A method and system for intelligent deployment of hydrological monitoring station network based on Internet of Things

Abstract

This invention relates to an intelligent deployment method and system for a hydrological monitoring station network based on the Internet of Things (IoT), comprising: deploying multimodal sensor arrays at hydrological stations to collect real-time data on energy levels, power supply status, and sensor loss indicators; quantifying risk scores using a random forest algorithm; triggering backup energy switching and optimizing sensor operating modes to reduce losses when the score exceeds a threshold; then, modeling network blind spots using a support vector machine based on correction coefficients and generating temporary buffer queues; adjusting deployment points to cover blind spots through iterative calculations; activating edge computing units to perform real-time denoising based on noise distribution maps to obtain purified emergency datasets; classifying extreme environmental factors using a random forest algorithm to determine the linkage effect between energy difficulties and sensor vulnerability and outputting adjustment commands; and finally, fusing feedback data on unstable power supply and decreased accuracy through a dynamic configuration interface to determine the final system optimization scheme.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to a method and system for the intelligent deployment of a hydrological monitoring station network based on the Internet of Things. Background Technology

[0002] Hydrological monitoring, as a crucial pillar of water conservancy management and disaster prevention and mitigation, plays an irreplaceable role in ensuring water resource security and responding to natural disasters. Especially in complex and ever-changing field environments, the deployment and operation of hydrological stations directly affect the timeliness and accuracy of data, which is of vital importance for flood warnings, ecological protection, and water resource allocation. However, current technologies still face numerous challenges in dealing with extreme environments and complex needs, requiring urgent breakthroughs.

[0003] Existing methods have revealed significant shortcomings in practical applications, particularly in equipment reliability and environmental adaptability. Many hydrological stations are located in deep mountains, valleys, or remote areas, where equipment frequently fails due to unstable energy supply, extreme weather, or lack of communication signals. More importantly, existing technologies often fail to balance equipment cost with long-term operational stability, leading to frequent hardware failures under harsh conditions, high maintenance costs, and impacting the overall effectiveness of the monitoring network.

[0004] Focusing on core technological challenges, the reliability of equipment in field environments has become the most pressing issue. Particularly regarding energy acquisition and hardware durability, solar power is limited by insufficient sunlight or severe weather, and battery life is insufficient to support high-frequency data acquisition during critical moments. Furthermore, sensors are easily damaged by silt or impacts from floating debris, leading to data interruptions or distortion. This reliability issue not only affects the operation of individual stations but also weakens the integrity of the entire watershed monitoring network due to data loss caused by equipment failure, thus impacting the accuracy of crucial decisions such as flood warnings.

[0005] Specifically, during periods of high flood risk, a hydrological station located on a tributary in a mountainous area might experience a situation where heavy rains prevent its solar panels from charging properly. Once the batteries are depleted, the equipment stops working and cannot transmit real-time information about rising water levels. Furthermore, if the sensors are damaged by debris in the water flow, the station will completely lose its monitoring capabilities, critical data cannot be uploaded, and the optimal time for flood control and dispatch may ultimately be missed.

[0006] Therefore, how to ensure the energy stability and hardware durability of hydrological monitoring equipment in extreme environments, while maintaining a balance between low cost and high reliability, has become a key issue that this research urgently needs to address. Summary of the Invention

[0007] The purpose of this invention is to provide an intelligent deployment method and system for hydrological monitoring station networks based on the Internet of Things, in order to solve the technical problems of low system reliability and poor data accuracy caused by unstable energy supply, easy sensor wear and tear and communication blind spots in extreme environments.

[0008] The technical solution of the present invention is as follows:

[0009] This invention provides a method for intelligent deployment of a hydrological monitoring station network based on the Internet of Things, mainly including:

[0010] A multimodal sensor array deployed at a hydrological station collects real-time energy levels, power supply status, and sensor loss indicators under extreme environments. The data is then fused and classified using a random forest algorithm to obtain a quantitative risk score for energy difficulties and unstable power supply. When the quantitative risk score exceeds a preset threshold, a backup energy module is switched on, and the sensor operating mode is optimized to reduce vulnerability. Noise filtering is applied to the switched data stream to determine a correction coefficient for accuracy degradation. This correction coefficient is applied to the network blind zone detection process. A support vector machine is used to model the signal strength and data island distribution within the blind zone, determine the blind zone coverage area, and generate a temporary data buffer queue. This yields a preliminary adjustment strategy to address the combined problems of communication blind zones, data islands, and sensor accuracy degradation. For the initial adjustment strategy, optimization parameters related to spatiotemporal redundancy are extracted from the mathematical model library. The deployment points are adjusted through iterative calculations to cover areas with both blind spots and noise, thus determining the noise source distribution map. Based on this distribution map, when a high-concentration area is shown, the edge computing unit is activated to perform a real-time denoising algorithm on the locally collected data, resulting in a purified emergency processing dataset. This purified dataset is then input into the overall system reliability assessment module, where a random forest algorithm is used to reclassify extreme environmental impact factors, determine the linkage effect between energy shortages and sensor vulnerability, and output adjustment instructions. These instructions are then applied to the hydrological station's dynamic configuration interface, integrating feedback data on unstable power supply and decreased accuracy to determine the final system optimization scheme.

[0011] This invention provides an intelligent deployment system for a hydrological monitoring station network based on the Internet of Things, mainly comprising:

[0012] The data acquisition and risk scoring module is used to collect real-time energy levels, power supply status, and sensor loss indicators under extreme environments through a multimodal sensor array deployed at hydrological stations. After fusion, the data is classified using a random forest algorithm to obtain a quantitative risk score for energy difficulties and unstable power supply.

[0013] The energy and mode optimization module is used to trigger the backup energy module to switch and optimize the sensor working mode to reduce the vulnerability rate when the quantitative risk score exceeds a preset threshold, and to filter noise in the data stream after switching to determine the correction coefficient for the decrease in accuracy.

[0014] The blind zone detection module is used to apply the correction coefficient to the network blind zone detection process. It uses a support vector machine to model the signal strength and data island distribution in the blind zone, determines the coverage of the blind zone and generates a temporary data buffer queue, and obtains a preliminary adjustment strategy to solve the joint problems of communication blind zones, data islands and sensor accuracy degradation.

[0015] The point adjustment module is used to extract optimization parameters related to spatiotemporal redundancy from the mathematical model library for the initial adjustment strategy, adjust the layout points through iterative calculation to cover blind areas and coexisting regions, and determine the source distribution map of data noise.

[0016] The edge denoising module is used to activate the edge computing unit to perform a real-time denoising algorithm on the locally collected data when the noise source distribution map shows a high concentration area, in order to obtain a purified emergency treatment dataset.

[0017] The reliability assessment module is used to input the purified emergency treatment dataset into the overall system reliability assessment module, use the random forest algorithm to reclassify the extreme environmental impact factors, determine the linkage effect between energy difficulties and sensor vulnerability, and output adjustment instructions.

[0018] The dynamic configuration interface module is used to obtain the dynamic configuration interface applied to the hydrological station after the adjustment command is received, integrate feedback data on unstable power supply and decreased accuracy, and determine the final system optimization scheme.

[0019] The beneficial effects of this application are as follows: An intelligent deployment method and system for a hydrological monitoring station network based on the Internet of Things (IoT) addresses the interconnected problems of energy shortages, unstable power supply, and sensor vulnerability in extreme environments at hydrological stations. It utilizes a multi-modal sensor array to collect real-time data on energy levels, power supply status, and loss indicators. The invention employs a random forest algorithm for data classification and risk scoring, switching to backup energy modules when thresholds are exceeded to optimize sensor operating modes. Simultaneously, it uses support vector machines to model blind zone signals, adjusting deployment points to cover blind zones. Combined with edge computing for noise reduction and reliability assessment, it dynamically generates optimization instructions, ultimately fusing feedback data to form an optimized system solution. Through multi-algorithm collaboration and data purification, this invention significantly improves system stability and data accuracy in extreme environments at hydrological stations, effectively mitigating the interconnected risks of energy shortages and sensor failures. Attached Figure Description

[0020] Figure 1This is a flowchart illustrating a specific embodiment of the intelligent deployment method for a hydrological monitoring station network based on the Internet of Things according to the present invention.

