A landslide risk early warning rule generation method based on multi-source data fuzzy weight

Abstract

The application discloses a landslide risk early warning rule generation method based on multi-source data fuzzy weight, and the method provided by the application trains the initially established risk early warning rule library by using historical monitoring data to obtain a risk early warning rule library with relatively strong fitness; compared with the early warning rules obtained by a traditional geological team according to self experience and on-site investigation, the risk early warning rule generation efficiency is high, and the cost is low; and based on the performance of historical data, the generated risk early warning rules have stronger applicability.

Description

Technical Field

[0001] This invention belongs to the field of landslide prediction and early warning technology, and in particular relates to a method for generating landslide risk early warning rules based on fuzzy weights of multi-source data. Background Technology

[0002] Landslides refer to the natural phenomenon where soil or rock masses on a slope, influenced by natural or human factors, slide down the slope along a certain weak surface or zone under the influence of gravity. Landslides include collapses and debris flows. Landslides are a highly destructive geological hazard, most commonly found in low mountain and hilly areas. During my country's period of rapid economic development and urban construction, human activities are altering the topography, leading to increased artificial slope cutting and a greater risk of landslides. Monitoring and early warning systems, as a crucial measure for mitigating geological hazard risks, are receiving increasing attention.

[0003] In monitoring and early warning work, issuing disaster risk levels according to certain early warning rules has been widely used. Currently, most landslides still rely on fixed triggering conditions provided by geological teams as the sole criterion for judgment. While the professionalism of such indicators is undeniable, they also suffer from drawbacks such as high cost and difficulty in large-scale application. With the increase in landslide monitoring points and the lengthening of monitoring time, improving the efficiency of early warning rule generation has become an urgent problem to be solved. Summary of the Invention

[0004] This invention provides a method for generating landslide risk early warning rules based on fuzzy weights of multi-source data, which helps to improve the generation efficiency of early warning rules to a certain extent, and the generated early warning rules have strong adaptability.

[0005] To achieve the objectives of this invention, the provided method for generating landslide risk early warning rules based on multi-source data fuzzy weights includes the following steps:

[0006] Step 1: Construct an initial risk warning rule base, which includes several sub-initial risk warning rule bases, each of which contains several risk warning rules;

[0007] Step 2: Obtain historical monitoring data, which includes displacement and rainfall. The corresponding displacement and rainfall are statistically analyzed on a daily basis, and the daily data is used as a sample data set.

[0008] Step 3: Train the sub-initial risk warning rule bases in the initial risk warning rule base constructed in Step 1 using the sample data from Step 2, and obtain the daily fitness score (MSE) of each sub-initial risk warning rule base. risk This includes the following steps:

[0009] Step 1: Obtain the displacement S for the day. iand rainfall A i Based on the constructed membership function, the displacement S i and rainfall A i The numerical values ​​are converted into corresponding risk levels;

[0010] Step 2: Matching the initial risk warning rule base using the displacement S i and rainfall A i The risk level is matched with the risk warning rules in the initial risk warning rule base. When the risk warning rule in the initial risk warning rule base matches the displacement and rainfall risk level with the displacement amount S... i and rainfall A i When the risk levels are equal, the risk warning rule is successfully matched.

[0011] Step 3: Using the constructed membership function, obtain the membership degree corresponding to the displacement risk level, rainfall risk level, and result risk level of each risk warning rule that was successfully matched in Step 2.

[0012] Step 4: Take the smaller value between the displacement risk level membership degree and the rainfall risk level membership degree obtained in Step 3 as the confidence level value of the risk warning rule in the corresponding sub-initial risk warning rule library. Use this confidence level value to truncate the membership function curves corresponding to the result risks of all successfully matched risk warning rules in the corresponding sub-initial risk warning rule library. Take the part smaller than the confidence level value as the output and superimpose it.

[0013] Step 5: Defuzzify the output of Step 4 to obtain the specific fuzziness level value of each sub-initial risk warning rule base, which is used as the warning score S of each sub-initial risk warning rule base. i ;

[0014] Step 6: Calculate the fitness score (MSE) of the initial risk warning rule base for the current day using the following formula. risk ;

[0015] MSE risk =(D i ′-S i ′) 2

[0016] S i ′=(S i -S min ) / (S max -S min )

[0017] D i ′=(D i -D min ) / (D max -D min )

[0018] S i ′ represents the normalized displacement value, S i S is the warning score for the day. min To determine the lowest historical warning score, when selecting data from the first day, S min The value is 0; S max S represents the highest historical warning score when selecting data from the first day. max The value is the warning score for the day; D i ′ represents the normalized displacement for the next day, D i D represents the displacement value for the next day. min D is the minimum displacement value among the historical data obtained. max The maximum displacement value among the historical data obtained;

[0019] Step 4: Calculate the final score MSE′ of each sub-initial risk warning rule base in the initial risk warning rule base established in Step 1 within the selected historical period according to the following formula. risk ;

[0020]

[0021] In the formula: N represents the historical period, in days;

[0022] Step 5: Select the final score MSE′ risk Sub-initial risk warning rule bases with scores less than or equal to a preset threshold are used as the final risk warning rule base for risk warning; or each sub-initial risk warning rule base and its final score are used as input, and a genetic algorithm is used to iteratively optimize the warning rules, and finally output sub-initial risk warning rule bases with final scores less than or equal to a preset threshold are used as the final risk warning rule base for risk warning.

[0023] In some implementations, step two involves cleaning the historical monitoring data after acquisition, removing outliers, smoothing spurious data, and then statistically analyzing the corresponding displacement and rainfall on a daily basis. This avoids the situation where abnormal and spurious data affect the generated results, and further ensures the adaptability of the final risk warning rule base.

[0024] A second object of the present invention is to provide a generating apparatus comprising:

[0025] The generation module is used to generate the initial risk warning rule base;

[0026] The acquisition module is used to acquire historical monitoring data, which includes displacement and rainfall. The corresponding displacement and rainfall are statistically analyzed on a daily basis, and the daily data is used as a set of sample data.

[0027] The processing module is used to train the sub-initial risk warning rule bases in the initial risk warning rule base constructed by the generation module using the sample data obtained by the acquisition module, so as to obtain the daily fitness score (MSE) of each sub-initial risk warning rule base. risk Then, according to the following formula, calculate the final score MSE′ of each sub-initial risk warning rule base in the initial risk warning rule base established by the generation module within the selected historical period. risk ;

[0028]

[0029] In the formula: N represents the historical period, in days;

[0030] The processing module is also used to select the final score MSE′. risk Sub-initial risk warning rule bases with scores less than or equal to a preset threshold are used as the final risk warning rule base for risk warning; or each sub-initial risk warning rule base and its final score are used as input, and a genetic algorithm is used to iteratively optimize the warning rules, and finally output sub-initial risk warning rule bases with final scores less than or equal to a preset threshold are used as the final risk warning rule base for risk warning.

[0031] The third objective of this invention is to provide a landslide risk early warning method, which uses a risk early warning rule library generated by the landslide risk early warning rule generation method based on multi-source data fuzzy weights provided in this application to predict landslide risks.

