Tower inclination collapse early warning system and method

Abstract

The invention belongs to the field of early warning and monitoring of power equipment, and particularly discloses an early warning system and method for inclination and collapse of a tower. The system comprises a sensor module, a data acquisition and preprocessing module, an edge calculation module, a data transmission and synchronization module, a central processing and intelligent analysis module, a visualization and early warning module, a power management module, a model training and optimization module and a system management and maintenance module. The system is high in data acquisition precision, strong in system robustness and high in early warning efficiency, can efficiently and accurately carry out real-time monitoring and inclination collapse early warning on the state of the tower of the transformer substation, and remarkably improves the operation safety and management efficiency of the transformer substation.

Description

Technical Field

[0001] The invention belongs to the field of early warning monitoring of electric power equipment, and in particular relates to an early warning system and method for tower tilt and collapse. Background Art

[0002] The safe and stable operation of the national power grid is an important guarantee for the national economic and social development. As an important part of the transmission line, the stability of the pole tower directly affects the safe operation of the power grid. As an important supporting structure for the transmission line, the pole tower is widely distributed in areas with complex environments, including mountainous areas, river valleys, and areas with unstable geological conditions. It often tilts or collapses due to changes in the external environment or the influence of geological disasters, thus threatening the safe operation of the transmission line. Especially in the case of unattended operation, the risk of the tilt and collapse of the pole tower is difficult to be discovered and handled in time, which may cause serious consequences such as power line interruption, damage to power equipment, and ground electric shock accidents. Therefore, there is an urgent need for a pole tower tilt and collapse warning system with real-time monitoring and early warning functions. Summary of the invention

[0003] In order to address the deficiencies in the prior art, the present invention aims to provide a tower tilt and collapse early warning system, which has high data acquisition accuracy, strong system robustness, and high early warning efficiency. It can efficiently and accurately monitor the status of substation towers in real time and provide tilt and collapse early warning, significantly improving the operating safety and management efficiency of substations.

[0004] In order to achieve the above object, the technical solution adopted by the present invention is:

[0005] A tower tilt collapse early warning system, comprising a sensor module, a data acquisition and preprocessing module, an edge computing module, a data transmission and synchronization module, a central processing and intelligent analysis module, a visualization and early warning module, a power management module, a model training and optimization module, and a system management and maintenance module;

[0006] The sensor module is used to collect physical parameters of the tower state in real time, including tilt angle, vibration amplitude and external environmental conditions, to provide a data basis for subsequent data analysis;

[0007] The data acquisition and preprocessing module is used to integrate, clean and preliminarily analyze the raw data collected by the sensor module to provide high-quality data input for subsequent data processing;

[0008] The edge computing module is used to quickly analyze the collected data and provide preliminary tower status assessment and real-time alarm capabilities;

[0009] The data transmission and synchronization module is used for data communication between the local device and the central processing system to ensure the real-time and integrity of the data;

[0010] The central processing and intelligent analysis module is used to conduct in-depth analysis of the large-scale data collected on the central server; use the long short-term memory network to analyze the time series data, predict the future tower tilt trend and assess the risk; and combine environmental data, vibration data and tilt angle changes to comprehensively assess the health status of the tower; accurately locate the fault area based on the location distribution of tilt angle and vibration anomalies;

[0011] The visualization and early warning module is used to display the real-time status and prediction results of the tower and send early warning information through various alarm methods;

[0012] The power management module is used to provide a stable power supply to ensure long-term stable operation of the system;

[0013] The model training and optimization module uses the latest data collected in real time to incrementally train the deep learning model and optimize the model performance; evaluates the prediction accuracy of the new model through cross-validation and test sets to ensure the applicability of the optimized model; and automatically adjusts the parameters of the Kalman filter and random forest model based on historical data trends;

[0014] The system management and maintenance module is used for the management and maintenance functions of the entire early warning system.

