Foundation pit monitoring system and method of self-repairing wireless sensor network

Abstract

The invention relates to the technical field of foundation pit monitoring, in particular to a foundation pit monitoring system of a self-repairing wireless sensor network and a method thereof, which realize high-precision (+ / -0.5 mm) displacement monitoring by integrating a multi-source heterogeneous data acquisition module, a self-diagnosis processing module, a predictive fault management module, a network topology management module and an intelligent resource scheduling module. The system adopts multi-dimensional information fusion and preprocessing, the perception capability is improved, nodes and network states are comprehensively monitored through a hierarchical health degree evaluation system, a foundation is laid for fault prediction and active maintenance, the system can dynamically adjust a network topology structure, resource allocation is optimized by using a reinforcement learning algorithm, and the system has a wide application prospect. Node task allocation, communication power control and sampling frequency adjustment are effectively executed, and the efficiency and reliability of foundation pit monitoring are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of foundation pit monitoring, and particularly to a foundation pit monitoring system and method of a self-repairing wireless sensor network. Background Art

[0002] With the rapid development of infrastructure construction in China, building structures are constantly becoming larger and more complex, posing higher requirements for the safety of building foundation pits. Although the construction safety and quality of foundation pits have been widely emphasized in China, due to the engineering complexity, construction accidents of foundation pit projects occur from time to time.

[0003] In the prior art, for example, the Chinese patent application with the publication number CN117073559A discloses a foundation pit deformation monitoring method and system based on machine vision. This method combines a binocular monitoring host and a positioning camera. The positioning camera is used to solve the pose disturbance problem of the binocular monitoring host, and by setting a main target and a positioning target on the target pole regarded as a rigid body, the problem of the deviation between the displacement of the main target and the actual displacement of the foundation pit is solved. Although this technology improves the monitoring accuracy and stability to a certain extent, there are still the following deficiencies:

[0004] 1. The system has a risk of single-point failure. The failure of any key device will cause the entire monitoring system to fail, lacking a redundant backup mechanism;

[0005] 2. The energy management efficiency is low. The target device is powered by a battery, and the battery life is limited. The battery needs to be replaced or charged frequently, increasing the maintenance cost and downtime;

[0006] 3. The communication reliability is insufficient. It only relies on a single 4G communication method, and data transmission interruption is likely to occur in areas with unstable signals;

[0007] 4. The data integrity guarantee mechanism is lacking. When data is lost due to equipment failure or communication interruption, it cannot be recovered, lacking data reconstruction and compensation algorithms;

[0008] 5. The system has a low degree of intelligence. It cannot automatically adjust monitoring parameters according to environmental conditions, lacking active monitoring and early warning of the health status of equipment;

[0009] 6. The network topology structure is fixed. Once deployed, it is difficult to adjust, and the communication network cannot be reconstructed according to changes in site conditions.

[0010] Therefore, there is an urgent need for a foundation pit monitoring system with self-repairing ability, high reliability, and high intelligence to overcome the deficiencies in the prior art. Summary of the Invention

[0011] The object of the present invention is to provide a foundation pit monitoring system and method for a self-healing wireless sensor network, so as to solve the technical problems existing in the prior art, such as single-point failure risk, low energy management efficiency, insufficient communication reliability, and lack of data integrity guarantee mechanism.

[0012] The present invention proposes a foundation pit monitoring system for a self-healing wireless sensor network, comprising:

[0013] A multi-source heterogeneous data acquisition module, configured to acquire displacement data, environmental parameters, and structural stress data within the foundation pit monitoring area;

[0014] A self-diagnosis processing module, communicatively connected to the multi-source heterogeneous data acquisition module, and configured to perform preprocessing, anomaly detection, and health assessment on the multi-source heterogeneous data;

[0015] A predictive fault management module, communicatively connected to the self-diagnosis processing module, and configured to predict equipment failures based on the health assessment results and generate data reconstruction strategies;

[0016] A network topology management module, communicatively connected to the predictive fault management module, and configured to dynamically adjust the network topology structure according to the equipment failure prediction results; and

[0017] An intelligent resource scheduling module, communicatively connected to the predictive fault management module and the network topology management module respectively, and configured to optimize the system resource allocation based on a reinforcement learning algorithm, and perform node task allocation, communication power control, and sampling frequency adjustment according to the network topology structure adjustment and the data reconstruction strategy.

[0018] Preferably, the multi-source heterogeneous data acquisition module comprises:

[0019] A displacement sensing unit, configured to acquire displacement data of foundation pit feature points;

[0020] An environmental parameter acquisition unit, configured to acquire temperature, humidity, air pressure, and light intensity;

[0021] A structural stress acquisition unit, configured to acquire vibration and strain data; and

[0022] A data preprocessing unit, configured to perform filtering, noise reduction, and normalization processing on the displacement data, the environmental parameters, and the structural stress data.

[0023] Preferably, the self-diagnosis processing module comprises:

[0024] A hardware self-check unit, configured to monitor the sensor calibration status, the communication link quality, and the power supply status;

[0025] A health calculation unit, configured to generate a health score based on multi-dimensional health indicators and a service life decay factor; and

[0026] Anomaly detection unit, which is used to detect data anomalies based on the adaptive threshold algorithm and the Isolation Forest method.

[0027] Preferably, the health degree calculation unit calculates the node health degree through the following formula:

[0028] H = (∑(W i ·S i ))·(1 - αD),

[0029] where H is the node health degree, W i is the weight of each dimension health index, S i is the score of each dimension, D is the service life attenuation factor, α is an adjustable parameter, and its value range is 0.05 - 0.15.

[0030] Preferably, the predictive fault management module includes:

[0031] A fault prediction unit, which is used to generate future health degree trajectories and fault probabilities based on the time series prediction network of the Transformer architecture;

[0032] A knowledge distillation unit, which is used to convert the complex model trained in the cloud into a lightweight model suitable for edge deployment; and

[0033] A data reconstruction unit, which is used to reconstruct missing data based on spatial correlation modeling and the multivariable spatio-temporal tensor decomposition algorithm.

