IoT-based landslide pore water pressure prediction method

Abstract

This invention discloses an IoT-based method for predicting pore water pressure in landslides. The method includes sensor deployment, data collection, data processing, pore water pressure prediction, and adaptive early warning. This invention relates to the field of landslide pore water pressure prediction technology, specifically to an IoT-based method for predicting landslide pore water pressure. This solution employs the deployment of multiple sensors to collect various data indicators, and comprehensively performs multimodal feature decomposition and processing. Then, it uses an improved long short-term neural network combining adaptive noise and variational mode decomposition to predict pore water pressure, improving the accuracy and depth of landslide pore water pressure prediction and enhancing the method's usability. An adaptive early warning mechanism is constructed by designing early warnings based on the predicted values ​​of multiple modes and indicators.

Description

Technical Field

[0001] This invention relates to the field of landslide pore water pressure prediction technology, specifically to an IoT-based method for predicting landslide pore water pressure. Background Technology

[0002] The landslide pore water pressure prediction method based on the Internet of Things (IoT) is an advanced geological disaster monitoring and early warning method. It deploys a high-precision sensor network in landslide-prone areas to collect key data such as pore water pressure and moisture content in the soil in real time. This data is transmitted to a central data processing platform via wireless communication technology. Using big data analysis and machine learning algorithms, the probability of landslide occurrence is dynamically assessed and predicted. Once abnormal pore water pressure is detected or reaches a critical value, the system can quickly issue an early warning signal to notify relevant departments and residents to take emergency response measures, thereby effectively reducing the risks and losses caused by landslides and significantly improving the intelligence and scientific level of geological disaster management.

[0003] However, existing methods for predicting landslide pore water pressure suffer from several drawbacks. Traditional methods focus solely on collecting water pressure data or using other environmental data as auxiliary features, neglecting complex and potential factors influencing the prediction process. This leads to insufficient prediction results and poor interpretability. Furthermore, traditional methods suffer from poor data quality due to complex environments and susceptibility to extreme weather, soil erosion, and sensor interference, making subsequent predictions difficult. Finally, existing methods struggle with quantification and implementation of predicted landslide pore water pressure, resulting in limited usability of the predictions. Summary of the Invention

[0004] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a landslide pore water pressure prediction method based on IoT. (1) In the existing landslide pore water pressure prediction methods, there is a problem that the traditional methods only focus on collecting water pressure data itself or only perform auxiliary feature prediction on other environmental data, thus ignoring various complex potential factors affecting the landslide pore water pressure prediction process, resulting in insufficient prediction results and poor interpretability. This solution creatively adopts the deployment of multiple sensors to collect multiple data indicators, and comprehensively performs multimodal feature decomposition and processing. Then, it adopts an improved long short-term neural network combining adaptive noise and variational mode decomposition to predict pore water pressure, thereby improving the prediction accuracy and prediction depth of landslide pore water pressure and improving the usability of the method. In view of the existing landslide pore water pressure prediction methods, Traditional landslide pore water pressure prediction suffers from poor data quality due to complex environments and susceptibility to extreme weather, soil erosion, and sensor interference, making it difficult to support subsequent prediction tasks. This solution creatively employs multi-directional redundant sensor deployment and data integration, and stores the data in the cloud via an IoT platform, improving data availability from multiple angles. Addressing the issue that existing landslide pore water pressure prediction methods struggle with quantification and implementation for subsequent use, resulting in poor usability of prediction results, this solution creatively constructs an adaptive early warning mechanism. By designing early warnings for predicted values ​​of various modes and indicators, it enhances the overall practicality and scalability of the method, providing a theoretical and practical foundation for landslide pore water pressure prediction and deeper IoT applications.

[0005] The technical solution adopted by this invention is as follows: This invention provides a landslide pore water pressure prediction method based on IoT, which includes the following steps:

[0006] Step S1: Sensor deployment;

[0007] Step S2: Data collection;

[0008] Step S3: Data processing;

[0009] Step S4: Pore water pressure prediction;

[0010] Step S5: Adaptive early warning.

[0011] Further, in step S1, the sensor deployment is used to design a sensor layout scheme for landslide-prone areas and to deploy the sensors. Specifically, it involves constructing a sensor cluster and deploying the sensors in the landslide-affected areas according to the soil characteristics and topographic information based on the sensor cluster, and using the sensors for data collection.

[0012] The sensor cluster specifically includes a pore water pressure sensor, a soil moisture sensor, a rainfall sensor, a temperature sensor, a humidity sensor, and a landslide acceleration sensor.

