Intelligent debris flow monitoring and early warning system

Abstract

The invention discloses a debris flow intelligent monitoring and early warning system. The system comprises a sensing layer which is responsible for deploying a multi-parameter sensor matrix, constructing a space-air-ground integrated monitoring system, collecting debris flow formation environment key parameters in real time, performing preliminary data preprocessing, generating an encrypted compressed data packet and sending the encrypted compressed data packet to a transmission layer; the transmission layer is used for receiving the encrypted and compressed data packet through a 5G and Beidou short message dual-channel redundancy transmission technology and forwarding the encrypted and compressed data packet to the analysis layer; and the analysis layer is responsible for constructing a hybrid early warning model driven by a physical mechanism and data, fusing decrypted and decompressed data, dynamically capturing a coupling relationship among rainfall, soil humidity and terrain gradient parameters, learning a multi-factor evolution mode of a historical debris flow event, generating a dynamically adjusted critical threshold curved surface, and performing multi-parameter joint early warning. The early warning information is output to the application layer; and the application layer is responsible for triggering graded early warning and driving multiple departments to cooperatively respond based on the early warning information.

Description

Technical Field

[0001] The present invention relates to the technical field of monitoring and early warning, and in particular to an intelligent monitoring and early warning system for debris flows. Background Art

[0002] Debris flows, a frequent geological disaster, pose a significant threat to human society due to their sudden and destructive nature. These exceptionally large torrents of solid matter can surge rapidly down slopes or valleys in a short period of time, causing catastrophic damage to downstream residential areas, transportation networks, agricultural land, and the ecological environment. To address the severe challenges posed by debris flows to the safety of life and property, a debris flow monitoring and early warning technology system has emerged.

[0003] However, traditional intelligent debris flow monitoring and early warning systems suffer from incomplete data coverage. Debris flow formation involves the dynamic coupling of multiple factors, including rainfall, terrain slope, soil moisture content, and vegetation cover. Traditional monitoring systems overly rely on rainfall as a single indicator to trigger early warnings. This technical limitation makes it difficult to capture the critical state of disasters under the synergistic effects of multiple parameters. For example, in the summer of 202X, a mountainous area experienced moderate rainfall, but rainfall did not reach the threshold preset by traditional systems. However, field monitoring revealed that the previous period of persistent drought had caused soil cracks to develop, the steep terrain exacerbated gravity, and recent wildfires had caused a sharp drop in vegetation cover, reducing soil and water conservation capacity. The deterioration of these key parameters formed a compound amplification mechanism, ultimately triggering large-scale debris flows under normal rainfall conditions, resulting in significant casualties and infrastructure damage. Therefore, a new intelligent debris flow monitoring and early warning system was proposed. Summary of the Invention

[0004] The purpose of the present invention is to solve the shortcomings of the existing technology and to propose an intelligent monitoring and early warning system for debris flow.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] An intelligent debris flow monitoring and early warning system, comprising:

[0007] The perception layer is responsible for building an integrated air-space-ground monitoring system by deploying a multi-parameter sensor matrix. It collects key environmental parameters related to debris flow formation, such as rainfall, terrain slope, soil moisture, and vegetation cover, in real time. It also performs preliminary data preprocessing (such as filtering and normalization). It then uses lightweight compression algorithms (such as LZ4) to reduce the data volume and encrypts the data using the AES-128 encryption algorithm. The encrypted and compressed data packets are then sent to the transport layer.

[0008] Transport layer: Responsible for receiving encrypted and compressed data packets uploaded by the perception layer through 5G and Beidou short message dual-channel redundant transmission technology, ensuring data security while reducing transmission delays, and forwarding encrypted and compressed data packets to the analysis layer;

[0009] The analysis layer is responsible for decrypting and decompressing data packets forwarded by the transport layer. It then constructs a hybrid early warning model driven by physical mechanisms and data to integrate the decrypted and decompressed data in real time. It dynamically captures the coupling relationship between parameters such as rainfall, soil moisture, and terrain slope. It also uses random forests and LSTM networks to learn the multi-factor evolution patterns of historical debris flow events, generate dynamically adjusted critical threshold surfaces, perform multi-parameter joint early warnings, and output warning information to the application layer.

[0010] Application layer: Responsible for building a closed loop of monitoring, early warning, and response. Based on the early warning information output by the analysis layer, it triggers hierarchical early warnings and drives coordinated responses from multiple departments. At the same time, it pushes customized risk avoidance instructions to the public through a multimodal early warning information release platform.

[0011] The above technical solution further includes:

[0012] Furthermore, the construction of an integrated air-ground-space monitoring system by deploying a multi-parameter sensor matrix includes the following steps:

[0013] Deploy the ground sensor matrix:

[0014] Deploy multi-parameter sensor nodes within the monitoring area. Each node integrates rainfall, terrain slope, soil moisture, and vegetation cover sensors. The sensor nodes are arranged in a grid pattern to fully cover areas prone to debris flows.

[0015] Integration of aerospace monitoring means:

[0016] Access satellite remote sensing data to obtain large-scale, high-resolution surface information, and use drones for regular or event-triggered aerial photography to capture local details.

[0017] Fusion and collaborative data:

[0018] The ground sensor data, satellite remote sensing data and drone aerial photography data are aligned in time and space, and through the data fusion algorithm, the multi-source data of air, space and land are complemented and coordinated to build a three-dimensional monitoring system.