[0034] Preferably, the network topology management module includes:

[0035] A topology awareness unit, which is used to discover the network topology structure based on the link quality evaluation matrix and the node criticality calculation;

[0036] A dynamic topology optimization unit, which is used to generate an optimized network topology scheme based on the node health degree and the link quality; and

[0037] A topology migration execution unit, which is used to achieve seamless topology switching through the two-phase commit mechanism.

[0038] Preferably, the intelligent resource scheduling module includes:

[0039] A resource status awareness unit, which is used to monitor the status of computing resources, communication resources, and energy resources;

[0040] A multi-objective resource optimization unit, which is used to generate a resource scheduling strategy based on the deep reinforcement learning network; and

[0041] A self-optimization execution unit, which is used to execute resource allocation decisions through a hierarchical scheduling mechanism and an adaptive control loop.

[0042] Preferably, the resource scheduling strategy includes sampling frequency control, computing task allocation, communication power control, and sleep scheduling, and balances monitoring accuracy, energy efficiency, and system lifespan through a multi-objective optimization function.

[0043] Preferably, the reconstructed data generated by the data reconstruction unit is subjected to displacement continuity and maximum deformation rate constraints through a Bayesian inference engine to ensure that the reconstructed data conforms to physical laws.

[0044] The foundation pit monitoring method for a self-healing wireless sensor network includes the following steps:

[0045] Collect displacement data, environmental parameters, and structural stress data within the foundation pit monitoring area;

[0046] Preprocess, perform anomaly detection, and evaluate the health of the displacement data, the environmental parameters, and the structural stress data;

[0047] Predict equipment failures and generate a data reconstruction strategy based on the health assessment results;

[0048] Dynamically adjust the network topology according to the equipment failure prediction results; and

[0049] Optimize system resource allocation based on a reinforcement learning algorithm, and perform node task allocation, communication power control, and sampling frequency adjustment according to the network topology adjustment and the data reconstruction strategy.

[0050] The beneficial effects of the present invention include:

[0051] 1. Through a multi-source heterogeneous data fusion and preprocessing mechanism, multi-dimensional information collection and analysis are realized, the system perception ability is enhanced, and the displacement monitoring accuracy is improved from the traditional ±2 mm to ±0.5 mm;

[0052] 2. By adopting a hierarchical health assessment system, comprehensive monitoring of the node and network states is realized, providing a basis for fault prediction and proactive maintenance;

[0053] 3. By introducing a predictive fault management and data reconstruction framework, equipment failures can be predicted 3 hours in advance and corresponding measures can be automatically taken to ensure that the data integrity rate reaches 99.9%;

[0054] 4. By constructing an adaptive network topology reconstruction mechanism, the system can dynamically adjust the communication structure, effectively improving the reliability and anti-interference ability of the network;

[0055] 5. By adopting an intelligent resource scheduling and self-optimization system, resource allocation is continuously optimized based on the reinforcement learning method, the average power consumption of nodes is reduced by 40%, and the battery life is extended by 65%;

[0056] 6. The continuous working ability of the overall system has been improved from 85% of the traditional solution to 99.7%, the on-site maintenance frequency has been reduced by 68%, and the total cost of ownership over the entire life cycle has been reduced by 35%. Description of the Drawings

[0057] Figure 1 It is a schematic diagram of the overall architecture of the foundation pit monitoring system of the self-healing wireless sensor network of the present invention.

[0058] Figure 2 It is a schematic diagram of the structure of the multi-source heterogeneous data acquisition module of the present invention.

[0059] Figure 3 It is a schematic diagram of the structure of the self-diagnosis processing module of the present invention.

[0060] Figure 4 It is a flowchart of the working process of the health calculation unit of the present invention.

[0061] Figure 5 It is a schematic diagram of the structure of the predictive fault management module of the present invention.

[0062] Figure 6 It is a schematic diagram of the structure of the network topology management module of the present invention.

[0063] Figure 7 It is a schematic diagram of the structure of the intelligent resource scheduling module of the present invention.

[0064] Figure 8 It is a flowchart of the foundation pit monitoring method of the self-healing wireless sensor network of the present invention.

[0065] Figure 9 It is a comparison diagram of the monitoring effects of the system of the present invention in a certain project. Detailed Embodiments

[0066] Please refer to the attached Figures 1-9 drawings. Hereinafter, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention.

[0067] Embodiment 1

[0068] As Figure 1 shown, the foundation pit monitoring system of the self-healing wireless sensor network provided by the present invention includes a multi-source heterogeneous data acquisition module 1, a self-diagnosis processing module 2, a predictive fault management module 3, a network topology management module 4, and an intelligent resource scheduling module 5.

[0069] The multi-source heterogeneous data acquisition module 1 is used to acquire displacement data, environmental parameters, and structural stress data within the foundation pit monitoring area. The self-diagnosis and processing module 2 is communicatively connected to the multi-source heterogeneous data acquisition module 1 and is used to preprocess the multi-source heterogeneous data, detect anomalies, and evaluate the health status. The predictive fault management module 3 is communicatively connected to the self-diagnosis and processing module 2 and is used to predict equipment failures based on the health status evaluation results and generate data reconstruction strategies. The network topology management module 4 is communicatively connected to the predictive fault management module 3 and is used to dynamically adjust the network topology structure according to the equipment failure prediction results. The intelligent resource scheduling module 5 is communicatively connected to the predictive fault management module 3 and the network topology management module 4 respectively and is used to optimize the system resource allocation based on the reinforcement learning algorithm, and perform node task allocation, communication power control, and sampling frequency adjustment according to the network topology structure adjustment and data reconstruction strategy.