[0021] Figure 2 This is a schematic diagram of the sub-processes for backup energy switching and sensor operating mode optimization of the present invention;

[0022] Figure 3 This is a schematic diagram of the sub-processes for network blind spot monitoring and data silo modeling in this invention;

[0023] Figure 4 This is a schematic diagram of the structure of an intelligent deployment system for a hydrological monitoring station network based on the Internet of Things according to the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only for explaining the invention and are not intended to limit the invention; that is, the described embodiments are merely some embodiments of the invention, and not all embodiments. The components of the embodiments of the invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0025] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0026] It should be noted that relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0027] The features and performance of the present invention will be further described in detail below with reference to embodiments.

[0028] A specific embodiment of the intelligent deployment method and system for a hydrological monitoring station network based on the Internet of Things according to the present invention:

[0029] like Figures 1-3 This embodiment of an intelligent deployment method for a hydrological monitoring station network based on the Internet of Things may specifically include:

[0030] S101 collects real-time energy levels, power supply status, and sensor loss indicators under extreme environments through a multimodal sensor array deployed at hydrological stations. After fusion, the collected data is classified and processed using a random forest algorithm to obtain a quantitative risk score for energy difficulties and unstable power supply.

[0031] Multimodal sensing data is acquired from sensor arrays at hydrological stations to collect energy levels, power supply status, and loss data under extreme environments, completing preliminary data collection and obtaining a raw dataset. Based on the raw dataset, standardization methods are used to preprocess the energy level, power supply status, and loss data (such as Min-Max normalization or Z-score) to eliminate dimensional differences between different data points, resulting in a standardized dataset.

[0032] For the normalized dataset, the random forest algorithm is used for classification processing to analyze anomalies in energy levels and power supply status, and to determine the category distribution of potential energy difficulties and power supply instability.

[0033] In a preferred embodiment, a normalized three-dimensional feature vector is first obtained, including real-time energy level, power supply status characteristics, and sensor data indicators. During the training phase, historical sample data under extreme conditions are pre-collected and manually labeled into four categories: normal, energy difficulty, unstable power supply, and multiple concurrent anomalies. This labeled data is used to train a random forest model, which is formed by ensembled decision trees. In the classification phase, the real-time collected and normalized feature vector is input into the trained random forest model. Each decision tree independently outputs a predicted category. The voting results of all decision trees are statistically analyzed, and the category with the most votes is taken as the anomaly classification result for the current data point. In anomaly analysis, each decision tree splits the data by setting a splitting threshold for the features; for example, when the energy level is <20%, it is divided into the energy difficulty branch. A time window is set, and the frequency of data points classified as normal or energy difficulty within this window is statistically analyzed to determine the severity and duration of the anomaly. In the category distribution calculation, the total amount of data collected within the current time window and the number of samples classified into each category are counted. The percentage of each category's sample count in the total amount of data collected is calculated, and finally, the distribution vector of each category is obtained to characterize the distribution status of abnormal situations such as energy difficulties and unstable power supply.

[0034] If the proportion of anomalies in a certain category exceeds a preset threshold in the classification results, deep feature extraction is performed on the data of that category to obtain the specific distribution pattern of the abnormal data and determine the source of the anomaly.

[0035] In a preferred embodiment, the preset threshold is typically set based on the safe operation boundary of the hydrological station. For example, the threshold for the abnormal proportion of the energy difficulty category is set at 15%, and the threshold for the unstable power supply category is set at 20%. When the distribution proportion of a certain abnormal category obtained from the classification process exceeds the corresponding preset threshold, deep feature extraction of the time series data corresponding to that category is triggered. This involves calculating the variance, peak-to-peak value, skewness, and kurtosis of the data in the time domain; extracting the dominant frequency component of the power spectral density and energy concentration in the frequency domain using a fast Fourier transform; and combining the above time-frequency features into a deep feature vector. Subsequently, kernel density estimation is used to fit the aggregation degree of the extracted deep features on the time axis to obtain the specific distribution pattern of the abnormal data, including sporadic pulse distribution, continuously decreasing distribution, or periodic oscillation distribution. Finally, the source of the anomaly is determined based on the decision tree discrimination matrix based on root cause analysis: when the power spectral density has a large amplitude in the low frequency region and the kernel density estimate shows a continuously decreasing distribution, the source of the anomaly is determined to be long-term shading or aging of the solar panel; when the kurtosis is extremely high and the kernel density estimate shows an occasional pulse distribution, the source of the anomaly is determined to be lightning electromagnetic interference or instantaneous short circuit of the equipment.

[0036] By analyzing the distribution patterns of anomaly sources and combining them with environmental data from hydrological stations, we can identify the key environmental factors affecting energy levels and power supply status through correlation analysis of the impacts of extreme environments.

[0037] This embodiment provides a method for identifying key environmental factors of energy anomalies based on grey relational analysis: First, time-series data of unstable energy supply or abnormal power supply status are used as a reference sequence. The data was compared with data from multiple environmental sensors collected simultaneously, including rainfall data. Light intensity Wind speed and ambient temperature Secondly, calculate the grey relational coefficient using the following formula:

[0038]

[0039] in, The resolution coefficient is set to 0.5. This is a comparison sequence, specifically the measurement value of the i-th environmental factor at time t. Then, the grey relational degree between each environmental factor and the energy anomaly sequence is calculated using the following formula:

[0040]

[0041] Where T represents the total number of time points. The correlation between various environmental factors. By sorting environmental factors from largest to smallest, one or more of the top-ranked factors are identified as the key environmental factors causing this energy anomaly. Based on the distribution pattern of the anomaly source and combined with environmental data from hydrological stations, the key environmental factors affecting energy levels and power supply status are obtained.

[0042] Based on key environmental factors, the loss data of the sensor array is compared and analyzed to determine the correlation between the loss data and environmental factors, and to ascertain the degree of impact of loss on data acquisition accuracy. If the correlation between the loss data and environmental factors is higher than a preset threshold, the acquisition frequency of the sensor array is dynamically adjusted to obtain more accurate energy level and power supply status data, thereby updating the risk score.

[0043] This embodiment provides a method for dynamically adjusting sensor acquisition frequency and updating risk scores based on the correlation with environmental factors: First, based on the identified key environmental factors, the loss data of the sensor array is compared and analyzed. The loss data includes the sensor zero-point drift and the thickness of surface sediment cover. The Spearman rank correlation coefficient is used to quantify the correlation between the loss data and the key environmental factors.

[0044]

[0045] in, The difference in rank between the two sets of data. For the sample size, The absolute value of the coefficients ranges from -1 to 1. Then, the impact of each environmental factor on the accuracy of data collection is determined using standardized regression coefficients from multiple linear regression or feature importance from random forest. A correlation threshold is set. .when At that time, it was determined that the loss data was highly correlated with environmental factors, indicating that environmental factors had a significant impact on sensor accuracy. The system then initiated an on-demand frequency reduction mechanism to dynamically adjust the sensor array's acquisition frequency to obtain more accurate energy level and power supply status data. Finally, a risk score RS was constructed and updated in real time.

[0046]

[0047] in, This represents the normal probability of the system. The probability of various anomalies occurring. The weighting coefficients for the impact of various anomalies on the overall flood control situation. This is the current loss value. This is the maximum allowable loss value. , , Normalized weight parameters and satisfying The system recalculates the Risk Ratio (RS) every 5 minutes based on newly collected data, achieving dynamic updates to the risk score. The RS formula is a quantitative model constructed in this invention for the fusion assessment of sensor loss and anomaly probability.

[0048] S102, based on the obtained quantitative risk score, if the risk score exceeds the preset threshold, the backup power module is switched and the sensor working mode is optimized to reduce the vulnerability rate. The data stream after the switch is filtered for noise through the information processing link to determine the correction coefficient for the decrease in accuracy.

[0049] If the risk score exceeds a preset threshold, the switchover process for the backup energy module is triggered. The system automatically detects the current energy status, obtains the switchover command, and completes the module replacement, thus acquiring the backup energy's operational data. The preset threshold is determined based on system simulation experiments. Example: In a 100-point scale, it is typically set to 75 points. When the risk score RS > 75, it means the current site faces severe energy depletion (e.g., remaining power below 15%) or the main sensor has suffered irreversible physical wear, necessitating a forced switchover.

[0050] Based on the operating data of the backup power source, the working mode of the sensor is adjusted. By reducing the sampling frequency and optimizing the acquisition range, equipment wear is reduced, and the output signal of the sensor after adjustment is determined.

