Slope displacement monitoring method based on UWB

Abstract

The invention relates to the technical field of slope displacement monitoring, in particular to a UWB-based slope displacement monitoring method. According to the technical scheme, the slope displacement monitoring method based on the UWB comprises the following steps that S1, equipment deployment is carried out, at least four anchor nodes are deployed in a slope stable area, each anchor node is provided with a double-MCU framework, a first MCU-MCU1 is specially used for UWB distance measurement and closes peripheral interruption, and a second MCU-MCU2 is provided with communication interfaces of a LoRa module, a 5G / 4G module, a WiFi / Bluetooth module and an OLED module; deploying at least one label node in a landslide monitoring area, and constructing a star-shaped or grid structure with the anchor nodes; and S2, data acquisition, processing and transmission: UWB ranging data of the label node and each anchor node are acquired through an MCU1 at an adjustable interval of 100 milliseconds to 60 minutes. The slope displacement monitoring technology is improved in a breakthrough mode, distance measurement interference is isolated through a double-MCU framework, the capturing capacity of infinitesimal displacement is remarkably improved, and the misjudgment problem caused by signal fluctuation of traditional equipment is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of slope displacement monitoring, and in particular to a slope displacement monitoring method based on UWB. Background Art

[0002] Landslides are common and destructive natural disasters worldwide. They can be categorized into two types: slow landslides and fast landslides. Although slow landslides move slowly, they can rapidly develop into collapses under certain triggering conditions, resulting in devastating consequences. Therefore, landslide monitoring is crucial. However, due to the destructive nature and unpredictability of landslides, traditional emergency response measures often struggle to effectively address them. Against this backdrop, developing novel landslide monitoring technologies and establishing low-cost, efficient, and real-time landslide monitoring and early warning systems are crucial.

[0003] Currently, landslide monitoring technologies are primarily categorized into three categories: surface monitoring, drone monitoring, and remote sensing. Global Navigation Satellite Systems (GNSS) are a representative technology for surface monitoring, enabling real-time spatial information acquisition at monitoring points. However, due to their high cost, GNSS systems can typically only be deployed at key locations within a landslide, making it difficult to monitor the entire landslide area. In recent years, unmanned aerial vehicles (UAVs) have gained widespread application in landslide monitoring, particularly for rapidly acquiring terrain information over a large area within a landslide zone. However, UAV monitoring is limited by weather conditions and can only be performed periodically, making it difficult to provide real-time monitoring. Synthetic Aperture Radar (InSAR), a representative technology for remote sensing, can cover vast areas and conduct long-term monitoring, but it has limitations in temporal and spatial resolution. For example, the long revisit period of satellites makes it impossible to monitor rapidly changing landslides in real time. In summary, existing landslide monitoring methods fail to meet the low-cost, high-real-time requirements of practical applications. Therefore, a new monitoring technology is urgently needed to fill this gap. Summary of the Invention

[0004] The present invention proposes a slope displacement monitoring method based on UWB, which solves the problem in the prior art that it is impossible to perform real-time monitoring of rapidly changing landslides.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0006] A slope displacement monitoring method based on UWB includes the following steps:

[0007] Step S1: Equipment deployment: Deploy at least four anchor nodes in the slope stability area. Each anchor node is configured with a dual MCU architecture, where the first MCU (MCU1) is dedicated to UWB ranging and has peripheral interrupts disabled, and the second MCU (MCU2) has communication interfaces for LoRa modules, 5G / 4G modules, WiFi / Bluetooth modules, and OLED modules. Deploy at least one tag node in the landslide monitoring area to form a star or mesh structure with the anchor nodes.

[0008] Step S2: Data collection, processing, and transmission: MCU1 collects UWB ranging data from the tag node and each anchor node at an adjustable interval of 100 milliseconds to 60 minutes to generate an original distance sequence; MCU2 performs multi-tasking processing, including real-time display of device status, local caching of abnormal data, and multi-channel data transmission;

[0009] Step S3, displacement prediction and early warning, performs 3σ outlier removal and Kalman filtering on the original distance data, calculates the periodic differential distance and cumulative differential distance; inputs the processed data into the UWBformer neural network model, and outputs the moving distance, horizontal angle, and pitch angle of the monitoring point; when the spatial displacement exceeds a set threshold of 1-50mm, an alarm message is sent through the cloud to the reserved mobile phone number and email address.

[0010] Furthermore, the UWB ranging in step S2 includes:

[0011] UWB data acquisition equipment uses TCXO crystal oscillators to compensate for temperature drift errors, and the operating temperature range is -20°C to 60°C;

[0012] The antenna delay error is eliminated by periodic differential calculation. The calculation formula is:

[0013]

[0014] in, is the ranging value of the jth anchor node in the nth period.