[0019] Furthermore, the preliminary data preprocessing comprises the following steps:

[0020] Data cleaning:

[0021] Based on physical rationality judgment (e.g., parameters such as rainfall and terrain slope should be within a reasonable range), outlier data points that significantly deviate from normal values ​​are removed; missing or invalid data are filled in using linear interpolation or based on the mean of historical data to make the data continuous;

[0022] Data filtering:

[0023] Apply low-pass filtering algorithm and moving average filtering to eliminate high-frequency noise interference:

[0024] Where y(n) represents the output after filtering, x(n) represents the input signal, and N represents the filter window size;

[0025] Data normalization:

[0026] Scaling the data to a uniform range (e.g., between 0 and 1) eliminates dimension effects and speeds up subsequent processing:

[0027]

[0028] Among them, x represents the original data, x min and x max Represents the minimum and maximum values ​​of the data respectively.

[0029] Furthermore, reducing the amount of data by using a lightweight compression algorithm comprises the following steps:

[0030] Data chunking:

[0031] Divide the data stream to be compressed into fixed-size data blocks (e.g., 128KB) to facilitate parallel processing;

[0032] Dictionary encoding:

[0033] Using sliding window technology, a dynamic dictionary is constructed to record recently appeared data fragments, and repeated data fragments are replaced with dictionary indexes to reduce data redundancy.

[0034] Match Replacement:

[0035] Scan the data block, looking for the longest data segment that matches the dictionary; replace the matching segment with a tuple of "(match length, backreference distance)":

[0036] Token = (Length, Offset), where Length represents the matching length and Offset represents the reverse reference distance;

[0037] Entropy coding:

[0038] Perform Huffman coding on the generated tuples and unmatched data fragments to further compress the data volume:

[0039] Where L(x) represents the encoding length of symbol x, and P(x) represents the probability of occurrence of symbol x.

[0040] Furthermore, the data is encrypted using the AES-128 encryption algorithm, comprising the following steps:

[0041] Based on the initial key (K0), a key expansion algorithm is used to generate multiple rounds of sub-keys (K1, K2, ..., K10). The initial data block (plaintext) is XORed with the first round sub-key (K0):

[0042] Among them, X1 is the first round input, P is the plaintext;

[0043] Then, an S-box (Substitution Box) is applied to each byte for nonlinear substitution to enhance the obfuscation of the encryption. The state matrix is ​​circularly shifted left, with the i-th row shifted left by i bytes for diffusion. Matrix multiplication is performed on each column of the state matrix, and a fixed matrix (M) is used for obfuscation:

[0044]

[0045] Among them, Y i represents the output after column confusion, X i Represents the input before column confusion;

[0046] Repeat the byte substitution, row shift, column confusion, and round key addition operations. In the last round, the column confusion operation is omitted and the round key addition is performed directly to generate the final ciphertext (C):

[0047] Furthermore, the method of receiving the encrypted compressed data packets uploaded by the perception layer through the dual-channel redundant transmission technology of 5G and Beidou short messages, thereby reducing transmission delay while ensuring data security, includes the following steps:

[0048] Dual-channel parallel reception:

[0049] Deploy 5G communication modules and Beidou short message receiving modules, and simultaneously monitor the encrypted compressed data packets sent by the perception layer; the 5G channel uses the TCP / IP protocol, and the Beidou short message channel uses a custom binary protocol;

[0050] Verify and deduplicate data packets:

[0051] Checksum verification of received data packets:

[0052] Among them, D i is the i-th byte of the data packet, and n is the length of the data packet;

[0053] Deduplication is performed by sequence number, retaining the first arriving data packet and discarding duplicate packets;

[0054] Dynamic priority scheduling:

[0055] Dynamically assign transmission priority based on data packet type (such as real-time monitoring data, control instructions).

[0056] Using the Weighted Fair Queuing (WFQ) algorithm:

[0057] Among them, W i is the weight of the i-th category data, P i is the priority of the i-th category data, B is the total bandwidth, and N is the total number of data categories;

[0058] Security enhancements:

[0059] The TLS1.3 protocol is implemented at the transport layer, and the 5G channel is encrypted twice with a key length of 256 bits. The Beidou short message channel is encrypted using the national secret SM4 algorithm.

[0060] Latency optimization strategy:

[0061] The pre-scheduling mechanism is enabled for the 5G channel to reduce signaling interaction time, and a short frame format is adopted for the Beidou short message channel to shorten the single transmission time.

[0062] Furthermore, the decrypting and decompressing the data packet forwarded by the transport layer includes the following steps:

[0063] Receive and verify data packets:

[0064] The receiving transport layer forwards encrypted compressed data packets through 5G and Beidou short message dual channels.

[0065] Perform integrity check on the data packet using the CRC-32 check algorithm:

[0066] CRC=(D0·x 32 +D1·x 31 +…+D n-1 ·x 32-n )modG(x), where D i is the i-th byte of the data packet, G(x) is the generating polynomial (such as G(x) = x 32 +x 26 +x 23 +…+x+1);

[0067] AES-128 decryption:

[0068] Use the same key (K) as used for encryption to perform AES-128 decryption, including key expansion, round key addition, inverse byte substitution, inverse row shift, and inverse column confusion operations. The round key addition formula is:

[0069] Among them, X i is the input of round i, K i is the key of round i;

[0070] LZ4 decompression:

[0071] The decrypted data packet is decompressed with LZ4 to restore the data format before compression. The decompression process is based on dictionary matching, replacing the "(match length, backreference distance)" tuple with the original data fragment:

[0072] Data=Decode(Token), where Token is the tuple generated during compression and Decode is the decoding function.