[0032] In some implementations, after a risk warning period ends, the displacement and rainfall of that warning period are used as historical monitoring data to generate a risk warning rule base according to the generation method provided in this application, thereby updating the risk warning rule base and ensuring its timeliness.

[0033] The technical effects that can be achieved by adopting the technical solution of this application include at least the following:

[0034] 1) The method provided by this invention uses historical monitoring data to train the initially established risk warning rule base to obtain a risk warning rule base with strong adaptability. Compared with the warning rules obtained by traditional geological teams based on their own experience and field investigations, the risk warning rule generation is more efficient and less costly; and based on the performance of historical data, the generated risk warning rules have stronger applicability.

[0035] 2) By regularly updating the historical monitoring data used to train risk warning rules, the timeliness of the risk warning rule base has been improved.

[0036] 3) This method can be applied on a large scale to different landslide monitoring points, and a risk warning rule library adapted to the actual prediction area can be generated, which meets the needs of universal landslide monitoring. Attached Figure Description

[0037] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention. It is obvious that the drawings described below are merely some embodiments of the invention, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort. In the drawings:

[0038] Figure 1 This is a schematic diagram of a risk warning rule matrix as described in this invention;

[0039] Figure 2 This is a schematic diagram of the improved Gaussian function curve described in this invention;

[0040] Figure 3 This is a schematic diagram of another risk warning rule matrix described in this invention;

[0041] Figure 4 This is a flowchart of the generation method described in this invention;

[0042] Figure 5 This is a membership function curve diagram as described in this invention;

[0043] Figure 6 This is a schematic diagram of the FIS principle described in this invention;

[0044] Figure 7 This is a schematic diagram of the Mean of Maximum (mmo) method described in this invention;

[0045] Figure 8 This is a schematic diagram of the Center of Mass(com) method described in this invention;

[0046] Figure 9 This is a schematic diagram of the lower part of the first slope zone on the right bank of the Bailongjiang Wudu urban section as described in this invention;

[0047] Figure 10 This is a slope diagram;

[0048] Figure 11 This is a distribution diagram of the monitoring data before cleaning 09-GP01 as described in this invention;

[0049] Figure 12 This is a distribution diagram of the monitoring data after cleaning of 09-GP01 as described in this invention;

[0050] Figure 13This is a distribution map of the daily variation of surface displacement after 09-GP01 cleaning as described in this invention;

[0051] Figure 14 This invention records the daily rainfall distribution of the experimental subjects;

[0052] Figure 15 This is a Gaussian membership function curve of the 3-day cumulative rainfall of the experimental object described in this invention;

[0053] Figure 16 This is a Gaussian membership function curve of the daily displacement of the experimental object described in this invention;

[0054] Figure 17 This is a Gaussian membership function curve of the comprehensive danger warning of the experimental object described in this invention;

[0055] Figure 18 This is a fitness distribution diagram for each individual in each generation involved in the experiments described in this invention;

[0056] Figure 19 The optimal fitness distribution diagram for each generation involved in the experiments described in this invention;

[0057] Figure 20 This is a fitness distribution diagram of each individual in each generation after incorporating constraint rules into the experiment described in this invention;

[0058] Figure 21 This is a graph showing the optimal fitness distribution of each generation involved in the experiments described in this invention after incorporating constraint rules.

[0059] Figure 22 This is an example diagram illustrating the integrated early warning system for surface displacement and rainfall of the experimental object described in this invention;

[0060] In the attached diagram: nan represents no risk level, low represents low risk level, mid represents medium risk level, high represents high risk level, and v-hight represents extremely high risk level; x, y, x0, and y0 represent input quantities (displacement and rainfall), rulestrength represents the risk warning rule, and uotputdistribution represents the output distribution (the truncated result). Detailed Implementation

[0061] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided to make the invention more comprehensive and complete, and to fully convey the concept of the exemplary embodiments to those skilled in the art.

[0062] For ease of understanding, the following is an explanation of the concepts related to this application.

[0063] 1. Sub-initial risk warning rule base

[0064] This includes several risk warning rules, which can be presented in the following forms: Figure 1 The risk matrix shown will output the corresponding risk level and issue an early warning when the monitored data, such as displacement and rainfall, meet the corresponding conditions in the matrix.

[0065] 2. Historical monitoring data

[0066] Monitoring data taken within a certain period before the warning period, such as data from one month, half a month, or three months before the warning period (since the weather, rainfall, displacement, and other data from adjacent months are similar, data from adjacent months are used as historical data for more accurate results), includes the displacement and rainfall of landslides at the monitoring points, and the monitoring data is statistically analyzed on a daily basis.

[0067] The early warning period refers to the specific time period during which monitoring stations need to conduct early warning monitoring.

[0068] 3. Membership function

[0069] If for any element x in the universe of discourse (the scope of study) U, there is a number A(x) ∈ [0, 1] corresponding to it, then A is called a fuzzy set on U, and A(x) is called the membership degree of x to A. When x changes in U, A(x) is a function, called the membership function of A. The closer the membership degree A(x) is to 1, the higher the degree to which x belongs to A; the closer A(x) is to 0, the lower the degree to which x belongs to A. The membership function A(x) taking values ​​in the interval (0, 1) is used to characterize the degree to which x belongs to A. Membership degree is a concept in fuzzy evaluation functions: Fuzzy comprehensive evaluation is a very effective multi-factor decision-making method for making a comprehensive evaluation of things affected by multiple factors. Its characteristic is that the evaluation result is not an absolute affirmation or negation, but is represented by a fuzzy set.

[0070] Membership transformation functions are used to convert specific numerical values ​​of the input system into fuzzy sets. Commonly used membership transformation functions include the triangular function, trapezoidal function, and Gaussian function. The Gaussian function has two key parameters, c and σ. c is the expected value of the sample, representing the center point of the function, and σ is the standard deviation of the sample, representing the width of the curve. Compared to the triangular and trapezoidal function graphs, the Gaussian function better reflects the distribution characteristics of ground features and can better represent the continuous changes in data.

[0071] This disclosure uses a Gaussian fuzzy transformation function. Since the membership function is not necessarily strictly symmetrical, this disclosure uses curves with different standard deviations to represent it. Specifically, a certain value is used as a dividing point, and the Gaussian curves on the left and right sides take different standard deviations, thereby controlling the curve shape. Figure 2 As shown, the horizontal axis represents variables, such as displacement, rainfall, and risk level, while the vertical axis represents the degree of membership. Figure 2 The curve representation shown is more flexible; the traditional Gaussian function is symmetrical, while... Figure 2 As shown, asymmetry can be achieved, making the degree to which the membership degree changes with the value more flexible.

[0072] 4. Deblurring

[0073] To convert the fuzziness level into a clear and specific output result, common defuzzification methods include Mean of Maximum (mmo) and Center of Mass (coM). The Center of Mass (coM) method takes the centroid of the truncated region and uses the x-coordinate of the centroid as the fuzzy value. The coM method has smoother output inference control.