[0015] Furthermore, the sensor module includes an acceleration sensor, an angle sensor, a vibration sensor and an environmental sensor; the acceleration sensor uses three-axis acceleration to measure the inclination angle of the tower and calculates the change in the tower posture in real time; the angle sensor is used to detect the rotation rate of the tower and cooperates with the acceleration sensor to provide more accurate inclination angle change data; the vibration sensor is used to monitor the vibration frequency and amplitude of the tower to determine whether the tower is in an abnormal state due to external force; the environmental sensor includes a temperature and humidity sensor and an anemometer to record the impact of external environmental changes on the tower state and assist in the analysis of the cause of the inclination;

[0016] The data acquisition and preprocessing module includes an analog-to-digital converter, a multi-channel data collector and a filtering module; the analog-to-digital converter is used to convert analog sensor signals into digital signals; the multi-channel data collector is used to synchronize the data streams of multiple sensors to ensure the consistency of acquisition timing; the filtering module is used to denoise the data to eliminate the influence of external electromagnetic interference and environmental noise.

[0017] Furthermore, the edge computing module includes an embedded microprocessor, an algorithm module and a short-term storage unit. The embedded microprocessor runs a Kalman filter algorithm to smooth the tilt angle data. The algorithm module integrates a random forest algorithm to classify the tower status into three categories: "normal", "slightly tilted" and "severely tilted". When the classification model detects a severe tilt state, it immediately triggers a local alarm signal and sends an early warning message to the central system. The short-term storage unit is used to store the most recent monitoring data to facilitate retrospective analysis of abnormal changes.

[0018] The data transmission and synchronization module includes a data segmentation and compression module and a wireless communication module; the data segmentation and compression module is used to divide the data collected by the sensor into small blocks and compress them to optimize the transmission efficiency, and use the network time protocol to synchronize the time of multiple devices to ensure the consistency and comparability of the data; the wireless communication module includes a LoRa module, a 4G / 5G module and a WiFi module, the LoRa module is used for low-power consumption and long-distance transmission; the 4G / 5G module is used for real-time transmission of high-frequency and high-capacity data; the WiFi module is used for monitoring tower groups in a local area network environment.

[0019] Furthermore, the central processing and intelligent analysis module includes a central processing unit, a graphics processing unit, a database module and a decision module; the graphics processing unit runs a long short-term memory network LSTM prediction model to perform time series analysis on the tilt trend of the tower and predict future risk status; the database module stores historical data for model optimization and trend comparison analysis; the decision module combines real-time data and prediction results to generate early warning signals and locate the fault area.

[0020] Furthermore, the visualization and early warning module includes an audible and visual alarm, a remote notification system and a visualization terminal; the audible and visual alarm emits audible and visual signals to attract the attention of the staff when the tower is tilted or vibrates abnormally; the remote notification system sends early warning information to maintenance personnel through SMS, APP push and email, including a detailed fault analysis report; the visualization terminal intuitively displays the real-time status, data trend and risk classification information of the tower on the monitoring platform, and uses a graphical interface to display the sensor data trend chart, tower status chart and the output results of the prediction model; the tower status is marked with different colors according to the risk level to facilitate quick judgment; maintenance personnel can query historical data and algorithm analysis results through the system interface to support quick decision-making;

[0021] The power management module includes a solar power supply system, a backup battery module and a power monitoring module; the solar power supply system utilizes the outdoor conditions of the tower to provide clean energy; the backup battery module provides backup power when there is insufficient light to ensure continuous operation of the system; the power monitoring module monitors the power supply status in real time to prevent system interruption due to power failure.

[0022] Furthermore, the system management and maintenance module includes a log management part, a remote update part, a health diagnosis part and a permission management part; the log management part is used to record the system operation status, data transmission status and fault occurrence time, providing a basis for problem tracking; the remote update part updates the system software and algorithm models online through the network to improve the system function; the health diagnosis part is used to perform health checks on hardware devices and software modules, and promptly discover and repair potential problems; the permission management part is used to perform hierarchical management of user access rights to ensure data security and the standardization of system operations.

[0023] Correspondingly, the present invention also proposes a tower tilt collapse early warning method based on the tower tilt collapse early warning system, the method comprising:

[0024] S1. Data collection: Use the sensor module to monitor the tower status in real time, including tilt angle, vibration intensity and environmental parameters. The data is transmitted to the edge computing module through a multi-channel interface.