[0070] The above-mentioned modules are interconnected through a wireless communication network or a wired connection to form a complete monitoring system. The system adopts a hierarchical distributed architecture, and each module can be flexibly configured and deployed according to actual needs. At the foundation pit monitoring site, the multi-source heterogeneous data acquisition module 1 is deployed at key monitoring points and is connected to the self-diagnosis and processing module 2 and the predictive fault management module 3 through a self-organizing network. The network topology management module 4 and the intelligent resource scheduling module 5 can be deployed in on-site edge computing devices and are responsible for the overall coordination and management of the network and resources.

[0071] The working process of the system is as follows: The multi-source heterogeneous data acquisition module 1 continuously acquires foundation pit monitoring data; the self-diagnosis and processing module 2 processes the acquired data in real time and evaluates the health status of the nodes; the predictive fault management module 3 predicts possible faults based on the health status and formulates countermeasures; the network topology management module 4 dynamically adjusts the network structure to adapt to changes; the intelligent resource scheduling module 5 optimizes the resource allocation to ensure the efficient operation of the system. This intelligent closed-loop architecture enables the system to have the capabilities of self-monitoring, self-diagnosis, and self-repair, significantly improving the reliability and efficiency of foundation pit monitoring.

[0072] Embodiment 2

[0073] As Figure 2 shown, the multi-source heterogeneous data acquisition module 1 includes a displacement sensing unit 11, an environmental parameter acquisition unit 12, a structural stress acquisition unit 13, and a data preprocessing unit 14.

[0074] The displacement sensing unit 11 uses a micro-millimeter wave displacement sensor array to replace the traditional vision target. Three sensors with a 120° angle are configured for each foundation pit feature point. The working frequency is 76 - 81 GHz, the measurement accuracy reaches ±0.5 mm, and the measurement range is 0 - 50 mm. Preferably, the displacement sensing unit 11 uses a TI IWR6843 chip, which has the characteristics of being resistant to light changes and still being able to work normally under harsh weather conditions.

[0075] The environmental parameter acquisition unit 12 is used to acquire environmental parameters such as temperature, humidity, air pressure, and light intensity. Preferably, the measurement range of the temperature sensor is -40°C to 85°C, and the accuracy is ±0.5°C; the measurement range of the humidity sensor is 0 - 100%RH, and the accuracy is ±3%RH; the measurement range of the air pressure sensor is 30 - 110 kPa, and the accuracy is ±0.1 kPa; the measurement range of the light intensity sensor is 0 - 65535 lux.

[0076] The structural stress acquisition unit 13 is used to acquire vibration and strain data. Preferably, the measurement range of the triaxial accelerometer is ±16g, and the resolution is 0.001g; the range of the strain gauge is ±3000 με, and the resolution is 1 με. These data can be used to evaluate the dynamic response characteristics of the foundation pit structure and provide supplementary information for monitoring.

[0077] The data preprocessing unit 14 performs filtering, noise reduction, and normalization processing on the acquired displacement data, environmental parameters, and structural stress data. Preferably, the wavelet transform denoising algorithm is used to eliminate environmental interference, and the sliding window is used for data smoothing processing. The window size can be adaptively adjusted between 7 and 31. At the same time, missing value imputation based on local linear regression is implemented, and time-domain features (mean, variance, kurtosis), frequency-domain features (power spectral density), and change rate features are extracted.

[0078] The data sampling frequency adopts an adaptive adjustment mechanism: it is 5 minutes per time in the normal state, and can be increased to a maximum of 10 seconds per time when an abnormal state is detected. The triggering methods of data acquisition include three modes: timing trigger, event trigger, and remote command trigger, which can be flexibly switched according to on-site requirements.

[0079] Each acquisition unit is connected to the data preprocessing unit 14 through an internal bus, and the preprocessed data is transmitted to the self-diagnosis processing module 2 through a low-power wireless communication module (such as LoRa or ZigBee) for further analysis.

[0080] Embodiment 3

[0081] As Figure 3 shown, the self-diagnosis processing module 2 includes a hardware self-check unit 21, a health calculation unit 22, and an anomaly detection unit 23.

[0082] The hardware self-check unit 21 is used to monitor the sensor calibration status, communication link quality, and power supply status. Preferably, the hardware layer self-check includes sensor calibration status check, analog-to-digital converter (ADC) self-check, and communication module link test; resource monitoring includes CPU load (threshold 80%), memory usage (threshold 75%), and storage space (threshold 85%); power management includes voltage fluctuation monitoring (±0.2V alarm), charge and discharge cycle count, and power prediction algorithm; communication quality assessment includes received signal strength indication (RSSI) monitoring, packet loss rate statistics, and channel quality index (CQI) analysis. The self-check period is configurable, and is generally set to a full self-check once per hour and a quick self-check once per minute.

[0083] The health calculation unit 22 generates a health score based on multi-dimensional health indicators and the service life attenuation factor. As Figure 4 shown, the health calculation process includes four steps: index collection weight assignment, health calculation, and time series update.

[0084] Preferably, the health calculation unit 22 uses a multi-dimensional health indicator weight assignment matrix W = [0.35, 0.25, 0.20, 0.20], corresponding to the four dimensions of sensing function, computing resources, power supply status, and communication ability respectively. The health calculation formula is:

[0085] H = (W i - S i )(1 - D),

[0086] where H is the node health, W i is the weight of each dimension's health indicator, S i is the score of each dimension (value range 0 - 1), D is the service life attenuation factor, which is an adjustable parameter, and the preferred value range is 0.05 - 0.15.

[0087] To ensure the temporal coherence of the health assessment, an exponential moving average update mechanism is adopted:

[0088] H t = αH t-1 + (1 - α)H new ,

[0089] where α is preferably set to 0.7 - 0.9, H t is the health score at time t, H t -1 is the health score at the previous moment, and H new is the newly calculated health value.

[0090] The health score ranges from 0 to 1 and is divided into five levels: excellent (0.9 - 1.0), good (0.8 - 0.9), fair (0.7 - 0.8), attention (0.6 - 0.7), and warning (below 0.6).