[0074] 5. Iterative Genetic Algorithm

[0075] Genetic algorithms are designed based on the evolutionary laws of organisms in nature. They are computational models that simulate the biological evolutionary process through natural selection and genetic mechanisms, as described in Darwin's theory of evolution. Genetic algorithms are a method for optimizing parameters by simulating natural evolution. They mainly consist of encoding, fitness function construction, genetic operations, and termination condition settings. Encoding involves modeling the actual problem to be solved using chromosomes in a genetic space. A fitness function is constructed to determine the quality of an individual within the population. Genetic operations include selection, crossover, and mutation. Iteration terminates when the termination condition is met or the fitness function converges. Compared to some conventional optimization algorithms, genetic algorithms can achieve better optimization results more quickly when solving complex combinatorial optimization problems.

[0076] The genetic algorithm process is as follows: (1) Initialize the population: Set the maximum number of generations and randomly generate the initial population in the genetic space. (2) Individual evaluation: Calculate the fitness of individuals in the population. (3) Genetic operations: Run selection, crossover, and mutation operations, and select high-quality individuals to be inherited by the next generation through fitness evaluation. (4) Determine whether to terminate: If the number of generations reaches the set condition or the fitness value tends to converge, the iteration is terminated, and the optimal solution is the individual with the highest fitness.

[0077] This disclosure utilizes a genetic algorithm to iterate through the initial risk warning rule base. The initial risk warning rule base and its score are used as input to the genetic algorithm. The genetic algorithm performs selection, crossover, mutation, and other operations based on the score to obtain a risk warning rule base with higher fitness.

[0078] Genetic algorithms can also incorporate constraints during the iteration process. Therefore, certain human experience can be added to the search system in the form of constraints. This disclosure sets that the warning level must not be reduced when the rainfall or displacement increases; at the same time, it sets that the warning level is the lowest level when both rainfall and displacement are at the lowest level, and the warning level is the highest level when both are at the highest level.

[0079] 6. Each generation refers to a risk warning rule base derived from specific historical monitoring data. For example, a risk warning rule base derived from historical monitoring data from June 30, 2020 to December 31, 2020 can be called a generation; a risk warning rule base derived from historical monitoring data from March 1, 2021 to May 30, 2021 can also be called a generation. Each generation yields at least one risk warning rule base.

[0080] Monitoring and early warning are crucial measures for reducing landslide risk. For rainfall-induced landslides, rainfall and landslide displacement are key parameters for monitoring and early warning. Critical rainfall warnings can indirectly reflect the temporal and spatial displacement failure patterns and common characteristics of displacement evolution in landslides to some extent, but their spatial accuracy is insufficient. Integrated early warning methods based on rainfall and displacement can fully incorporate the unique deformation characteristics of the slope while integrating the common characteristics of landslide deformation evolution, making them an effective measure to reduce false alarm rates. However, current integrated rainfall and displacement early warning rules are mostly based on expert experience, resulting in significant subjective uncertainty. Furthermore, rule generation is inefficient and time-consuming, failing to meet the increasing number of landslide monitoring points and monitoring time.

[0081] This invention provides a landslide risk early warning rule generation method based on multi-source data fuzzy weights to achieve automatic generation of landslide early warning rules, improve rule generation efficiency and cost, and effectively ensure the applicability of the generated early warning rules. This method is based on fuzzy principles; fuzzy theory can realize complex nonlinear mapping relationships. Specifically, the reasoning process involves first fuzzifying the clear system input, then executing the fuzzy input in parallel through a fuzzy rule base, obtaining the total fuzzy output set based on the reasoning mechanism, and finally defuzzifying the fuzzy output set into specific numbers to obtain the fuzzy inference result. This generation method includes steps such as data processing, quantitative evaluation of early warning rules, and adaptive generation of the early warning rule base. Figure 4 As shown, specific implementation methods are described in the following embodiments.

[0082] Step 1: Construct an initial risk warning rule base, including several sub-initial risk warning rule bases. Each sub-initial risk warning rule base contains several risk warning rules. The warning rules in each constructed sub-initial risk warning rule base are different, but all sub-initial risk warning rule bases can contain all warning rules. That is, for each set of sample data (displacement, rainfall), a corresponding risk warning rule can be found in each sub-initial risk warning rule base. For example, the constructed initial risk warning rule base includes two sub-initial risk warning rule bases, labeled as Sub-initial Risk Warning Rule Base A and Sub-risk Warning Rule Base B. The warning rules in Sub-initial Risk Warning Rule Base A can be as follows: Figure 1 As shown, the warning in the initial risk warning rule base B can be set as follows: Figure 3 As shown; when rainfall > 50mm and displacement is between 50 and 100mm, both sub-initial risk warning rule base A and sub-risk warning rule base B have corresponding warning rules. The warning rule corresponding to sub-initial risk warning rule base A is: rainfall > 50mm, displacement between 50 and 100mm, risk level - extremely high; the warning rule corresponding to sub-initial risk warning rule base B is: rainfall > 50mm, displacement between 50 and 100mm, risk level - high.

[0083] Each sub-initial risk warning rule base contains all risk warning situations to ensure the completeness of each sub-initial risk warning rule base. Those skilled in the art should understand that although this disclosure describes the initial risk warning rules as including two sub-initial risk warning rule bases, it does not imply that the initial risk warning rule base includes only two sub-initial risk warning rule bases. Rather, it may include 10, 20, 50, 100, or more sub-initial risk warning rule bases depending on the needs.

[0084] Step Two: Acquire historical monitoring data, including landslide displacement and rainfall. The corresponding displacement and rainfall are statistically analyzed daily to form training samples for training the adaptability of the risk warning rule base. In landslide monitoring, various monitoring data are generated by monitoring instruments. Instrument instability, malfunctions, and line faults are unavoidable, leading to data loss, spikes, and noise. Furthermore, instrument maintenance can cause abrupt changes in monitoring data values, resulting in outliers and data loss. To ensure the adaptability of the generated results, this disclosure preprocesses the historical monitoring data after acquisition.

[0085] The data preprocessing disclosed in this publication mainly includes anomaly identification and smoothing. Anomaly identification is primarily aimed at removing abrupt changes in monitoring data caused by instrument instability, equipment malfunction, circuit failure, or instrument maintenance. Various monitoring instruments on the landslide have their measurable ranges. For data collected that exceeds the range or outputs special codes indicating instrument malfunction (such as negative values ​​from rain gauges), these can be removed and cleaned according to the instrument's instructions.

[0086] After removing obvious instrument anomalies, it is necessary to identify anomalies within the normal observation range. First, the landslide displacement data is segmented according to the different displacement rates to obtain the stable stage and the rapid deformation stage of the landslide. Then, anomalies are identified within each stage, and monitoring data that are significantly greater than or less than the average value of that stage are removed.