[0051] In this embodiment, when the backup energy module is triggered, the system adjusts the working mode of the sensor array based on the backup energy's operating data to reduce equipment wear and extend the system's lifespan. Specifically, this includes: 1. Reducing the sampling frequency: Switching the data acquisition mode from a high-frequency continuous mode to an intermittent pulse mode. For example, the sampling frequency of the water level sensor is reduced from the default 1Hz (once per second) to 0.0017Hz (once every 10 minutes). During critical early warning periods such as floods, the edge device dynamically adjusts the sampling frequency based on water level fluctuations. When the water level rises by more than 10cm per hour, it automatically returns to once per minute. 2. Optimizing the acquisition range: Closing high-energy-consuming, non-core auxiliary sensing channels. The multimodal sensor array includes water level, flow velocity, sediment concentration, water temperature, and air pressure sensors. The system temporarily shuts down the high-energy-consuming radar flow meter and laser sediment concentration analyzer, retaining only the low-power piezoresistive water level gauge core channel, reducing single-acquisition power consumption by more than 70%. Through the above adjustments, the adjusted sensor output signal is determined for subsequent energy status assessment.

[0052] The data flow direction is extracted from the output signal after sensor adjustment. The data flow is then preliminarily cleaned through information processing to remove outliers and redundant information, resulting in a cleaned data flow.

[0053] In one possible implementation, the sensor array output signals at the hardware layer are sequentially transmitted to a local FIFO queue, and then to the edge computing node. The edge computing unit reads the raw ADC code from the sensor register at a reduced frequency period via SPI or RS-485 industrial bus, and encapsulates it into a timestamped data frame. Then, the data stream is pre-cleaned by the data stream cleaning operator in the information processing stage to remove dead values ​​that exceed physical boundaries (such as negative water levels or zero voltage), remove outliers and redundant information, and obtain a cleaned data stream.

[0054] For the cleaned data stream, noise filtering technology is employed to smooth the data stream, reducing the impact of external interference and determining the filtered data sequence. Based on the filtered data sequence, the trend of accuracy degradation is analyzed. By comparing the data differences before and after the switchover through information processing, a correction coefficient for accuracy degradation is calculated. As a preferred example, a first-order moving least squares method is used to fit a sliding window to the time-series data sequence, calculating the local slope. If the slope monotonically increases or decreases with time, it is determined that the sensor is experiencing zero-point drift due to unstable power supply voltage. The average high-frequency standard data and the average low-frequency data for five cycles before and after the switchover to backup power are extracted and compared differentially.

[0055]

[0056] in, The difference between the mean values ​​of the data before and after switching to backup power is used for bias correction; The average of low-frequency data over five consecutive cycles after switching to backup power and stabilizing; This is the average of five consecutive high-frequency standard data cycles before the switch. The gain correction coefficient is calculated using precision comparison and calibration operators in the information processing flow. and bias correction matrix Thus, the correction factor for the decrease in accuracy is obtained:

[0057]

[0058] Wherein, the bias correction matrix . The standard deviation of the high-frequency standard data before the switch; The standard deviation of the low-frequency data after the switch; This is the voltage fluctuation compensation coefficient (an empirical constant determined by the sensor type). This represents the fluctuation of the current supply voltage relative to the nominal voltage. This is the sensor's nominal operating voltage.

[0059] After obtaining the correction coefficients, calibration is performed on the current data sequence. By applying the correction coefficients to the data output, accuracy compensation is completed, and the calibrated data result is obtained. Specifically, the correction coefficients are obtained... Subsequently, all low-frequency raw data collected under backup power source were processed. Perform real-time linear mapping compensation and output calibrated data. :

[0060]

[0061] Eliminate variance shrinkage caused by low-frequency sampling and systematic errors caused by unstable power supply voltage, complete accuracy compensation, and obtain calibrated data results.

[0062] Based on the calibrated data, the operating status of the backup energy module is continuously monitored. Potential anomalies are detected through information processing. If an anomaly signal is detected, an alarm process is triggered to determine the anomaly handling instructions.

[0063] In one possible implementation, anomalies are detected using a dual-window logic threshold method through steady-state monitoring and alarm operators in the information processing stage: if the absolute value of the difference between the current calibrated water level data and the previous moment exceeds the historical maximum surge rate (e.g., >0.5 m / min), it is determined to be a physical abrupt change anomaly; if the backup energy voltage change rate continuously exceeds the discharge slope threshold, it is determined to be a secondary collapse anomaly. When any anomaly persists for more than 3 sampling cycles, a three-level alarm process is triggered: the current data frame is latched and a fault code (e.g., E-03: backup energy overload) is generated, the backup communication link is woken up (BeiDou short message or low-power ultra-shortwave), and alarm data packets are broadcast to the central station and nearby hydrological stations. Anomaly handling instructions are dynamically generated based on the fault code. These instructions are register control words (e.g., 0xFF3A0012): at the hardware level, power supply to unnecessary peripherals is cut off, and a super sleep wake-up mode is forcibly entered; at the software level, a local temporary data cache queue is started, unsent historical data is encrypted and written to the onboard solid-state flash memory, and wireless signal transmission is stopped to save residual power and prepare data for subsequent processes.

[0064] S103, obtain the correction coefficients and apply them to the subsequent network blind spot detection process. Use support vector machine to model the signal strength and data island distribution in the blind spot, determine the coverage of the blind spot and generate a temporary data buffer queue, and obtain a preliminary adjustment strategy to solve the joint problem of communication blind spots, data islands and sensor accuracy degradation.

[0065] By collecting signal strength data within network blind spots, a support vector machine is used to model the signal distribution, resulting in a preliminary division of the blind spot coverage area.

[0066] In one possible implementation, firstly, signal strength data is collected within network blind spots. These blind spots refer to areas where the main communication network signal cannot reach or is extremely weak. The data collection device employs a multi-mode wireless network measurement module integrated within the hydrological station. This module supports RSSI signal strength indication and integrates a LoRa / UHF self-organizing network module and a multi-mode 4G / 5G module. The data collection methods include: active sniffing, which involves periodically sending access requests to surrounding cellular base stations to measure downlink received signal strength indication and reference signal reception quality; and neighbor station mutual measurement, which involves point-to-point communication with neighboring hydrological stations via LoRa or UHF when cellular network connectivity is unavailable, collecting the signal arrival strength of neighboring stations to obtain a three-dimensional radio wave feature vector at the given geographical coordinates. Then, a support vector machine is used to model the signal distribution, yielding preliminary results for the blind spot coverage area. The specific steps are as follows: Construct a sample set with longitude, latitude, elevation, and received signal strength indication as feature vectors and whether the signal meets the standard as a label; introduce a Gaussian radial basis kernel function to nonlinearly map the samples to a high-dimensional feature space; solve for the optimal hyperplane through a sequence minimum optimization algorithm to obtain a decision function; after gridding the coordinates of the area to be measured, input the decision function to determine the grid points of the blind zone, thereby completing the preliminary division of the blind zone coverage area.

[0067] Based on the results of the blind zone coverage division, the distribution of data silos is analyzed, a temporary cache queue is generated, and the correspondence between data silos and blind zone coverage is determined.

[0068] In one possible implementation, firstly, based on the blind zone coverage division results obtained from the aforementioned support vector machine, the distribution of data silos is analyzed. Data silos refer to isolated stations or clusters of stations where hydrological monitoring data cannot be uploaded and remains locally due to being located in network blind zones. Specifically, a density-based spatial clustering algorithm can be used, for example: the coordinates of all hydrological stations identified as blind zones and with communication interruptions are taken as input, and a neighborhood radius is set. The system uses the minimum number of core samples (MinPts) to identify single-point isolated data islands and multi-point clustered isolated data islands through clustering, and counts the total number of data packets stuck within each isolated data island. Secondly, when a site determines it is in an isolated state, a temporary cache queue is generated. The main control chip allocates a protected circular doubly linked list storage area in local random access memory and external flash memory as a temporary data cache queue. Each node in this queue contains a timestamp, water level data, priority label, checksum, and link pointer. The write mechanism follows the first-in, first-out principle; when communication is restored, the data at the head of the queue is sent first; if memory is full, a priority overwrite strategy or dynamic compression is performed. Finally, the correspondence between data islands and blind zone coverage is determined. This correspondence is a topological mapping between a specific data island cluster and a specific connected blind zone geometries defined by a support vector machine. The system uses spatial bounding box technology to calculate whether the coordinates of data island sites fall within the geometric polygons defined by the decision surface, and establishes an association matrix table of inclusion relationships. ,in This indicates that station i belongs to the blind zone geometry j, otherwise it is 0, thus completing the determination of the correspondence.