[0013] Furthermore, in step S2, the data collection is used to collect the raw data required for landslide pore water pressure prediction. Specifically, based on the sensor cluster, multiple sensors are redundantly deployed and data is redundantly collected. An integrated algorithm is used to process the sensor data to obtain the raw data for pore water pressure prediction. The real-time raw data for pore water pressure prediction is then sent to the Internet of Things cloud platform for data storage via wireless communication technology.

[0014] The aforementioned multi-sensor redundancy deployment specifically refers to deploying three identical sensors at the location where the sensors are arranged to perform redundant data acquisition.

[0015] The aforementioned redundant data acquisition specifically refers to the acquisition of repetitive types of sensor data, which is used to reduce the possibility of data inaccuracy or malfunction.

[0016] The ensemble algorithm specifically employs a weighted average algorithm;

[0017] The raw data for predicting pore water pressure specifically includes pore water pressure data, soil moisture data, rainfall data, temperature data, humidity data, and soil displacement data.

[0018] Further, in step S3, the data processing is used to optimize the original data, specifically to optimize the original pore water pressure prediction data to obtain optimized pore water pressure prediction data, which includes the following steps:

[0019] Step S31: Data cleaning, specifically, outlier detection and processing are performed on the sensor data in the original pore water pressure prediction data, and missing and outlier data are supplemented by mean imputation method to obtain a cleaned dataset;

[0020] Step S32: Timestamp synchronization, specifically, performing timestamp synchronization on the cleaned dataset to obtain a synchronized time-series dataset;

[0021] Step S33: Data standardization, specifically, performing Z-score standardization on the synchronous time-series dataset to obtain a standardized sensor dataset;

[0022] Step S34: Data optimization processing, specifically, through data cleaning, timestamp synchronization and data standardization, the original pore water pressure prediction data is optimized to obtain optimized pore water pressure prediction data;

[0023] The optimized pore water pressure prediction data specifically includes standardized pore water pressure data, standardized soil moisture data, standardized rainfall data, standardized temperature data, standardized humidity data, and standardized soil displacement data.

[0024] Further, in step S4, the pore water pressure prediction is used to predict pore water pressure based on optimized data. Specifically, based on the optimized pore water pressure prediction data, an improved long short-term neural network combining adaptive noise and variational mode decomposition is used to predict pore water pressure, obtaining pore water pressure prediction reference data. This specifically includes the following steps:

[0025] Step S41: Sensing data decomposition, specifically, using the complete set empirical mode decomposition method to decompose the pore water pressure prediction optimization data to obtain an intrinsic mode function set;

[0026] Step S42: Variational mode decomposition, specifically, applying the variational mode decomposition method, constructing an optimization objective function, performing variational mode decomposition, and obtaining high-frequency decomposed mode data;

[0027] Step S43: Decomposition mode feature extraction, specifically, using statistical feature extraction methods to extract multimodal features from the intrinsic mode function set and the high-frequency decomposition mode data to obtain a pore water pressure multimodal feature dataset;

[0028] Step S44: Construct an improved Long Short-Term Memory (LSTM) network. Specifically, based on the pore water pressure multimodal feature dataset, an improved LSM neural network is constructed by connecting the modal feature datasets in the pore water pressure multimodal feature dataset. This network is used for pore water pressure multimodal prediction, and the prediction results of each modality are merged to obtain the model prediction output. The calculation formula is as follows:

[0029] ;

[0030] In the formula, f t This is the unit identifier for a feature-improved Long Short-Term Memory (LSTM) neural network, used to represent input, output, and forget units. t is the time index, sig(·) is the sigmoid activation function, and W... f It is the weight of the long short-term memory network unit, h t-1 It is the output hidden state at the previous time t-1, (F1:F n F1 is the modal feature connection data, used to represent the connection of modal feature datasets in the pore water pressure multimodal feature dataset. F1 is the modal feature data of the first mode. M This is the modal feature data of the Mth mode, where M specifically takes the value 4, representing the total number of feature modes in the pore water pressure multimodal feature dataset, bf It is a cell bias term;

[0031] W m This is the classification prediction result for the m-th modality, where m is the modality index. It is a classification weight vector. It is the importance weight of the m-th mode;

[0032] P(t) is the final predicted output of the multi-modal model of pore water pressure, h t It is the output hidden state of the feature-improved long short-term memory neural network;

[0033] Step S45: Construct an adaptive optimization algorithm, specifically by adding adaptive weights to construct an eagle hunting adaptive weight search algorithm, update the model hyperparameter combination, and obtain the optimal model parameter combination;

[0034] The calculation formula for constructing the adaptive weighted search algorithm for eagle hunting is as follows:

[0035] ;

[0036] In the formula, W new W is the adaptive weight value of the updated eagle hunting search algorithm. old These are the adaptive weight values ​​of the eagle hunting search algorithm before the update, and rand is a random factor with a value range of [0,1]. It is an adaptive update of factor weights, X target It is the optimal solution in the current iteration search, used to represent the current optimal combination of hyperparameters for the model. T is the current iteration number, and T is the maximum iteration number.