[0087] Due to objective observation conditions, historical observation data often exhibits significant random fluctuations, resulting in spikes. In the actual construction of machine learning models, the models may overfit in an attempt to fit these locally unstable data. Therefore, to avoid overfitting and inaccurate results, this disclosure employs a moving average method for data smoothing. The moving average method averages 2n+1 observations before and after a selected historical data period (the middle time, the time corresponding to the maximum data point, or the time corresponding to the minimum data point) to obtain the filtered result for the current time. Here, n represents the number of preceding and following times, where n is the current time plus the previous n times and the next n times. The value of n is set according to the situation, such as 5, 8, 12, or greater than or equal to 15. When the actual data within the moving window does not change significantly, a large portion of the noise can be suppressed, and the filtered result approximates the true value. When the actual value within the moving window changes significantly, this filtering method will lose some accuracy, and the filtered result will approach the average expected value of the true value.

[0088] In this disclosure, the sample is calculated by taking days as the time unit, and statistically analyzing data such as displacement and rainfall to generate daily data. Each day's data is a set of sample data, as shown in Table 1.

[0089] Table 1. Data Sample Illustration

[0090]

[0091] Step 3: Train the sub-initial risk warning rule bases in the initial risk warning rule base constructed in Step 1 using the sample data from Step 2, and obtain the daily fitness score (MSE) of each sub-initial risk warning rule base. riskThis step requires using each set of sample data constructed in step two as input to train the sub-initial risk warning rule base. Each set of sample data input into each sub-initial risk warning rule base can obtain the corresponding warning score for that day, denoted as the warning score S. i If historical monitoring data is taken from 30 days, then 30 sets of sample data are formed. Each sub-initial risk warning rule base has 30 warning scores, denoted as S1, S2, S3, S4, S5, ..., S... 29 S 30 , representing the warning scores when 30 sets of sample data are used as input. It should be understood by those skilled in the art that this description uses 30 as an example only, and does not limit historical monitoring data to only 30 days.

[0092] When acquiring historical monitoring data, the rainfall can be the rainfall of the current day only, or it can be the cumulative amount, which is the sum of the rainfall of the previous few days and the rainfall of the current day. For example, if the historical monitoring data is taken for 10 days, the rainfall of the 3rd day can be the cumulative amount of the rainfall of the 1st day, the rainfall of the 2nd day, and the rainfall of the 3rd day; the rainfall of the 8th day can be the sum of the rainfall of the previous 7 days and the rainfall of the 8th day, or the sum of the rainfall of 2 or 3 of the previous 7 days and the rainfall of the 8th day.

[0093] Step three specifically includes the following steps:

[0094] Step 1: Obtain the displacement S for the day. i and rainfall A i (A set of sample data), based on the constructed membership function, the displacement S i and rainfall A i The numerical value is converted into the corresponding risk level.

[0095] Step 2: Matching the initial risk warning rule base using the displacement S i and rainfall A i The risk level is matched with the risk warning rules in the initial risk warning rule base. When the risk warning rule in the initial risk warning rule base matches the displacement and rainfall risk level with the displacement amount S... i and rainfall A i When the risk levels are equal, the risk warning rule is successfully matched.

[0096] Step 3: Using the constructed membership function, obtain the membership degree corresponding to the displacement risk level, rainfall risk level, and result risk level of each risk warning rule that was successfully matched in Step 2.

[0097] Step 4: Take the smaller value between the displacement risk level membership degree and the rainfall risk level membership degree obtained in Step 3 as the confidence level value of the risk warning rule in the corresponding sub-initial risk warning rule library. Use this confidence level value to truncate the membership function curves corresponding to the result risks of all successfully matched risk warning rules in the corresponding sub-initial risk warning rule library. Take the part smaller than the confidence level value as the output and superimpose it.

[0098] If two risk warning rules are successfully matched within a sub-initial risk warning rule base, it indicates that the displacement risk level, rainfall risk level, and result risk level (the final risk level obtained based on the displacement risk level and rainfall risk level; for example, if the displacement risk level is medium and the rainfall risk level is high, the corresponding risk level is high) in each risk warning rule correspond to their respective membership functions. Based on these membership functions, the membership degrees of the displacement risk level and rainfall risk level in each risk warning rule, as well as the risk level result membership function curve, are obtained and denoted as displacement risk membership degree a1, displacement risk membership degree a1, and displacement risk membership degree a2, respectively. Risk membership degree a2, rainfall risk membership degree b1, rainfall risk membership degree b2, membership degree function curve y1, membership degree function curve y2; select the smallest membership degree among displacement risk membership degree a1, displacement risk membership degree a2, rainfall risk membership degree b1, and rainfall risk membership degree b2 as the risk warning rule confidence value of the corresponding sub-initial risk warning rule base; use this confidence value to truncate the membership degree function curves y1 and y2, and take the part of the membership function curves y1 and y2 that is less than the confidence value as the output, and the two outputs are superimposed.

[0099] Step 5: Defuzzify the output of Step 4 to obtain the specific fuzziness level value of each sub-initial risk warning rule base, which is used as the warning score S of each sub-initial risk warning rule base. i ;

[0100] Step 6: Obtain the fitness score (MSE) of the initial risk warning rule base for the current day (for this set of sample data) using the following formula. risk ;

[0101] MSE risk =(D i ′-S i ′) 2

[0102] S i ′=(S i -S min ) / (S max -S min )

[0103] D i ′=(D i -D min) / (D max -D min )

[0104] S i ′ represents the normalized displacement value, S i S is the warning score for the day. min To determine the lowest historical warning score, when selecting data from the first day, S min The value is 0; S max The highest historical warning score, when selecting the data from the first day, is S. max The value is the warning score for the day; D i ′ represents the normalized displacement for the next day, D i D represents the displacement value for the next day. min D is the minimum displacement value among the historical data obtained. max This represents the maximum displacement value among the historical data collected.

[0105] Following steps 1 through 6 above, obtain the fitness score (MSE) for each set of sample data in each sub-initial risk warning rule base under all sample data. risk .

[0106] Step 4: Calculate the final score MSE′ of each sub-initial risk warning rule base in the initial risk warning rule base established in Step 1 within the selected historical period using the following formula. risk ;

[0107]

[0108] In the formula: N represents the historical period, in days;

[0109] Step 5: Select the final score MSE′ risk Sub-initial risk warning rule bases with scores less than or equal to a preset threshold are used as the final risk warning rule base for risk warning; or each sub-initial risk warning rule base and its final score are used as input, and a genetic algorithm is used to iteratively optimize the warning rules, and finally output sub-initial risk warning rule bases with final scores less than or equal to a preset threshold are used as the final risk warning rule base for risk warning.

[0110] In this disclosure, the preset threshold is set according to the requirements, such as 0.1, 0.05, 0.2, etc.

[0111] This disclosure presents a quantitative description of qualitative risk warning rules using fuzzy inference systems (FIS), specifically based on... Figure 1 , 3The risk warning rules shown can only provide warning levels. Using fuzzy inference, a quantitative warning level score can be given based on the risk matrix. Then, a genetic algorithm is used to iteratively obtain risk warning rules with suitable fitness, which can be used to predict landslide risks within the prediction period.

