An open-pit mine multi-element weather monitoring and disaster warning system based on NB-IoT

Abstract

A kind of open-pit mine multi-element weather monitoring and disaster warning system based on NB-IoT is composed of monitoring node and weather data cloud platform.In the process of NB-IoT communication, the monitoring node reads the sampling sequence of radio frequency link parameters, uses weather stable window multiple sampling for statistical smoothing processing, extracts statistical characteristic parameters to determine the terrain type, and corrects the weather data and reports accordingly.The platform constructs a space collaborative terrain correlation model based on the radio frequency link parameters of multiple nodes, corrects the disaster propagation path and generates a collaborative wake-up instruction.The present application uses the existing radio frequency link parameters of standard NB-IoT communication module to realize terrain inversion, reducing the cost of hardware customization.

Description

Technical Field

[0001] This invention relates to the intersection of meteorological monitoring instruments and narrowband wireless communication technology, specifically to a method for adaptive acquisition, compression coding, transmission, and early warning of multi-element meteorological data in the context of NB-IoT narrowband transmission characteristics and complex terrain environments in open-pit mines. Background Technology

[0002] Open-pit mine safety production requires continuous meteorological monitoring. Strong winds, heavy rainfall, and low visibility can easily trigger landslides, transportation accidents, and falls from heights.

[0003] Current mine meteorological monitoring primarily relies on wired networking or 4G public network backhaul. The wired mode depends on RS485 bus and fiber optic ring network; however, the mine terrain dynamically changes with mining progress, requiring frequent cable re-laying, resulting in high maintenance costs, and cabling is impossible at the bottom of deep pits. The 4G public network mode depends on operator base station coverage, but base station density is insufficient in remote mining areas, and signal attenuation is severe in deep pits; 4G modules typically operate at tens of milliamps of current, requiring frequent battery replacements in outdoor scenarios without mains power, making deployment difficult.

[0004] NB-IoT narrowband IoT technology features low power consumption, wide coverage, and deep penetration. Its coverage is about 20 dB better than traditional cellular networks, and its standby current can be reduced to below 5 microamps in PSM power-saving mode, making it suitable for battery-powered outdoor scenarios. However, directly applying NB-IoT to mine meteorological monitoring is not a simple matter of replacing the communication standard. NB-IoT has narrow uplink bandwidth, with a single packet capacity typically below 512 bytes. Traditional JSON text encoding results in low transmission efficiency and high cost per-traffic billing. Fixed-period reporting strategies cannot simultaneously meet the demands of long battery life during periods of no risk and high-frequency monitoring during disasters. The complex terrain of mining areas creates signal blind spots such as deep pits and leeward slopes of spoil heaps, making the reliability of a single operator's network insufficient and leading to the easy loss of monitoring data.

[0005] The existing early warning response chain also has significant shortcomings. Each node operates independently, and downstream nodes remain dormant with low frequency during disaster propagation, only being activated when local meteorological elements exceed limits, resulting in significant monitoring delays. Early warning delivery is mostly achieved through independent alarms from single nodes or push notifications from the platform, with asynchronous alarm times at each node, easily causing confusion on-site. More critically, the existing solution relies on dispatchers manually forwarding early warning information, resulting in a long and delayed delivery chain from the platform to the dispatch system and then to the operational site.

[0006] In existing technologies, NB-IoT communication modules generate radio frequency (RF) link parameters during the attach and data transmission processes, including Reference Signal Received Power (RSRP), Received Signal Strength Indication (RSSI), Reference Signal Received Quality (RSRQ), and Signal-to-Noise Ratio (SNR). These parameters can be read using standard AT commands; for example, the Quectel BC26 module can obtain RSRP and RSRQ using the AT+QENG command and RSSI and SNR using the AT+CSQ command. However, existing mine meteorological monitoring solutions only use these parameters for communication quality assessment and network optimization, without establishing a mapping relationship between RF link parameters and terrain type or meteorological data quality. The main reason for this is that RF link parameters at a single moment are subject to random factors such as dynamic adjustments to base station transmit power, antenna patterns, daily variations in atmospheric refractive index, and temporary electromagnetic interference, resulting in significant fluctuations. Furthermore, dynamic adjustments to base station transmit power can cause systematic drift in RF link parameters that are independent of terrain. Those skilled in the art generally believe that a stable mapping relationship between RF link parameters and terrain type cannot be established; therefore, existing solutions only use them for communication quality assessment. In addition, the existing scheme uses fixed-cycle full reporting, which consumes the same communication resources even if the meteorological elements are stable and unchanged. It does not utilize the spatiotemporal continuity of meteorological data for compression and optimization, resulting in unnecessary traffic overhead. Summary of the Invention

[0007] The purpose of this invention is to provide a multi-element meteorological monitoring and disaster early warning system for open-pit mines based on NB-IoT, which solves the technical problems of efficient backhaul of meteorological data under narrowband transmission of NB-IoT, long endurance of field nodes and high frequency of disaster monitoring, enhanced signal coverage in complex terrain of mining areas, reliable topographic quantification of meteorological data, semantic-level compression of communication resources, synchronous delivery of early warning information, and reuse of radio frequency link parameters in meteorological monitoring.

[0008] To achieve the above objectives, the present invention adopts the following technical solution.

[0009] S1, System Architecture Setup.

[0010] The system comprises multi-element meteorological monitoring nodes deployed in the mining area and a remote meteorological data cloud platform. The monitoring nodes are responsible for seven-element meteorological data acquisition, radio frequency link parameter sampling, statistical smoothing sequence generation, differential semantic compression coding, NB-IoT wireless backhaul, local audible and visual alarm driving, and RTC synchronization control. The meteorological data cloud platform is responsible for data parsing, statistical smoothing sequence assistance, multi-node spatial collaborative terrain association model construction, multi-source fusion, disaster propagation prediction, collaborative wake-up control, early warning generation, and business interface services.