[0025] S2. Data preprocessing: De-noise, calibrate and format the collected raw data to ensure data quality and consistency; and use the Kalman filter algorithm to smooth the tilt angle data to reduce measurement errors and noise effects;

[0026] S3. Data analysis: Use the edge computing module to run the random forest classification model to quickly classify the tower status into "normal", "slightly tilted" and "severely tilted". When the status is classified as "severely tilted", a local alarm is triggered and the data is uploaded to the central system.

[0027] S4. Data transmission: The sensor data and preliminary analysis results are transmitted to the central server through the wireless communication module to ensure integrity verification and time synchronization during data transmission;

[0028] S5. In-depth data analysis: The central server uses the long short-term memory network (LSTM) prediction model to analyze the time series data of the tower to predict the tilt trend and future risks. It also conducts a multi-dimensional assessment based on environmental parameters and historical data to generate a tower health status report.

[0029] S6. Risk assessment and fault location: Combined with the in-depth analysis results, the specific fault location of the tower is determined through the fault location algorithm, and the status is displayed according to the assessed risk level: normal, warning or dangerous;

[0030] S7. Warning and notification: The system triggers the warning mechanism according to the risk level, including:

[0031] Normal status: data recording, no notification required;

[0032] Warning status: Send yellow warning information to maintenance personnel and recommend inspection;

[0033] Dangerous state: Send red emergency warning information, notify relevant personnel through SMS and APP push, and trigger sound and light alarm equipment;

[0034] S8. Visual display: The tower status is displayed in real time on the monitoring platform, including the tilt angle trend graph, vibration intensity change graph and risk assessment results, and the predicted tower tilt trend and specific fault location are displayed to facilitate quick decision-making;

[0035] S9. Model optimization: Based on the latest data collected in real time, incremental training is performed on the LSTM prediction model and the random forest classification model to improve prediction accuracy; Kalman filter parameters and other algorithm configurations are automatically adjusted to adapt to environmental changes;

[0036] S10. System management and maintenance: The system automatically monitors the operating status of hardware and the health of software modules, and promptly alerts when abnormalities are found; maintenance personnel perform software updates, log viewing and data management through remote operations to ensure that the system continues to operate efficiently.

[0037] Among them, when step S2 cleans and fuses the data, the Kalman filter algorithm is used to comprehensively organize the data of multiple sensors, and the noise and abnormal values ​​are filtered to improve the smoothness of the data; specifically, it includes:

[0038] Set the system state to x k , represents the actual tilt angle of the tower at time k; the sensor measurement value z k For state x k The observed value of; then the state transition model is expressed as:

[0039] x k =Fx k-1 +ω k (1);

[0040] Where F is the state transfer matrix, which represents the change of the system over time. It is set to 1, assuming that the angle changes smoothly; x k-1 represents the actual tilt angle of the tower at time k-1; ωk is process noise, which follows a Gaussian distribution Represents the process noise of the system in each time step, that is, the dynamic changes of the system are not completely predictable, the change of the error follows a Gaussian distribution with a mean of 0 and a covariance matrix of Q;

[0041] The observation model is expressed as:

[0042] z k =Hx k +ν k (2);

[0043] In the formula, z k is the actual measurement value at the current moment; H is the observation matrix, which represents the relationship between the sensor measurement value and the true state; ν k is the measurement noise, which follows a Gaussian distribution represents a Gaussian distribution with a mean of zero and a covariance matrix of R;

[0044] The state prediction and covariance prediction formulas of the Kalman filter algorithm in the prediction stage are expressed as:

[0045]

[0046] In the formula, Represents the predicted state at the current moment, which is calculated based on the estimated value at the previous moment; Indicates the estimated state at the previous moment; is the covariance matrix predicted at the current moment; P k-1 is the estimation error covariance matrix of the previous moment, which represents the uncertainty of the system state estimation; F ┬ represents the transpose of the state transfer matrix F;