[0069] For the isolated data distribution in the temporary buffer queue, a correction coefficient is obtained to adjust the signal strength and identify signal loss areas within the blind zone coverage. If the signal loss area exceeds a preset threshold, the temporary buffer queue is prioritized to obtain a high-priority data island processing sequence.

[0070] In one possible implementation, firstly, for the isolated data distribution in the temporary buffer queue, an accuracy reduction correction coefficient calculated due to the change in power supply status caused by the switching between primary and backup power sources is obtained. Since a drop in power supply voltage reduces the RF transmission power of the wireless network monitoring module, causing a systematic downward drift in the original signal strength, the correction coefficient is used to perform inverse gain compensation adjustment on the original signal strength: the original signal strength is added to the product of the RF attenuation constant, the deviation of the correction coefficient, and the maximum transmission power to obtain the adjusted signal strength, thereby eliminating the impact of voltage drops in the equipment itself and restoring the true signal strength of the blind zone caused by external terrain obstruction. Then, the signal loss area within the blind zone coverage is determined: the adjusted signal strength is compared with the minimum demodulation threshold for communication establishment; if it is lower than this threshold, the spatial grid is marked as a signal loss point. When the total area of ​​consecutive signal loss grids exceeds a preset threshold (e.g., the area of ​​consecutive blind zones reaches or exceeds 5 square kilometers), the signal loss area is determined to be out of limit. Finally, the temporary buffer queue is prioritized. The ranking is based on a comprehensive weighted model, which considers the following factors: the temporal fluctuation of the current water level (sites with rapidly rising water levels have higher priority), the length of time data is retained (the longer the retention, the higher the priority), and the fullness rate of the local cache queue (the closer to full, the higher the priority). These factors are multiplied by their corresponding weight coefficients and then summed to obtain a priority score for each data silo. Based on the scores, the higher-priority data silos are determined and processed in a specific order, with the corresponding data being sent or processed first.

[0071] By processing high-priority data silos, the resource allocation scheme in the detection process is adjusted to determine the signal supplementation points within the blind zone coverage area.

[0072] In one possible implementation, the resource allocation scheme in the detection process is adjusted according to the determined high-priority data island processing sequence. This resource allocation scheme includes the reallocation of energy and network bandwidth resources. Specifically, it involves notifying edge relay sites not connected to the blind zone to expand their receiving time slot windows, opening up surplus channel bandwidth for receiving burst data from high-priority data island transmissions; simultaneously, reducing the inspection power consumption of non-core blind zone sites, allocating surplus computing and communication power to bridge sites located at the edge of high-risk blind zones, allowing them to temporarily act as routing forwarding nodes. Then, signal supplementation points within the blind zone coverage area are determined. These signal supplementation points refer to the optimal physical coordinates of temporary communication relays or IoT gateways needed to eliminate the blind zone. Using the maximum coverage fitting operator, spatial coordinates that simultaneously satisfy the following conditions are found around the geometric centroid of the blind zone defined by the support vector machine: highest terrain elevation, best visibility to surrounding data island sites, and closest to the main base station. These spatial coordinates (X, Y, Z) are then determined as signal supplementation points.

[0073] Based on the distribution of signal supplementation points, a preliminary adjustment strategy is generated to address the combined problem of communication blind spots, data silos, and sensor accuracy degradation, resulting in an optimized network blind spot detection scheme. The combined problem refers to the systemic network disconnection contradiction arising from the interplay of continuous spatial communication blind spots, locally scattered data silos, and dynamic accuracy degradation of multimodal sensors in extremely harsh outdoor environments. The preliminary adjustment strategy is generated by extracting spatiotemporal redundancy optimization parameters from a mathematical model library and using a variant of the shortest path algorithm to plan an emergency multi-hop communication topology path that uses signal supplementation points as stepping stones, connects various data silos, and ultimately connects to an external main base station. This serves as the preliminary adjustment strategy. Then, the preliminary adjustment strategy is converted into a configuration parameter matrix executable by the device to obtain the optimized network blind spot detection scheme. The core optimized parameters include: the three-dimensional geographic coordinates and antenna gain downtilt angle of the newly added supplementation points; the spatiotemporal redundancy dynamic observation period, which extends the ineffective blind network search period of sites within the blind spot to reduce redundant power consumption; and the spreading factor and transmit power of the LoRa self-organizing network, trading space for the distance to penetrate the blind spot.

[0074] An optimized network blind spot detection scheme is adopted to continuously monitor the signal strength within the blind spot coverage area and determine the dynamic trend of data island distribution. In this embodiment, a Markov chain state transition matrix is ​​introduced to calculate the dynamic evolution of the distribution state based on the rate of change of the number of island sites that successfully clear the cache queue in each time period. When the number of islands decreases and the spatial clustering shrinks towards the center, it is determined that the network is recovering and the blind spot is being resolved; when the number of islands increases against the trend, it is determined that the extreme disaster is still expanding, and this trend is fed back to subsequent steps to trigger higher-level edge computing real-time denoising and secondary optimization of system reliability.

[0075] S104. Based on the obtained preliminary adjustment strategy, optimization parameters related to spatiotemporal redundancy are extracted from the mathematical model library. The deployment points are adjusted through iterative calculation to cover the blind zone and coexisting area, and the source distribution map of data noise is determined.

[0076] By obtaining optimization parameters related to spatiotemporal redundancy from a mathematical model library, an initial deployment scheme is constructed, and preliminary deployment data is obtained.

[0077] In a preferred embodiment of the present invention, the Kriging variogram and mutual information algorithm are first called from a mathematical model library. Based on the historical hydrological observation time series within the monitoring basin, the mutual information and variogram between two stations are calculated, and the spatiotemporal redundancy optimization parameters, including the spatial correlation radius, are dynamically calculated. Time masking window Spatial overlap coverage (Typically set at 15%–25%). Then, a mathematical model library is constructed, which is embedded in the hydrological monitoring system server or edge gateway. This library includes at least ordinary kriging, inverse distance weighting, and other spatial interpolation models, a Thiessen polygon-based grid cover model, and a Longley-Rice radio propagation fading model. Finally, an initial deployment point scheme is constructed: the digital elevation model of the monitored watershed is discretized into... The spatial grid points are used to call the Thiessen polygon cover model from the mathematical model library, with spatially related radii. Geometric tessellation is performed using the circumcircle radius of the polygons, and the centroid coordinates of each polygon are extracted as the initial layout point scheme. This allows us to obtain preliminary data on the distribution of points.

[0078] Based on the preliminary location distribution data, the location of the points is adjusted using an iterative calculation method, and areas with insufficient blind spot coverage are supplemented to determine the adjusted location layout information.

[0079] In a preferred embodiment of the present invention, the initial layout point scheme is obtained in the aforementioned steps. Building upon this foundation, a virtual force-guided algorithm is further employed to iteratively adjust the location of monitoring points. Specifically, each hydrological monitoring point is treated as a physical particle, and the spatial overlap coverage rate is considered... Repulsive forces are applied between points to prevent over-aggregation, while attractive forces are applied to points based on the geometric centroid of the insufficiently covered blind zone, guiding the points to move towards the signal blind zone. The resultant force on each point is calculated in each iteration. and according to the formula ( The coordinates are updated using a step size factor, and the distribution of points tends to the optimal equilibrium after multiple iterations. For areas with insufficient blind zone coverage, the radio signal coverage envelope of all current points is calculated based on the blind zone boundaries defined by the support vector machine. If a sub-region belongs to a key flood control area (such as a gully prone to debris flows), and the received signal strength (RSSI) under the Longley-Rice propagation model is lower than the communication demodulation threshold (such as -115dBm), and there is no redundant site coverage in the surrounding area, it is determined to be an area with insufficient blind zone coverage. Supplementation is performed for the above-mentioned areas: firstly, the geometric center coordinates of the area with insufficient coverage are extracted ( , A temporary emergency monitoring point is added directly at the central location; then the new point is incorporated into the virtual force model, and 3 to 5 adjacent stations in the local area are fine-tuned and iterated 10 times until the joint coverage probability of the area reaches more than 95%, thereby locking in and determining the adjusted point layout information.