[0112] When combining fuzzy reasoning systems and genetic algorithms, we first construct a risk warning rule base by combining human experience and historical data, and obtain historical monitoring data. The detailed steps are as described in steps one and two above, and will not be repeated here.

[0113] After generating samples in step two, the inference process using the FIS algorithm involves the following key steps:

[0114] 1) Fuzzify the input variables

[0115] In this step, membership functions for each variable need to be defined. For a specific value of an input variable, it is fuzzified to become a "degree variable." The purpose of the membership function is to characterize the confidence level to which a variable belongs.

[0116] In this method, the input variables are the displacement and rainfall on the target day. A membership function is defined as follows: if for any element x in the universe of discourse (the scope of study) U, there is a corresponding number A(x) ∈ [0, 1], then A is called a fuzzy set on U, and A(x) is called the membership degree of x to A. When x varies in U, A(x) is a function, called the membership function of A. The closer the membership degree A(x) is to 1, the higher the degree to which x belongs to A; the closer A(x) is to 0, the lower the degree to which x belongs to A. The membership function A(x) taking values ​​in the interval (0, 1) is used to characterize the degree to which x belongs to A. Figure 5 This is a schematic diagram of the membership function, representing the membership function corresponding to the risk result. One risk value may correspond to two risk levels.

[0117] like Figure 5 There are five risk levels corresponding to the values ​​of the independent variable: no risk, low risk, medium risk, high risk, and extremely high risk. Based on the specific values, the values ​​can be converted into levels using membership functions. In this example, extremely high risk has the highest membership score, so it can be identified as extremely high risk. This process of converting specific risk scores into risk levels is called "fuzzification." Both the independent variable (displacement, rainfall) and the dependent variable (risk level) have fuzzy functions used for the conversion between levels and values.

[0118] 2) Formulate reasoning rules

[0119] Inference rules are rules that derive specific evaluation levels from input values. For example:

[0120] Internally, the FIS algorithm consists of a "rule antecedent" and a "behavior consequent." The part from "If" to "Then" is the rule antecedent, composed of multiple fuzzy variables connected by the logical operators "and" or "or." The part after "Then" is the behavior consequent, representing the action to be taken under this rule. The logical operators "and" and "or" are used to represent the relationships between the fuzzy variables. Here, "and" and "or" differ slightly from their usual understanding. Normally, "and" and "or" operate on 0 and 1 (i.e., 1 and 0 = 0, 1 or 0 = 1). In FIS, the values ​​of fuzzy variables are not simply 0 or 1 (because it represents confidence levels), but can be a float value (between 0 and 1). Therefore, "and" refers to the minimum value of the two variables, and "or" refers to the maximum value.

[0121] 3) Inference using behavioral fuzzy variables: Defuzzification of behavioral variables

[0122] After determining the membership function and inference rules, given an input, the input can be fuzzified to obtain its classification level and confidence level. Then, a rule matching process is performed from the rule base (the input independent variables are precipitation and displacement, and many rules are formulated based on precipitation and displacement conditions, such as: If "high precipitation level" and "high displacement level" then "high risk level," etc.). These correspond to the previous rule matrix, where each cell corresponds to a rule. After inputting the independent variables, the levels of the independent variables are obtained, and then the corresponding rules are searched in the rule base. For example, if the input independent variables are "low" precipitation level and "medium" displacement level, then the rule base is searched for the result corresponding to "If "low precipitation level" and "medium displacement level," let's assume it's "If "low precipitation level" and "medium displacement level" then "low risk level." This is the rule matching process). The corresponding result level and its confidence level are obtained. The confidence level is then used to truncate the corresponding rating level, such as... Figure 6 As shown in the image below, the top left corner illustrates the cropping process. Because each independent variable has only one numerical value, after fuzzification, each level has a membership degree. For example... Figure 5Although it can be identified as an extremely high level, there is still a certain degree of membership for high-level risks. For example, the membership degree for high-level precipitation is 0.8, for medium-level precipitation it is 0.9, for high-level displacement it is 1, and for all others it is 0. Therefore, the rules corresponding to this independent variable include two rules: "high precipitation level and high displacement level then high risk level" and "medium precipitation level and high displacement level then medium risk level." These two rules can respectively obtain the scores for high-level and medium-level risks. In practical application, logical operations are still needed. The "and" in fuzzy logic operations takes the smaller value. After obtaining the result, it is then compared with the risk membership function to obtain the risk output distribution. Figure 6 The upper half shows the output distribution of the results after the two rules are processed. Following the example above, the first rule, after logical operation, yields a high risk level score of 0.8, and the second rule yields a medium risk level score of 0.9. Each risk level (high or medium) corresponds to a membership function curve. Using this value as the truncation point on the curve, we can obtain a distribution. Superimposing multiple distributions yields the final distribution. The truncation then produces the output distribution. The specific process is as follows... Figure 6 As shown.

[0123] After obtaining the output distribution, it is necessary to determine the specific degree of fuzziness for the evaluation level based on this distribution (that is, to calculate the score (x value) using a specific distribution), i.e., to defuzzify. The Mean of Maximum (mmo) method is as follows... Figure 7 As shown, the Center of Mass(com) method is as follows Figure 8 As shown.

[0124] The MMO method involves finding the highest straight line segment within the cut region and using the x-coordinate of the midpoint of that line as the deblurred value for the behavior. The COM method involves finding the centroid of the cut region and using the x-coordinate of the centroid as the deblurred value for the behavior.

[0125] After obtaining the early warning score using fuzzy inference, a fitness function needs to be defined to evaluate the degree of consistency between the early warning score and the actual situation. The mean square error (MSE) of the early warning score and the second-day displacement value, both normalized, is used as the evaluation metric. A smaller MSE indicates a better risk matrix performance. The MSE threshold is set according to requirements, such as 0.1. When the obtained MSE is less than or equal to 0.1, it indicates that the risk early warning rule is effective.

[0126] Then, by combining the genetic algorithm iteration, the sub-risk warning rule base with the best fitness is obtained as the final risk rule base; or the score is directly judged to determine whether it is less than or equal to the threshold, and the sub-risk warning rule base with the best fitness is obtained as the final risk rule base.

[0127] The early warning rule generation method proposed in this invention can automatically generate risk early warning rules for different slopes based on historical monitoring data, thus better integrating the actual conditions of the slopes. Furthermore, the results obtained by this method can be subject to human constraints, meaning that human experience can be used to assist in rule generation. Compared to directly using machine learning methods for early warning rating, the early warning rules obtained in this study are clearer and more intuitive.

[0128] After establishing early warning rules, each day's monitoring records will generate an early warning level, i.e., a predicted value, based on those rules. For example, the early warning level might be determined based on actual observation data and previously established rules, such as specifying a red warning level if the displacement is greater than 100mm and the rainfall is greater than 50mm. Determining the appropriateness of this predicted value is crucial; that is, defining the actual value of the hazard level. This scheme uses the displacement of the following day as a reference for the actual value (based on actual early warning and forecasting work; the predicted value given on the current day is an assessment of the subsequent hazard level, so the displacement on the second day is used for evaluation). In other words, a larger displacement on the following day indicates a higher hazard level for that day.