[0015] Furthermore, the multi-channel data transmission in step S2 includes:

[0016] Uploaded data is transmitted between nodes through the LoRaMesh network, and the data is aggregated to the LoRa gateway node that supports 4G / 5G communication. The gateway node uploads the collected data to the cloud through the cellular data of the base station;

[0017] The downlink data is transmitted from the cloud to the LoRa gateway node via the 4G / 5G network. The gateway node then transmits the data to the target device based on the routing table information.

[0018] On-site debugging parameters are written to the UWB data acquisition device through the communication interface of the WiFi / Bluetooth module.

[0019] Furthermore, the UWBformer neural network model construction in step S3 includes:

[0020] Dual-channel feature extraction of time domain channel and frequency domain channel. The input of time domain channel is anchor node coordinate information, periodic differential distance and cumulative differential distance. The input of frequency domain channel is periodic differential distance and cumulative differential distance after FFT transformation.

[0021] Cross attention fusion module, the calculation formula is:

[0022]

[0023] where Q time is the time domain feature, K freq 、V freq is the frequency domain feature.

[0024] Furthermore, the cross attention fusion module further includes:

[0025] The spatial offset term generation module extracts the spatial information matrix S from the cumulative differential distance and obtains it through linear projection

[0026] Corrected attention score calculation:

[0027]

[0028] where Q h , K h 、V h is the query, key, and value vector of the h-th attention head.

[0029] Furthermore, after the warning signal in step S3 is triggered, the following steps are further performed:

[0030] Start high-density sampling mode and shorten the data collection interval to 100 milliseconds-1 second;

[0031] Generate a three-dimensional displacement trajectory map and display it through the cloud-based human-computer interaction interface;

[0032] When the displacement exceeds 120% of the threshold for three consecutive cycles, the emergency response protocol is triggered.

[0033] Furthermore, the method also includes a model online optimization step:

[0034] 20% of the real-time data is used as a validation set to calculate the MAE value of the moving distance;

[0035] When MAE exceeds 4mm for three consecutive times, incremental training is started and the update formula is:

[0036]

[0037] Where η is the dynamic learning rate, is the online loss function.

[0038] Furthermore, the incremental training includes:

[0039] Adopting a sliding window mechanism to retain the historical data of the last 30 days;

[0040] The newly added data is weighted and the weight distribution formula is:

[0041]

[0042] where t i is the data generation time, t current is the current time.

[0043] Furthermore, the method further includes a device remote maintenance step:

[0044] Receive configuration instructions through an AES-256 encrypted channel and dynamically adjust the UWB communication frequency band and transmission power;

[0045] When it is detected that the device is offline for more than 2 hours, it will automatically switch to solar power mode and start the self-test program;

[0046] When connecting to the on-site debugging device via the RS485 interface, the execution of remote configuration commands is suspended.

[0047] Furthermore, the training method of the UWBformer neural network includes:

[0048] The weighted root mean square error loss function is used:

[0049]

[0050] Using Xavier initialization and Adam optimizer, the learning rate is dynamically adjusted according to the validation loss, and the adjustment range does not exceed ±20% of the initial value.

[0051] The positive effects of the present invention are: the present invention achieves a breakthrough improvement in slope displacement monitoring technology, the dual MCU architecture isolates ranging interference, combines time series difference and frequency domain analysis, significantly improves the ability to capture tiny displacements, and overcomes the misjudgment problem caused by signal fluctuations in traditional equipment; it is the first to integrate three-dimensional displacement prediction of distance, direction, and angle, breaking through the limitations of single-point ranging and accurately depicting the spatial motion trajectory of the landslide body; the dynamic learning mechanism enables the system to autonomously optimize model parameters and maintain stable monitoring performance in complex environments (such as rain, fog, and vegetation obstruction); multi-mode communication collaboration and remote configuration technology realize cloud-based visual management of equipment status, greatly reducing field maintenance costs; the end-to-end processing time from data collection to early warning triggering is shortened to seconds, striving for a golden window for disaster emergency response.

[0052] While completing high-precision ranging tasks, it can also simultaneously complete data management and peripheral interface data transmission tasks. Compared with traditional UWB devices, it has higher ranging accuracy and sampling frequency, and more complete peripheral management functions, making it more suitable for applications in landslide monitoring scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 Physical photos of the UWB monitoring device provided in the embodiment of the present invention;

[0054] Figure 2 A hardware architecture diagram of a UWB monitoring device provided in an embodiment of the present invention;

[0055] Figure 3 A schematic diagram of the deployment of UWB monitoring equipment provided by an embodiment of the present invention;

[0056] Figure 4 The network topology provided by the embodiment of the present invention;

[0057] Figure 5 The UWBformer network structure provided by the embodiment of the present invention. DETAILED DESCRIPTION

[0058] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described 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 any creative efforts are within the scope of protection of the present invention.