[0073] Furthermore, the construction of a hybrid early warning model driven by physical mechanisms and data, which integrates decrypted and decompressed data in real time and dynamically captures the coupling relationship between parameters such as rainfall, soil moisture, and terrain slope, includes the following steps:

[0074] Constructing a physical mechanism model:

[0075] Based on the theory of debris flow formation, the rainfall intensity and rainfall duration (ID) threshold curves are constructed:

[0076] I=αD β , where I is the rainfall intensity, D is the rainfall duration, and α and β are model parameters. The density of the water-soil mixture is determined by numerical simulation (such as the infinite slope model and the one-dimensional Richard water motion equation). A fluid-solid coupling model is introduced to describe the interaction between debris flow and slope, including the displacement coordination condition and the traction equation:

[0077] u s =u f , t s =t f , where u s and u f denote the displacement fields of solid and fluid respectively, t s and t f represent the stress fields of solid and fluid respectively;

[0078] Building data-driven models:

[0079] Using random forest or LSTM networks, with historical debris flow event data (including rainfall, soil moisture, terrain slope and other parameters) as input, a multi-factor evolution pattern recognition model is trained. Model hyperparameters, such as the number of trees in the random forest and the number of hidden layer nodes in the LSTM, are optimized through grid search:

[0080] Hyperparameters = argmax θ Accuracy(θ), where θ represents the hyperparameter set and Accuracy represents the accuracy of the model on the validation set;

[0081] Multi-source data fusion:

[0082] The decrypted and decompressed data is preprocessed, including geometric calibration, coordinate registration, and format conversion. A feature-level fusion method is used to extract feature information of rainfall, soil moisture, and terrain slope parameters: F = Concat(Fr, Fs, Ft), where Fr, Fs, and Ft are the feature vectors of remote sensing data, ground sensor data, and terrain data, respectively, and Concat is a vector concatenation operation.

[0083] Analyze parameter coupling relationship:

[0084] Construct a multi-factor coupling model to analyze the interaction of parameters such as rainfall, soil moisture, and terrain slope:

[0085] λ = f(R, S, T), where λ is the probability of debris flow, R is rainfall, S is soil moisture, T is terrain slope, and f is a coupling function. The risk of debris flow is determined by setting critical values ​​for fluid bulk density and flow velocity and combining them with real-time monitoring data:

[0086]

[0087] Where γ and v represent the fluid density and flow rate, respectively, and γthreshold and vthreshold represent the critical values, respectively.

[0088] Furthermore, the random forest or LSTM network is used to learn the multi-factor evolution pattern of historical debris flow events, generate a dynamically adjusted critical threshold surface, and perform multi-parameter joint early warning, including the following steps:

[0089] Prepare historical data:

[0090] Collect historical debris flow event data, including time series data on rainfall, soil moisture, terrain slope parameters, and information on the time, location, and scale of debris flows; and perform data preprocessing including data cleaning, normalization, and feature extraction.

[0091] Training the model:

[0092] Random Forest Model:

[0093] Construct multiple decision trees, each trained on a randomly sampled data subset and feature subset. The prediction results of each decision tree are combined through a voting mechanism to obtain the probability of debris flow occurrence.

[0094] LSTM network model:

[0095] Construct a neural network structure consisting of an input layer, an LSTM layer, a fully connected layer, and an output layer. Use historical time series data to train the LSTM network and learn how rainfall, soil moisture, and terrain slope parameters change over time. Use the backpropagation algorithm to optimize network weights and minimize prediction errors.

[0096] Generate a critical threshold surface:

[0097] Based on the trained random forest or LSTM network model, the contribution of each parameter to the occurrence of debris flow is analyzed. Multiple parameter combinations (such as rainfall and soil moisture, rainfall and terrain slope, etc.) are set and the probability of debris flow occurrence is calculated for each combination. Based on the probability values, a dynamically adjusted critical threshold surface is generated:

[0098] P(debris flow occurrence) = f(R, S, T)

[0099] Where P(debris flow occurrence) is the probability of debris flow occurrence, R is rainfall, S is soil moisture, T is terrain slope, and f is the functional relationship learned by random forest or LSTM network;

[0100] Critical threshold surface:

[0101] R × S × T = Threshold, where Threshold is learned from historical data and is dynamically adjusted as new data is added.

[0102] Multi-parameter joint early warning:

[0103] Receive decrypted and decompressed data in real time, extract the current values ​​of rainfall, soil moisture, and terrain slope parameters; substitute the current parameter values ​​into the critical threshold surface equation to calculate the probability of debris flow; judge the risk level of debris flow based on the probability value, and generate corresponding warning information, and output the warning information to the application layer so that relevant departments and personnel can take timely response measures.

[0104] Furthermore, the debris flow formation potential index and warning information output by the analysis layer automatically triggers a graded warning and drives a coordinated response from multiple departments, including the following steps:

[0105] Assessment of debris flow formation potential index:

[0106] The analysis layer calculates the debris flow formation potential index (L) through a hybrid warning model, and the potential index integrates the coupling relationships of multiple parameters such as rainfall (R), soil moisture (S), and terrain slope (T):

[0107] L = w 1 ·R + w 2 ·S + w 3 ·T + w 4 ·(R × S) + w 5 ·(R × T) + w 6 ·(S × T)

[0108] Among them, wi is the weight coefficient of each parameter and its coupling term, which is determined through random forest feature importance analysis;

[0109] Trigger hierarchical warning:

[0110] Set four-level warning thresholds (L1, L2, L3, L4), corresponding to blue, yellow, orange, and red warning levels: when L < L1, it is blue; when L1 ≤ L < L2, it is yellow; when L2 ≤ L < L3, it is orange; when L ≥ L3, it is red; when the potential index (L) exceeds the preset threshold, the corresponding warning signal is automatically triggered;

[0111] Multi-department collaborative response drive:

[0112] Based on the warning level, response measures are automatically matched through a preset emergency plan, and warning information and response instructions are sent to the emergency management, natural resources, transportation, and meteorological departments through a unified communication interface. Each department executes the corresponding tasks according to the received instructions, such as the emergency management department starting the emergency plan, the natural resources department strengthening geological monitoring, and the transportation department implementing traffic control.