[0011] S2, monitoring node hardware configuration.

[0012] The monitoring node comprises a main control unit, a sensing unit, an NB-IoT communication module, a positioning and timing unit, a storage unit, and a power supply unit, all integrated within an IP67-rated equipment enclosure and mounted on a 2.5-meter galvanized steel pipe pole. The sensing unit covers seven meteorological elements: temperature, humidity, air pressure, wind speed, wind direction, precipitation, and visibility. The main control unit uses an ARM Cortex-M4 microcontroller operating at 3.3 volts. It connects to the temperature, humidity, and pressure sensors via an I2C bus, reads the wind direction magnetic encoder output via a GPIO interface, connects to the ultrasonic anemometer transducer via a timer capture channel, connects to a visibility meter via an RS485 interface, and connects to the NB-IoT communication module via a UART interface. The main control unit reads the RF link parameter sampling sequence through the NB-IoT communication module's standard AT command interface. The NB-IoT communication module supports Band 3, Band 5, and Band 8 frequency bands, has a built-in CoAP protocol stack, and supports PSM power-saving mode and eDRX extended discontinuous reception mode. The battery voltage is stepped down to a 5V bus via a DC-DC converter. The module's power supply voltage is 3.8V, provided by an independent LDO stepping down from the 5V bus. A large-capacity energy storage capacitor is connected in parallel at the power supply end to absorb current spikes during RF transmission. The power unit uses a 20W 18V monocrystalline silicon solar panel in conjunction with an MPPT controller to charge a 12.8V 30Ah lithium iron phosphate battery. The MPPT controller uses a perturbation-observation method to track the maximum power point, performing a voltage perturbation and power comparison every 30 seconds. The battery is equipped with overcharge protection, over-discharge protection, and temperature protection.

[0013] S3 is a low-power architecture deployment that interleaves RF link parameter sampling and meteorological sampling timing.

[0014] During each NB-IoT uplink transmission, the monitoring node's main control unit reads the RF link parameter sampling sequence through the standard AT command interface of the NB-IoT communication module. RF link parameter sampling and meteorological sampling employ a time-interleaved architecture. After RTC wake-up, the main control unit first powers on the meteorological sensor and performs seven-element sampling. The main control unit packages the meteorological data into a CoAP message and sends it to the NB-IoT communication module via UART. The NB-IoT communication module automatically updates the RF link parameters during network interactions in uplink transmission transactions. The main control unit is in a waiting-for-interrupt state during module RF transmission, rather than running at full speed. After module transmission is complete, it notifies the main control unit via a UART interrupt, and the main control unit reads the RF link parameter sampling data from the module interface. The main control unit jointly encapsulates and reports the meteorological data and RF link parameter sampling data. The main control unit then shuts down the sensor and module, calculates the next wake-up time, and enters STOP mode. RF link parameter sampling is fully embedded in the existing NB-IoT communication interaction process without triggering additional transmissions; the main control unit is in a waiting-for-interrupt state during RF transmission, achieving task-level power sharing.

[0015] S4, NB-IoT RF link parameter statistical smoothing sequence generation.

[0016] Nodes utilize multiple transmissions within a meteorological stability window to perform statistical smoothing on the RF link parameter sequence. A node determines it is currently within a meteorological stability window if consecutive sampling of wind speed is below the wind speed stability threshold and wind direction change is below the wind direction change threshold, marking this period as the RF link parameter statistical period. Upon entering a meteorological stability window, the node temporarily shortens its reporting cycle, completing a set number of data acquisitions and transmissions within the window. During the statistical period, the near-surface scatterer change rate is below the environmental change threshold, and the main contributor to RF link parameter fluctuations shifts from environmental scatterer changes to system noise. The platform performs synchronization analysis on the RF link parameter sequences of multiple nodes under the same base station coverage. If multiple nodes exhibit a systematic shift in the same direction and amplitude of RSRP within the same time period, it is determined that the shift is caused by base station power control. The platform issues a statistical period removal instruction to the multiple nodes. The nodes cache the sampling data within the statistical period. Upon receiving the statistical period removal instruction from the platform, or if no removal instruction is received after the waiting period expires, the sampling data for the corresponding time period is removed from the statistical period, retaining only the sampling sequence dominated by terrain scattering characteristics for statistical smoothing. During the statistical period, each time the node performs an uplink transmission in NB-IoT, it reads the RF link parameter sampling sequence from the module interface. The node performs extremum filtering on each sampling sequence to remove abnormal sampling points caused by sudden electromagnetic interference. The node then performs point-by-point weighted smoothing on the filtered multiple RF link parameter sampling sequences, outputting a statistically smoothed RF link parameter sequence. This statistical smoothing process reduces the variance of random noise through multiple sampling fusion, ensuring that the stability of the RF link parameter estimation meets the requirements of terrain inversion. The node extracts the parameter fluctuation range, mean level index, and parameter sequence fluctuation entropy from the statistically smoothed sequence for terrain inversion. The parameter fluctuation range is the difference between the maximum and minimum RSRP values ​​during the statistical period; the mean level index is the ratio of the mean RSRP to the reference mean of the open platform during the statistical period; and the parameter sequence fluctuation entropy is the information entropy of the amplitude distribution of the RF link parameter sequence during the statistical period. The sequence amplitude interval is evenly divided into a set number of sub-intervals, and the percentage of sampling points falling into each sub-interval is calculated using the information entropy formula. The entropy value is positively correlated with the terrain scattering complexity.

[0017] S5, terrain inversion and adaptive correction of meteorological data based on radio frequency link parameters.