[0080] Based on the adjusted location layout information, corresponding regional data is acquired, and the distribution characteristics of data noise are analyzed to determine the concentrated areas of noise sources. In one possible implementation, firstly, an autoregressive moving average model residual analysis combined with wavelet packet transform is used to perform three-level wavelet packet decomposition on the time-series signals collected from each measurement point, extracting high-frequency coefficients representing noise. The variance, energy spectral density, and kurtosis of these high-frequency coefficients at each point are calculated to form the spatial distribution characteristics of the data noise. Then, the local Moran's index in spatial autocorrelation analysis is used to identify the concentrated areas of noise sources. The calculation formula is as follows:

[0081]

[0082] in, Let be the local Moran index of the i-th station; Let be the noise energy value of the i-th station (e.g., the high-frequency coefficient variance or energy spectral density obtained from wavelet packet decomposition). This is the arithmetic mean of the noise energy values ​​at all sites. The sample variance of noise energy values ​​for all sites is typically calculated as follows: (Or divide by n-1, depending on the specific implementation); This represents the total number of stations within the monitoring area. This is the spatial weight matrix; Let be the noise energy value of the j-th station. When the noise energy of a certain station is... The calculated value is a significantly high positive value and exceeds the preset Moran threshold. When the noise source is identified, the connected network area consisting of the station and its neighboring stations is determined to be the concentrated area of ​​noise source.

[0083] If the concentrated area of ​​the noise source overlaps with the blind spot, the area is supplemented by increasing the density of the points to obtain an updated deployment scheme.

[0084] In a preferred embodiment of the present invention, after identifying the concentrated area of ​​the noise source, the spatial overlap between this area and the network blind zone coverage area obtained in the aforementioned steps is further determined. The overlap criterion uses the inter-grid intersection ratio (IoU), calculated using the following formula:

[0085]

[0086] When the overlap area between the two reaches 40% or more of their respective total combined area, it is identified as a high-risk overlap area. This area has both signal blind spots and is subject to high noise pollution. For such overlap areas, the system adopts a strategy of "spatial spacing reduction and density doubling" to supplement coverage: increasing the deployment point density in this local area to 1.5 to 2.0 times that of the initial plan (i.e., point density multiplier). At the same time, the average communication distance between stations is forcibly shortened from the standard 5km to 1.5km to 2km. Through high-density mutual transmission and environmental noise, an updated deployment scheme is obtained to ensure the spatiotemporal redundancy of data.

[0087] Based on the updated deployment plan, noise signals are extracted from the regional data and filtered using preset thresholds to determine the preliminary distribution range of noise sources.

[0088] In one possible implementation, a preset threshold is first set: the standard deviation of the noise signal in decibels. This can be equivalent to a 5% error threshold at full scale of the sensor. Then, an adaptive Wiener filter operator is used for filtering, locking in high-frequency noise components through a preset threshold, and adjusting the transfer function in the frequency domain.

[0089]

[0090] in, The frequency response of the original system. For their conjugate, The power spectral density of the signal is used. Interference peaks exceeding the threshold are clipped and filtered out, thus completely separating the pure hydrological main signal from the chaotic environmental noise, and finally determining the preliminary distribution range of the noise source.

[0091] By performing spatial mapping analysis on the initial distribution range, a distribution map of data noise sources is generated, completing a comprehensive assessment of regional coverage and noise sources.

[0092] In one possible implementation, after determining the initial distribution range of noise sources, the system further performs spatial mapping analysis using a GIS spatial topology mapping engine to generate a distribution map of the data noise sources. Specifically: First, the system collects the pure numerical values ​​of residual noise intensity at each discrete deployment point after filtering. Then, bicubic spline interpolation or Gaussian process regression (GPR) from the mathematical model library is used to map discrete point noise levels into a continuous two-dimensional spatial scalar field. Finally, according to The numerical values ​​are automatically rendered to generate a contour heatmap with warm and cool tones. Dark red areas represent high-concentration areas with noise energy density greater than 30dB, while blue areas represent low-noise empty areas. This rendered two-dimensional digital matrix and visualization layer serve as a distribution map of data noise sources, directly used as a guiding base map for edge computing units to perform precise denoising, thereby completing a comprehensive assessment of area coverage and noise sources.

[0093] S105. Based on the determined source distribution map, if the noise source distribution map shows a high concentration area, the edge computing unit is activated to perform a real-time denoising algorithm on the locally collected data to obtain the purified emergency treatment dataset.

[0094] By analyzing the noise source distribution map, we can determine whether there are high-concentration areas and obtain preliminary results on the regional distribution. In this embodiment, noise concentration is first defined as the noise concentration at the region coordinates (x, y). This is the integral value of the high-frequency noise power spectral density (PSD) at that location. and : Indicates the range of noise frequencies of interest (e.g., the start and end frequencies of high-frequency noise in hydrological monitoring). The noise power spectral density at location (x,y) and frequency f. Then, a connected component analysis algorithm is used for determination: a noise scalar threshold value is set. The contour heat map Pixels marked as 1 (high noise) are assigned a 0, and the rest are marked as 0. Connected pixels marked as 1 are clustered, and the geometric area of ​​the closed connected region is calculated. If the area of ​​a continuous high-noise connected domain exceeds a set spatial threshold (e.g., 5% of the total area of ​​the watershed), it is determined that a high-concentration area exists, and its geographical boundary is officially locked, thereby obtaining the regional distribution judgment result.

[0095] If a high-concentration area is confirmed, the edge computing unit is activated to perform initialization processing on the locally collected data stream to obtain the raw dataset to be processed. In a preferred embodiment of the invention, when a high-concentration noise area is determined, the system activates the local edge computing unit (e.g., a low-power Cortex-M4 or embedded DSP) at the hydrological station to perform initialization processing on the locally collected data stream to obtain the raw dataset to be processed. Specifically: First, the GPS / BeiDou satellite timing clock is retrieved to perform millisecond-level timestamp alignment on the heterogeneous asynchronous data stream uploaded by the multimodal sensor array (water level, flow velocity, rainfall, etc.) to achieve clock synchronization and phase alignment; Second, the input data is coarsely screened using hardware registers. If the water level sensor reading exceeds the historical limit water level of the flood control section (e.g., H<0m or H>50m), it is directly identified as a hard fault and discarded to prevent filter divergence; Third, a circular buffer is opened in the RAM of the edge chip, and discrete time-series data segments of length N=128 or 256 are extracted using a Hanning window to form a continuous raw dataset to be processed.

[0096] A real-time denoising algorithm is used to filter noise from the original dataset, generating a purified intermediate dataset. In this embodiment, for the initialized original dataset, the edge computing unit uses a composite algorithm of adaptive Kalman filtering and wavelet threshold denoising for real-time filtering to generate a purified intermediate dataset. First, the time-series signal to be processed is decomposed into three levels using the db4 wavelet basis, breaking it down into low-frequency approximation coefficients (representing the actual hydrological change trend) and high-frequency detail coefficients (representing noise) at each level. Second, a noise level is introduced based on the current noise concentration. A dynamically adjusted adaptive threshold is used for improved soft threshold quantization of high-frequency coefficients. Specifically, this adaptive threshold... Wavelet universal threshold Multiply by the noise concentration adjustment function Composition, that is Less than The high-frequency coefficients are set to zero, retaining useful abrupt changes greater than a threshold. Finally, the wavelet-reconstructed signal is input into a one-dimensional standard Kalman filter, and three iterative steps—state prediction, Kalman gain calculation, and state update—are executed sequentially. The measurement noise variance in the Kalman filter is dynamically adjusted using a precision degradation correction coefficient. The iterative output state estimate is the purified high-precision hydrological dataset.

[0097] Based on the purified intermediate dataset and the triggering mechanism's conditional judgment, if the data quality reaches a preset threshold, an emergency dataset is output. In a preferred embodiment of this invention, after obtaining the purified intermediate dataset, the system performs a comprehensive quantitative evaluation of its data quality: data quality is assessed on a percentage basis, measured by calculating the variance convergence and statistical confidence of the denoised signal. The preset quality threshold is, for example, 85 points. The judgment criteria are that the residual sequence must pass the multidimensional normal distribution test (residuals are independent and have no obvious periodic bias) and the effective signal energy percentage must be no less than 85%. When the data quality score reaches or exceeds 85 points, and the system simultaneously determines that the current hydrological station is in a network blind zone / isolated state, an emergency interruption mechanism is triggered. The edge computing unit labels the data with a high-priority "emergency" tag, removes it from the general memory area, and stores it in a dedicated emergency data queue, thus outputting the emergency dataset. This emergency dataset is a minimized golden survival dataset, specifically including: station code, standard timestamps, and calibrated water level values, among other core flood control elements. Non-core redundant logs are compressed and removed to utilize extremely narrow bandwidth links such as BeiDou satellites for maximum transmission.

