and removed from the indicator data; S6: Use synthetic minority class oversampling technology to balance indicator data: x n =x i +Rand(0,1)×(x iz -x i ) In the formula, x i is a sample in the minority class; x iz According to the sampling rate N, from x i Randomly select the zth sample x from the K nearest neighbors of iz ;x n is a newly generated sample; Rand(0,1) is a random number within (0,1). S7: Using random forest to predict the health of hydropower station slopes: The random forest algorithm is an ensemble learning algorithm based on decision trees. For classification problems, it is a classifier composed of multiple decision trees. For indicator data, K decision trees {l1(x),l2(x),…,l k (x)}, construct a random forest model based on the obtained decision tree model: Where L(x) is the random forest model finally constructed; i (x) is a single decision tree model; Y is the target variable; H(.) is the classification function of the decision tree model. S8: Use grid search method to optimize the parameters of random forest: In the random forest model, the number of decision trees, the maximum number of features, and the depth of the decision tree are key factors affecting the growth of the decision tree, and are also important guarantees for the performance and accuracy of the random forest model. The grid search method is used to find the optimal parameter combination of the model.

2. The method for predicting the comprehensive health of a hydropower station slope considering ecological environment index data according to claim 1 is characterized in that: In step S1, the detection of the topography includes the detection of the slope direction, pore water pressure, slope and curvature at the disaster observation point A and the disaster observation point B.

3. The method for predicting the comprehensive health of a hydropower station slope considering ecological environment index data according to claim 1 is characterized in that: In step S1, the detection of meteorological conditions includes the detection of rainfall, humidity, and rock and soil moisture content at disaster monitoring point A and disaster observation point B.

4. The method for predicting the comprehensive health of a hydropower station slope considering ecological environment index data according to claim 1 is characterized in that: In step S1, the detection of the ecological environment includes the detection of the normalized difference vegetation index and the normalized difference water index at the disaster observation point A and the disaster observation point B.

5. The method for predicting the comprehensive health of a hydropower station slope considering ecological environment index data according to claim 4 is characterized in that: In step S1, the normalized difference vegetation index (NDVI) among the ecological environment indicators reflects the growth density and health of plants. The higher the value, the more luxuriant the plant growth and the more developed the root system, which can better fix the soil, enhance the stability of the slope, and make the slope healthier.

6. The method for predicting the comprehensive health of a hydropower station slope considering ecological environment index data according to claim 4 is characterized in that: In step S1, the normalized difference water index among the ecological environment indicators reflects the existence of surface water bodies. The higher the value, the greater the risk of erosion and geological disasters on the slope, which will reduce the stability of the slope and the health of the slope.

7. The method for predicting the comprehensive health of a hydropower station slope considering ecological environment index data according to claim 1 is characterized in that: In step S4, the distance between the index data and its adjacent points is shown in formula (1-1): The distance from the indicator data to the neighboring points is shown in formula (1-2): reach_dist(p,q)=max{d(p,q),k_distance(q)}(1-2) The local reachable density of the indicator data is shown in formula (1-3): The outlier calculation of indicator data is shown in formula (1-4):

8. The method for predicting the comprehensive health of a hydropower station slope considering ecological environment index data according to claim 1 is characterized in that: In step S5, the maximum outlier set to be eliminated is shown in formula (2-1): O={o1,o2,…,o n }(2-1); In step S6, the over-synthesized minority class oversampling technique of the index data is shown in formula (3-1): x n =x i +Rand(0,1)×(x iz -x i )(3-1) In the formula, x i is a sample in the minority class; x iz According to the sampling rate N, from x i Randomly select the zth sample x from the K nearest neighbors of iz ;x n is a newly generated sample; Rand(0,1) is a random number within (0,1).

9. The method for predicting the comprehensive health of a hydropower station slope considering ecological environment index data according to claim 1 is characterized in that: In step S7, if there are N samples, the constructed feature vector is as shown in formula (4-1): {(x i ,y i )|i=1,2,…,N} (4-1) where y i The slope of the hydropower station i The factors affecting the health of the hydropower station slope, namely, the slope aspect, pore water pressure, slope, curvature rainfall, humidity, rock and soil moisture content, normalized difference vegetation index, and normalized difference water index selected in step S1 are used as feature vectors; The established single decision tree randomly selects m feature variables for splitting, and i and health i , search for split variables and split values, and divide the entire vector space into m partitions, as shown in formula (4-2): {R1,R2,…,R m }(4-3) For any partition, the partition is classified into two parts, as shown in formulas (4-4) and (4-5): R1(j,s)={I|I j ≤s}(4-4) R2(j,s)={I|I j >s}(4-5) In the formula: j represents the variable type, and s represents the value at the time of splitting. The objective function for searching for split variables and split values ​​in vector space is shown in formula (4-6): in: c1=ave(y i |x i ∈R1(j,s))(4-7) c2=ave(y i |x i ∈R2(j,s))(4-8) A random forest is constructed based on the constructed decision tree, as shown in formula (4-9):

10. The method for predicting the comprehensive health of a hydropower station slope considering ecological environment index data according to claim 1 is characterized in that: In step S8, the grid search method is used to find the best parameter combination with a sample training set ratio of 0.8, a number of decision trees of 50, a maximum number of features of 4, and a maximum depth of 17.

Images