[0018] A mapping model between RF link parameters and meteorological data quality is established. During each NB-IoT data transmission or network attach process, the node reads the RF link parameter sampling sequence from the module interface. The node locally calculates the RF link statistical characteristic parameters. The parameter fluctuation range reflects the terrain scattering complexity, the mean level index reflects the proportion of direct sunlight, and the parameter sequence fluctuation entropy reflects the degree of disorder in the scattering environment. The node inputs the parameter fluctuation range, mean level index, and parameter sequence fluctuation entropy into a pre-calibrated terrain classification decision model to determine the local terrain type. The determination priority is: first, determine the deeply recessed, heavily shaded areas; second, determine the leeward slope area of ​​the spoil heap; and finally, determine the open platform. For samples with ambiguous boundaries, they are marked as transition areas and interpolation correction is applied. The wind speed correction coefficient for deeply recessed, heavily shaded areas is obtained through field calibration, restoring the elevated measurements under the funnel effect to the equivalent open wind speed; the wind direction offset angle and temperature compensation value are obtained through field calibration to restore the equivalent open platform reference temperature. The wind speed correction coefficient for the leeward slope area of ​​the spoil heap was obtained through on-site calibration. Wind direction was decomposed into parallel and vertical slope components, each with its own correction coefficient obtained through on-site calibration. These components were then combined and superimposed with the leeward vortex offset angle. Nodes encapsulated and reported the corrected meteorological data, terrain labels, and radio frequency link statistical characteristic parameters. When the platform received data from each node, it extracted the radio frequency link statistical characteristic parameters and terrain labels, using the terrain labels as a crucial input for spatial interpolation confidence weights. For nodes with deeply concave, heavily shaded terrain or on the leeward slope of the spoil heap, the weight of wind speed and direction data in the fusion process was further reduced compared to conventional methods. When generating the fusion results, the platform labeled the data confidence level of each grid, which was used to adjust the confidence interval of the early warning rule engine. The platform continuously tracked the changing trends of the radio frequency link statistical characteristic parameters of each node. If the parameter fluctuation range showed a trend of increasing, it indicated a change in the surrounding terrain, triggering an update of the platform's terrain database.

[0019] S6, Multi-node RF link parameter spatial collaborative terrain association model construction.

[0020] The platform collects statistically smoothed radio frequency link parameter sequences from all nodes in the mining area to construct a spatial collaborative terrain association model. The platform establishes a node spatial topology graph, with nodes as vertices and node visibility as edge weights. The platform extracts the cross-correlation function of the radio frequency link parameter time series within the same meteorological stability window only for adjacent node pairs with a distance less than a distance threshold. The zero-latency cross-correlation coefficient is used as a measure of terrain connectivity, reflecting the similarity of the scattering environment between node pairs. If the zero-latency cross-correlation coefficient is higher than the connectivity threshold, the nodes are considered to be openly connected; if the zero-latency cross-correlation coefficient is lower than the occlusion threshold, significant terrain occlusion exists between nodes. Based on the cross-correlation matrix of all node pairs, the platform divides terrain blocks using a graph cut algorithm. The graph cut algorithm maps the node topology graph to a weighted undirected graph, using the elements of the cross-correlation matrix as the weights of the edges between nodes, and the spatial difference of the statistical feature vectors of the node's radio frequency link parameters as node attributes. Terrain block division is achieved by minimizing the sum of cross-block edge weights and maximizing intra-cluster density. The cross-block cost term is cut(A,B)=Σw(i,j), i∈A, j∈B, and the intra-cluster density term is assoc(A,V)=Σw(i,j), i∈A, j∈V. The objective function is to minimize cut(A,B) / assoc(A,V)+cut(A,B) / assoc(B,V). The optimal partition is solved through iterative optimization. The iteration terminates when the change in the number of nodes within a block is less than a node number threshold or the decrease in the objective function is less than a function decrease threshold. The platform calculates the distance from each node to the base station based on the node's location coordinates and the base station coordinates. It performs log-linear regression on the statistically smoothed RSRP mean and logarithmic distance of nodes within each block to obtain the attenuation rate exponent, which characterizes the density distribution of terrain scatterers within the block. The platform performs joint estimation of radio frequency link parameters for nodes within the same block, using spatial smoothing to suppress independent noise and outputting representative radio frequency link parameters at the block level for terrain correction of meteorological data within that block.

[0021] S7, adaptive differential semantic compression driven by RF link parameter stability.

[0022] The node uses the stability of radio frequency link parameters as the second control parameter in differential compression, and makes joint decisions with meteorological element differentials. The node locally maintains the baseline values ​​of each element from the last successfully reported data, as well as the radio frequency link parameter feature vector at the time of the last report. After each sampling, the node calculates the deviation between the current radio frequency link parameter feature vector and the feature vector from the last report. The deviation is the ratio of the weighted sum of squares of the differences between the feature vector components to the weighted sum of squares of the feature vector components from the last report. If the deviation is lower than the propagation environment deviation threshold, the propagation environment is considered stable, and the meteorological element differential thresholds are relaxed. The temperature differential threshold is relaxed from the standard temperature threshold to a more lenient temperature threshold, and the wind speed differential threshold is relaxed from the standard wind speed threshold to a more lenient wind speed threshold. If the deviation is between the propagation environment deviation threshold and the environmental change threshold, the standard meteorological element differential thresholds are maintained. If the deviation is higher than the environmental change threshold, the propagation environment is considered to have changed, and the differential thresholds for each meteorological element are tightened by a fixed proportion, requiring immediate reporting even for minor changes in meteorological elements. If the node is in a disaster warning state, the radio frequency link parameter stability assessment is frozen, and full high-frequency reporting is forced. During periods of stable weather, if the differences of all elements are less than the relaxed threshold, nodes only report heartbeat frames containing timestamps, or choose not to report. The platform extrapolates based on the previous baseline value and the physical continuity of meteorological elements. When the difference of any element exceeds the threshold, or the node enters a disaster warning state, or receives a baseline value reset command, the node automatically switches to full high-frequency reporting. The platform sends a baseline value confirmation command to the node every 6 hours, and the node replies with the current full data as the new baseline to prevent long-term accumulation of differences and errors.