[0098] The emergency dataset is structured to obtain formatted data sets suitable for subsequent analysis. As a preferred example, C language structure alignment and encapsulation are used, and Protocol Buffers lightweight serialization technology is utilized to map discrete hydrological elements (such as station codes, timestamps, and calibrated water level values) into fixed-byte-length structures, compressing them into compact binary data streams, reducing the size of a single data frame to less than 16 bytes.

[0099] By hierarchically storing formatted data groups, the final emergency data storage scheme is determined. As a preferred example, based on the timeliness of data and the power consumption characteristics of hardware media, the latest high-frequency emergency data collected within the current hour is temporarily stored in the on-chip SRAM (high-speed cache layer) built into the MCU; when the data is retained for more than 1 hour or the SRAM is 80% full, the DMA controller is activated to batch write the data to the emergency dedicated sector (persistent non-volatile layer) of the low-power external SPI NAND Flash, ensuring that the data is not lost after power failure, thus forming the final emergency data storage scheme.

[0100] Once the storage scheme is complete, an index is built on the data storage path to obtain a fast-access data index table. As a preferred example, an implicit linear B+ tree index structure based on timestamps is used, with a standard Unix timestamp of type uint32_t as the index key and the Flash physical page address as the value. A lightweight two-level B+ tree is maintained in memory. When data at a specific historical moment needs to be retrieved, a binary search is used to search the tree index at most three times using hash mappings to directly locate the physical sector address, resulting in a fast-access data index table.

[0101] S106 inputs the purified emergency response dataset into the overall system reliability assessment module, uses the random forest algorithm to reclassify the extreme environmental impact factors, determines the linkage effect between energy shortages and sensor vulnerability, and outputs adjustment instructions.

[0102] The purified emergency response dataset is imported into the system reliability assessment module. The data is then preliminarily processed using a preset workflow to obtain structured input data.

[0103] In this embodiment, the system reliability assessment module consists of four sub-modules: a dynamic data access kernel, a spatiotemporal feature association engine, an ensemble learning classification matrix, and a linkage effect risk assessor. The preset processing flow is as follows: First, using the timestamp T and site code ST_ID in the emergency dataset as the index key, environmental parameters (such as rainfall intensity) from the local cache are retrieved. Ambient temperature Sediment content of water flow Secondly, feature differences are constructed, and the first-order difference of water level is calculated. To characterize the impact force of the flood peak; finally, max-min normalization is used to scale all heterogeneous features to the [0,1] interval to eliminate the influence of dimensions and obtain a structured input data matrix. The initial data processing is complete. Here, X represents the original feature values, such as heterogeneous data like water level, flow velocity, and rainfall intensity at a specific station at a given time. , These are the minimum and maximum values ​​of the feature in the sample dataset, respectively.

[0104] Based on the structured input data, the random forest algorithm is invoked to perform secondary classification of the influencing factors under extreme environments, and the weight distribution of each type of factor is determined.

[0105] In one possible implementation, firstly, the structured input data... Input a random forest of 200 decision trees, use the Gini index for branching, and output a multi-label probability vector. This is used to identify the subcategories of complex extreme environments (such as kinetic impact, physical depletion, and energy limitation). Then, the feature importance mechanism built into random forests is employed, and the weight distribution of each environmental factor is determined using out-of-bag permutation tests: for each tree in the forest, the classification error using out-of-bag data is calculated. After randomly adding noise to a certain environmental factor characteristic (such as sediment content in water flow), the out-of-bag error is calculated again. Calculate the weight of this factor using the following formula. Where M is the total number of decision trees in the random forest, and in this embodiment, M = 200. A higher weight indicates a stronger decisive influence of that factor on the reliability of the system. Finally, the weights of all factors are normalized to obtain the weight distribution vector W.

[0106] Based on the weighted distribution results, the linkage effect between energy shortages and sensor vulnerability is analyzed. If the correlation of the linkage effect exceeds a preset threshold, it is marked as a high-risk combination, and detailed characteristics of the high-risk combination are obtained. For example, by extracting the time series of the rate of change of the remaining power of the backup power supply and the rate of change of the zero-point drift of the sensor signal within a certain time window, the correlation between the two is calculated using the Pearson correlation coefficient method. The absolute value threshold of the correlation is set to 0.70. When the calculated correlation coefficient reaches or exceeds 0.70, it is determined that there is a strong feedback linkage between energy consumption and sensor loss, and it is marked as a high-risk combination. At the same time, the system captures and outputs a multi-dimensional spatial feature snapshot of the combination, including the correlation coefficient (e.g., 0.85) and the weights of the dominant environmental factors (e.g., high sand content 0.62, weak light 0.28), to characterize the causes of system collapse and the speed of the vicious cycle.

[0107] By analyzing the detailed characteristics of high-risk combinations and considering the influencing factors of extreme environments, targeted environmental adaptability parameters are generated to determine whether the parameters meet the system reliability requirements.

[0108] As a preferred implementation of the present invention, based on the detailed characteristics of high-risk combinations, the system generates targeted environmental adaptability parameters. Including dynamic power budget cap Sensor excitation current correction value and the maximum number of fault-tolerant retransmissions The generation logic is as follows: if the dominant environmental factor is high sand content (leading to signal attenuation), then the signal strength will be automatically increased. This amplifies the physical signal source. Then, it determines whether the parameters meet the system reliability requirements: the system calculates the predicted mean time between failures (MTBF) under the current conditions. It is required that it be no less than the minimum time boundary required for flood control emergency rescue (quantitative standard is...). (hours). If the calculated result is less than 72 hours, it is determined that the preset standard has not been met, and the system automatically converts the difference into a low-level hardware register configuration control word (such as a hexadecimal instruction). This represents "forcing hardware power consumption reduction and channel decoupling", and the execution scope is determined by the spatial topology network adjacency matrix: it not only acts on this site (such as cutting off the power bus of non-core peripherals), but also spreads to the spatially adjacent 2-Hop backbone relay sites through low-power self-organizing network (LoRa), notifying them to increase the data forwarding priority in order to receive the dying burst transmission of this site.

[0109] Based on the environmental adaptability parameter assessment results, if the parameters do not meet the preset standards, i.e., the system reliability requirements, a corresponding adjustment command is generated, and the specific execution range of the adjustment command is determined. By adjusting the execution range of the command, the energy allocation scheme under sensor vulnerability conditions is simulated to obtain an optimized resource allocation strategy. Using the optimized resource allocation strategy, the data records in the system reliability assessment module are updated to determine whether the updated system stability meets the expected requirements.

[0110] As a preferred implementation of the present invention, if the environmental adaptability parameters are determined not to meet the system reliability requirements (i.e.) If the sensor wear level is low (e.g., 1 hour), a corresponding adjustment instruction is generated and the execution range is determined. Based on this, an energy allocation scheme under sensor vulnerability conditions is simulated to obtain an optimized resource allocation strategy. Specifically, the system divides a shadow memory area in the edge microcontroller and runs a lightweight Markov decision process reinforcement learning simulation operator. The state space is defined as the remaining power and sensor wear level, and the action space includes full-channel operation, shutting down the flow rate and leaving only the water level, and entering an ultra-long sleep interval. A transition probability matrix is ​​established, and the reward function is set as a weighted combination of data validity and power consumption rate. The Bellman equation is used to perform 50 policy iterations in the shadow memory. Simulation results show that under the conditions of sensor vulnerability and increased power consumption from repeated data sampling, the action "shutting down high-power peripherals such as radar flow rate and concentrating 90% of the energy for high-frequency multiple sampling noise reduction of the low-power water level gauge" can obtain the highest reward. The system solidifies the simulation results into an optimized resource allocation strategy and executes it in hardware. Subsequently, the data records in the system reliability assessment module are updated using this strategy, and the predicted mean time between failures (MTBF) is recalculated. If the updated predicted value is not less than 72 hours, the system stability is determined to meet the expected requirements.

[0111] S107, after obtaining the output adjustment command, is applied to the dynamic configuration interface of the hydrological station. Through the information processing stage, feedback data on unstable power supply and decreased accuracy are integrated to determine the final system optimization scheme.

[0112] By using the operational status data of hydrological stations, real-time feedback information from the dynamic configuration interface is obtained to determine the specific manifestations of unstable power supply and decreased accuracy.