[0047] The Kalman gain, state update and covariance update formulas in the update phase are expressed as:

[0048]

[0049] In the formula, K k represents the Kalman gain; H ┬ is the transpose of the measurement matrix H; is the updated state at the current moment; is the predicted state obtained in the prediction phase; is the measured value prediction calculated based on the predicted state; P k is the updated covariance matrix at the current moment; I is the identity matrix;

[0050] Set the acceleration sensor's measurement value to z acc , the noise covariance is R acc ; The angle sensor's measurement value is zangle , the noise covariance is R angle ; Through weighted average fusion, the comprehensive tilt angle estimation value is obtained:

[0051]

[0052] The Kalman filter algorithm effectively filters out high-frequency noise through model prediction and measurement update; for detected abnormal values, it uses the measurement residual Statistical characteristics of , set the threshold:

[0053]

[0054] In the formula, α is a constant used to adjust the size of the tolerated residual. This constant is adjusted according to the noise level in the application scenario. After the outliers are removed, the predicted value is used instead. Make a substitution;

[0055] Formula (9) This condition is used to determine the measurement residual |e k | is greater than a certain threshold, the threshold is determined by α and covariance P k - Determined by both; if the condition is met: it means that the current measurement residual is large, there may be outliers or measurement errors, then the measurement value is considered abnormal and ignored or further processed; if the condition is not met, it means that the current measurement residual is within a reasonable range, and the filter continues to update.

[0056] The step S5 of detecting the inclination of the tower specifically includes:

[0057] Set the safety thresholds of the tower inclination angle and acceleration. When the monitoring value exceeds the threshold, trigger an alarm. Use the random forest algorithm to classify according to historical data to determine whether the tower is in a dangerous state. The specific steps are as follows:

[0058] 1) The original training set is N, and the bootstrap sampling method is used to randomly select m new bootstrap sample sets with replacement, and then construct m classification trees;

[0059] 2) If there are D features in the feature space, then in each round of decision tree generation, d features are randomly selected from the D features to form a new feature set. By using the new feature set to generate a decision tree, a total of m independent decision trees are generated in m rounds;

[0060] 3) The generated multiple trees are combined into a random forest. The importance of several independent decision trees is equal, and there is no need to consider their weights.

[0061] When analyzing the collapse of a tower, the long short-term memory network (LSTM) prediction model is used to capture the long-term dependencies in the data and perform the tower tilt data prediction task, including:

[0062] The forget gate is used to control which information needs to be discarded from the cell state. The specific formula is as follows:

[0063] f t =σ(W f ·[h t-1 ,x t ]+b f ) (10);

[0064] In the formula, f t is the output of the forget gate, W f and b f are weights and biases, σ is the activation function sigmoid, h t-1 is the hidden state of the previous time step, x t is the input of the current time step;

[0065] The input gate is used to determine which information needs to be updated to the cell state. The specific formula is as follows:

[0066] i t =σ(W i ·[h t-1 ,x t ]+b i ) (11);

[0067]

[0068] In the formula, i t is the output of the input gate, indicating the degree of selection of the input information at the current moment. The value is between 0 and 1. The closer it is to 1, the more the input information at the current moment is updated to the unit state. The closer it is to 0, the less influence the input information has on the current state. i is the weight matrix of the input gate, which is used to transform the previous hidden state h t-1 and the current input x t Mapped to the space of the input gate; [h t-1 ,x t ] means to change the hidden state h of the previous moment t-1 and the current input x t concatenate into a vector as input to the input gate; b i is the bias term of the input gate; is the candidate unit state, which represents the potential update value of the current time step; tanh is the hyperbolic tangent activation function, whose output range is [-1,1], which is used to generate nonlinear transformations of candidate unit states; W Cis the weight matrix of the candidate unit state, which transforms the hidden state h at the previous moment t-1 and the current input x t Mapped into the space of candidate unit states; b C is the bias term of the candidate unit state;

[0069] The specific formula for updating the unit status is as follows:

[0070]