[0113] The present invention has the following beneficial effects:

[0114] In the present invention, by deploying a multi-parameter sensor matrix, an integrated air-space-ground monitoring system is constructed to collect key environmental parameters of debris flow formation, such as rainfall, terrain slope, soil moisture, and vegetation coverage in real time; this helps to comprehensively capture the multi-factor precursors of debris flow formation, rather than relying on a single parameter; a hybrid early warning model driven by physical mechanism and data is constructed to integrate the decrypted and decompressed data in real time, and dynamically capture the coupling relationship between parameters such as rainfall, soil moisture, and terrain slope; through the combination of physical mechanism models and data-driven models, the system can more accurately analyze the interaction between multiple factors; random forest or LSTM networks are used to learn the multi-factor evolution pattern of historical debris flow events , generate a dynamically adjusted critical threshold surface, and conduct multi-parameter joint early warning; use machine learning algorithms to capture multi-factor evolution patterns to achieve more accurate early warnings; based on the debris flow formation potential index and early warning information output by the analysis layer, automatically trigger graded early warnings and drive multi-department coordinated responses; according to the results of the multi-parameter joint early warning, the system can automatically trigger early warnings of corresponding levels and drive multi-department coordinated responses to ensure timely response measures; effectively solve the problem of incomplete data coverage of traditional debris flow intelligent monitoring and early warning systems, and make it difficult to capture disaster precursors under the coupling of multiple factors, greatly improve the reliability of the monitoring and early warning system, and effectively provide protection for people's lives and property. BRIEF DESCRIPTION OF THE DRAWINGS

[0115] Figure 1 This is a system block diagram of a debris flow intelligent monitoring and early warning system proposed by the present invention. DETAILED DESCRIPTION

[0116] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0117] See also Figure 1 As shown, the present invention is an intelligent monitoring and early warning system for debris flow, comprising:

[0118] The perception layer is responsible for building an integrated air-space-ground monitoring system by deploying a multi-parameter sensor matrix. It collects key environmental parameters related to debris flow formation, such as rainfall, terrain slope, soil moisture, and vegetation cover, in real time. It also performs preliminary data preprocessing (such as filtering and normalization). It then uses lightweight compression algorithms (such as LZ4) to reduce the data volume and encrypts the data using the AES-128 encryption algorithm. The encrypted and compressed data packets are then sent to the transport layer.

[0119] Transport layer: Responsible for receiving encrypted and compressed data packets uploaded by the perception layer through 5G and Beidou short message dual-channel redundant transmission technology, ensuring data security while reducing transmission delays, and forwarding encrypted and compressed data packets to the analysis layer;

[0120] The analysis layer is responsible for decrypting and decompressing data packets forwarded by the transport layer. It then constructs a hybrid early warning model driven by physical mechanisms and data to integrate the decrypted and decompressed data in real time. It dynamically captures the coupling relationship between parameters such as rainfall, soil moisture, and terrain slope. It also uses random forests and LSTM networks to learn the multi-factor evolution patterns of historical debris flow events, generate dynamically adjusted critical threshold surfaces, perform multi-parameter joint early warnings, and output warning information to the application layer.

[0121] Application layer: Responsible for building a closed loop of monitoring, early warning, and response. Based on the early warning information output by the analysis layer, it triggers hierarchical early warnings and drives coordinated responses from multiple departments. At the same time, it pushes customized risk avoidance instructions to the public through a multimodal early warning information release platform.

[0122] The above technical solution further includes:

[0123] In one embodiment, the construction of an integrated air-ground-air monitoring system by deploying a multi-parameter sensor matrix includes the following steps:

[0124] Deploy the ground sensor matrix:

[0125] Deploy multi-parameter sensor nodes within the monitoring area. Each node integrates rainfall, terrain slope, soil moisture, and vegetation cover sensors. The sensor nodes are arranged in a grid pattern to fully cover areas prone to debris flows.

[0126] Integration of aerospace monitoring means:

[0127] Access satellite remote sensing data to obtain large-scale, high-resolution surface information, and use drones for regular or event-triggered aerial photography to capture local details.

[0128] Fusion and collaborative data:

[0129] The ground sensor data, satellite remote sensing data and drone aerial photography data are aligned in time and space, and through the data fusion algorithm, the multi-source data of air, space and land are complemented and coordinated to build a three-dimensional monitoring system.

[0130] In one embodiment, the preliminary data preprocessing comprises the following steps:

[0131] Data cleaning:

[0132] Based on physical rationality judgment (e.g., parameters such as rainfall and terrain slope should be within a reasonable range), outlier data points that significantly deviate from normal values ​​are removed; missing or invalid data are filled in using linear interpolation or based on the mean of historical data to make the data continuous;

[0133] Data filtering:

[0134] Apply low-pass filtering algorithm and moving average filtering to eliminate high-frequency noise interference:

[0135] Where y(n) represents the output after filtering, x(n) represents the input signal, and N represents the filter window size;

[0136] Data normalization:

[0137] Scaling the data to a uniform range (e.g., between 0 and 1) eliminates dimension effects and speeds up subsequent processing:

[0138]

[0139] Among them, x represents the original data, x min and x max Represents the minimum and maximum values ​​of the data respectively.

[0140] In one embodiment, reducing the amount of data by using a lightweight compression algorithm comprises the following steps:

[0141] Data chunking:

[0142] Divide the data stream to be compressed into fixed-size data blocks (e.g., 128KB) to facilitate parallel processing;

[0143] Dictionary encoding:

[0144] Using sliding window technology, a dynamic dictionary is constructed to record recently appeared data fragments, and repeated data fragments are replaced with dictionary indexes to reduce data redundancy.