[0091] The anomaly detection unit 23 detects data anomalies based on the adaptive threshold algorithm and the Isolation Forest method. Preferably, the adaptive threshold calculation uses the formula μ + k·σ, where μ is the historical mean, σ is the standard deviation, and k is an adjustable coefficient with a value range of 1.5 - 3.5.

[0092] The anomaly levels are divided into three levels: attention (0.7 - 0.85), warning (0.85 - 0.95), and emergency (0.95). In addition, Z - score normalization processing and the Isolation Forest anomaly point detection algorithm are used, and the contamination rate parameter is set to 0.03 to effectively identify anomaly patterns in multi - dimensional data.

[0093] The self - diagnosis processing module 2 exchanges data with the multi - source heterogeneous data acquisition module 1 through a secure wireless communication protocol, supporting data encryption and integrity verification. The processing results are transmitted to the predictive fault management module 3 in the form of a health report and an anomaly event, providing basic data for subsequent fault prediction and management.

[0094] Embodiment 4

[0095] The health calculation unit 22 calculates the node health through the following formula:

[0096] H = (W i - S i )(1 - D),

[0097] In actual implementation, each parameter of this formula has a clear physical meaning and value - taking method. The scores Si of each dimension are normalized so that their value ranges are unified to 0 - 1. The specific calculation method is as follows:

[0098] The sensing function score S1 consists of three sub - indicators: calibration status score (accounting for 40%), signal quality score (accounting for 35%), and responsiveness score (accounting for 25%).

[0099] The calibration status score is based on the time interval and drift amount of the most recent calibration; the signal quality score is based on the signal - to - noise ratio and stability; the responsiveness score is based on the sensor response time and sampling consistency.

[0100] The computing resource score S2 consists of three indicators: CPU load, memory usage rate, and storage status, and each indicator has an equal weight. The calculation formula is:

[0101]

[0102] Among them, CPUload, MEMuse, and STORuse respectively represent the usage percentages of CPU, memory, and storage.

[0103] The power status score S3 takes into account three factors: battery level, number of charge cycles, and voltage stability. The calculation formula is:

[0104]

[0105] Among them, BAT level is the battery level percentage, CYCcount is the current number of charge cycles, CYCmax is the maximum number of charge cycles (usually 500 - 1000 times), and VOLstability is the voltage stability score (0 - 1).

[0106] The communication ability score S4 comprehensively considers signal strength, packet loss rate, and channel quality. The calculation formula is:

[0107] Among them, RSSI score is the normalized RSSI score (0 - 1), PLOSS rate is the packet loss rate (0 - 1), and CQI is the channel quality index (0 - 15).

[0108] The service life attenuation factor D represents the ratio of the device usage time to the design life. The calculation formula is:

[0109] Among them, T current is the current service time (hours), and T design is the design life (hours). The parameter α is used to adjust the influence degree of the service life on the health degree. The larger the α value, the more obvious the attenuation effect of the usage time on the health degree.

[0110] During the long - term operation of the system, the parameters for health degree calculation can be fine - tuned according to the actual situation to adapt to different engineering environments and requirements. The health degree score, as an important basis for the system's self - repair decision - making, can effectively reflect the overall working state and potential risks of the node.

[0111] Example 5

[0112] As Figure 5 shown, the predictive fault management module 3 includes a fault prediction unit 31, a knowledge distillation unit 32, and a data reconstruction unit 33.

[0113] The fault prediction unit 31 generates future health trajectories and fault probabilities based on a time series prediction network with a Transformer architecture. Preferably, this network consists of an input layer, an encoding layer, a decoding layer, and an output layer. The input layer receives the health time series of the nodes, with an input window length of 168 hours (one week); the encoding layer contains 8 attention heads, and the hidden layer dimension is 512; the decoding layer consists of 3 layers of GRU (Gated Recurrent Unit), with 128 neurons in each layer; the output layer predicts the health trajectory and fault probability for the next 72 hours. In addition, the fault prediction unit 31 also includes a fault type classifier that classifies 7 common fault types (sensor drift, communication failure, power problem, memory corruption, physical damage, software defect, unknown) using the SoftMax function. The early warning mechanism is triggered when the predicted health is lower than 0.6 or when it drops by more than 20% within 3 hours.

[0114] The knowledge distillation unit 32 converts the complex model trained in the cloud into a lightweight model suitable for edge deployment. Preferably, the teacher model is a complex Transformer network trained in the cloud, with approximately 10M parameters; the student model is a lightweight network suitable for edge deployment, with approximately 300K parameters. The distillation temperature T is set to 2.5, and the soft label weight α is set to 0.7. Through model pruning and quantization (4-bit fixed-point quantization), the computational complexity is reduced by 85%, while keeping the loss of prediction performance within an acceptable range (the accuracy reduction does not exceed 5%). This enables the complex fault prediction algorithm to run efficiently on resource-constrained edge devices.

[0115] The data reconstruction unit 33 reconstructs missing data based on spatial correlation modeling and a multivariable spatio-temporal tensor decomposition algorithm. Preferably, first, a proximity relationship matrix is constructed based on the Euclidean distance and measurement correlation between nodes to identify donor nodes suitable as references for data reconstruction. Then, a low-rank tensor completion algorithm based on CP decomposition is used to process multi-dimensional missing data, and the formula can be expressed as:

[0116]

[0117] where X is the original data tensor, W is the mask tensor indicating missing values, is the reconstructed complete tensor, ⊙ represents the Hadamard product, ||·|| F is the Frobenius norm, ||·|| * is the nuclear norm, and λ is the regularization parameter.

[0118] In addition, the data reconstruction unit 33 also includes a Bayesian inference engine, which combines prior knowledge of physical constraints (displacement continuity, maximum deformation rate) to ensure that the reconstructed data conforms to physical laws. The evaluation of the reconstructed data quality is based on the consistency test of historical data patterns, and the confidence interval is set at 95%. For data points that do not meet the requirements of the reconstruction quality, the system will mark and notify the operation and maintenance personnel for on-site inspection.