[0071] In the formula, C t is the cell state at the current time step, C t-1 is the cell state at the previous time step;

[0072] The output gate is used to control the output of the current time step. The specific formula is as follows:

[0073] o t =σ(W o ·[h t-1 ,x t ]+b o ) (14);

[0074] h t =o t *tanh(C t ) (15);

[0075] In the formula, o t is the output of the output gate, controlling the hidden state h t What information needs to be output in the hidden state? Its value is also between 0 and 1. The closer it is to 1, the more information in the hidden state will be output. o is the weight matrix of the output gate, and the hidden state h of the previous moment is t-1 and the current input x t Mapped into the space of the output gate; b o is the bias term of the output gate; h t is the hidden state at the current moment, which represents the current output of the model and is usually passed to the next moment or used as the final output; tanh(C t ) means that the current cell state Ct is activated by the hyperbolic tangent function to ensure that the hidden state h t In the range [-1,1], avoid numerical overflow.

[0076] The beneficial effects of the present invention are:

[0077] The present invention uses a sensor module to collect real-time data such as the tilt angle and vibration of the tower, and transmits it efficiently through a wireless communication module, so as to realize real-time monitoring of the tower status. Combined with the Kalman filter algorithm, it is possible to optimize sensor data under noise interference, eliminate signal noise and interference, and improve data accuracy. The LSTM model is used to dynamically predict the tilt trend of the tower, which can capture the nonlinear and time-dependent characteristics in long time series and realize early warning of future risks. By combining sensor technology, wireless communication, Kalman filtering, random forest and LSTM model, real-time monitoring and accurate prediction of the risk of tower tilt collapse are realized. It has significant advantages such as high data acquisition accuracy, strong system robustness, high warning efficiency and high intelligence level. It can efficiently and accurately realize real-time monitoring of the status of substation towers and tilt collapse warning, which significantly improves the operation safety and management efficiency of substations. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] Figure 1 It is a time series diagram of Kalman filter data preprocessing of the present invention;

[0079] Figure 2 The random forest algorithm flow chart of the present invention;

[0080] Figure 3 It is a prediction diagram of the test results of the tower tilt collection data of the present invention;

[0081] Figure 4 It is a schematic diagram of the LSTM structure of the present invention;

[0082] Figure 5 It is the training and test loss curve diagram of the present invention;

[0083] Figure 6 This is a graph showing the prediction results of the LSTM of the present invention for the inclination angle of the tower;

[0084] Figure 7 The present invention is a flow chart of the tower tilt collapse early warning method. DETAILED DESCRIPTION

[0085] The principles and features of the present invention are described below in conjunction with the accompanying drawings. The examples given are only used to explain the present invention and are not intended to limit the scope of use of the present invention.

[0086] The present invention proposes a tower tilt and collapse early warning system, which includes a sensor module, a data acquisition and preprocessing module, an edge computing module, a data transmission and synchronization module, a central processing and intelligent analysis module, a visualization and early warning module, a power management module, a model training and optimization module, and a system management and maintenance module.

[0087] Among them, the sensor module is used to collect the physical parameters of the tower state in real time, including the tilt angle, vibration amplitude and external environmental conditions, providing a data basis for subsequent data analysis. The sensor module includes an accelerometer, an angle sensor, a vibration sensor and an environmental sensor. The accelerometer uses three-axis acceleration to measure the tilt angle of the tower and calculate the tower posture change in real time. The angle sensor is used to detect the rotation rate of the tower and cooperates with the accelerometer to provide more accurate tilt angle change data; the vibration sensor is used to monitor the vibration frequency and amplitude of the tower to determine whether the tower is in an abnormal state due to external forces; the environmental sensor includes a temperature and humidity sensor and an anemometer, which records the impact of external environmental changes on the tower state and assists in the analysis of the cause of the tilt.

[0088] Specifically, in order to accurately detect the tilt change of the tower, the acceleration sensor uses a high-precision ADXL355 three-axis accelerometer, and the technical parameters are shown in Table 1.