[0145] Match Replacement:

[0146] Scan the data block, looking for the longest data segment that matches the dictionary; replace the matching segment with a tuple of "(match length, backreference distance)":

[0147] Token = (Length, Offset), where Length represents the matching length and Offset represents the reverse reference distance;

[0148] Entropy coding:

[0149] Perform Huffman coding on the generated tuples and unmatched data fragments to further compress the data volume:

[0150] Where L(x) represents the encoding length of symbol x, and P(x) represents the probability of occurrence of symbol x.

[0151] In one embodiment, encrypting data using the AES-128 encryption algorithm includes the following steps:

[0152] Based on the initial key (K0), a key expansion algorithm is used to generate multiple rounds of sub-keys (K1, K2, ..., K10). The initial data block (plaintext) is XORed with the first round sub-key (K0):

[0153] Among them, X1 is the first round input, P is the plaintext;

[0154] Then, an S-box (Substitution Box) is applied to each byte for nonlinear substitution to enhance the obfuscation of the encryption. The state matrix is ​​circularly shifted left, with the i-th row shifted left by i bytes for diffusion. Matrix multiplication is performed on each column of the state matrix, and a fixed matrix (M) is used for obfuscation:

[0155]

[0156] Among them, Y i represents the output after column confusion, X i Represents the input before column confusion;

[0157] Repeat the byte substitution, row shift, column confusion, and round key addition operations. In the last round, the column confusion operation is omitted and the round key addition is performed directly to generate the final ciphertext (C):

[0158] In one embodiment, the method of receiving the encrypted compressed data packet uploaded by the perception layer through the 5G and Beidou short message dual-channel redundant transmission technology to reduce transmission delay while ensuring data security includes the following steps:

[0159] Dual-channel parallel reception:

[0160] Deploy 5G communication modules and Beidou short message receiving modules, and simultaneously monitor the encrypted compressed data packets sent by the perception layer; the 5G channel uses the TCP / IP protocol, and the Beidou short message channel uses a custom binary protocol;

[0161] Verify and deduplicate data packets:

[0162] Checksum verification of received data packets:

[0163] Among them, D i is the i-th byte of the data packet, and n is the length of the data packet;

[0164] Deduplication is performed by sequence number, retaining the first arriving data packet and discarding duplicate packets;

[0165] Dynamic priority scheduling:

[0166] Dynamically assign transmission priority based on data packet type (such as real-time monitoring data, control instructions).

[0167] Using the Weighted Fair Queuing (WFQ) algorithm:

[0168] Among them, W i is the weight of the i-th category data, P i is the priority of the i-th category data, B is the total bandwidth, and N is the total number of data categories;

[0169] Security enhancements:

[0170] The TLS1.3 protocol is implemented at the transport layer, and the 5G channel is encrypted twice with a key length of 256 bits. The Beidou short message channel is encrypted using the national secret SM4 algorithm.

[0171] Latency optimization strategy:

[0172] The pre-scheduling mechanism is enabled for the 5G channel to reduce signaling interaction time, and a short frame format is adopted for the Beidou short message channel to shorten the single transmission time.

[0173] In one embodiment, the decrypting and decompressing the data packet forwarded by the transport layer includes the following steps:

[0174] Receive and verify data packets:

[0175] The receiving transport layer forwards encrypted compressed data packets through 5G and Beidou short message dual channels.

[0176] Perform integrity check on the data packet using the CRC-32 check algorithm:

[0177] CRC=(D0·x 32 +D1·x 31 +…+D n-1 ·x 32-n )modG(x), where D i is the i-th byte of the data packet, G(x) is the generating polynomial (such as G(x) = x 32 +x 26 +x 23 +…+x+1);

[0178] AES-128 decryption:

[0179] Use the same key (K) as used for encryption to perform AES-128 decryption, including key expansion, round key addition, inverse byte substitution, inverse row shift, and inverse column confusion operations. The round key addition formula is:

[0180] Among them, X i is the input of round i, K i is the key of round i;

[0181] LZ4 decompression:

[0182] The decrypted data packet is decompressed with LZ4 to restore the data format before compression. The decompression process is based on dictionary matching, replacing the "(match length, backreference distance)" tuple with the original data fragment:

[0183] Data=Decode(Token), where Token is the tuple generated during compression and Decode is the decoding function.

[0184] In one embodiment, the construction of a hybrid early warning model driven by physical mechanisms and data, integrating decrypted and decompressed data in real time, and dynamically capturing the coupling relationship between parameters such as rainfall, soil moisture, and terrain slope, includes the following steps:

[0185] Constructing a physical mechanism model:

[0186] Based on the theory of debris flow formation, the rainfall intensity and rainfall duration (ID) threshold curves are constructed:

[0187] I=αD β , where I is the rainfall intensity, D is the rainfall duration, and α and β are model parameters. The density of the water-soil mixture is determined by numerical simulation (such as the infinite slope model and the one-dimensional Richard water motion equation). A fluid-solid coupling model is introduced to describe the interaction between debris flow and slope, including the displacement coordination condition and the traction equation:

[0188] u s =u f , t s =t f , where u s and u f denote the displacement fields of solid and fluid respectively, t s and t f represent the stress fields of solid and fluid respectively;

[0189] Building data-driven models:

[0190] Using random forest or LSTM networks, with historical debris flow event data (including rainfall, soil moisture, terrain slope and other parameters) as input, a multi-factor evolution pattern recognition model is trained. Model hyperparameters, such as the number of trees in the random forest and the number of hidden layer nodes in the LSTM, are optimized through grid search:

[0191] Hyperparameters = argmax θ Accuracy(θ), where θ represents the hyperparameter set and Accuracy represents the accuracy of the model on the validation set;

[0192] Multi-source data fusion:

[0193] The decrypted and decompressed data is preprocessed including geometric calibration, coordinate registration, and format conversion; the feature level fusion method is used to extract the feature information of rainfall, soil moisture, and terrain slope parameters: F = Concat (Fr, Fs, Ft), where F r, Fs, Ft are the feature vectors of remote sensing data, ground sensor data, and terrain data, respectively, and Concat is a vector concatenation operation;

[0194] Analyze parameter coupling relationship:

[0195] Construct a multi-factor coupling model to analyze the interaction of parameters such as rainfall, soil moisture, and terrain slope:

[0196] λ = f(R, S, T), where λ is the probability of debris flow, R is rainfall, S is soil moisture, T is terrain slope, and f is a coupling function. The risk of debris flow is determined by setting critical values ​​for fluid bulk density and flow velocity and combining them with real-time monitoring data:

[0197]

[0198] Where γ and v represent the fluid density and flow rate, respectively, and γthreshold and vthreshold represent the critical values, respectively.