[0119] The predictive fault management module 3 establishes two-way communication connections with the self-diagnosis processing module 2, the network topology management module 4, and the intelligent resource scheduling module 5. On the one hand, it receives the health assessment results provided by the self-diagnosis processing module 2; on the other hand, it sends the fault prediction and data reconstruction strategies to the network topology management module 4 and the intelligent resource scheduling module 5 to cooperate in completing the system self-repair.

[0120] Embodiment 6

[0121] As Figure 6 shown, the network topology management module 4 includes a topology awareness unit 41, a dynamic topology optimization unit 42, and a topology migration execution unit 43.

[0122] The topology awareness unit 41 discovers the network topology structure based on the link quality assessment matrix and node criticality calculation. Preferably, a distributed network topology discovery protocol is adopted, based on the improved RPL (Routing Protocol for Low-Power and Lossy Networks) protocol, supporting the 6LoWPAN (IPv6 Low-Power Wireless Personal Area Network) standard. The link quality assessment matrix combines the ETX (Expected Transmission Count) and PRR (Packet Reception Rate) metrics, and the calculation formula is:

[0123]

[0124] where LQI is the link quality index (0 - 1), ω is the weight coefficient, and the preferred value is 0.6.

[0125] The node criticality is calculated based on betweenness centrality and degree centrality, and is used to identify critical nodes in the network. The network bottleneck identification algorithm is based on the maximum flow minimum cut theorem and is used to discover potential network weak links.

[0126] The dynamic topology optimization unit 42 generates an optimized network topology scheme based on node health and link quality. Preferably, a graph reconstruction algorithm based on health and link quality is adopted, and the objective function is:

[0127] min(w1·E + w2·(1 - R)),

[0128] where E represents network energy consumption, R represents data transmission reliability, and w1 and w2 are weight coefficients (w1 + w2 = 1).

[0129] The constraints include: the node healthiness is not lower than the threshold (default 0.6), the hop count limit (maximum 5 hops), and the latency requirement (lower than 250 ms). In addition, the dynamic topology optimization unit 42 also implements a role dynamic switching mechanism, supporting the promotion strategy from ordinary node → relay node → cluster head node, and calculating the role fitness score based on the energy level, processing capacity, and location. The redundant path planning ensures that any node has at least 2 independent paths connected to the gateway, improving the fault tolerance of the network.

[0130] The topology migration execution unit 43 realizes seamless topology switching through a two-phase commit mechanism. Preferably, the topology migration is divided into a preparation stage and an execution stage. The preparation stage verifies the feasibility of the new topology to ensure that all nodes can correctly receive and process the topology update instructions; the execution stage adopts a synchronous switching strategy to uniformly update the routing tables and role settings of all nodes at a preset time point. The rollback guarantee mechanism can automatically restore the original topology in case of a switching failure, avoiding network paralysis. The software-defined network control plane realizes centralized policy distribution and distributed execution, improving the flexibility and efficiency of topology adjustment.

[0131] The network topology management module 4 maintains real-time data exchange with the predictive fault management module 3 and the intelligent resource scheduling module 5. It receives the fault warning information from the predictive fault management module 3 and sends the topology adjustment results to the intelligent resource scheduling module 5 for corresponding resource optimization. The frequency of topology adjustment is dynamically determined according to the network state, checking once every 12 hours under normal circumstances and triggering inspections and adjustments immediately when potential faults or network performance degradation are detected.

[0132] Embodiment 7

[0133] As Figure 7 shown, the intelligent resource scheduling module 5 includes a resource status perception unit 51, a multi-objective resource optimization unit 52, and a self-optimization execution unit 53.

[0134] The resource status perception unit 51 is used to monitor the status of computing resources, communication resources, and energy resources. Preferably, the computing resource monitoring includes CPU usage, memory occupancy, and storage space; the communication resource monitoring includes bandwidth occupancy, channel utilization, and communication window allocation; the energy resource monitoring includes battery power, energy harvesting efficiency, and power consumption distribution. The task load modeling constructs a resource consumption prediction model based on historical task execution data. The environment perception adaptation mechanism dynamically adjusts the resource strategy according to environmental factors such as light and temperature, for example, increasing the sampling frequency when the light is sufficient and reducing the energy consumption when the light is insufficient.

[0135] The multi-objective resource optimization unit 52 generates a resource scheduling policy based on a deep reinforcement learning network. Preferably, a method based on deep Q-learning is adopted. The state space includes node health, network topology status, environmental conditions, and task queues. The action space includes sampling frequency adjustment, processing task allocation, communication power control, and sleep strategy. The reward function is:

[0136] R = w1·Accuracy + w2·EnergyEfficiency + w3·SystemLifespan - w4·RiskFactor,

[0137] where w1, w2, w3, and w4 are weight coefficients, preferably set to [0.45, 0.30, 0.20, 0.05]. The deep Q-learning network structure includes: input layer (state vector) → 3 fully connected layers (256 → 128 → 64) → output layer (action value). The capacity of the experience replay buffer is 10,000 entries, the batch training size is 64, and the target network update frequency is 100 steps. Initially, the system adopts a rule-based scheduling policy and gradually transitions to an optimization policy based on reinforcement learning as experience accumulates.

[0138] The self-optimizing execution unit 53 executes resource allocation decisions through a hierarchical scheduling mechanism and an adaptive control loop. Preferably, the hierarchical scheduling mechanism includes three levels: system level (global resource allocation and load balancing, 5-minute cycle), cluster level (intra-region node cooperation and task sharing, 1-minute cycle), and node level (fine regulation of internal functional modules, 10-second cycle). The adaptive control loop implements closed-loop feedback regulation based on a PID controller, and the control parameters are dynamically adjusted according to the system response. The conflict resolution mechanism solves scheduling conflicts based on priorities and resource dependency graphs to ensure the priority execution of critical tasks. The dynamic policy evaluation continuously monitors the policy execution effect and makes real-time adjustments to policies that do not meet expectations.