[0037] Step S46: Pore water pressure prediction model training, specifically, involves training the model using the sensor data decomposition, variational mode decomposition, decomposition mode feature extraction, improved long short-term memory network, and adaptive optimization algorithm to obtain the pore water pressure prediction model. WP ;

[0038] Step S47: Pore water pressure prediction, specifically using the pore water pressure prediction model. WP Based on the pore water pressure prediction optimization data, pore water pressure is predicted to obtain pore water pressure prediction reference data.

[0039] The pore water pressure prediction reference data specifically includes predicted pore water pressure values, predicted soil moisture values, predicted rainfall values, and predicted soil displacement values.

[0040] Further, in step S5, the adaptive early warning is used to construct an adaptive early warning mechanism based on the predicted pore water pressure result. Specifically, it is to construct an adaptive early warning mechanism based on the pore water pressure prediction optimization data and the pore water pressure prediction reference data, obtain adaptive early warning reference data, and perform adaptive early warning.

[0041] The adaptive early warning mechanism specifically involves constructing a four-dimensional adaptive early warning threshold group, and based on the four-dimensional adaptive early warning threshold group, reading the predicted values ​​in the pore water pressure prediction reference data in real time and comparing the data. When the early warning conditions are met, an adaptive early warning is issued.

[0042] The four-dimensional adaptive early warning threshold group specifically includes an early warning threshold group and an early warning category group;

[0043] The warning threshold group specifically includes pore water pressure threshold, soil moisture threshold, rainfall threshold and soil displacement threshold;

[0044] The aforementioned warning category groups specifically include normal warnings, general warnings, and emergency warnings;

[0045] The adaptive early warning mechanism is constructed by adopting an adaptive fuzzy logic control method, including the following steps:

[0046] Step S51: Input initialization, specifically, based on the four-dimensional adaptive early warning threshold group, the input variables are initialized to obtain the output early warning detection data;

[0047] Step S52: Design fuzzy membership functions, specifically designing fuzzy membership functions for each early warning threshold group in the four-dimensional adaptive early warning threshold group;

[0048] Step S53: Construct a fuzzy rule base, specifically by constructing a fuzzy rule base based on the pore water pressure prediction reference data and the historical records of manual early warnings, which is used to determine the comprehensive early warning type;

[0049] Step S54: Fuzzy reasoning, specifically, using Mamdani reasoning to construct a fuzzy inference engine and obtain fuzzy inference output;

[0050] Step S55: Defuzzification transformation, specifically, using the centroid method to defuzzify the fuzzy inference output into a warning value, thereby obtaining the warning output;

[0051] Step S56: Real-time adaptive early warning, specifically, real-time prediction and early warning are performed based on the early warning output, and the fuzzy membership function is dynamically adjusted during the early warning process to perform adaptive early warning updates and maintenance.

[0052] The beneficial effects achieved by the present invention using the above solution are as follows:

[0053] (1) In view of the existing landslide pore water pressure prediction methods, the traditional methods only focus on collecting water pressure data itself or only make auxiliary features for other environmental data to predict landslide pore water pressure, thus ignoring the various complex potential factors affecting the landslide pore water pressure prediction process, resulting in insufficient prediction results and poor interpretability. This solution creatively adopts the method of deploying multiple sensors to collect multiple data indicators, and comprehensively performs multimodal feature decomposition and processing. Then, it adopts an improved long short-term neural network that combines adaptive noise and variational mode decomposition to predict pore water pressure, thereby improving the prediction accuracy and prediction depth of landslide pore water pressure and improving the usability of the method.

[0054] (2) In view of the technical problem that the existing landslide pore water pressure prediction methods have poor data quality due to the complex environment and susceptibility to extreme weather, soil corrosion and sensor interference, which makes it difficult to support subsequent prediction tasks, this solution creatively adopts multi-directional redundant sensor deployment and data integration, and uses the Internet of Things platform for cloud storage, which improves the availability of data from multiple perspectives.