[0059] Example 1

[0060] A slope displacement monitoring method based on UWB includes the following steps:

[0061] Step S1: Equipment deployment: Deploy at least four anchor nodes in the slope stability area. Each anchor node is configured with a dual MCU architecture, where the first MCU (MCU1) is dedicated to UWB ranging and has peripheral interrupts disabled. The second MCU (MCU2) has communication interfaces for LoRa modules, 5G / 4G modules, WiFi / Bluetooth modules, and OLED modules, and each module is connected through each communication interface. Deploy at least one tag node in the landslide monitoring area to form a star or mesh structure with the anchor nodes.

[0062] At least four anchor nodes are deployed in the slope stability area. These anchor nodes need to be deployed in the up, down, left and right directions of the landslide area. When the number of anchor node devices increases, they need to be gradually inserted between the two existing adjacent anchor nodes to increase the density of peripheral anchor nodes, so as to achieve an encircled deployment in the peripheral area of ​​the landslide; at least one tag node is deployed in the landslide monitoring area. When the number of tag nodes increases, the coverage density in the landslide area needs to be gradually increased to achieve a mesh deployment in the landslide area; among them, the first MCU-MCU1 is dedicated to UWB ranging and turns off peripheral interrupts, and the second MCU-MCU2 has a LoRa communication interface, a WiFi communication interface and a 5G / 4G communication interface.

[0063] Step S2: Data collection, processing, and transmission: MCU1 collects UWB ranging data from the tag node and each anchor node at an adjustable interval of 100 milliseconds to 60 minutes to generate an original distance sequence; MCU2 performs multi-tasking processing, including real-time display of device status, local caching of abnormal data, and multi-channel data transmission;

[0064] Step S3, displacement prediction and early warning, performs 3σ outlier removal and Kalman filtering on the original distance data, calculates the periodic differential distance and cumulative differential distance; inputs the processed data into the UWBformer neural network model, outputs the moving distance, horizontal angle and pitch angle of the monitoring point; when the spatial displacement exceeds the set threshold of 1-50mm, an alarm message is sent through the cloud to the reserved mobile phone number and email address.

[0065] The overall process of the monitoring method is as follows: The dual MCU architecture realizes the parallel processing of precise ranging and data transmission. MCU1 uses the interrupt mask mode to execute the UWB ranging process, transmits / receives UWB pulses with a width of 6.5ns through the DWM1000 module, uses the double-sided two-way (DS-TWR) method to record the timestamps of multiple communications, and calculates the time of flight (ToF) of the signal. Finally, the distance measurement value is calculated based on ToF and the speed of light. The ranging data is directly written into the memory space corresponding to UART1 of MCU1 through the DMA channel to avoid the data transmission of the peripheral interface occupying the operating resources of the MCU. MCU2 reads the raw ranging data sent by MCU1 through UART in interrupt mode and executes synchronously:

[0066] OLED display: Using an OLED display module with an SPI interface, with a refresh cycle of 200ms, it dynamically displays the distance information between the monitoring point and each anchor node and the device power information (battery voltage is collected through ADC);

[0067] LoRa networking: Using LoRaMesh networking, path finding and data connection establishment are completed through the preset routing table information after device deployment, and the data is finally aggregated to the LoRa gateway node;

[0068] 5G backhaul: The transport layer protocol uses the TCP protocol, and the application layer protocol uses a custom protocol to enable data converged to the LoRa gateway node to be uploaded to the cloud or downloaded to the device.

[0069] WiFi / Bluetooth: After the device is deployed, you need to use the WiFi / Bluetooth interface to initialize and configure the device parameters based on the actual situation on site.

[0070] The three-dimensional displacement prediction module runs on the cloud server and triggers the UWBformer inference process after receiving data. The prediction results will be displayed through the human-computer interaction interface and displayed in real time through three-dimensional images. When the set threshold is exceeded, the early warning function will be triggered and text messages and emails will be automatically sent to managers.

[0071] The UWB ranging in step S2 includes:

[0072] The UWB data acquisition equipment uses a TCXO crystal oscillator to compensate for temperature drift errors. The operating temperature range is -20°C to 60°C. The TCXO model used is ECS-2520SMV, which maintains a frequency deviation of ±0.5ppm within the range of -20°C to 60°C.

[0073] The antenna delay error is eliminated by periodic differential calculation. The calculation formula is:

[0074]

[0075] in, is the ranging value of the j-th anchor node in the n-th period.