[0139] The intelligent resource scheduling module 5 is the core execution unit of the system self-healing engine. It receives the fault management policy from the predictive fault management module 3 and the topology adjustment result from the network topology management module 4, generates an optimal resource allocation plan after comprehensively considering various factors, and issues execution instructions to each functional module. Through continuous self-optimization, the system can maximize energy efficiency and system lifespan while ensuring monitoring accuracy.

[0140] Embodiment 8

[0141] The resource scheduling policy includes sampling frequency control, computing task allocation, communication power control, and sleep scheduling, and balances monitoring accuracy, energy efficiency, and system lifespan through a multi-objective optimization function.

[0142] In specific implementation, the sampling frequency control strategy dynamically adjusts the data acquisition frequency for different types of sensors and different monitoring scenarios. The basic sampling period is set to 300 seconds (5 minutes), supporting event-driven adaptive adjustment, and the sampling frequency range is from 60 seconds (high frequency) to 1800 seconds (low frequency). The triggering conditions include: increasing the sampling rate when the node health is lower than 0.7; increasing the sampling rate when the displacement change rate exceeds 0.2 mm / h; reducing the sampling rate when the battery power is lower than 30%. This strategy can not only ensure the acquisition density of key data but also effectively extend the battery life.

[0143] The computing task allocation strategy reasonably allocates data processing tasks according to the computing power and health status of the nodes. Compute-intensive tasks (such as tensor completion) are offloaded to the edge gateway for processing when the CPU utilization exceeds 80%; lightweight tasks (such as anomaly detection) are processed locally when the node health is higher than 0.75 and the battery power is higher than 40%. This differential allocation strategy can balance the system load and avoid overloading a single node.

[0144] The communication power control strategy dynamically adjusts the transmission power according to the link quality, minimizing energy consumption while ensuring communication reliability. When the link quality is high, the power is reduced to 60% of the maximum power; when the link quality is medium, the power is set to 80% of the maximum power. At the same time, adaptive power boosting is supported, and the power is temporarily increased when an increase in the packet loss rate is detected. This strategy saves an average of 25% of the communication energy consumption compared to the fixed power scheme.

[0145] The sleep scheduling strategy minimizes the standby power consumption by reasonably arranging the active and sleep cycles of the nodes. The typical sleep cycle parameters are: 3 seconds for the active period, 27 seconds for the sleep period, and 0.1 second for pre-sleep listening. When a critical event is detected, the system will temporarily extend the active period (up to 300 seconds); when a low activity state is detected, the sleep period will be increased (extension factor 1.5). This strategy reduces the average energy consumption of the nodes by more than 40%.

[0146] The above four types of strategies are coordinated and adjusted through a multi-objective optimization function to ensure the optimal overall performance of the system. The weights of the optimization objectives can be dynamically adjusted according to engineering requirements. For example, the weight of monitoring accuracy is increased during the critical monitoring period, and the weight of system life is increased during the long-term unattended period. After testing, compared with the fixed strategy, this adaptive resource scheduling scheme reduces the power consumption by an average of 22.7%, increases the data transmission rate by 6.2%, and reduces the communication delay by 35.9%.

[0147] Example 9

[0148] The reconstructed data generated by the data reconstruction unit is subjected to displacement continuity and maximum deformation rate constraints through the Bayesian inference engine to ensure that the reconstructed data conforms to physical laws.

[0149] In specific implementation, when the system predicts that a node is about to fail or has detected data loss, the data reconstruction process includes four steps: spatial correlation modeling, multi-variable spatio-temporal tensor decomposition, physical constraint application, and reconstruction quality assessment.

[0150] Spatial correlation modeling first calculates the positional relationship and data correlation between nodes in the monitoring network. The positional relationship is quantified by the Euclidean distance; the data correlation is calculated by the Pearson correlation coefficient, and nodes with a correlation coefficient greater than 0.7 are selected as candidate donor nodes. Preferably, the system selects 3 - 5 donor nodes for each node to build the basis for data reconstruction.

[0151] Multi-variable spatio-temporal tensor decomposition uses an improved tensor CP decomposition algorithm to process a third-order tensor containing dimensions of time, space, and monitoring parameters. Preferably, the tensor rank is set to where I, J, and K are the sizes of the three dimensions of the tensor respectively. To improve the calculation efficiency, the alternating least squares method is used to solve the decomposition problem, and the number of iterations is set to 100 or the convergence error is less than 1e -4 。

[0152] For the third-order tensor its CP decomposition form is:

[0153]

[0154] where, λ r is the weight coefficient, a r ∈R I , b r ∈R J , c r ∈R K are the factor vectors obtained by decomposition, represents the outer product of vectors, and R is the rank of the tensor.

[0155] The optimization objective is to minimize the reconstruction error:

[0156]

[0157] where, ||·|| F is the Frobenius norm, is the tensor reconstructed by CP decomposition.

[0158] The solution process uses the alternating least squares method (ALS) to alternately optimize each factor matrix (such as A, B, C) until one of the following conditions is met: the number of iterations reaches 100; the convergence error is less than 1×10 -4 。

[0159] This setting can effectively improve the calculation efficiency while ensuring the decomposition accuracy, and is applicable to the analysis and processing of large-scale multi-variable spatio-temporal data.

[0160] The physical constraints are implemented through a Bayesian inference engine, and physical laws are introduced as prior information into the reconstruction process. The displacement continuity constraint ensures that the displacement changes smoothly at adjacent time points, and the maximum change rate is usually set to 0.3 mm / h; the maximum deformation rate constraint is set based on engineering experience, usually 0.1%-0.3% of the foundation pit depth. The Bayesian inference formula is:

[0161] P(D|X) ∝ P(X|D)·P(D),

[0162] where D is the reconstructed data, X is the observed data, P(D) is the prior distribution (including physical constraints), P(X|D) is the likelihood function, and P(D|X) is the posterior distribution. The reconstructed data that conforms to physical laws is obtained through maximum a posteriori (MAP) estimation.