[0023] S8, Disaster propagation path correction based on radio frequency link parameter terrain connectivity.

[0024] The platform utilizes a spatial collaborative terrain association model to correct disaster propagation paths. Upon receiving an emergency frame from the first node, the platform extracts the disaster type and initial propagation direction. It then queries the node spatial topology map, searches for downstream nodes along the initial direction, and determines terrain connectivity based on the cross-correlation of radio frequency link parameters between node pairs. If terrain obstruction exists between node pairs, the platform reduces the probability of disaster propagation along that path and searches for alternative routes. The platform calculates the propagation speed after terrain connectivity correction and only sends collaborative wake-up commands to downstream nodes determined to be terrain-connected.

[0025] S9 is an NB-IoT node collaborative wake-up mechanism based on the propagation characteristics of meteorological disaster chains.

[0026] A disaster propagation prediction engine is deployed on the meteorological data cloud platform, and a node collaborative wake-up mechanism is established by combining the downlink communication capabilities of NB-IoT. When the first node detects a meteorological anomaly, such as wind speed exceeding a preset threshold within a certain time period, it immediately enters a disaster warning state and reports a customized high-priority emergency message to the platform. Upon receiving the emergency message from the first node, the platform extracts the wind speed, wind direction, and air pressure data, combines them with the node's GPS coordinates and the mine area's DEM elevation data, and runs a disaster propagation model. The platform calculates the disaster propagation velocity vector. For strong winds or sandstorms, the horizontal component of the propagation velocity vector is composed of the measured wind speed projection and a terrain acceleration term. The terrain acceleration term is multiplied by a corresponding acceleration coefficient based on the different terrain types of the mine pit pass or platform. Based on the propagation velocity vector, the platform calculates the set of downstream nodes along the disaster propagation path and the estimated arrival time for each node. Nodes in normal monitoring mode periodically open their downlink receiving windows in eDRX mode to listen for downlink commands from the platform. The platform sends collaborative wake-up commands to nodes in the downstream node set whose estimated arrival time is less than the disaster arrival time threshold via NB-IoT node-by-node downlink control messages. The command includes the disaster type, estimated arrival time in seconds, and duration of continuous monitoring. Upon receiving the coordinated wake-up command, downstream nodes immediately enter disaster warning mode, even if local meteorological elements have not yet exceeded limits. They increase the sampling frequency to a preset high frequency, and if local limits are not exceeded after the continuous monitoring period expires, they revert to normal monitoring mode or deep dormancy mode. The platform continuously tracks the spread of the disaster, dynamically updates the set of downstream nodes and estimated arrival times, and sends wake-up commands to newly entered threat areas.

[0027] S10, a method for synchronizing distributed audible and visual alarms in mining areas based on NB-IoT.

[0028] Leveraging the downlink communication capabilities of NB-IoT, synchronized alarms are achieved across all nodes in a specific area. After generating an alert, the platform identifies the affected area and queries a list of all NB-IoT nodes within that area. The platform then sequentially sends synchronized alarm commands via node-by-node downlink control messages. Each command includes an alarm identifier, alarm type, level, unified trigger time, and duration. Nodes periodically calibrate their local RTCs using their positioning and timing units to maintain synchronization with the platform clock. Upon receiving the synchronized alarm command, each node parses the unified trigger time and calculates the time difference between its local RTC and the unified trigger time. The node sets an RTC alarm to trigger after the time difference, simultaneously driving a GPIO output to a high level to control the external audible and visual alarm to sound and flash. If a node receives a cancellation command before the unified trigger time, the alarm is canceled. If no cancellation command is received, the alarm continues until the duration expires. During the alarm period, nodes maintain a disaster warning state, continuously monitoring and reporting, achieving near-synchronous alarms across the area.

[0029] S11, Dynamic Threshold Warning Rule Engine.

[0030] The early warning rule structure is as follows: if the element conditions, spatiotemporal conditions, and auxiliary conditions are all met, the early warning level, affected area, and suggested measures are output. Dynamic thresholds are calculated based on a multi-dimensional coupling of season, terrain, and operational status. The seasonal correction coefficient is determined based on the current month: winter season coefficient for winter, summer season coefficient for summer, and transitional season coefficient for other periods. The terrain correction coefficient is determined based on the terrain type of the early warning target grid: mining platform terrain coefficient, spoil heap terrain coefficient, and transport pass terrain coefficient. The operational status coupling coefficient is determined based on the currently active operational status fed back by the mine scheduling system: regular operation coefficient, blasting operation coefficient, hoisting operation coefficient, and emergency operation coefficient. If multiple statuses are active simultaneously, the product of the coefficients of each active status is taken; if the product is lower than the lower threshold, it is truncated to the lower threshold. The platform distributes the calculated dynamic thresholds to each node, and the nodes replace their local static thresholds with the dynamic thresholds for state machine switching determination.

[0031] Compared with the prior art, the present invention has the following beneficial effects.