[0076] Cycle difference processing: Calculate the distance difference between adjacent cycles in real time in the FPGA embedded logic to eliminate the inherent deviation of antenna delay:

[0077]

[0078] where μ j is the factory-calibrated delay compensation value for the jth anchor point. The variance of the processed data is reduced by 62% (measured data).

[0079] The multi-channel data transmission in step S2 includes:

[0080] Uploaded data is transmitted between nodes through the LoRaMesh network, and the data is aggregated to the LoRa gateway node that supports 4G / 5G communication. The gateway node uploads the collected data to the cloud through the cellular data of the base station;

[0081] The downlink data is transmitted from the cloud to the LoRa gateway node via the 4G / 5G network. The gateway node then transmits the data to the target device based on the routing table information.

[0082] On-site debugging parameters are written to the UWB data acquisition device through the communication interface of the WiFi / Bluetooth module.

[0083] Uploaded data supports adaptive adjustment from 1 second to 60 minutes or manually set time intervals. The cloud service can obtain real-time weather information in the device deployment area. When severe weather such as rain or snow is about to occur, the cloud service will send a command to the UWB data acquisition device ( / device) to set the data upload interval to a high-frequency transmission of 1 second. After the rain or snow passes, it will gradually return to the normal 60-minute transmission interval, thus achieving the best balance between power consumption and real-time performance.

[0084] Uploaded data includes regular ranging data, abnormal cache data, device status information, etc.; downlinked data includes remote setting parameters and remote control commands. The former mainly includes device parameter setting, LoRa parameter setting and UWB parameter setting, and the latter mainly includes device status control, device firmware upgrade and maintenance; on-site debugging parameters are the same as remote and setting parameter items, but after the on-site equipment is deployed, the device parameters need to be initialized and set based on actual measurements on site; device parameters mainly include device access and connection management, device name, device number, device type, deployment information, etc.; UWB parameters mainly include channel number, pulse frequency, preamble length, data transmission rate, etc.; LoRa parameters mainly include spreading factor, coding rate, debugging bandwidth and routing table information, etc.

[0085] The communication between the LoRa gateway node and the cloud service uses the TCP protocol, which is divided into four channel ports: 10001-10004; 10001 is used to upload regular ranging data, 10002 is used to transmit other uploaded data, 10003 is used to download remote setting parameters, and 10004 is used to download remote control commands; the priority of each channel is arranged from large to small as follows: 10004, 10003, 10001, 10002;

[0086] MCU2 uses serial port AT commands to communicate with the LoRa module, 5G / 4G module, BLE module and WiFi module.

[0087] The LoRa module uses Ebit EMW290, which is controlled by AT commands; the center frequency is 433MHz, the bandwidth BW is 250MHz, the spreading factor SF is 7, the coding rate CR is 4 / 5, the transmission power is 22dBm, the receiving sensitivity is -123dBm, and CRC cyclic redundancy check is used.

[0088] The 5G module uses the SIMCom A8230, which is controlled via AT commands. The module supports both 5G NR and 4G LTE modes, with a transmit power of 23dBm and a receive sensitivity of -85dBm (5G) and -97dBm (4G).

[0089] The WiFi module used was the Acred ESP-01F, controlled via AT commands. To ensure better anti-interference performance during on-site implementation, the WiFi used the 2.4 GHz frequency band, the IEEE 802.11ac standard, and WPA2 encryption.

[0090] The Bluetooth module uses the Daxia Longque BX-BT24, which is controlled by AT commands. It uses the low-power Bluetooth 5.0 protocol standard, the operating frequency uses the 2.4GHz band, the pairing method uses a PIN code and NFC, and the encryption method uses AES-128 / 256.

[0091] Data collection and preprocessing:

[0092] In this example, a sampling period of 100ms and a processing period of 5 minutes were selected. Each processing period collected 3000 sets of ranging data, each containing four ranging values. The data collected in each period was first filtered for outliers using the 3σ criterion, and then iteratively estimated using a Kalman filter with a window size of 100. Finally, the preprocessed ranging data for that period was obtained. The sampled data from each period was aligned, and the differential ranging value for that period and the accumulated differential ranging value up to that period were calculated.

[0093] The UWBformer neural network model construction in step S3 includes:

[0094] Dual-channel feature extraction of time domain channel and frequency domain channel. The input of time domain channel is anchor node coordinate information, periodic differential distance and cumulative differential distance. The input of frequency domain channel is periodic differential distance and cumulative differential distance after FFT transformation.

[0095] Cross attention fusion module, the calculation formula is:

[0096]

[0097] where Q time is the time domain feature, K freq 、Vfreq is the frequency domain feature.