[0113] The operational status data includes electrical status data and service operation data: electrical status data includes at least the remaining battery power, charging current, discharging current, bus voltage fluctuation rate, and the switching status of each low-power peripheral power supply branch; service operation data includes at least the current sensor sampling frequency, local temporary buffer queue fullness rate, data packet transmission success rate, and BeiDou / 4G signal strength. The above data is processed by the power management chip built into the embedded microcontroller. Alternatively, the data can be obtained by periodically reading the sensor's physical layer registers via the SPI industrial bus and encapsulating it into a timing status frame. The dynamic configuration interface is a set of software function APIs or abstract control register mappings reserved in the hydrological station's main control component, such as (Set_Sensor_Power_Duty(uint8_t channel, float duty) and Update_Sampling_Rate(uint16_t freq)), allowing upper-layer algorithms to dynamically modify configuration words during operation to adjust hardware power consumption and behavior. The criterion for determining power supply insufficiency is: the bus voltage continuously fluctuates within 1 second. Rated voltage, specifically manifested as frequent flashing of low power supply alarms in the system log, reference voltage deviation of the digital-to-analog converter, and reset of the communication module due to instantaneous high current taps. The criteria for judging decreased accuracy are: the statistical variance or bias of the denoised data exceeds 1.5 times the historical average for the same period, or the data quality score is below 85%, specifically manifested as numerous outlier abrupt changes or flat dead lines in the water level time series curve, and a deterioration in the signal-to-noise ratio.

[0114] Based on the feedback information obtained, the data on unstable power supply is classified and processed through information processing to obtain a preliminary judgment on the abnormal mode.

[0115] In one possible implementation, the system invokes the information processing stage, namely the time-domain feature stream pattern recognition engine embedded in the firmware, and uses a dynamic time warping algorithm or a lightweight decision tree for classification. First, a time-series segment of the current power supply voltage fluctuation is captured using a sliding window (window length set to 60 sampling points). Then, the peak-to-peak value, information entropy, and zero-crossing rate of this segment are calculated as feature vectors. Finally, these features are compared with a preset standard anomaly pattern library, and assigned to the closest classification category. Through feature comparison, the following three typical power supply anomaly modes can be identified: Mode A is a leakage short-circuit mode (characterized by a sudden surge in current and a precipitous drop in voltage, caused by water ingress into the circuit board or sensor damage); Mode B is an energy depletion mode (characterized by a monotonous, slow voltage decrease that cannot recover, caused by battery depletion); and Mode C is a pulse electromagnetic interference mode (characterized by high-frequency periodic voltage oscillations, caused by lightning or nearby large high-power equipment). Based on this, the system outputs a preliminary anomaly mode judgment result for power supply instability.

[0116] By classifying and processing abnormal patterns, correlation analysis is performed on data with decreased accuracy, and a preset threshold is used for comparison to identify key factors affecting system operation.

[0117] In one possible implementation, let the supply voltage ripple sequence be... The accuracy of the residual sequence is reduced. ,in The time lag factor is used. The Pearson correlation coefficient is calculated under different lag windows. And find the maximum correlation coefficient. The system has a preset correlation threshold. .when At that time, a deterministic causal chain was determined to exist between unstable power supply and accuracy degradation. Subsequently, the system sorted the correlation coefficients between all candidate environmental factors (including leakage short-circuit mode, energy depletion mode, and pulse electromagnetic interference mode) and their corresponding accuracy degradation indicators in descending order. When the calculation results showed that the correlation coefficient between pulse electromagnetic interference mode and the accuracy degradation of high-frequency clutter in the water level gauge was the highest, this factor was identified as the key factor affecting the current system operation, namely, the penetration of external pulse electromagnetic interference into the power supply link.

[0118] Based on the identified key factors, obtain the output adjustment instructions from the dynamic configuration interface, determine whether the instructions conform to the preset optimization range, and if so, generate the corresponding adjustment parameters.

[0119] In a preferred embodiment of the present invention, the optimization range is a dynamically configured, modifiable safety physical boundary hard constraint, including: a minimum sampling frequency of 0.00028Hz (i.e., at least once per hour), and a sensor excitation power supply voltage adjustment range of 3.3V to 5.0V. When the adjustment command exceeds the above boundaries (e.g., setting the sampling frequency lower or adjusting the voltage below 3.3V), execution is rejected to prevent hardware damage. When the command conforms to the optimization range, specific adjustment parameters are generated based on key factors through a lookup table adaptive control matrix embedded in the firmware, including: peripheral channel shutdown mask, hardware sampling interval time, and digital filter cutoff frequency. For example, when the key factor is the penetration of pulse electromagnetic interference into the power supply link, the adjustment parameters output by the lookup table mapping are: lowering the digital filter cutoff frequency of the level gauge to 0.1Hz to filter out high-frequency glitches, and simultaneously reducing the sensor power supply duty cycle by 20%. The system then dynamically adjusts the hardware operating parameters accordingly.

[0120] By generating adjustment parameters, commands are issued to address system optimization needs, resulting in real-time response data for interface management.

[0121] In a specific embodiment of the present invention, after generating adjustment parameters that conform to the preset optimization range, the system issues instructions through an internal inter-process communication (IPC) mechanism. Specifically, the field-level microcontroller converts the adjustment parameters into hexadecimal control words via a message queue or hardware interrupt service routine, and directly writes them into the SPI or I2C control register of the hardware driver layer, forcing the hardware to perform a configuration rewrite. After the instruction is issued, the dynamic configuration interface immediately sends back real-time response data, including at least: a status code (e.g., 0x00 indicating successful writing) and the current adjusted voltage value (Current_Voltage). The system confirms the instruction execution result based on the real-time response data, completing the dynamic optimization of the hardware operating state.

[0122] Based on the response data from the interface management, the random forest algorithm is used to predict the effect of system optimization and determine the applicability of the final solution.

[0123] In one possible implementation, the currently generated adjustment parameters, real-time response data, and current environmental features are combined into a new feature vector, which is then input into a pre-trained regression random forest model to predict the data quality score and system power consumption change value for the next hour. The dual-indicator joint judgment method is adopted: if the predicted data quality score is not lower than 85 points and the predicted overall system power consumption increment is not greater than 0 (i.e., power consumption does not increase or decreases), the adjustment scheme is judged to be "fully applicable"; otherwise, the system rolls back the current adjustment, regenerates and issues new configuration instructions.

[0124] Based on the applicability results obtained from the predictions, data integration is performed on the final solution to obtain a stable configuration for system operation. Specifically, the integrated data includes: historical denoising features (S101-S105), system reliability MTBF estimation results (S106), and applicability optimization parameters validated by random forest (S107).

[0125] As a preferred example, spatiotemporal metadata encapsulation technology is employed to structurally concatenate the aforementioned parameters according to a unified spatiotemporal context semantic standard, eliminating garbage variables generated during intermediate iterations and compressing them into a global configuration description block. This configuration block is stored in the system EEPROM or non-volatile flash memory in the form of a hexadecimal parameter configuration file; at the software logic layer, it is represented as a standard static JSON data structure or a C language global parameter configuration structure. Example configuration content includes: main bus mode with standby battery power supply, excitation voltage 3300mV, peripheral sleep mask 0x00FF3A11; water level sampling rate 0.0017Hz, radar flow velocity channel closed, digital filter cutoff frequency 0.1Hz; LoRa spreading factor 11, cache priority strategy with water level priority increase, and index tree depth 2.

[0126] During subsequent operation in harsh environments, the IoT terminal of the hydrological station will lock onto and continuously operate with this configuration as the bottom line, no longer blindly searching the network on a large scale or sampling at high frequency, thereby achieving self-healing and energy limit extension in network blind spots.

[0127] like Figure 4 As shown, this invention provides an intelligent deployment system for a hydrological monitoring station network based on the Internet of Things, mainly comprising:

[0128] The data acquisition and risk scoring module is used to collect real-time energy levels, power supply status, and sensor loss indicators under extreme environments through a multimodal sensor array deployed at hydrological stations. After fusion, the data is classified using a random forest algorithm to obtain a quantitative risk score for energy difficulties and unstable power supply.

[0129] The energy and mode optimization module is used to trigger the backup energy module to switch and optimize the sensor working mode to reduce the vulnerability rate when the quantitative risk score exceeds a preset threshold, and to filter noise in the data stream after switching to determine the correction coefficient for the decrease in accuracy.

[0130] The blind zone detection module is used to apply the correction coefficient to the network blind zone detection process. It uses a support vector machine to model the signal strength and data island distribution in the blind zone, determines the coverage of the blind zone and generates a temporary data buffer queue, and obtains a preliminary adjustment strategy to solve the joint problems of communication blind zones, data islands and sensor accuracy degradation.