[0055] (3) In view of the technical problem that the subsequent use of landslide pore water pressure prediction methods is difficult to quantify and implement, resulting in poor usability of landslide pore water pressure prediction results, this solution creatively constructs an adaptive early warning mechanism. By designing early warning for the predicted values ​​of various modes and indicators, the overall practicality and scalability of the method are improved, and a theoretical and practical foundation is provided for landslide pore water pressure prediction and deeper Internet of Things applications. Attached Figure Description

[0056] Figure 1 A schematic flowchart of an IoT-based landslide pore water pressure prediction method provided by the present invention;

[0057] Figure 2 A schematic diagram of an IoT-based landslide pore water pressure prediction system provided by the present invention;

[0058] Figure 3 This is a flowchart illustrating the data processing in step S3.

[0059] Figure 4 This is a schematic diagram of the process for predicting pore water pressure in step S4.

[0060] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0061] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0062] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0063] Example 1, see Figure 1 This invention provides an IoT-based method for predicting pore water pressure in landslides, which includes the following steps:

[0064] Step S1: Sensor deployment;

[0065] Step S2: Data collection;

[0066] Step S3: Data processing;

[0067] Step S4: Pore water pressure prediction;

[0068] Step S5: Adaptive early warning.

[0069] By performing the above operations, this solution addresses the technical problem in existing landslide pore water pressure prediction methods. Traditional methods focus solely on collecting water pressure data or using other environmental data as auxiliary features for prediction, thus ignoring various complex potential factors affecting the prediction process and resulting in insufficient prediction results and poor interpretability. This solution creatively employs multiple sensors to collect various data indicators, comprehensively performs multimodal feature decomposition and processing, and then uses an improved long short-term neural network combining adaptive noise and variational mode decomposition to predict pore water pressure. This improves the accuracy and depth of landslide pore water pressure prediction and enhances the usability of the method.

[0070] Example 2, see Figure 1 and Figure 2In step S1, the sensor deployment is used to design a sensor layout scheme for landslide-prone areas and to deploy the sensors. Specifically, it involves building a sensor cluster and deploying the sensors in the landslide-prone areas according to the soil characteristics and terrain information based on the sensor cluster, and using them for data collection.

[0071] The sensor cluster specifically includes a pore water pressure sensor, a soil moisture sensor, a rainfall sensor, a temperature sensor, a humidity sensor, and a landslide acceleration sensor.

[0072] The pore water pressure sensor is used to monitor pore water pressure, water saturation and soil stability in the soil in real time.

[0073] The soil moisture sensor is used to measure soil moisture content to help determine the water accumulation situation;

[0074] The rainfall sensor is used to measure rainfall.

[0075] The temperature sensor is used to monitor ambient temperature and help characterize changes in pore water pressure.

[0076] The humidity sensor is used to monitor ambient humidity and help characterize changes in pore water pressure.

[0077] The landslide acceleration sensor is used to monitor the displacement and vibration of soil or rock and to provide landslide early warning information support.

[0078] Example 3, see Figure 1 , Figure 2 and Figure 3 This embodiment is based on the above embodiment. In step S2, the data collection is used to collect the raw data required for landslide pore water pressure prediction. Specifically, based on the sensor cluster, multiple sensors are redundantly deployed and data is redundantly collected. An integrated algorithm is used to process the sensor data to obtain the raw data for pore water pressure prediction. The real-time raw data for pore water pressure prediction is then sent to the Internet of Things cloud platform for data storage via wireless communication technology.

[0079] The aforementioned multi-sensor redundancy deployment specifically refers to deploying three identical sensors at the location where the sensors are arranged to perform redundant data acquisition.

[0080] The aforementioned redundant data acquisition specifically refers to the acquisition of repetitive types of sensor data, which is used to reduce the possibility of data inaccuracy or malfunction.

[0081] The ensemble algorithm specifically employs a weighted average algorithm;

[0082] The raw data for predicting pore water pressure specifically includes pore water pressure data, soil moisture data, rainfall data, temperature data, humidity data, and soil displacement data.

[0083] By performing the above operations, this solution addresses the technical problem in existing landslide pore water pressure prediction methods, which suffer from poor data quality due to complex environments and susceptibility to extreme weather, soil erosion, and sensor interference, making it difficult to support subsequent prediction tasks. This solution creatively employs multi-directional redundant sensor deployment and data integration, and stores the data in the cloud through an IoT platform, thereby improving data availability from multiple perspectives.