[0199] In one embodiment, the method of using a random forest or LSTM network to learn the multi-factor evolution pattern of historical debris flow events, generate a dynamically adjusted critical threshold surface, and perform multi-parameter joint early warning includes the following steps:

[0200] Prepare historical data:

[0201] Collect historical debris flow event data, including time series data on rainfall, soil moisture, terrain slope parameters, and information on the time, location, and scale of debris flows; and perform data preprocessing including data cleaning, normalization, and feature extraction.

[0202] Training the model:

[0203] Random Forest Model:

[0204] Construct multiple decision trees, each trained on a randomly sampled data subset and feature subset. The prediction results of each decision tree are combined through a voting mechanism to obtain the probability of debris flow occurrence.

[0205] LSTM network model:

[0206] Construct a neural network structure consisting of an input layer, an LSTM layer, a fully connected layer, and an output layer. Use historical time series data to train the LSTM network and learn how rainfall, soil moisture, and terrain slope parameters change over time. Use the backpropagation algorithm to optimize network weights and minimize prediction errors.

[0207] Generate a critical threshold surface:

[0208] Based on the trained random forest or LSTM network model, the contribution of each parameter to the occurrence of debris flow is analyzed. Multiple parameter combinations (such as rainfall and soil moisture, rainfall and terrain slope, etc.) are set and the probability of debris flow occurrence is calculated for each combination. Based on the probability values, a dynamically adjusted critical threshold surface is generated:

[0209] P(debris flow occurrence) = f(R, S, T)

[0210] Where P(debris flow occurrence) is the probability of debris flow occurrence, R is rainfall, S is soil moisture, T is terrain slope, and f is the functional relationship learned by random forest or LSTM network;

[0211] Critical threshold surface:

[0212] R × S × T = Threshold, where Threshold is learned from historical data and is dynamically adjusted as new data is added.

[0213] Multi-parameter joint early warning:

[0214] Receive decrypted and decompressed data in real time, extract the current values ​​of rainfall, soil moisture, and terrain slope parameters; substitute the current parameter values ​​into the critical threshold surface equation to calculate the probability of debris flow; judge the risk level of debris flow based on the probability value, and generate corresponding warning information, and output the warning information to the application layer so that relevant departments and personnel can take timely response measures.

[0215] In one embodiment, the debris flow formation potential index and warning information output by the analysis layer automatically triggers a graded warning and drives a multi-department coordinated response, including the following steps:

[0216] Evaluate the debris flow formation potential index:

[0217] The analysis layer calculates the debris flow formation potential index (L) through a hybrid warning model. The potential index integrates the coupling relationships of multiple parameters such as rainfall (R), soil moisture (S), and terrain slope (T):

[0218]

[0219] Among them, wi is the weight coefficient of each parameter and its coupling term, which is determined through random forest feature importance analysis;

[0220] Trigger hierarchical warning:

[0221] Set four-level warning thresholds (L1, L2, L3, L4), corresponding to blue, yellow, orange, and red warning levels: When L < L1, it is blue; when L1 ≤ L < L2, it is yellow; when L2 ≤ L < L3, it is orange; when L ≥ L3, it is red; when the potential index (L) exceeds the preset threshold, the corresponding warning signal is automatically triggered;

[0222] Multi-sector collaborative response drive:

[0223] Based on the warning level, response measures are automatically matched through a preset emergency plan. Warning information and response instructions are sent to the emergency management, natural resources, transportation, and meteorological departments through a unified communication interface. Each department executes corresponding tasks according to the received instructions, such as the emergency management department starting the emergency plan, the natural resources department strengthening geological monitoring, and the transportation department implementing traffic control.

[0224] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent monitoring and early warning system for debris flow, characterized in that: include: The perception layer is responsible for building an integrated air-space-ground monitoring system by deploying a multi-parameter sensor matrix. It collects key environmental parameters related to debris flow formation, such as rainfall, terrain slope, soil moisture, and vegetation cover, in real time, and performs preliminary data preprocessing. It then uses a lightweight compression algorithm to reduce the data volume and encrypts the data using the AES-128 encryption algorithm. The encrypted and compressed data packets are then sent to the transport layer. Transport layer: Responsible for receiving encrypted and compressed data packets uploaded by the perception layer through 5G and Beidou short message dual-channel redundant transmission technology, ensuring data security while reducing transmission delays, and then forwarding the encrypted and compressed data packets to the analysis layer; The analysis layer is responsible for decrypting and decompressing data packets forwarded by the transport layer. It then constructs a hybrid early warning model driven by physical mechanisms and data to integrate the decrypted and decompressed data in real time. It dynamically captures the coupling relationship between rainfall, soil moisture, and terrain slope parameters. It also uses random forests and LSTM networks to learn the multi-factor evolution patterns of historical debris flow events, generate dynamically adjusted critical threshold surfaces, perform multi-parameter joint early warnings, and output warning information to the application layer. Application layer: Responsible for building a closed loop of monitoring, early warning, and response. Based on the early warning information output by the analysis layer, it triggers hierarchical early warnings and drives coordinated responses from multiple departments. At the same time, it pushes customized risk avoidance instructions to the public through a multimodal early warning information release platform.