[0129] Therefore, for each day's observation data, the rainfall and displacement of that day are used as independent variables, and the displacement of the next day is used as the dependent variable to generate an experimental sample. One sample can be generated for each day, as shown in Table 1.

[0130] The method disclosed herein generates early warning rules automatically based on historical data, which is highly efficient. It can be applied on a large scale to a large number of landslide monitoring points and can continuously self-optimize in actual operation, meeting the needs of universal landslide monitoring.

[0131] After determining the warning period, risk warning generation rules adapted to the target monitoring points can be generated according to the generation method of this disclosure, resulting in stronger adaptability. The risk warning rule base generated using the production method of this disclosure is used for risk warnings at monitoring points. The risk warning rule base remains unchanged during the monitoring period to ensure monitoring stability. Based on the warning situation, the latest historical data can be used to generate an updated risk warning rule base for the next monitoring cycle according to the production method of this disclosure. If the set monitoring period has ended, the risk warning rule base can be generated using the data from that monitoring period according to the method of this disclosure to update the risk warning rule base and adjust its adaptability; alternatively, historical monitoring data from the same period of previous years can be used to generate the risk warning rule base.

[0132] This disclosure uses fuzzy logic to convert specific numerical values ​​into risk levels. A single numerical value can be converted to correspond to several risk levels. For example, rainfall of 50mm can be converted to correspond to medium-risk and high-risk levels through the membership principle of fuzzy logic. Alternatively, each risk level can be converted to several corresponding numerical values ​​through fuzzy logic. This transformation from a single qualitative value to a quantitative value is beneficial for generating more adaptable early warning rules.

[0133] This method is applicable to rainfall-induced landslides. It generates early warning rules based on rainfall-induced landslide data, and these rules are suitable for early warning of rainfall-induced landslides.

[0134] This disclosure uses a landslide at Zhaoba Village, Jiangnan Subdistrict, Wudu District, Longnan City, Gansu Province (104°55′42″, 33°22′15″) as the experimental object, combined with the method provided in this application, to illustrate the application in the middle and lower part of the first slope zone on the right bank of the Bailong River in the Wudu urban area (see...). Figure 9 It consists of 5 secondary landslides and 2 unstable slopes (see...). Figure 10 H1 landslide, once unstable, could form a high-speed long-distance landslide, potentially sliding into the Bailong River and generating swells, causing secondary disasters. It is included in key monitoring.

[0135] 1. Landslide Overview

[0136] The formation of the Xieliupo landslide is closely related to its surrounding topography, fault structure, and material composition. Rainfall and groundwater accumulation accelerate its development. Favorable topographical conditions for landslide development include: the landslide slope dips towards the Bailongjiang Riverbed; the landslide profile is generally straight with a steep embankment at the rear and a bulge at the front; and the exposed slope topography is conducive to landslide development. Complex geological tectonic processes: the landslide development is mainly influenced by the large thrust fault along the south bank of the Bailongjiang River in Wudu, which caused the Carboniferous limestone at the rear edge of the old Xieliupo landslide to thrust over Silurian phyllite. The strata between these two layers are highly fractured, forming a fault fracture zone, which promotes landslide formation. Easily landslide-prone strata: the exposed strata are mainly Carboniferous, Silurian, and Quaternary strata, with the Silurian strata being soft rock and the main component, exhibiting well-developed fissures and faults, and localized small folds. The surface rocks are mostly severely weathered and strongly weathered layers, severely fractured, making them easily landslide-prone strata. Groundwater accumulation reduces soil strength: This is significantly influenced by seasonality. During the rainy season, rainfall replenishes the soil, causing large amounts of groundwater to accumulate along the top of the bedrock, reducing soil strength and increasing the risk of landslides. ⑤ Rainfall is one of the main factors inducing landslides. Wudu's average annual rainfall is 494.8 mm, with over 65% of the annual rainfall occurring from June to September, often in the form of torrential rain. The erosion and infiltration of rainwater into the gullies increase the weight of the slope, triggering landslides. From August 11th to 17th, 2020, Longnan City experienced a prolonged rainfall event, with rainfall ranging from 189.4 to 221.2 mm, and a maximum daily rainfall of 68.6 to 77 mm, resulting in significant deformation of the runoff slope landslide.

[0137] 2. Monitoring data preprocessing

[0138] In this experiment, deformation characterization was performed using surface displacement data. The experiment was conducted based on monitoring data from rain gauges (number: 10-YL01) and BeiDou surface displacement detectors (number: 09-GP01) deployed on the H1 landslide body of the spillway slope. The surface displacement was calculated as a three-dimensional sum displacement. The monitoring data period was from June 18, 2020 to August 31, 2021. The original monitoring curves are shown below. Figure 11 As shown, due to factors such as the field observation environment, instrument stability, and data communication, the monitoring data contains spikes and noise. Therefore, data cleaning is the first step, including the identification and removal of abnormal monitoring data and the filling of missing data. In this experiment, data conforming to the following formula were identified as potential outliers:

[0139]

[0140] In the formula, A represents a data point at a certain time in the segmented data; Q(i), i∈[0,1] indicates that the data with a proportion of i in the segmented data is less than Q(i); 0.9 and 0.1 are the set sample confidence ratio intervals, and 1.5 represents the ratio multiple of the difference between the feature values ​​of the two confidence ratio intervals.

[0141] Landslide monitoring anomalies often exhibit short-term and sudden characteristics. Potential anomalies require further confirmation using adjacent time-series data. If the differences with adjacent time-series data are in the same direction (i.e., simultaneously greater than or simultaneously less than both the preceding and following time-series data), they are identified as anomalies and removed. For short-term, small-volume data gaps caused by insufficient power supply, communication interruptions, or equipment offline, a linear interpolation method is used to fill the gaps with data from preceding and following time series. The reconstructed monitoring data is continuous and conforms to the statistical regularity of the complete dataset. The cleaned cumulative surface displacement data and daily displacement data are shown below. Figure 12 , 13 As shown, Figure 14 This represents the distribution of daily rainfall over time.

[0142] 3. Membership function construction and Mamdani fuzzy inference system establishment

[0143] The experiment involves three variables: 3-day cumulative rainfall, daily displacement, and the combined hazard (referred to as combined hazard). First, the corresponding membership function needs to be constructed. Then, Mamdani fuzzy inference and defuzzification after combination are performed according to the combination rule matrix to obtain the comprehensive warning score and the comprehensive warning level with confidence.

[0144] When constructing membership functions for the three variables, Gaussian membership function curves with different standard deviations were used. Therefore, each combined Gaussian curve required three segment points for combined control, and each combined Gaussian curve represented the membership degree change for a certain classification level. The specific method for constructing the membership functions was as follows: the 3-day cumulative rainfall, daily displacement, and combined hazard were divided into five levels: extremely low, low, medium, high, and extremely high. The corresponding segment points are shown in Table 2. The segment points for the 3-day cumulative rainfall were mainly based on historical experience of historical disaster events and corresponding rainfall data. The segment points for daily displacement were mainly based on the distribution characteristics of the monitoring data. The segment points for combined hazard had an even distribution attribute. The membership function images for the warning levels corresponding to the 3-day cumulative rainfall, daily displacement, and combined hazard are shown in Table 2. Figure 15-17 As shown in the figure, the curves of different line segments correspond to different warning levels.