[0032] This invention extends the application of radio frequency (RF) link parameters from communication quality assessment to terrain inversion by using meteorological stability window filtering and statistical smoothing. It eliminates the need for customized firmware or additional RF front-ends, reducing hardware costs and deployment barriers for monitoring nodes. Through multiple sampling and statistical averaging within the meteorological stability window, random fluctuations in RF link parameters are effectively suppressed. Furthermore, the platform identifies and eliminates systematic offsets caused by base station power control by identifying and eliminating synchronous changes across multiple nodes, providing stable input for terrain classification. Based on this, the monitoring node utilizes the stability characteristics of the RF link parameters to drive adaptive differential compression. This reduces communication traffic when the propagation environment is stable and triggers encrypted monitoring in advance during sudden changes, balancing long endurance and disaster response sensitivity.

[0033] The platform-side spatial collaborative terrain association model utilizes a distributed node network to achieve blind zone complementarity and automatic terrain block division, offering higher reliability than single-node judgment. Disaster propagation path correction based on radio frequency link parameter terrain connectivity reflects dynamic terrain changes caused by temporary engineering activities, avoiding false wake-ups to nodes behind obstructions. The collaborative wake-up mechanism enables downstream nodes to enter high-frequency monitoring mode minutes before disaster arrival, significantly earlier than the local over-limit wake-up mode. The synchronous alarm method achieves quasi-synchronous triggering of regional nodes through NB-IoT node-by-node downlink control messages, eliminating alarm time lag caused by node-by-node push. The dynamic threshold engine couples seasonality, terrain, and operational status in calculations, adapting early warning rules to mine production plans.

[0034] By tracking the changing trends of statistical characteristic parameters of the radio frequency link over a long period of time, the dynamic monitoring of the terrain in the mining area and the adaptive updating of meteorological correction parameters can be achieved. Attached Figure Description

[0035] Figure 1 This is a schematic diagram of the overall system architecture.

[0036] Figure 2 This is a schematic diagram of the hardware structure of the monitoring node.

[0037] Figure 3 Timing diagram for a low-power architecture that interleaves RF link parameter sampling and meteorological sampling timing.

[0038] Figure 4 This is a flowchart of terrain inversion and adaptive correction of meteorological data based on radio frequency link parameters.

[0039] Figure 5 A flowchart of adaptive differential semantic compression driven by RF link parameter stability.

[0040] Figure 6 A flowchart for correcting disaster propagation paths based on terrain connectivity parameters of radio frequency links.

[0041] Figure 7 This is a flowchart of the NB-IoT node collaborative wake-up mechanism based on the propagation characteristics of meteorological disaster chains. Detailed Implementation

[0042] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings. Figure 1 As shown, the system includes a multi-element meteorological monitoring node 100 deployed in the mining area and a meteorological data cloud platform 200. The monitoring node 100 communicates with the meteorological data cloud platform 200 through an NB-IoT base station 300. This embodiment uses an NB-IoT communication module that supports the standard AT command interface, such as Quectel BC26 or a similar model. RSRP and RSRQ are read using the AT+QENG command, and RSSI and SNR are read using the AT+CSQ command.

[0043] Example 1: Node hardware implementation.

[0044] like Figure 2As shown, monitoring node 100 includes a main control unit 1, an NB-IoT communication module 2, a meteorological sensor unit 3, a positioning and timing unit 4, a storage unit 5, and a power supply unit 6. The main control unit 1 uses an ARM Cortex-M4 microcontroller with an 80 MHz operating frequency. Temperature, humidity, and pressure sensors are connected via an I2C bus with a 100 kHz bus speed. Sensor power is controlled by the main control GPIO, and is pulled low to power off during sleep mode. The 16-channel output of the wind direction magnetic encoder is filtered by RC and then connected to the GPIO for software-timed scanning and reading. The ultrasonic transducer for wind speed is connected to a timer capture channel, measuring the propagation time difference between headwind and tailwind to calculate wind speed. The tipping bucket rain gauge pulse is connected to an external interrupt pin, configured for falling edge triggering. Software debouncing is performed in the interrupt service routine to confirm valid tipping. The visibility meter's RS485 interface is connected to the main control UART via a transceiver. The transceiver control terminal is connected to the GPIO, with a high level before transmission and a low level after transmission. The communication protocol uses Modbus RTU. The battery voltage is stepped down to a 5V bus via a DC-DC converter. The NB-IoT module is powered by 3.8V, provided by an independent LDO stepping down from the 5V bus. A 470µF tantalum capacitor and a 100µF ceramic capacitor are connected in parallel at the power supply end. The module's PSM_EINT pin is connected to the main control GPIO, and a low-level pulse wakes it up. The RESET pin is connected to the main control GPIO, and a low-level pulse resets it. The 20W 18V solar panel charges the 12.8V 30Ah lithium iron phosphate battery via the MPPT controller. The MPPT performs a voltage perturbation and power comparison every 30 seconds to track the maximum power point. The battery is equipped with overcharge protection at 14.8V, over-discharge protection at 10.8V, and temperature protection at 60°C and -20°C.

[0045] Example 2: Implementation of a time-interleaved architecture.

[0046] like Figure 3 As shown, after the RTC timed wake-up, the main control unit first powers on the meteorological sensor and performs seven-element sampling, which takes approximately 500 milliseconds. The main control unit packages the meteorological data into a CoAP message and sends it to the NB-IoT module via UART. During the network interaction of uplink transmission transactions, the module automatically updates the RF link parameters, and the main control unit enters the WFI wait-for-interrupt state during this period. After the module completes transmission, it wakes up the main control unit via a UART interrupt, and the main control unit reads the RF link parameter sampling data from the module interface. The main control unit encapsulates the meteorological data and RF link parameter data together and reports them as a single frame. Subsequently, it shuts down the sensor and the module, calculates the next wake-up time, and enters STOP mode.

[0047] Example 3: Implementation of Statistical Smoothing Sequence Generation.