2. The intelligent monitoring and early warning system for debris flow according to claim 1 is characterized in that: The construction of an integrated air-ground-air monitoring system by deploying a multi-parameter sensor matrix includes the following steps: Deploy the ground sensor matrix: Deploy multi-parameter sensor nodes within the monitoring area. Each node integrates rainfall, terrain slope, soil moisture, and vegetation cover sensors. The sensor nodes are arranged in a grid pattern to fully cover areas prone to debris flows. Integration of aerospace monitoring means: Access satellite remote sensing data to obtain large-scale, high-resolution surface information, and use drones for regular or event-triggered aerial photography to capture local details. Fusion and collaborative data: The ground sensor data, satellite remote sensing data and drone aerial photography data are aligned in time and space, and through the data fusion algorithm, the multi-source data of air, space and land are complemented and coordinated to build a three-dimensional monitoring system.

3. The intelligent monitoring and early warning system for debris flow according to claim 1 is characterized in that: The preliminary data preprocessing comprises the following steps: Data cleaning: Based on physical rationality judgment, abnormal data points that obviously deviate from normal values ​​are removed; for missing or invalid data, linear interpolation is used to fill in the data based on the mean of historical data to make the data continuous; Data filtering: Apply low-pass filtering algorithm and moving average filtering to eliminate high-frequency noise interference: Where y(n) represents the output after filtering, x(n) represents the input signal, and N represents the filter window size; Data normalization: Scaling data to a uniform scale eliminates dimension effects and speeds up subsequent processing: Among them, x represents the original data, x min and x max Represents the minimum and maximum values ​​of the data respectively.

4. The intelligent debris flow monitoring and early warning system according to claim 1 is characterized in that: The method of reducing the amount of data by using a lightweight compression algorithm comprises the following steps: Data chunking: Divide the data stream to be compressed into fixed-size data blocks for parallel processing; Dictionary encoding: Using sliding window technology, a dynamic dictionary is constructed to record recently appeared data fragments, and repeated data fragments are replaced with dictionary indexes to reduce data redundancy. Match Replacement: Scan the data block, looking for the longest data segment that matches the dictionary; replace the matching segment with a tuple of "(match length, backreference distance)": Token = (Length, Offset), where Length is the matching length and Offset is the reverse reference distance; Entropy coding: Perform Huffman coding on the generated tuples and unmatched data fragments to further compress the data volume: L(x)=-log 2 (P(x)), where L(x) is the encoding length of symbol x and P(x) is the probability of occurrence of symbol x.

5. The intelligent monitoring and early warning system for debris flow according to claim 1 is characterized in that: The data is encrypted using the AES-128 encryption algorithm, including the following steps: Based on the initial key (K0), a key expansion algorithm is used to generate multiple rounds of sub-keys (K1, K2, ..., K10). The initial data block is XORed with the first round sub-key (K0): Among them, X1 is the first round input, P is the plaintext; Then, an S-box is applied to each byte for nonlinear substitution to enhance the obfuscation of the encryption. The state matrix is ​​circularly shifted left, with the i-th row shifted left by i bytes for diffusion. A matrix multiplication operation is performed on each column of the state matrix, using a fixed matrix for obfuscation: Among them, Y i represents the output after column confusion, X i Represents the input before column confusion; Repeat the byte substitution, row shift, column confusion, and round key addition operations. In the last round, the column confusion operation is omitted and the round key addition is performed directly to generate the final ciphertext:

6. The intelligent debris flow monitoring and early warning system according to claim 1 is characterized in that: The method of receiving encrypted compressed data packets uploaded by the perception layer through 5G and Beidou short message dual-channel redundant transmission technology, while ensuring data security and reducing transmission delay, includes the following steps: Dual-channel parallel reception: Deploy 5G communication modules and Beidou short message receiving modules, and simultaneously monitor the encrypted compressed data packets sent by the perception layer; the 5G channel uses the TCP / IP protocol, and the Beidou short message channel uses a custom binary protocol; Verify and deduplicate data packets: Checksum verification of received data packets: Among them, D i is the i-th byte of the data packet, and n is the length of the data packet; duplicate packets are removed by sequence number, the first arriving packet is retained, and duplicate packets are discarded; Dynamic priority scheduling: Dynamically assigns transmission priority based on packet type; uses the Weighted Fair Queuing (WFQ) algorithm: Among them, W i is the weight of the i-th category data, P i is the priority of the i-th category data, B is the total bandwidth, and N is the total number of data categories; Security enhancements: The TLS1.3 protocol is implemented at the transport layer, and the 5G channel is encrypted twice with a key length of 256 bits. The Beidou short message channel is encrypted using the national secret SM4 algorithm. Latency optimization strategy: Enable the pre-scheduling mechanism for 5G channels to reduce signaling interaction time, and adopt the short frame format for Beidou short message channels to shorten the duration of single transmission.

7. The intelligent debris flow monitoring and early warning system according to claim 1 is characterized in that: The decryption and decompression of the data packet forwarded by the transport layer comprises the following steps: Receive and verify data packets: The receiving transport layer forwards encrypted compressed data packets via the 5G and BeiDou short message dual channels; the integrity of the data packets is checked using the CRC-32 checksum algorithm: CRC=(D0·x 32 +D1·x 31 +…+D n-1 ·x 32-n )modG(x), where D i is the i-th byte of the data packet, G(x) is the generating polynomial; AES-128 decryption: Use the same key as used for encryption to perform AES-128 decryption, including key expansion, round key addition, inverse byte substitution, inverse row shift, and inverse column confusion. The round key addition formula is: Among them, X i is the input of round i, K i is the key of round i; LZ4 decompression: The decrypted data packet is decompressed with LZ4 to restore the data format before compression. The decompression process is based on dictionary matching, replacing the "(match length, backreference distance)" tuple with the original data fragment: Data=Decode(Token), where Token is the tuple generated during compression and Decode is the decoding function.