[0084] Example 4, see Figure 1 , Figure 2 and Figure 4 This embodiment is based on the above embodiment. In step S3, the data processing is used to optimize the original data, specifically to optimize the original pore water pressure prediction data to obtain optimized pore water pressure prediction data. This includes the following steps:

[0085] Step S31: Data cleaning, specifically, outlier detection and processing are performed on the sensor data in the original pore water pressure prediction data, and missing and outlier data are supplemented by mean imputation method to obtain a cleaned dataset;

[0086] Step S32: Timestamp synchronization, specifically, performing timestamp synchronization on the cleaned dataset to obtain a synchronized time-series dataset;

[0087] Step S33: Data standardization, specifically, performing Z-score standardization on the synchronous time-series dataset to obtain a standardized sensor dataset;

[0088] Step S34: Data optimization processing, specifically, through data cleaning, timestamp synchronization and data standardization, the original pore water pressure prediction data is optimized to obtain optimized pore water pressure prediction data;

[0089] The optimized pore water pressure prediction data specifically includes standardized pore water pressure data, standardized soil moisture data, standardized rainfall data, standardized temperature data, standardized humidity data, and standardized soil displacement data.

[0090] Example 5, see Figure 1 and Figure 2This embodiment is based on the above embodiment. In step S4, the pore water pressure prediction is used to predict pore water pressure based on optimized data. Specifically, based on the optimized pore water pressure prediction data, an improved long short-term neural network combining adaptive noise and variational mode decomposition is used to predict pore water pressure, obtaining pore water pressure prediction reference data. The specific steps include:

[0091] Step S41: Sensor data decomposition, specifically, using the complete ensemble empirical mode decomposition method to decompose the pore water pressure prediction optimization data to obtain an intrinsic mode function set. The calculation formula is as follows:

[0092] ;

[0093] In the formula, X(t) is the original sensing signal, used to represent the signal data in the pore water pressure prediction optimization data, I is the total number of decomposed modes, i is the decomposed mode index, and IMF is the index of the decomposed mode. i r(t) is the intrinsic mode function corresponding to the i-th decomposed mode, t is the time index, and r(t) is the residual signal after decomposition.

[0094] Step S42: Variational mode decomposition, specifically, applying the variational mode decomposition method, constructing an optimization objective function, performing variational mode decomposition, and obtaining high-frequency decomposed mode data;

[0095] The formula for calculating the optimization objective function is as follows:

[0096] ;

[0097] In the formula, F is the variational mode decomposition objective function, min is the function for finding the minimum value, and u k It is the value of the k-th variational high-frequency mode function, w k X(t) is the center frequency value of the k-th variational high-frequency mode, K is the total number of high-frequency modes obtained from variational mode decomposition, k is the high-frequency mode index, and X(t) is the original sensing signal. It is the penalty factor, and j is the imaginary unit;

[0098] Step S43: Decomposition mode feature extraction, specifically, using statistical feature extraction methods to extract multimodal features from the intrinsic mode function set and the high-frequency decomposition mode data to obtain a pore water pressure multimodal feature dataset;

[0099] The pore water pressure multimodal feature dataset specifically includes a pore water pressure modal feature dataset, a soil moisture modal feature dataset, a rainfall modal feature dataset, and a soil displacement modal feature dataset;

[0100] The statistical feature extraction specifically extracts the mean feature, standard deviation feature, extreme value feature, skewness feature, and kurtosis feature;

[0101] Step S44: Construct an improved Long Short-Term Memory (LSTM) network. Specifically, based on the pore water pressure multimodal feature dataset, an improved LSM neural network is constructed by connecting the modal feature datasets in the pore water pressure multimodal feature dataset. This network is used for pore water pressure multimodal prediction, and the prediction results of each modality are merged to obtain the model prediction output. The calculation formula is as follows:

[0102] ;

[0103] In the formula, f t This is the unit identifier for a feature-improved Long Short-Term Memory (LSTM) neural network, used to represent input, output, and forget units. t is the time index, sig(·) is the sigmoid activation function, and W... f It is the weight of the long short-term memory network unit, h t-1 It is the output hidden state at the previous time t-1, (F1:F n F1 is the modal feature connection data, used to represent the connection of modal feature datasets in the pore water pressure multimodal feature dataset. F1 is the modal feature data of the first mode. M This is the modal feature data of the Mth mode, where M specifically takes the value 4, representing the total number of feature modes in the pore water pressure multimodal feature dataset, b f It is a cell bias term;

[0104] W m This is the classification prediction result for the m-th modality, where m is the modality index. It is a classification weight vector. It is the importance weight of the m-th mode;

[0105] P(t) is the final predicted output of the multi-modal model of pore water pressure, h t It is the output hidden state of the feature-improved long short-term memory neural network;

[0106] Step S45: Construct an adaptive optimization algorithm, specifically by adding adaptive weights to construct an eagle hunting adaptive weight search algorithm, update the model hyperparameter combination, and obtain the optimal model parameter combination;

[0107] The calculation formula for constructing the adaptive weighted search algorithm for eagle hunting is as follows:

[0108] ;

[0109] In the formula, W new W is the adaptive weight value of the updated eagle hunting search algorithm.old These are the adaptive weight values ​​of the eagle hunting search algorithm before the update, and rand is a random factor with a value range of [0,1]. It is an adaptive update of factor weights, X target It is the optimal solution in the current iteration search, used to represent the current optimal combination of hyperparameters for the model. T is the current iteration number, and T is the maximum iteration number.