[0048] A meteorological stability window is defined as three consecutive sampling periods where the wind speed is below 1 m / s, there is no precipitation, and visibility is above the visibility threshold. The wind speed stability threshold is 1 m / s, and the wind direction change threshold is 15 degrees. These thresholds are determined based on the statistical characteristics of near-surface wind turbulence intensity in the mining area; that is, when the wind speed is below these thresholds and the wind direction change is less than these thresholds, the impact of near-surface scatterer motion on RF link parameters is negligible. After entering the meteorological stability window, the node temporarily shortens its reporting cycle from the regular reporting cycle to the window reporting cycle. The regular reporting cycle is 30 to 60 minutes, and the window reporting cycle is 2 to 5 minutes, completing 16 data acquisitions and transmissions within the window period. Within this window, the near-surface scatterer change rate is below the environmental change threshold, and the main contributor to RF link parameter fluctuations shifts from environmental scatterer changes to system noise. The platform performs synchronization analysis on the RF link parameter sequences of multiple nodes under the coverage of the same base station. If the RSRP of multiple nodes shows a systematic shift in the same direction and amplitude during the same time period, it is determined that the shift during that time period is caused by the base station power control. The platform sends a statistical period removal instruction to each node. The node caches the sampling data of each time within the statistical period. After receiving the statistical period removal instruction from the platform, or if it does not receive the removal instruction after the waiting time expires, it removes the sampling data of the corresponding time period from the statistical period. During each NB-IoT uplink transmission, the master controller reads the RF link parameter sampling sequence from the module interface. The master controller performs extreme value filtering on each sequence to remove abnormal sampling points. The master controller performs point-by-point weighted smoothing on the 16 filtered sequences and outputs the statistically smoothed RF link parameter sequence. The master controller extracts the parameter fluctuation range, mean level index, and parameter sequence fluctuation entropy from the sequence for subsequent terrain inversion.

[0049] Example 4: Implementation of terrain classification decision model calibration, terrain inversion, and correction.

[0050] The calibration experiment selected three standard locations each in the mining area: an open platform, a deep pit, and the leeward slope of the spoil heap. At each location, a monitoring node and a calibrated portable weather station were deployed as true references, with continuous data collection for at least 72 hours. Using the portable weather station measurements as the dependent variable and the original node measurements as the independent variable, correction coefficients for each terrain type were obtained through least squares fitting. During the calibration period, base station power adjustment events were recorded. Data from periods of power adjustment were removed before refitting to ensure that the correction coefficients only reflected the influence of terrain.

[0051] like Figure 4As shown, the nodes input the parameter fluctuation range, mean level index, and parameter sequence fluctuation entropy into the terrain classification decision model. The decision priority is as follows: First, determine the deep concave strong shading area. If the parameter fluctuation range is greater than the deep concave fluctuation threshold, the mean level index is less than the deep concave mean threshold, and the parameter sequence fluctuation entropy is greater than the deep concave entropy threshold, then it is marked as a deep concave strong shading area. Second, determine the leeward slope area of ​​the spoil heap. If the parameter fluctuation range is greater than the leeward slope fluctuation threshold, the mean level index is within the leeward slope mean range, and the parameter sequence fluctuation entropy is greater than the leeward slope entropy threshold, then it is marked as a spoil heap leeward slope. Finally, determine the open platform. If the parameter fluctuation range is less than the open platform fluctuation threshold, the mean level index is greater than the open platform mean threshold, and the parameter sequence fluctuation entropy is less than the open platform entropy threshold, then it is marked as an open platform. For boundary samples that do not meet any of the above conditions, they are marked as transition areas and interpolation correction is applied. In the deeply recessed, heavily shaded area, the wind speed is multiplied by a correction factor obtained through field calibration to restore the elevated measurement value under the funnel effect to the equivalent open wind speed. The wind direction is superimposed with the offset angle obtained through field calibration, and temperature compensation uses the compensation value obtained through field calibration to restore the equivalent open platform reference temperature. In the leeward slope area of ​​the spoil heap, the wind speed is multiplied by a correction factor obtained through field calibration. The wind direction is decomposed into parallel slope aspect components and vertical slope aspect components, which are then multiplied by correction factors obtained through field calibration, and finally superimposed with the leeward eddy offset angle. The node encapsulates and reports the corrected meteorological data, terrain labels, and RF link statistical characteristic parameters together.

[0052] Example 5: Implementation of the spatial collaborative terrain association model.

[0053] After receiving data from each node, the platform extracts the statistically smoothed RF link parameter sequence. The platform establishes a node spatial topology graph and calculates the cross-correlation function of the RF link parameter time series within the same meteorological stability window only for adjacent node pairs with a distance threshold. If the zero-latency cross-correlation coefficient is higher than the connectivity threshold, it is considered open connectivity; if the zero-latency cross-correlation coefficient is lower than the occlusion threshold, terrain occlusion is identified. Based on the cross-correlation matrix, the platform divides terrain blocks using a graph cut algorithm. It calculates the distance from each node to the base station based on the node's location coordinates and the base station coordinates. Logarithmic linear regression is performed on the statistically smoothed RSRP mean and logarithmic distance of each node to obtain the attenuation rate exponent. The platform performs joint estimation for nodes within the same block and outputs representative RF link parameters at the block level.

[0054] Example 6: Adaptive Differential Compression Implementation.