[0098] Time-frequency dual-channel processing flow:

[0099] Time domain channel: Input the coordinate data of the anchor node and the differential distance sequence of length 11600 ((3000-100)*4) and the accumulated differential distance sequence, and extract the local change features through 1D convolution (kernel=4, stride=1), and the output dimension is 256;

[0100] Frequency domain channel: perform 1024-point FFT on the differential distance sequence and the accumulated differential distance sequence, take the power spectrum of the 0.1-10 Hz frequency band, and reduce the dimension to 256 dimensions through the Mel filter bank;

[0101] Cross-attention fusion: time domain features are used as query, frequency domain features are used as key / value, and the calculation formula is:

[0102]

[0103] Among them, d k Set to 256, output 256*d v The fusion features, in this case, d v The dimension is reduced to 64, and the final output is 256*64 fusion features.

[0104] The cross attention fusion module also includes:

[0105] The spatial bias term generation module extracts the spatial information matrix S from the cumulative distance and obtains it through linear projection

[0106] Corrected attention score calculation:

[0107]

[0108] where Q h , K h 、V h is the query, key, and value vector of the h-th attention head.

[0109] Spatial offset generation process:

[0110] Cumulative distance coding: The cumulative displacement of each anchor point is normalized, and after the full connection layer and reconstruction processing, the spatial distribution matrix S in the range of 0-1 is obtained;

[0111] Linear projection: Map S to the attention head dimension through a fully connected layer:

[0112]

[0113] in is the trainable parameter of the h-th head;

[0114] Attention correction: A spatial bias term is added to the standard attention score to strengthen the consistency of the displacement direction:

[0115]

[0116] Where α=0.3 is a learnable scaling factor. Experiments show that this design reduces the direction prediction error by 41%.

[0117] After the early warning signal in step S3 is triggered, the following steps are also included:

[0118] Start high-density sampling mode and shorten the data collection interval to 100 milliseconds-1 second;

[0119] Generate a three-dimensional displacement trajectory map and display it through the cloud-based human-computer interaction interface;

[0120] When the displacement exceeds 120% of the threshold for three consecutive cycles, the emergency response protocol is triggered.

[0121] Three-level emergency response process:

[0122] Primary warning (displacement > threshold): start high-density sampling (data sampling rate increased to 10Hz);

[0123] Intermediate response (three consecutive threshold violations): The human-computer interface flashes an alarm, and the platform sends a text message and email to the administrator's reserved mobile phone number and email address;

[0124] Advanced response (3 consecutive times exceeding the threshold by 120%):

[0125] Trigger an audible and visual alarm (LED flashes at 5Hz, buzzer emits 1kHz pulse tone); the platform continues to alarm, and sends text messages and emails to the platform's reserved mobile phone number and email address every 1 minute until the event is processed on the platform;

[0126] Generate 3D trajectory map: Use OpenGL ES 3.0 to render the displacement vector field and generate the landslide deformation cloud map through the isosurface extraction algorithm;

[0127] Push emergency messages to designated contacts.

[0128] The method further comprises a model online optimization step:

[0129] 20% of the real-time data is used as a validation set to calculate the MAE value of the moving distance;

[0130] When MAE exceeds 4mm for three consecutive times, incremental training is started and the update formula is:

[0131]

[0132] Where η is the dynamic learning rate, is the online loss function.

[0133] Incremental learning implementation steps:

[0134] Validation set construction: Randomly sample 20% from the real-time data stream as the validation set and calculate the MAE of the moving distance:

[0135]

[0136] Trigger condition: When MAE is > 4mm for three consecutive cycles, the optimization thread is started;

[0137] Parameter update: Adopt the elastic weight consolidation (EWC) algorithm to retain important parameters:

[0138]

[0139] Among them F i is the diagonal element of the Fisher information matrix, and λ = 0.5 is the penalty coefficient to prevent catastrophic forgetting.

[0140] The incremental training includes:

[0141] Adopting a sliding window mechanism to retain the historical data of the last 30 days;

[0142] The newly added data is weighted and the weight distribution formula is:

[0143]

[0144] where t i is the data generation time, t current is the current time.

[0145] Historical data management mechanism:

[0146] Sliding window: maintains a 30-day ring buffer, with new data overwriting the oldest record;

[0147] Time decay weighting: The weight of each data is adjusted according to exponential decay:

[0148]

[0149] Where β = 0.1 is the attenuation rate, T max = 2592000 seconds (30 days), ensuring that the weight of recent data accounts for ≥ 70%;

[0150] Gradient calculation: During backpropagation, the loss of each sample is multiplied by its weight, giving priority to fitting recent displacement patterns.

[0151] The method further comprises the step of remote maintenance of the equipment:

[0152] Receive configuration instructions through an AES-256 encrypted channel and dynamically adjust the UWB communication frequency band and transmission power;

[0153] When connecting to the on-site debugging device via the RS485 interface, the execution of remote configuration commands is suspended.