8. The intelligent debris flow monitoring and early warning system according to claim 1 is characterized in that: The hybrid early warning model driven by physical mechanisms and data is constructed to integrate decrypted and decompressed data in real time and dynamically capture the coupling relationship between parameters such as rainfall, soil moisture, and terrain slope. The model includes the following steps: Constructing a physical mechanism model: Based on the theory of debris flow formation, the rainfall intensity and rainfall duration threshold curves are constructed: I=αD β , where I is the rainfall intensity, D is the rainfall duration, and α and β are model parameters, which are determined by numerical simulation of the water-soil mixture density. A fluid-solid coupling model is introduced to describe the interaction between debris flow and bank slope, including the displacement coordination condition and traction equation: u s =u f , t s =t f , where u s and u f denote the displacement fields of solid and fluid respectively, t s and t f represent the stress fields of solid and fluid respectively; Building data-driven models: Using random forest or LSTM networks, with historical debris flow event data as input, a multi-factor evolution pattern recognition model is trained. Grid search is used to optimize model hyperparameters, such as the number of trees in the random forest and the number of hidden layer nodes in the LSTM. Hyperparameters = argmax θ Accuracy(θ), where θ represents the hyperparameter set and Accuracy represents the accuracy of the model on the validation set; Multi-source data fusion: The decrypted and decompressed data is preprocessed, including geometric calibration, coordinate registration, and format conversion. A feature-level fusion method is used to extract feature information of rainfall, soil moisture, and terrain slope parameters: F = Concat(Fr, Fs, Ft), where Fr, Fs, and Ft are the feature vectors of remote sensing data, ground sensor data, and terrain data, respectively, and Concat is a vector concatenation operation. Analyze parameter coupling relationship: Construct a multi-factor coupling model to analyze the interaction of parameters such as rainfall, soil moisture, and terrain slope: λ = f(R, S, T), where λ is the probability of debris flow, R is rainfall, S is soil moisture, T is terrain slope, and f is a coupling function. The risk of debris flow is determined by setting critical values ​​for fluid bulk density and flow velocity and combining them with real-time monitoring data: Where γ and v represent the fluid density and flow rate, respectively, and γthreshold and vthreshold represent the critical values, respectively.

9. The intelligent debris flow monitoring and early warning system according to claim 1 is characterized in that: The random forest and LSTM networks are used to learn the multi-factor evolution pattern of historical debris flow events, generate a dynamically adjusted critical threshold surface, and perform multi-parameter joint early warning, including the following steps: Prepare historical data: Collect historical debris flow event data, including time series data on rainfall, soil moisture, terrain slope parameters, and information on the time, location, and scale of debris flows; and perform data preprocessing including data cleaning, normalization, and feature extraction. Training the model: Random Forest Model: Construct multiple decision trees, each trained on a randomly sampled data subset and feature subset. The prediction results of each decision tree are combined through a voting mechanism to obtain the probability of debris flow occurrence. LSTM network model: Construct a neural network structure consisting of an input layer, an LSTM layer, a fully connected layer, and an output layer. Use historical time series data to train the LSTM network and learn how rainfall, soil moisture, and terrain slope parameters change over time. Use the backpropagation algorithm to optimize network weights and minimize prediction errors. Generate a critical threshold surface: Based on the trained random forest or LSTM network model, the contribution of each parameter to the occurrence of debris flow is analyzed; multiple parameter combinations are set and the probability of debris flow occurrence under each combination is calculated; and a dynamically adjusted critical threshold surface is generated based on the probability value: P(O)=f(R,S,T) Where P(O) is the probability of debris flow, R is rainfall, S is soil moisture, T is terrain slope, and f is the functional relationship learned by random forest or LSTM network; Critical threshold surface: R×S×T=Threshold, where Threshold is the critical threshold learned from historical data and is dynamically adjusted as new data is added; Multi-parameter joint early warning: Receive the data after real-time decryption and decompression, and extract the current values of rainfall, soil humidity, and terrain slope parameters; substitute the current parameter values into the critical threshold surface equation to calculate the probability of debris flow occurrence; judge the risk level of debris flow occurrence according to the probability value, generate corresponding early warning information, and output the early warning information to the application layer so that relevant departments and personnel can take corresponding measures in a timely manner.

10. The intelligent debris flow monitoring and early warning system according to claim 1, characterized in that: Based on the early warning information output by the analysis layer, trigger hierarchical early warning and drive multi-department collaborative response, including the following steps: Evaluate the debris flow formation potential index: The analysis layer calculates the debris flow formation potential index (L) through a hybrid early warning model, and the potential index integrates the multi-parameter coupling relationship of rainfall (R), soil humidity (S), and terrain slope (T): L=w 1 ·R+w 2 ·S+w 3 ·T+w 4 ·(R×S)+w 5 ·(R×T)+w 6 ·(S×T) Among them, wi is the weight coefficient of each parameter and its coupling term, which is determined by random forest feature importance analysis; Trigger hierarchical early warning: Set four-level early warning thresholds (L1, L2, L3, L4), corresponding to blue, yellow, orange, and red early warning levels: when L < L1, it is blue; when L1 ≤ L < L2, it is yellow; when L2 ≤ L < L3, it is orange; when L ≥ L3, it is red; when the potential index exceeds the preset threshold, automatically trigger the corresponding level of early warning signal; Drive multi-department collaborative response: Based on the early warning level, automatically match response measures through a preset emergency plan, send early warning information and response instructions to the emergency management, natural resources, transportation, and meteorological departments through a unified communication interface, and each department executes corresponding tasks according to the received instructions.

Images