[0110] Step S46: Pore water pressure prediction model training, specifically, involves training the model using the sensor data decomposition, variational mode decomposition, decomposition mode feature extraction, improved long short-term memory network, and adaptive optimization algorithm to obtain the pore water pressure prediction model. WP ;

[0111] Step S47: Pore water pressure prediction, specifically using the pore water pressure prediction model. WP Based on the pore water pressure prediction optimization data, pore water pressure is predicted to obtain pore water pressure prediction reference data.

[0112] The pore water pressure prediction reference data specifically includes predicted pore water pressure values, predicted soil moisture values, predicted rainfall values, and predicted soil displacement values.

[0113] By performing the above operations, this solution addresses the technical problem in existing landslide pore water pressure prediction methods. Traditional methods focus solely on collecting water pressure data or using other environmental data as auxiliary features for prediction, thus ignoring various complex potential factors affecting the prediction process and resulting in insufficient prediction results and poor interpretability. This solution creatively employs multiple sensors to collect various data indicators, comprehensively performs multimodal feature decomposition and processing, and then uses an improved long short-term neural network combining adaptive noise and variational mode decomposition to predict pore water pressure. This improves the accuracy and depth of landslide pore water pressure prediction and enhances the usability of the method.

[0114] Example 7, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. In step S5, the adaptive early warning is used to construct an adaptive early warning mechanism based on the predicted pore water pressure result. Specifically, it is to construct an adaptive early warning mechanism based on the pore water pressure prediction optimization data and the pore water pressure prediction reference data, obtain adaptive early warning reference data, and perform adaptive early warning.

[0115] The adaptive early warning mechanism specifically involves constructing a four-dimensional adaptive early warning threshold group, and based on the four-dimensional adaptive early warning threshold group, reading the predicted values ​​in the pore water pressure prediction reference data in real time and comparing the data. When the early warning conditions are met, an adaptive early warning is issued.

[0116] The four-dimensional adaptive early warning threshold group specifically includes an early warning threshold group and an early warning category group;

[0117] The warning threshold group specifically includes pore water pressure threshold, soil moisture threshold, rainfall threshold and soil displacement threshold;

[0118] The aforementioned warning category groups specifically include normal warnings, general warnings, and emergency warnings;

[0119] The adaptive early warning mechanism is constructed by adopting an adaptive fuzzy logic control method, including the following steps:

[0120] Step S51: Input initialization, specifically, based on the four-dimensional adaptive early warning threshold group, the input variables are initialized to obtain the output early warning detection data;

[0121] Step S52: Design fuzzy membership functions. Specifically, design fuzzy membership functions for each warning threshold group in the four-dimensional adaptive warning threshold group. The calculation formula is as follows:

[0122] ;

[0123] In the formula, f N (·) is the normal early warning fuzzy membership function, f W (·) is the fuzzy membership function for warning and alerting, f D (·) represents the fuzzy membership function for emergency warnings. A value of 0 indicates no warning is issued, while a value of 1 indicates a warning is issued. x represents the output warning detection data. This is the normal lower threshold for early warning. This is the normal upper limit threshold for early warning. It is the lower threshold for warnings and alerts. It is the upper limit threshold for warnings and alerts. It is the lower threshold for emergency warning. This is the upper limit threshold for emergency warnings;

[0124] Step S53: Construct a fuzzy rule base, specifically by constructing a fuzzy rule base based on the pore water pressure prediction reference data and the historical records of manual early warnings, which is used to determine the comprehensive early warning type;

[0125] Step S54: Fuzzy reasoning, specifically, using Mamdani reasoning to construct a fuzzy inference engine and obtain fuzzy inference output;

[0126] Step S55: Defuzzification transformation, specifically, using the centroid method to defuzzify the fuzzy inference output into a warning value, thereby obtaining the warning output;

[0127] Step S56: Real-time adaptive early warning, specifically, real-time prediction and early warning are performed based on the early warning output, and the fuzzy membership function is dynamically adjusted during the early warning process to perform adaptive early warning updates and maintenance.