[0145] Table 2. Warning Level Classification and Gaussian Function Segmentation Points (3-day Cumulative Rainfall, Daily Displacement, and Combined Hazard)

[0146]

[0147] 4. Evolution and Analysis of Comprehensive Early Warning Rules

[0148] When using a combined early warning system based on rainfall and surface displacement, since both rainfall and surface displacement are divided into 5 levels, there are 125 possible rules when combining them. To facilitate early warning calculations, the early warning levels are coded, with extremely low, low, medium, high, and extremely high levels corresponding to codes 0, 1, 2, 3, and 4, respectively.

[0149] The experiment first conducted a purely data-driven evolution of combined early warning rules. The initial population size was defined as 50, the number of generations as 500, and the mutation probability as 0.001. After calculations involving selection, crossover, and mutation, the optimal rule matrix was obtained. The fitness distribution for each individual in each generation and the optimal fitness distribution for each generation are shown in Figures 18 and 19. The evolved optimal rule matrix is ​​shown in Table 3. The fitness value corresponding to the optimal matrix is ​​0.008. According to the meaning of the fitness function design, the optimal combined early warning rule has an early warning accuracy of approximately 91%, which is quite good. However, the early warning rule matrix showed that when the rainfall warning level or the surface displacement warning level increased, the overall early warning level decreased. This contradicts existing understanding of landslide deformation. Therefore, the results of the purely data- and model-driven combined early warning decision are not entirely reasonable.

[0150] Table 3 Matrix of Optimal Combination Early Warning Rules

[0151]

[0152] To address the issue of inconsistent results from purely data-driven and model-driven calculations, domain knowledge constraints are incorporated into the evolutionary comprehensive early warning rules. This means that expert experience is added as constraints to the search system during the genetic algorithm iteration process. For example, the following constraint rule is added: the warning level must not be lowered when rainfall or displacement increases. These domain knowledge constraints align with our understanding of practical work, improving both the efficiency of iterative calculations and the reasonableness of the comprehensive early warning results.

[0153] Under the same experimental conditions, the optimal rule matrix obtained by incorporating the above constraint rules is shown in Table 4. It can be seen that: (1) When the surface displacement level is low, although the rainfall level continues to increase, the comprehensive warning level does not upgrade. The reason may be that the landslide is not sensitive to rainfall response when it is in the early stage of deformation; (2) When the surface displacement warning level is high, the comprehensive warning level upgrades rapidly with the continuous increase of rainfall level. The reason may be that when the landslide body is in a faster deformation stage, the deformation is more sensitive to the increase of rainfall and the deformation response is more positive; (3) When the rainfall level is extremely low, the comprehensive warning level will upgrade slowly with the continuous increase of surface displacement level. This may be because the rapid deformation of the landslide requires a certain amount of rainfall to trigger it. The correlation between weak rainfall or no rain and rapid landslide deformation is weak. The fitness distribution of each individual in each generation and the optimal fitness distribution of each generation are shown in Table 4. Figure 20 , 21 As shown

[0154] Table 4. Matrix of Early Warning Rules for the Optimal Combination of Domain Knowledge Constraints

[0155]

[0156] The fitness function value corresponding to the optimal integrated early warning rule matrix that integrates domain knowledge is 0.023. According to the definition of the fitness function, the early warning accuracy of the optimal integrated early warning rule is approximately 85%, slightly lower than the integrated early warning result under pure data-driven conditions. This may be because the data sample is limited under pure data-driven conditions, and the model is prone to overfitting. In practical applications, it is necessary to comprehensively consider the model's accuracy and generalization ability.

[0157] Based on the evolved optimal rules, the corresponding comprehensive early warning level, warning score, and confidence level can be automatically obtained for surface displacement and rainfall monitoring data. For example, on August 20, 2020, the daily change in surface displacement was 77 mm, and the cumulative rainfall over 3 days was 99.84 mm. The combined hazard warning score was 92.77, with a confidence level of 62.4% for an extremely high hazard warning and 11.9% for a high hazard warning. (See...) Figure 22As shown, the daily changes in surface displacement on August 21st and 22nd were 66mm and 100mm respectively, indicating that issuing an extremely high comprehensive warning level on August 20th was reasonable. Furthermore, with the continuous increase in monitoring data, more evolutionary sample inputs can be provided, and the optimal comprehensive warning rule can be continuously optimized through a self-learning mechanism.

[0158] To achieve the functions described above, this application discloses a generation apparatus, which includes hardware structures and / or software modules corresponding to each function. Those skilled in the art will readily recognize that the exemplary units and method steps described in conjunction with the embodiments disclosed in this application can be implemented in hardware or a combination of hardware and software programs. Whether a function is implemented in hardware or by software program-driven hardware depends on the specific circumstances.

[0159] The generating apparatus disclosed herein includes:

[0160] The generation module is used to generate an initial risk warning rule base, which is as described above;

[0161] The acquisition module is used to acquire historical monitoring data, which includes displacement and rainfall. The corresponding displacement and rainfall are statistically analyzed on a daily basis, and the daily data is used as a set of sample data.

[0162] The processing module is used to train the sub-initial risk warning rule bases in the initial risk warning rule base constructed by the generation module using the sample data obtained by the acquisition module, and to obtain the daily fitness score (MSE) of each sub-initial risk warning rule base. risk Then, according to the following formula, calculate the final score MSE′ of each sub-initial risk warning rule base in the initial risk warning rule base established by the generation module within the selected historical period. risk ;

[0163]

[0164] In the formula: N represents the historical period, and the unit is days.

[0165] The processing module is also used to select the final score MSE′. risk Sub-initial risk warning rule bases with scores below a preset threshold are used as the final risk warning rule base for risk warning; or each sub-initial risk warning rule base and its final score are used as input, and a genetic algorithm is used to iteratively optimize the warning rules, and finally outputs a sub-initial risk warning rule base with a final score below a preset threshold as the final risk warning rule base for risk warning.

[0166] The acquisition module of the generating device is also used to clean the acquired historical monitoring data according to the aforementioned method before generating sample data.

[0167] The processing module obtains the daily fitness score (MSE) of each sub-initial risk warning rule base using the aforementioned method. risk .

[0168] This disclosure has been described with reference to the foregoing embodiments; however, these embodiments are merely examples for implementing this disclosure. It must be noted that the disclosed embodiments do not limit the scope of this disclosure. On the contrary, any changes and modifications made without departing from the spirit and scope of this disclosure are within the scope of patent protection of this disclosure.