[0163] The reconstruction quality is evaluated using the cross-validation method. Some points are randomly masked from the known data, and the deviation between the reconstructed value and the true value is compared. The evaluation indicators include root mean square error (RMSE), mean absolute error (MAE), and coefficient of determination (R 2 ). Preferably, when RMSE is less than 0.5 mm, MAE is less than 0.3 mm, and R 2 is greater than 0.9, it is considered that the reconstruction quality meets the requirements.

[0164] Practice shows that the data reconstruction method combined with physical constraints has about 35% higher reconstruction accuracy than the pure statistical method, especially when the data missing rate is relatively high (30%). This method has been verified effective in multiple practical projects and can ensure the continuity and reliability of monitoring data in the case of sensor node failures.

[0165] Example 10

[0166] As Figure 8 shown, the foundation pit monitoring method of the self-healing wireless sensor network of the present invention includes the following steps:

[0167] Step S1: Collect displacement data, environmental parameters, and structural stress data in the foundation pit monitoring area. Specifically, use a micro-millimeter wave displacement sensor array to collect displacement data of foundation pit feature points; use temperature and humidity sensors, barometric pressure sensors, and light intensity sensors to collect environmental parameters; use accelerometers and strain gauges to collect vibration and strain data. The collection frequency is adaptively adjusted according to the system state, and is 5 minutes / time under normal conditions.

[0168] Step S2: Preprocess the displacement data, environmental parameters, and structural stress data, perform anomaly detection, and evaluate the health status. The preprocessing includes filtering, noise reduction, and normalization; the anomaly detection uses an adaptive threshold algorithm and the IsolationForest method; the health status evaluation is based on a multi-dimensional health index weight matrix and a service life decay factor to generate a node health score. The health status calculation formula is H = (∑(W i ·S i ))·(1 - αD), where W i is the weight of each dimension's health index, S i is the score of each dimension, D is the service life decay factor, and α is an adjustable parameter. This formula clearly shows how to calculate the node health status by weighted summing the scores of each dimension and considering the service life decay factor.

[0169] Step S3: Predict equipment failures and generate data reconstruction strategies based on the health status evaluation results. Use a time series prediction network with a Transformer architecture to generate future health status trajectories and failure probabilities; convert the complex model into a lightweight edge model through knowledge distillation technology; formulate a data reconstruction strategy according to spatial correlation and the multi-variable spatio-temporal tensor decomposition algorithm, and impose physical constraints through a Bayesian inference engine. The prediction time window is set to 72 hours to provide sufficient early warning time.

[0170] Step S4: Dynamically adjust the network topology according to the equipment failure prediction results. Discover the current network topology based on the link quality evaluation matrix and node criticality calculation; generate an optimized network topology scheme according to the node health status and link quality; achieve seamless topology switching through a two-phase commit mechanism and maintain rollback capabilities. The network topology optimization goal is to minimize energy consumption while maximizing data transmission reliability.

[0171] Step S5: Optimize the system resource allocation based on the reinforcement learning algorithm, and adjust the node task allocation, communication power control, and sampling frequency according to the network topology structure adjustment and data reconstruction strategy. Generate a resource scheduling strategy through a deep Q-learning network; adopt a hierarchical scheduling mechanism and an adaptive control loop to execute resource allocation decisions; continuously evaluate the strategy execution effect and optimize it. The resource scheduling strategy includes four aspects: sampling frequency control, computing task allocation, communication power control, and sleep scheduling.

[0172] The above steps constitute a closed-loop self-repair process. The system continuously executes these five steps to form an adaptive monitoring mechanism of perception-diagnosis-prediction-adjustment-optimization. There are close data dependencies among the steps: the multi-source data generated in step S1 is input into step S2; the health assessment results of step S2 are input into step S3; the fault prediction and data reconstruction strategies of step S3 are input into steps S4 and S5; the topology adjustment results of step S4 are input into step S5; the resource optimization results of step S5 in turn affect the data collection process of step S1, forming a complete feedback loop.

[0173] According to engineering requirements, the system can be configured in full-automatic mode or semi-automatic mode. In full-automatic mode, the system autonomously completes fault prediction, topology adjustment, and resource optimization; in semi-automatic mode, the system generates a recommended plan, which is executed after being confirmed by the administrator. This flexible configuration mode adapts to the safety levels and management requirements of different projects.

[0174] The foundation pit monitoring system of the self-repairing wireless sensor network of the present invention has been applied in a large-scale underground project. The depth of the foundation pit of this project is 36 meters, the perimeter is about 580 meters, and the surrounding area is a densely built-up area. 58 monitoring nodes are deployed in the system, covering key positions around the foundation pit. The application results are as follows:

[0175] 1. Significantly improved monitoring accuracy: The displacement monitoring accuracy of the system reaches ±0.5mm, which is 75% higher than that of the traditional vision scheme (±2mm). During a heavy rainstorm, the system successfully captured the tiny displacement change (1.2mm / 12h) at the northeast corner of the foundation pit and early warned of potential risks.

[0176] 2. Verification of self-repair ability: During the 6-month monitoring period, the system predicted and processed 13 node failures in total, and the success rate reached 92%. One typical case is that the system predicted a power failure of a certain node 3.5 hours in advance, automatically triggered network topology reconstruction, and started the data reconstruction algorithm. There was no manual intervention throughout the process, and the continuity of the monitoring data remained 100%.

[0177] 3. Optimization of energy efficiency: The average power consumption of the system is reduced by 42% compared with the traditional scheme, and the battery life of the nodes is extended from the original 45 days to 78 days. Through intelligent power control and sleep scheduling, while ensuring data quality, the system significantly reduces the maintenance frequency, and the maintenance workload is reduced by 65%.