[0154] Safety Maintenance Agreement:

[0155] Encryption configuration: AES-256-GCM mode is used, and the key is updated every 24 hours through ECDH protocol negotiation to prevent replay attacks;

[0156] Memory self-test: Detect SRAM errors through March C-algorithm;

[0157] Local debugging is preferred: When the RS485 interface is connected to the computer's serial debugging window, device debugging and abnormal problem location are completed through log information and interactive information.

[0158] The training method of the UWBformer neural network includes:

[0159] The weighted root mean square error loss function is used:

[0160]

[0161] Using Xavier initialization and Adam optimizer, the learning rate is dynamically adjusted according to the validation loss, and the adjustment range does not exceed ±20% of the initial value.

[0162] Training optimization details:

[0163] Weighted loss calculation:

[0164] The distance loss uses Huber loss, δ = 1.0:

[0165]

[0166] The angle loss uses cosine similarity:

[0167]

[0168] Dynamic learning rate: Initial lr = 1e-4, evaluate validation loss every 5 epochs, and multiply by γ = 0.5 if it does not decrease, down to a minimum of 1e-6;

[0169] Weight initialization: The convolutional layer uses He normal initialization, and the attention layer uses Xavier uniform initialization to ensure the stability of the variance of the forward propagation activation value.

[0170] Example 2

[0171] In landslide monitoring scenarios, monitoring equipment must have the characteristics of high precision, high efficiency, low cost, long battery life and easy installation. Traditional UWB devices are mainly designed for indoor positioning, with a single peripheral interface. Only one MCU is used to collect UWB data and output it through a serial interface. When the data download function is enabled, the device will affect the ranging accuracy due to the MCU entering a receive interrupt. This effect is not obvious in indoor dynamic positioning with low accuracy requirements, but in landslide monitoring applications with high accuracy requirements, it will cause abnormal fluctuations in ranging data and is difficult to eliminate. Compared with traditional UWB devices, this design significantly improves data accuracy and processing efficiency, and at the same time has more complete system management functions, which is more suitable for the actual needs of landslide monitoring. Empowered by AI technology, combined with the UWBformer model proposed in this invention, slope displacement monitoring can be achieved with the advantages of low cost and high real-time performance.

[0172] The main sources of UWB distance measurement errors are crystal oscillator temperature drift, antenna delay parameter errors, and external environmental interference. To address these three influencing factors, the UWB device of the present invention proposes the following measures:

[0173] (1)TCXO has been applied to the UWB device we designed, which can effectively reduce the impact of temperature changes on the crystal oscillator frequency and improve the accuracy of distance measurement.

[0174] (2) Antenna delay error is difficult to completely eliminate. In order to obtain accurate distance measurement values, precise antenna delay parameters are required. Usually, a special calibration method is used to calibrate the antenna delay to reduce the error. Despite this, the distance measurement accuracy can only reach the centimeter level, which is difficult to meet the millimeter-level accuracy requirements of landslide monitoring. Since the antenna delay error is additive noise, the present invention proposes a new processing method that uses the distance difference between adjacent cycles as the input of the UWBformer instead of using the distance value. This processing method can not only reduce the impact of antenna delay, but also improve the generalization ability of the UWBformer, so that it is not affected by the device deployment location.

[0175] (3) External environmental interference factors are complex, including humidity changes, weather changes, and noise interference, and there are many uncertainties. Therefore, they cannot be completely eliminated, and the only option is to suppress the interference components as much as possible. This study first used the 3σ criterion to remove outliers in the UWB data, and then used Kalman filtering to denoise the data. Through these preprocessing, the impact of external interference was reduced.

[0176] In landslide monitoring, mainstream surface displacement monitoring methods include synthetic aperture radar (InSAR), ground-based radar (GB-SAR), unmanned aerial vehicles (UAVs), and GNSS. InSAR offers the widest monitoring range, but is weaker in other areas, is very expensive, has a long acquisition cycle, and lacks real-time performance. It is only used for preliminary screening of landslide areas and cannot effectively monitor small slope displacements. SAR is suitable for monitoring entire landslides and offers high accuracy, but also has a high cost. It is typically used for short-term emergency monitoring and cannot be applied to long-term monitoring scenarios. UAVs use image modeling to monitor landslides, but their accuracy is lower, typically only reaching the centimeter level, and they cannot provide real-time monitoring. They are usually used for periodic monitoring of landslide areas. GNSS offers high accuracy and can achieve real-time monitoring, but due to cost constraints, it is generally only suitable for single-point monitoring of key landslide points. The data sampling rate is at most in seconds, and the data processing time is long, typically requiring at least half an hour of processing for RTK data to achieve millimeter-level accuracy. However, the UWB equipment used in this study costs approximately one-tenth of GNSS, is smaller, and has a higher sampling rate, reaching milliseconds. Furthermore, its power consumption is significantly lower than that of GNSS devices, enabling longer battery life and preventing data loss caused by equipment downtime during continuous rainy days. These advantages make it easier to deploy UWB devices at multiple locations within landslide areas, enabling networked monitoring of landslide areas and providing a novel approach to practical slope displacement monitoring.