[0131] The point adjustment module is used to extract optimization parameters related to spatiotemporal redundancy from the mathematical model library for the initial adjustment strategy, adjust the layout points through iterative calculation to cover blind areas and coexisting regions, and determine the source distribution map of data noise.

[0132] The edge denoising module is used to activate the edge computing unit to perform a real-time denoising algorithm on the locally collected data when the noise source distribution map shows a high concentration area, in order to obtain a purified emergency treatment dataset.

[0133] The reliability assessment module is used to input the purified emergency treatment dataset into the overall system reliability assessment module, use the random forest algorithm to reclassify the extreme environmental impact factors, determine the linkage effect between energy difficulties and sensor vulnerability, and output adjustment instructions.

[0134] The dynamic configuration interface module is used to obtain the dynamic configuration interface applied to the hydrological station after the adjustment command is received, integrate feedback data on unstable power supply and decreased accuracy, and determine the final system optimization scheme.

[0135] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. The scope of patent protection of the present invention shall be determined by the claims. Similarly, any equivalent structural changes made based on the description and drawings of the present invention shall also be included within the scope of protection of the present invention.

Claims

1. A method for intelligent deployment of a hydrological monitoring station network based on the Internet of Things, characterized in that, The method includes: By collecting real-time energy levels, power supply status, and sensor loss indicators under extreme environments through a multimodal sensor array deployed at hydrological stations, the data are fused and classified using a random forest algorithm to obtain quantitative risk scores for energy difficulties and unstable power supply. When the quantitative risk score exceeds a preset threshold, the backup power module is switched and the sensor working mode is optimized to reduce the vulnerability rate. The data stream after the switch is noise filtered to determine the correction coefficient for the decrease in accuracy. The correction coefficient is applied to the network blind spot detection process. Support vector machine is used to model the signal strength and data island distribution in the blind spot, determine the coverage of the blind spot and generate a temporary data buffer queue, and obtain a preliminary adjustment strategy to solve the joint problem of communication blind spot, data island and sensor accuracy degradation. For the aforementioned preliminary adjustment strategy, optimization parameters related to spatiotemporal redundancy are extracted from the mathematical model library, and the deployment points are adjusted through iterative calculations to cover areas with both blind spots and noise, thereby determining the distribution map of data noise sources. According to the source distribution map, when the noise source distribution map shows a high concentration area, the edge computing unit is activated to perform a real-time denoising algorithm on the locally collected data to obtain the purified emergency processing dataset. The purified emergency response dataset is input into the overall system reliability assessment module. The random forest algorithm is used to reclassify the extreme environmental impact factors, determine the linkage effect between energy shortages and sensor vulnerability, and output adjustment instructions. After obtaining the adjustment command, it is applied to the dynamic configuration interface of the hydrological station. By integrating feedback data on unstable power supply and decreased accuracy, the final system optimization scheme is determined.

2. The method according to claim 1, characterized in that, The triggering of the backup power module to switch and optimize the sensor operating mode to reduce vulnerability includes: The system automatically detects the current energy status, obtains switching instructions, and completes module replacement to obtain the operating data of the backup energy source. Based on the operational data of the backup power source, the working mode of the sensor is adjusted by reducing the sampling frequency and optimizing the acquisition range, and the output signal of the sensor after adjustment is determined. The data flow direction is extracted from the output signal and preliminary cleaning and noise filtering are performed. The trend of decreased accuracy is analyzed and the correction coefficient is calculated. The correction factor is applied to the data output to complete the accuracy compensation, and the operating status of the backup energy module is continuously monitored.

3. The method according to claim 1, characterized in that, The method of using support vector machines to model the signal strength and data island distribution within the blind zone includes: Signal strength data within network blind zones are collected, and a support vector machine is used to model the signal distribution to obtain preliminary results of the blind zone coverage area. Analyze the distribution of data silos and generate a temporary cache queue; To address the isolated data distribution in the temporary buffer queue, a correction coefficient is used to adjust the signal strength and identify areas with missing signals. When the signal missing area exceeds a preset threshold, the temporary buffer queue is prioritized to obtain a high-priority processing sequence. Adjust the resource allocation plan to determine signal supplementation points and generate a preliminary adjustment strategy.

4. The method according to claim 1, characterized in that, The method of adjusting the deployment points through iterative calculation to cover areas with both blind spots and existing blind spots includes: Optimization parameters related to spatiotemporal redundancy are obtained from a mathematical model library to construct preliminary point distribution data; An iterative calculation method was used to adjust the location of the monitoring points, supplementing areas with insufficient coverage in blind spots, and determining the adjusted layout information of the monitoring points. Based on the adjusted location layout information, obtain the corresponding regional data, analyze the data noise distribution characteristics, and determine the concentrated areas of noise sources; When the noise concentration area overlaps with the blind spot, the density of the points is increased to obtain an updated deployment scheme. Noise signals are extracted from regional data and filtered using preset thresholds to determine the preliminary distribution range of noise sources; Spatial mapping analysis is performed on the initial distribution range to generate a distribution map of the sources of data noise.

5. The method according to claim 1, characterized in that, The activated edge computing unit performs a real-time denoising algorithm on the locally acquired data, including: Analyze the noise source distribution map to identify high-concentration areas; Activate the edge computing unit to initialize and process the locally acquired data stream; Noise filtering is performed using a real-time denoising algorithm; When the data quality reaches a preset threshold, an emergency dataset is output. The emergency dataset is structured, stored in layers, and a data index table is constructed.

6. The method according to claim 1, characterized in that, The process of using the random forest algorithm to reclassify extreme environmental impact factors, determine the linkage effect, and output adjustment instructions includes: The purified emergency response dataset was imported into the reliability assessment module for initial processing. The random forest algorithm is used to perform secondary classification of the factors affecting extreme environments to determine the weight distribution; The correlation between energy shortages and sensor vulnerability is analyzed, and combinations exceeding a preset threshold are marked as high-risk. Generate environmental adaptability parameters, and generate adjustment instructions when the parameters do not meet the preset standards; By simulating the energy allocation scheme under the vulnerable state of the sensor, an optimized resource allocation strategy is obtained.

7. The method according to claim 1, characterized in that, The feedback data on unstable power supply and decreased accuracy are used to determine the final system optimization scheme, including: Obtain real-time feedback information from the dynamic configuration interface to determine the specific manifestations of unstable power supply and decreased accuracy; The abnormal patterns are obtained by classifying and processing the unstable power supply data. Correlation analysis was performed on the data showing decreased accuracy to identify key factors; Generate adjustment parameters based on the output adjustment instructions; By generating adjustment parameters, instructions are issued to address the system optimization needs. The random forest algorithm is used to predict the optimization effect of the system, determine the applicability of the final solution, and integrate them to obtain a stable system configuration.

8. An intelligent deployment system for a hydrological monitoring station network based on the Internet of Things, characterized in that, include: The data acquisition and risk scoring module is used to collect real-time energy levels, power supply status, and sensor loss indicators under extreme environments through a multimodal sensor array deployed at hydrological stations. After fusion, the data is classified using a random forest algorithm to obtain a quantitative risk score for energy difficulties and unstable power supply. The energy and mode optimization module is used to trigger the backup energy module to switch and optimize the sensor working mode to reduce the vulnerability rate when the quantitative risk score exceeds a preset threshold, and to filter noise in the data stream after switching to determine the correction coefficient for the decrease in accuracy. The blind zone detection module is used to apply the correction coefficient to the network blind zone detection process. It uses a support vector machine to model the signal strength and data island distribution in the blind zone, determines the coverage of the blind zone and generates a temporary data buffer queue, and obtains a preliminary adjustment strategy to solve the joint problems of communication blind zones, data islands and sensor accuracy degradation. The point adjustment module is used to extract optimization parameters related to spatiotemporal redundancy from the mathematical model library for the initial adjustment strategy, adjust the layout points through iterative calculation to cover blind areas and coexisting regions, and determine the source distribution map of data noise. The edge denoising module is used to activate the edge computing unit to perform a real-time denoising algorithm on the locally collected data when the noise source distribution map shows a high concentration area, in order to obtain a purified emergency treatment dataset. The reliability assessment module is used to input the purified emergency treatment dataset into the overall system reliability assessment module, use the random forest algorithm to reclassify the extreme environmental impact factors, determine the linkage effect between energy difficulties and sensor vulnerability, and output adjustment instructions. The dynamic configuration interface module is used to obtain the dynamic configuration interface applied to the hydrological station after the adjustment command is received, integrate feedback data on unstable power supply and decreased accuracy, and determine the final system optimization scheme.

Images