[0128] Table 1 shows the specific contents of the four-dimensional adaptive early warning threshold group. As shown in the table, the pore water pressure threshold (unit: kPa) includes normal early warning, warning early warning and critical early warning.

[0129] Soil moisture thresholds include normal warning, alert warning, and critical warning.

[0130] Rainfall thresholds (unit: mm / h) include normal warnings, warnings, and emergency warnings;

[0131] Soil displacement thresholds (unit: mm) include normal warning, warning, and critical warning.

[0132] Table 1. Detailed Contents of the Four-Dimensional Adaptive Early Warning Threshold Group

[0133]

[0134] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0135] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

[0136] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A landslide pore water pressure prediction method based on IoT, characterized in that: The method includes the following steps: Step S1: Sensor deployment, building a sensor cluster, and deploying sensors; Step S2: Data collection. Based on the sensor cluster, multiple sensors are redundantly deployed and data is redundantly collected. An integrated algorithm is used to process the sensor data to obtain the raw data for pore water pressure prediction. The multiple sensor redundancy deployment specifically refers to deploying three sensors of the same type at the sensor locations for redundant data collection. The redundant data collection specifically refers to collecting repetitive types of sensor data to reduce data inaccuracies or malfunctions. The integrated algorithm specifically uses a weighted average algorithm. Step S3: Data processing, specifically, optimizing the original pore water pressure prediction data to obtain optimized pore water pressure prediction data; Step S4: Pore water pressure prediction. Based on the optimized pore water pressure prediction data, an improved long short-term neural network combining adaptive noise and variational mode decomposition is used to predict pore water pressure, obtaining pore water pressure prediction reference data. This specifically includes the following steps: Step S41: Sensing data decomposition, specifically, using the complete set empirical mode decomposition method to decompose the pore water pressure prediction optimization data to obtain an intrinsic mode function set; Step S42: Variational mode decomposition, specifically, applying the variational mode decomposition method, constructing an optimization objective function, performing variational mode decomposition, and obtaining high-frequency decomposed mode data; The formula for calculating the optimization objective function is as follows: ； In the formula, F is the variational mode decomposition objective function, min is the function for finding the minimum value, and u k It is the value of the k-th variational high-frequency mode function, w k X(t) is the center frequency value of the k-th variational high-frequency mode, K is the total number of high-frequency modes obtained from variational mode decomposition, k is the high-frequency mode index, and X(t) is the original sensing signal. It is the penalty factor, and j is the imaginary unit; Step S43: Decomposition mode feature extraction, specifically, using statistical feature extraction methods to extract multimodal features from the intrinsic mode function set and the high-frequency decomposition mode data to obtain a pore water pressure multimodal feature dataset; The pore water pressure multimodal feature dataset specifically includes a pore water pressure modal feature dataset, a soil moisture modal feature dataset, a rainfall modal feature dataset, and a soil displacement modal feature dataset; Step S44: Construct an improved long short-term memory network. Specifically, based on the pore water pressure multimodal feature dataset, construct an improved long short-term memory neural network by connecting the modal feature datasets in the pore water pressure multimodal feature dataset, perform pore water pressure multimodal prediction, and obtain the model prediction output by merging the prediction results of each modality. Step S45: Construct an adaptive optimization algorithm, specifically by adding adaptive weights to construct an eagle hunting adaptive weight search algorithm, update the model hyperparameter combination, and obtain the optimal model parameter combination; Step S46: Pore water pressure prediction model training, specifically, involves training the model using the sensor data decomposition, variational mode decomposition, decomposition mode feature extraction, improved long short-term memory network, and adaptive optimization algorithm to obtain the pore water pressure prediction model. WP ; Step S47: Pore water pressure prediction, specifically using the pore water pressure prediction model. WP Based on the pore water pressure prediction optimization data, pore water pressure is predicted to obtain pore water pressure prediction reference data. The pore water pressure prediction reference data specifically includes pore water pressure prediction values, soil moisture prediction values, rainfall prediction values, and soil displacement prediction values. Step S5: Adaptive early warning. Construct an adaptive early warning mechanism, obtain adaptive early warning reference data, and perform adaptive early warning.

2. The landslide pore water pressure prediction method based on IoT according to claim 1, characterized in that: In step S1, the sensor deployment is used to design a sensor layout scheme for landslide-prone areas and to deploy the sensors. Specifically, it involves building a sensor cluster and deploying the sensors in the landslide-prone areas according to the soil characteristics and topographic information based on the sensor cluster, and using them for data collection. The sensor cluster specifically includes a pore water pressure sensor, a soil moisture sensor, a rainfall sensor, a temperature sensor, a humidity sensor, and a landslide acceleration sensor.