[0055] like Figure 5As shown, each node locally maintains a register of seven meteorological element baseline values ​​and a feature vector of the previous radio frequency link parameters. After each sampling, the deviation between the current feature vector and the previous feature vector is calculated. The deviation is the ratio of the weighted sum of squares of the differences between the components of the feature vector to the weighted sum of squares of the components of the previous feature vector. If the deviation is less than the propagation environment deviation threshold, the meteorological element difference threshold is relaxed. If the deviation is between the propagation environment deviation threshold and the environmental change threshold, the conventional meteorological element difference threshold is maintained. If the deviation is greater than the environmental change threshold, the difference thresholds of each meteorological element are tightened by a fixed proportion. The propagation environment deviation threshold is set to 5%, and the environmental change threshold is set to 20%, which are determined based on the statistical fluctuation range of the radio frequency link parameters within the meteorological stability window. The fixed proportion is 50%, meaning that the difference thresholds of all elements are tightened to half of the original thresholds. If the system is in a disaster warning state, the judgment is frozen, and full high-frequency reporting is forced. During meteorologically stable periods, if the differences of all elements are less than the relaxed thresholds, the node only sends heartbeat frames. The platform calculates the current value based on the previous baseline value and the physical continuity of the elements. If the difference of any element exceeds the threshold, the node immediately reports all data and updates the baseline value.

[0056] Example 7: Implementation of Disaster Propagation Path Correction.

[0057] like Figure 6 As shown, after receiving an emergency frame from the first node, the platform extracts the disaster type and initial propagation direction. The platform queries the node spatial topology map, searches for downstream nodes along the initial direction, and determines terrain connectivity based on the cross-correlation of radio frequency link parameters between node pairs. If terrain obstruction exists, the platform reduces the propagation probability of that path and searches for alternative routes. The platform calculates the propagation speed after terrain connectivity correction and only sends coordinated wake-up commands to downstream nodes determined to be terrain-connected.

[0058] Example 8: Collaborative wake-up implementation.

[0059] like Figure 7 As shown, the platform sends a coordinated wake-up command to downstream nodes whose estimated arrival time is less than the disaster arrival time threshold. The disaster arrival time threshold is set at 30 minutes, determined based on the average propagation speed of strong wind disasters in the mining area and the distance between nodes, ensuring that downstream nodes have sufficient time to enter high-frequency monitoring mode before the disaster arrives. Upon receiving the command, the downstream nodes parse the estimated arrival time in seconds and the continuous monitoring duration, immediately enter disaster warning mode, and shorten the sampling period to the preset high-frequency sampling period. If local elements exceed the limit during continuous monitoring, the disaster warning mode is maintained and the high-frequency monitoring and reporting process is executed. If the limit expires and is not exceeded, the module is shut down, returning to the normal monitoring mode or deep dormancy mode. The platform reruns the propagation model every 5 minutes to update the downstream node set and estimated arrival time.

[0060] Example 9: Implementation of synchronous alarm and dynamic threshold.

[0061] The platform sequentially sends synchronous alarm commands via node-by-node downlink control messages, with a unified trigger time accurate to the second. Each node parses the unified trigger time, calculates the time difference with its local RTC, and sets an RTC alarm. When the alarm is triggered, the GPIO outputs a high level, and an external 12V audible and visual alarm sounds and flashes. If a clear command is received, the GPIO is pulled low, and the alarm stops. If no clear command is received, the alarm continues until the set duration expires. The platform reads the next day's production plan uploaded by the mine scheduling system at midnight each day, parsing the job type and time period. The early warning engine calculates a dynamic threshold every hour. The seasonal correction factor is determined based on the current month: 0.9 for winter, 1.1 for summer, and 1.0 for other periods. The terrain correction factor is determined based on the terrain type of the early warning target grid: 1.0 for mining platforms, 0.85 for spoil heaps, and 1.15 for transport passes. The job state coupling coefficient is determined based on the currently active job state fed back by the scheduling system. It is set to 1.0 for routine operations, 0.8 for blasting, 0.7 for hoisting, and 0.6 for emergency operations. If multiple states are active simultaneously, the product of the coefficients for each active state is taken; if the product is less than 0.6, it is truncated to 0.6. The platform distributes the calculated dynamic threshold to each node, and the nodes replace their local static threshold with the dynamic threshold for state machine switching determination.

[0062] Example 10: Experimental verification.

[0063] To verify the technical effectiveness of this invention, the following tests were conducted in actual mining area deployment. **Time-interleaved power consumption test:** Under the same battery capacity and solar energy configuration, compared with the independent triggering architecture, continuous monitoring for 72 hours showed that the experimental group's additional power consumption was significantly lower than that of the independent triggering architecture. **Statistical smoothing sequence stability test:** One node was deployed in both the open platform and the deep pit, continuously collecting RF link parameters for 72 hours. The standard deviation of the fluctuation range of single transmission parameters was large, but after multiple statistical smoothings, the standard deviation significantly decreased, and the reduction was in line with statistical theory expectations. **Terrain inversion correction accuracy test:** Three standard points were selected in the mining area: an open platform, a deep pit, and the leeward slope of the spoil heap. Monitoring nodes and a portable weather station were deployed for true reference. Continuous observation for 30 days showed that the average wind speed deviation in the uncorrected deep pit area was large, but the deviation was significantly reduced after terrain inversion correction using RF link parameters. **Multi-node spatial collaborative terrain division accuracy test:** Six nodes were deployed in the mining area. The automatic terrain division results based on the cross-correlation of RF link parameters were compared with manual field surveys, showing good agreement. In the fusion accuracy test, portable weather stations were deployed within the mining area grid as reference values. The control group showed a larger root mean square error in wind speed fusion, while the experimental group showed a significant reduction in error. In the differential compression flow test, continuous observations were conducted for 72 hours during a period of stable weather. The control group showed a higher average flow per node, while the experimental group showed a significant reduction in flow. In the disaster propagation path correction test, five nodes were deployed in the mining area to simulate the propagation of a strong wind disaster. The control group had a higher false wake-up rate, while the experimental group showed a significantly lower false wake-up rate. In the collaborative wake-up lead time test, the time difference between downstream nodes entering a disaster warning state and the actual arrival of the disaster was recorded. The experimental group achieved a lead time of several minutes. In the alarm synchronization accuracy test, three monitoring nodes were deployed in the affected area. The start-up time difference between nodes in the experimental group was within an acceptable range. In the endurance test, under the same battery capacity and solar power configuration, the control group had an endurance of 9 to 11 months, while the experimental group had a significantly longer endurance.