[0177] Example 3

[0178] like Figure 1 、 Figure 2As shown, the UWB monitoring device designed in this invention includes two MCUs, a UWB module for ranging, a LoRa module for networking, an OLED display module, a Flash memory module, a Bluetooth / WiFi communication module, an RS485 interface, and a 5G / 4G network card. After UWB data is collected by MCU1, it is transmitted to MCU2 via the UART interface for subsequent processing. To ensure ranging accuracy, all peripheral interrupts must be disabled during MCU1's operation to prevent interference with the timestamp recorded by DWM1000 when the program enters the interrupt state. MCU2 integrates multiple functional modules, including an OLED display, Flash memory, a Bluetooth / WiFi module, an RS485 interface, and a LoRa communication module. When the device functions as a LoRa gateway node, it also supports 5G / 4G / GPRS communication. The OLED display displays device status and ranging information in real time; the Flash memory module stores unsuccessfully transmitted ranging data and system configuration information; the Bluetooth and WiFi modules are used for parameter configuration; the RS485 interface is used for debugging; and the LoRa module is used for networking communication, achieving efficient transmission of monitoring point data. After receiving data via the LoRa network, the gateway node can upload the data to a cloud server using 5G and other communication methods. In addition, the device supports functions such as remote parameter configuration, enabling remote management.

[0179] In actual applications, a 5V lithium battery and a solar power supply system are required to provide stable endurance, and the equipment is installed in combination with an external bracket. The UWB tag node on the landslide body maintains continuous two-way ranging communication with the four surrounding anchor nodes, and collects four sets of high-precision distance values ​​in each monitoring cycle. These anchor nodes are fixed in geologically stable areas to ensure that their positions remain unchanged during the monitoring period. As the position of the monitoring point changes due to the movement of the landslide body, the distance between the tag node and the anchor node will also change accordingly. By monitoring the dynamic changes of these distances and combining the UWBformer neural network model proposed in the present invention, the displacement of the monitoring point in three-dimensional space can be accurately predicted, including the distance and direction of the displacement, providing reliable data support for landslide disaster warning.

[0180] like Figure 3 As shown in the figure, taking 4 anchor node devices and 1 tag device as an example, the deployment method of UWB monitoring equipment is demonstrated. When performing multi-node monitoring, it is only necessary to add anchor node devices in the peripheral stable area and add multiple monitoring node devices in the landslide area.

[0181] like Figure 4 As shown in the figure, the network topology structure collected by UWB monitoring equipment is displayed. The data collected by the UWB device in the figure is uploaded to the gateway node through LoRa networking, and finally uploaded to the cloud server through cellular data, completing the application layer's preprocessing of the data and outputting the monitoring point prediction information.

[0182] like Figure 5 As shown, the UWBformer network structure includes a dual-channel and feature fusion architecture, with an improved encoder layer and an improved attention module. Data is preprocessed using 3σ and Kalman filtering to obtain processed distance change information and cumulative distance change information. The coordinate information of the anchor node and the distance change information of each monitoring point are input, and the model predicts the spatial information of the monitoring points. The neural network structure includes, but is not limited to, the UWBformer. The input parameters include, but are not limited to, UWB ranging information and node coordinate information. The input results include, but are not limited to, coordinate information, displacement information, pitch angle, and horizontal angle information.

[0183] The above-mentioned embodiments are described in a relatively detailed and specific manner, expressing preferred embodiments of the present invention. They are only used to illustrate the technical ideas and features of the present invention. Their purpose is to enable those skilled in the art to understand the contents of the present invention and implement them accordingly. However, they are not limited to the present invention alone, and the patent scope of the present invention cannot be limited solely by these embodiments. That is, any equivalent changes or modifications made to the spirit disclosed by the present invention, for researchers or technicians in this field, without departing from the structure of the present invention, local improvements within the system and changes and conversions between subsystems, etc., are still within the patent scope of the present invention.