3. The landslide pore water pressure prediction method based on IoT according to claim 2, characterized in that: The data collection is used to collect the raw data required for landslide pore water pressure prediction. Specifically, based on the sensor cluster, multiple sensors are redundantly deployed and data is redundantly collected. An integrated algorithm is used to process the sensor data to obtain the raw data for pore water pressure prediction. The real-time raw data for pore water pressure prediction is then sent to the Internet of Things cloud platform for data storage via wireless communication technology. The raw data for predicting pore water pressure specifically includes pore water pressure data, soil moisture data, rainfall data, temperature data, humidity data, and soil displacement data.

4. The landslide pore water pressure prediction method based on IoT according to claim 3, characterized in that: In step S3, the data processing specifically includes the following steps: Step S31: Data cleaning, specifically, outlier detection and processing are performed on the sensor data in the original pore water pressure prediction data, and missing and outlier data are supplemented by mean imputation method to obtain a cleaned dataset; Step S32: Timestamp synchronization, specifically, performing timestamp synchronization on the cleaned dataset to obtain a synchronized time-series dataset; Step S33: Data standardization, specifically, performing Z-score standardization on the synchronous time-series dataset to obtain a standardized sensor dataset; Step S34: Data optimization processing, specifically, through data cleaning, timestamp synchronization and data standardization, the original pore water pressure prediction data is optimized to obtain optimized pore water pressure prediction data; The optimized pore water pressure prediction data specifically includes standardized pore water pressure data, standardized soil moisture data, standardized rainfall data, standardized temperature data, standardized humidity data, and standardized soil displacement data.

5. The landslide pore water pressure prediction method based on IoT according to claim 4, characterized in that: In step S44, the improved long short-term memory network is constructed using the following formula: ； In the formula, f t This is the unit identifier for a feature-improved Long Short-Term Memory (LSTM) neural network, used to represent input, output, and forget units. t is the time index, sig(·) is the sigmoid activation function, and W... f It is the weight of the long short-term memory network unit, h t-1 It is the output hidden state at the previous time t-1, (F1:F n F1 is the modal feature connection data, used to represent the connection of modal feature datasets in the pore water pressure multimodal feature dataset. F1 is the modal feature data of the first mode. M This is the modal feature data of the Mth mode, where M specifically takes the value 4, representing the total number of feature modes in the pore water pressure multimodal feature dataset, b f It is a cell bias term; W m This is the classification prediction result for the m-th modality, where m is the modality index. It is a classification weight vector. It is the importance weight of the m-th mode; P(t) is the final predicted output of the multi-modal model of pore water pressure, h t It is the output hidden state of the feature-improved long short-term memory neural network; In step S45, the calculation formula for constructing the eagle hunting adaptive weight search algorithm is as follows: ； In the formula, W new W is the adaptive weight value of the updated eagle hunting search algorithm. old These are the adaptive weight values ​​of the eagle hunting search algorithm before the update, and rand is a random factor with a value range of [0,1]. It is an adaptive update of factor weights, X target It is the optimal solution in the current iteration search, used to represent the current optimal combination of hyperparameters for the model. T is the current iteration number, and T is the maximum iteration number.

6. The landslide pore water pressure prediction method based on IoT according to claim 5, characterized in that: In step S5, the adaptive early warning is used to construct an adaptive early warning mechanism based on the predicted pore water pressure results. Specifically, it is to construct an adaptive early warning mechanism based on the pore water pressure prediction optimization data and the pore water pressure prediction reference data, obtain adaptive early warning reference data, and perform adaptive early warning.

7. The landslide pore water pressure prediction method based on IoT according to claim 6, characterized in that: In step S5, the adaptive early warning mechanism specifically involves constructing a four-dimensional adaptive early warning threshold group, and based on the four-dimensional adaptive early warning threshold group, reading the predicted values ​​in the pore water pressure prediction reference data in real time and comparing the data. When the early warning conditions are met, an adaptive early warning is issued. The four-dimensional adaptive early warning threshold group specifically includes an early warning threshold group and an early warning category group; The warning threshold group specifically includes pore water pressure threshold, soil moisture threshold, rainfall threshold and soil displacement threshold; The aforementioned warning category groups specifically include normal warnings, general warnings, and emergency warnings; The adaptive early warning mechanism is constructed by adopting an adaptive fuzzy logic control method, including the following steps: Step S51: Input initialization; Step S52: Design fuzzy membership function; Step S53: Construct fuzzy rule base; Step S54: Fuzzy inference; Step S55: Defuzzification transformation; Step S56: Real-time adaptive early warning.

Images