[0178] 4. Improvement of communication reliability: The data transmission success rate of the system reaches 99.8%, which is much higher than 92% of the traditional scheme. In a situation of severe electromagnetic interference caused by adjacent construction, the system maintains communication stability through adaptive communication power control and dynamic routing adjustment.

[0179] 5. Overall benefit analysis: The system deployment cost is reduced by about 25% compared with the traditional solution, and the total cost of ownership over the entire life cycle is reduced by 35%. At the same time, the quality and reliability of the monitoring data are significantly improved, providing strong support for project safety management, avoiding a potential safety accident, and generating significant economic and social benefits.

[0180] As Figure 9 shown in the performance comparison between the system of the present invention and the traditional monitoring system during a certain external interference event of this project. It can be seen that when communication interference and node failures occur, the data quality of the traditional system drops rapidly, while the system of the present invention maintains stable monitoring performance through the self-healing mechanism.

[0181] In summary, the foundation pit monitoring system and method of the self-healing wireless sensor network provided by the present invention construct a reliable, efficient, and intelligent foundation pit monitoring solution through five innovation points: multi-source heterogeneous data fusion and preprocessing, hierarchical health assessment, predictive fault management and data reconstruction, adaptive network topology reconstruction, and intelligent resource scheduling. This solution breaks through the technical bottlenecks existing in traditional monitoring systems, such as single-point failure risk, low energy management efficiency, and insufficient communication reliability, providing new ideas for the development of foundation pit monitoring technology.

[0182] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, various changes and modifications can be made to the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A self-repairing wireless sensor network foundation pit monitoring system, characterized in that: include: Multi-source heterogeneous data acquisition module, used to collect displacement data, environmental parameters and structural stress data within the foundation pit monitoring area; A self-diagnosis processing module, which is in communication connection with the multi-source heterogeneous data acquisition module and is used for pre-processing, anomaly detection and health evaluation of the multi-source heterogeneous data; a predictive fault management module, in communication with the self-diagnosis processing module, for predicting equipment faults and generating a data reconstruction strategy based on the health assessment result; A network topology management module, which is in communication with the predictive fault management module and is used to dynamically adjust the network topology structure according to the device fault prediction result; as well as The intelligent resource scheduling module is respectively connected to the predictive fault management module and the network topology management module for optimizing system resource allocation based on a reinforcement learning algorithm, and performing node task allocation, communication power control and sampling frequency adjustment according to the network topology adjustment and the data reconstruction strategy.

2. The system according to claim 1, characterized in that The multi-source heterogeneous data acquisition module includes: A displacement sensing unit, used to collect displacement data of characteristic points of the foundation pit; Environmental parameter collection unit, used to collect temperature, humidity, air pressure and light intensity; Structural stress acquisition unit for collecting vibration and strain data; and The data preprocessing unit is used to filter, reduce noise and standardize the displacement data, the environmental parameters and the structural stress data.

3. The system according to claim 1, characterized in that The self-diagnosis processing module includes: Hardware self-test unit to monitor sensor calibration status, communication link quality, and power supply status; A health calculation unit, for generating a health score based on multi-dimensional health indicators and a service time decay factor; and Anomaly detection unit, used to detect data anomalies based on adaptive threshold algorithm and Isolation Forest method.

4. The system according to claim 3, characterized in that The health calculation unit calculates the node health by the following formula: H=(∑(W i ·S i ))·(1-αD), Among them, H is the node health, W i is the weight of each dimension of health indicators, S i is the score of each dimension, D is the service time attenuation factor, and α is an adjustable parameter with a value range of 0.05-0.

15.

5. The system according to claim 1, characterized in that The predictive fault management module includes: Fault prediction unit, which is used to generate future health trajectory and fault probability based on the time series prediction network of Transformer architecture; Knowledge distillation unit, which converts complex models trained in the cloud into lightweight models suitable for edge deployment; and Data reconstruction unit, used to reconstruct missing data based on spatial correlation modeling and multivariate spatiotemporal tensor decomposition algorithm.

6. The system according to claim 1, characterized in that The network topology management module includes: A topology sensing unit, used to discover the network topology structure based on the link quality evaluation matrix and node criticality calculation; A dynamic topology optimization unit for generating an optimized network topology solution based on node health and link quality; and The topology migration execution unit is used to achieve seamless topology switching through a two-phase commit mechanism.

7. The system according to claim 1, characterized in that The intelligent resource scheduling module includes: Resource status sensing unit, used to monitor the status of computing resources, communication resources and energy resources; A multi-objective resource optimization unit for generating resource scheduling strategies based on a deep reinforcement learning network; and A self-optimizing execution unit is used to perform resource allocation decisions through a hierarchical scheduling mechanism and an adaptive control loop.

8. The system according to claim 7, characterized in that The resource scheduling strategy includes sampling frequency control, computing task allocation, communication power control and sleep scheduling, and balances monitoring accuracy, energy efficiency and system life through a multi-objective optimization function.

9. The system according to claim 5, characterized in that The reconstructed data generated by the data reconstruction unit is subjected to displacement continuity and maximum deformation rate constraints through a Bayesian reasoning engine to ensure that the reconstructed data conforms to physical laws.

10. A method for monitoring a foundation pit using a self-repairing wireless sensor network, using the system according to any one of claims 1 to 9, characterized in that: The following steps are involved: Collect displacement data, environmental parameters and structural stress data within the foundation pit monitoring area; Preprocessing, anomaly detection and health assessment of the displacement data, the environmental parameters and the structural stress data; Predicting equipment failures and generating data reconstruction strategies based on the health assessment results; Dynamically adjust the network topology according to the equipment failure prediction result; as well as The system resource allocation is optimized based on the reinforcement learning algorithm, and node task allocation, communication power control and sampling frequency adjustment are performed according to the network topology adjustment and the data reconstruction strategy.

Images