Claims

1. A method for generating landslide risk early warning rules based on fuzzy weights of multi-source data, characterized in that, The method includes the following steps: Step 1: Construct an initial risk warning rule base, which includes several sub-initial risk warning rule bases, each of which contains several risk warning rules; Step 2: Obtain historical monitoring data, which includes displacement and rainfall. The corresponding displacement and rainfall are statistically analyzed on a daily basis, and the daily data is used as a sample data set. Step 3: Train the sub-initial risk warning rule bases in the initial risk warning rule base constructed in Step 1 using the sample data from Step 2, and obtain the daily fitness score of each sub-initial risk warning rule base. This includes the following steps: Step 1: Obtain the displacement for the day and rainfall Based on the constructed membership function, the displacement is... and rainfall The numerical values ​​are converted into corresponding risk levels; Step 2: Matching the initial risk warning rule base using displacement. and rainfall The risk level is matched with the risk warning rules in the initial risk warning rule base. When the risk warning rules in the initial risk warning rule base match the displacement and rainfall risk level with the displacement amount... and rainfall When the risk levels are equal, the risk warning rule is successfully matched. Step 3: Using the constructed membership function, obtain the membership degree corresponding to the displacement risk level, rainfall risk level, and result risk level of each risk warning rule that was successfully matched in Step 2. Step 4: Take the smaller value between the displacement risk level membership degree and the rainfall risk level membership degree obtained in Step 3 as the confidence level value of the risk warning rule in the corresponding sub-initial risk warning rule library. Use this confidence level value to truncate the membership function curves corresponding to the result risks of all successfully matched risk warning rules in the corresponding sub-initial risk warning rule library. Take the part smaller than the confidence level value as the output and superimpose it. Step 5: Defuzzify the output of Step 4 to obtain the specific fuzziness level value of each sub-initial risk warning rule base, which is used as the warning score of each sub-initial risk warning rule base. ; Step 6: Calculate the fitness score of the initial risk warning rule base for the current day using the following formula. ; ； These are the normalized displacement values. The warning score for the day, The lowest historical warning score, when selecting data from the first day, The value is 0; This represents the highest historical warning score when selecting data from the first day. The value is the warning score for the day; The normalized displacement for the next day, This represents the displacement value for the next day. The minimum displacement value among the historical data collected. The maximum displacement value among the historical data obtained; Step 4: Calculate the final score of each sub-initial risk warning rule base in the initial risk warning rule base established in Step 1 within the selected historical period using the following formula. ; ； In the formula: N represents the historical period, in days; Step 5: Select the final score Sub-initial risk warning rule bases with scores less than or equal to a preset threshold are used as the final risk warning rule base for risk warning; or each sub-initial risk warning rule base and its final score are used as input, and a genetic algorithm is used to iteratively optimize the warning rules, and finally output sub-initial risk warning rule bases with final scores less than or equal to a preset threshold are used as the final risk warning rule base for risk warning.

2. The method for generating landslide risk early warning rules based on multi-source data fuzzy weights according to claim 1, characterized in that, Step two involves cleaning the historical monitoring data after acquisition, removing outliers, smoothing spurious data, and then statistically analyzing the corresponding displacement and rainfall on a daily basis.

3. The method for generating landslide risk early warning rules based on multi-source data fuzzy weights according to claim 2, characterized in that, The outliers include those caused by sudden changes in monitoring data due to instrument instability, equipment failure, line failure, and instrument maintenance during landslide detection, as well as outliers within the normal observation range.

4. The method for generating landslide risk early warning rules based on multi-source data fuzzy weights according to claim 3, characterized in that, The outlier removal process within the normal observation range first involves segmenting the landslide displacement data based on the different displacement rates, resulting in a stable landslide stage and a rapid deformation stage. Within each stage, outliers are identified, and monitoring data that are significantly greater than or less than the average value of that stage are removed.

5. The method for generating landslide risk early warning rules based on multi-source data fuzzy weights according to claim 2, characterized in that, The spike data was smoothed using a moving average method.

6. A generating apparatus, characterized in that, The device includes: The generation module is used to generate the initial risk warning rule base; The acquisition module is used to acquire historical monitoring data, which includes displacement and rainfall. The corresponding displacement and rainfall are statistically analyzed on a daily basis, and the daily data is used as a set of sample data. The processing module is used to train the sub-initial risk warning rule bases in the initial risk warning rule base constructed by the generation module using the sample data obtained by the acquisition module, and to obtain the daily fitness score of each sub-initial risk warning rule base. Then, the final score of each sub-initial risk warning rule base in the initial risk warning rule base established by the generation module within the selected historical period is calculated according to the following formula. ; ； In the formula: N represents the historical period, in days; The processing module is also used to select the final score. The initial risk warning rule base with a score less than or equal to a preset threshold is used as the final risk warning rule base for risk warning; or each initial risk warning rule base and its final score are used as input, and the warning rules are continuously optimized by using a genetic algorithm to iterate and finally output the initial risk warning rule base with a final score less than or equal to a preset threshold as the final risk warning rule base for risk warning. The processing module obtains the daily fitness score of each sub-initial risk warning rule base through the following processing. : Step 1: Obtain the displacement for the day and rainfall Based on the constructed membership function, the displacement is... and rainfall The numerical values ​​are converted into corresponding risk levels; Step 2: Matching the initial risk warning rule base using displacement. and rainfall The risk level is matched with the risk warning rules in the initial risk warning rule base. When the risk warning rules in the initial risk warning rule base match the displacement and rainfall risk level with the displacement amount... and rainfall When the risk levels are equal, the risk warning rule is successfully matched. Step 3: Using the constructed membership function, obtain the membership degree corresponding to the displacement risk level, rainfall risk level, and result risk level of each risk warning rule that was successfully matched in Step 2. Step 4: Take the smaller value between the displacement risk level membership degree and the rainfall risk level membership degree obtained in Step 3 as the confidence level value of the risk warning rule in the corresponding sub-initial risk warning rule library. Use this confidence level value to truncate the membership function curves corresponding to the result risks of all successfully matched risk warning rules in the corresponding sub-initial risk warning rule library. Take the part smaller than the confidence level value as the output and superimpose it. Step 5: Defuzzify the output of Step 4 to obtain the specific fuzziness level value of each sub-initial risk warning rule base, which is used as the warning score of each sub-initial risk warning rule base. ; Step 6: Calculate the fitness score of the initial risk warning rule base for the current day using the following formula. ; ； ； ； These are the normalized displacement values. The warning score for the day, The lowest historical warning score, when selecting data from the first day, The value is 0; This represents the highest historical warning score when selecting data from the first day. The value is the warning score for the day; The normalized displacement for the next day, This represents the displacement value for the next day. The minimum displacement value among the historical data collected. This represents the maximum displacement value among the historical data collected.

7. A landslide risk early warning method, characterized in that, This method uses a risk warning rule library generated by the landslide risk warning rule generation method based on multi-source data fuzzy weights as described in any one of claims 1-5 to predict landslide risks.

8. The landslide risk early warning method according to claim 7, characterized in that, After a risk warning is issued for a warning period, the displacement and rainfall of that warning period are used as historical monitoring data to generate a risk warning rule base according to any one of claims 1-5, so as to update the risk warning rule base and ensure the timeliness of the risk warning rule base.

Images