Claims

1. A slope displacement monitoring method based on UWB, characterized in that: The following steps are involved: Step S1: Equipment deployment: Deploy at least four anchor nodes in the slope stability area. Each anchor node is configured with a dual MCU architecture, where the first MCU (MCU1) is dedicated to UWB ranging and has peripheral interrupts disabled, and the second MCU (MCU2) has communication interfaces for LoRa modules, 5G / 4G modules, WiFi / Bluetooth modules, and OLED modules. Deploy at least one tag node in the landslide monitoring area to form a star or mesh structure with the anchor nodes. Step S2: Data collection, processing, and transmission: MCU1 collects UWB ranging data from the tag node and each anchor node at an adjustable interval of 100 milliseconds to 60 minutes to generate an original distance sequence; MCU2 performs multi-tasking processing, including real-time display of device status, local caching of abnormal data, and multi-channel data transmission; Step S3, displacement prediction and early warning, performs 3σ outlier removal and Kalman filtering on the original distance data, calculates the periodic differential distance and cumulative differential distance; inputs the processed data into the UWBformer neural network model, outputs the moving distance, horizontal angle and pitch angle of the monitoring point; when the spatial displacement exceeds the set threshold of 1-50mm, an alarm message is sent through the cloud to the reserved mobile phone number and email address.

2. The UWB-based slope displacement monitoring method according to claim 1, characterized in that: The UWB ranging in step S2 includes: UWB data acquisition equipment uses TCXO crystal oscillators to compensate for temperature drift errors, and the operating temperature range is -20°C to 60°C; The antenna delay error is eliminated by periodic differential calculation. The calculation formula is: in, is the ranging value of the j-th anchor node in the n-th period.

3. The UWB-based slope displacement monitoring method according to claim 2, characterized in that: The multi-channel data transmission in step S2 includes: Uploaded data is transmitted between nodes through the LoRaMesh network, and the data is aggregated to the LoRa gateway node that supports 4G / 5G communication. The gateway node uploads the collected data to the cloud through the cellular data of the base station; The downlink data is transmitted from the cloud to the LoRa gateway node via the 4G / 5G network. The gateway node then transmits the data to the target device based on the routing table information. On-site debugging parameters are written to the UWB data acquisition device through the communication interface of the WiFi / Bluetooth module.

4. The UWB-based slope displacement monitoring method according to claim 1, characterized in that: The UWBformer neural network model construction in step S3 includes: Dual-channel feature extraction of time domain channel and frequency domain channel. The input of time domain channel is anchor node coordinate information, periodic differential distance and cumulative differential distance. The input of frequency domain channel is periodic differential distance and cumulative differential distance after FFT transformation. Cross attention fusion module, the calculation formula is: where Q time is the time domain feature, K freq 、V freq is the frequency domain feature.

5. The UWB-based slope displacement monitoring method according to claim 4, characterized in that: The cross attention fusion module also includes: The spatial offset term generation module extracts the spatial information matrix S from the cumulative differential distance and obtains it through linear projection Corrected attention score calculation: where Q h , K h 、V h is the query, key, and value vector of the h-th attention head.

6. The UWB-based slope displacement monitoring method according to claim 1, characterized in that: After the early warning signal in step S3 is triggered, the following steps are also included: Start high-density sampling mode and shorten the data collection interval to 100 milliseconds-1 second; Generate a three-dimensional displacement trajectory map and display it through the cloud-based human-computer interaction interface; When the displacement exceeds 120% of the threshold for three consecutive cycles, the emergency response protocol is triggered.

7. The UWB-based slope displacement monitoring method according to claim 1, characterized in that: The method further comprises a model online optimization step: 20% of the real-time data is used as a validation set to calculate the MAE value of the moving distance; When MAE exceeds 4mm for three consecutive times, incremental training is started and the update formula is: Where η is the dynamic learning rate, is the online loss function.

8. The UWB-based slope displacement monitoring method according to claim 7, characterized in that: The incremental training includes: Adopting a sliding window mechanism to retain the historical data of the last 30 days; The newly added data is weighted and the weight distribution formula is: where t i is the data generation time, t current is the current time.

9. The UWB-based slope displacement monitoring method according to claim 1, characterized in that: The method further comprises the step of remote maintenance of the equipment: Receive configuration instructions through an AES-256 encrypted channel and dynamically adjust the UWB communication frequency band and transmission power; When it is detected that the device is offline for more than 2 hours, it will automatically switch to solar power mode and start the self-test program; When connecting to the on-site debugging device via the RS485 interface, the execution of remote configuration commands is suspended.

10. A slope displacement monitoring method based on UWB according to any one of claims 1 to 9, characterized in that: The training method of the UWBformer neural network includes: The weighted root mean square error loss function is used: Using Xavier initialization and Adam optimizer, the learning rate is dynamically adjusted according to the validation loss, and the adjustment range does not exceed ±20% of the initial value.

Images