Claims

1. A multi-element meteorological monitoring and disaster early warning system for open-pit mines based on NB-IoT, characterized in that, include: A multi-element meteorological monitoring node, the node including a main control unit, an NB-IoT communication module, a meteorological sensing unit, a positioning and timing unit, a storage unit and a power supply unit; The NB-IoT communication module is equipped with a standard AT command interface, which is used to output the radio frequency link parameter sampling sequence during communication. The main control unit is connected to the NB-IoT communication module and the meteorological sensing unit respectively, and is configured to read the sampling sequence of the radio frequency link parameters, perform statistical smoothing on the multiple collected sequences within the meteorological stability window, determine the local terrain type based on the statistical feature parameters of the processed sequence, adaptively correct the meteorological measurement data collected by the meteorological sensing unit according to the terrain type, and transmit the corrected meteorological data and terrain label back through the NB-IoT communication module. The meteorological data cloud platform is connected to the monitoring node to receive the corrected meteorological data and terrain labels, construct a spatial collaborative terrain association model based on multi-node radio frequency link parameters, correct the disaster propagation path, and generate a collaborative wake-up command.

2. A method for multi-element meteorological monitoring and disaster early warning in open-pit mines based on NB-IoT, characterized in that, include: During NB-IoT communication, the monitoring node reads the radio frequency link parameter sampling sequence through the standard AT command interface of the NB-IoT communication module. Within a stable meteorological window, the sampling sequences of the radio frequency link parameters collected multiple times are statistically smoothed to obtain a statistically smoothed radio frequency link parameter sequence. Based on the statistical characteristic parameters of the statistically smoothed radio frequency link parameter sequence, the local terrain type is determined; The meteorological measurement data are adaptively corrected according to the terrain type. The corrected meteorological data and terrain labels are transmitted back to the meteorological data cloud platform via NB-IoT.

3. The system according to claim 1, characterized in that, The main control unit is connected to the temperature, humidity and pressure sensor via I2C bus, reads the wind direction magnetic encoder output via GPIO interface, connects to the ultrasonic anemometer transducer via timer capture channel, connects to the visibility meter via RS485 interface, and connects to the NB-IoT communication module via UART interface.

4. The system according to claim 1 or 3, characterized in that, The monitoring node uses a time-interleaved architecture to collect the RF link parameter sampling sequence and meteorological data. During a single RTC wake-up cycle, the main control unit first performs meteorological sensor sampling and then sends the data to the NB-IoT module. The module automatically updates the RF link parameters during the network interaction of uplink transmission transactions. After transmission is completed, the main control unit reads the RF link parameter sampling data, and jointly encapsulates and reports it.

5. The method according to claim 2, characterized in that, The radio frequency link parameter sampling sequence includes at least one of the following: reference signal received power RSRP, received signal strength indication RSSI, reference signal received quality RSRQ, and signal-to-noise ratio SNR. The statistical smoothing process includes: performing extremum filtering and weighted smoothing on the multiple-acquired RF link parameter sampling sequences.

6. The method according to claim 2 or 5, characterized in that, The criteria for determining the meteorological stability window are: the wind speed is lower than the wind speed stability threshold and the wind direction change is lower than the wind direction change threshold for multiple consecutive samplings. The statistical characteristic parameters include parameter fluctuation range, mean level index, and parameter sequence fluctuation entropy; The terrain types include open platforms, deep, well-shaded areas, and leeward slopes of spoil heaps.

7. The method according to claim 6, characterized in that, The meteorological data cloud platform collects statistically smoothed radio frequency link parameter sequences from all nodes in the mining area to construct a spatial collaborative terrain association model. Establish a node spatial topology graph, extract the cross-correlation function of the time series of radio frequency link parameters within the same meteorological stability window only for adjacent node pairs with a spacing less than the distance threshold, divide the terrain into blocks based on the cross-correlation matrix using a graph cut algorithm, perform joint estimation on the nodes within the same block, and output the block-level representative radio frequency link parameters.

8. The method according to claim 6, characterized in that, The monitoring node uses the stability of radio frequency link parameters to control the differential compression threshold of meteorological data: calculate the deviation of the current radio frequency link parameter feature vector from the last reported value, where the deviation is the ratio of the weighted sum of squares of the differences of each component of the feature vector to the weighted sum of squares of each component of the feature vector at the time of the last reported value; If the deviation is lower than the propagation environment deviation threshold, the meteorological element difference threshold is relaxed; if the deviation is higher than the environmental change threshold, the difference thresholds of each meteorological element are tightened by a fixed proportion.

9. The method according to claim 7, characterized in that, The meteorological data cloud platform constructs a terrain connectivity model based on the terrain type and corrects the disaster propagation path: it queries the node spatial topology map, searches for downstream nodes along the initial propagation direction, determines terrain connectivity based on the cross-correlation of radio frequency link parameters between node pairs, and if there is terrain obstruction, it reduces the probability of disaster propagation along the path and searches for detour paths, and only sends a collaborative wake-up command to downstream nodes that are determined to be terrain connected. The collaborative wake-up command is sent through NB-IoT downlink control messages to each node. After receiving the command, the downstream node immediately enters the disaster warning state even if the local meteorological elements have not exceeded the limit, increases the sampling frequency, and continuously monitors for the set duration.

Images