[0089] Table 1 ADXL355 three-axis accelerometer technical parameters

[0090] parameter Range Value Operating voltage 2.0V to 3.6V Acceleration measurement range ±2g, ±4g, ±8g Resolution 20 people Output Data Rate (ODR) 4Hz to 4kHz Noise Density 2μg / √Hz Operating temperature range -40°C to +125°C Power consumption 200μA (normal mode) interface I2C, SPI Encapsulation LGA-14(3mm×3mm×0.8mm)

[0091] The accelerometer can accurately sense the tilt and displacement trends of the tower in the three directions of X, Y, and Z through highly sensitive three-axis acceleration measurement. When the tower is tilted due to external force, uneven foundation settlement, or environmental stress, the ADXL355 can quickly capture tiny acceleration changes and accurately calculate the tilt angle. With its low noise and high precision characteristics, the device can avoid the influence of external environmental interference and ensure the reliability and accuracy of the data. At the same time, the wide dynamic range and low power consumption design of the ADXL355 make it suitable for complex field working environments. It can not only provide tilt warnings, but also identify potential collapse hazards in combination with vibration monitoring. Combined with data processing algorithms, the accelerometer provides comprehensive support for early warning of tower tilt and collapse, laying a solid foundation for the safe operation of power grid facilities.

[0092] The angle sensor can accurately predict the time when the tower will tilt and collapse by monitoring the angle change of the tower in real time. The selected SCA100T digital inclinometer adopts a capacitive sensor based on MEMS technology, which has the characteristics of high precision, high stability and wide measurement range. Combined with low-noise electronic circuits and digital output, it can provide high-resolution angle measurement data.

[0093] Table 2 SCA100T digital inclinometer technical parameters

[0094] parameter Range Value Measuring range ±30° or ±90° Accuracy ±0.05° Resolution 0.001° Operating temperature range -40°C to +125°C Supply voltage 3.3V to 5.0V Communication interface SPI / CAN Impact resistance 20000g Package size 7.6mm×8.5mm×3.3mm

[0095] The angle sensor can timely capture the tilt trend of the subtle tower through high-precision data collection. When the tower tilts abnormally due to external force or foundation sinking, the angle sensor can accurately detect its angle deviation and analyze the data through the built-in algorithm to determine whether the warning threshold is reached. Once it is detected that the tilt exceeds the safe range, the system will trigger the alarm module and issue an alarm to the maintenance personnel in time to ensure a quick response. In addition, the high resolution and temperature compensation function of the angle sensor enable it to provide stable and reliable data in complex climatic environments, providing a solid guarantee for early warning and risk prevention and control of tower tilt and collapse, effectively reducing the accident rate and improving the safety and stability of power grid operation.

[0096] The data acquisition and preprocessing module is used to integrate, clean and preliminarily analyze the raw data collected by the sensor module, providing high-quality data input for subsequent data processing. Sensor data is usually collected at different frequencies, and the data of each sensor needs to be synchronized in time during data integration. Data preprocessing is a key step to ensure data quality and improve analysis accuracy. The Kalman filter algorithm is used for noise removal.

[0097] The data acquisition and preprocessing module includes an analog-to-digital converter, a multi-channel data collector, and a filtering module. The analog-to-digital converter (ADC) is used to convert analog sensor signals into digital signals. The multi-channel data collector is used to synchronize the data streams of multiple sensors to ensure the consistency of the acquisition timing. The filtering module is used to denoise the data to eliminate the influence of external electromagnetic interference and environmental noise.

[0098] After preprocessing, the data is stored and transmitted in a combination of local storage and cloud storage. Local storage is achieved by setting up local storage devices (such as SD cards, hard disks) at the substation site to store short-term data and real-time data. Cloud storage / remote servers upload data to remote servers or cloud platforms through wireless networks (such as 4G / 5G, Wi-Fi, LoRa, etc.) for long-term storage and analysis.

[0099] In order to further analyze the data through algorithms to achieve the evaluation and early warning of the tower stability, the tower tilt and vibration are analyzed by combining the sensor data through data fusion methods to determine whether the tower is in an abnormal state. The specific steps include:

[0100] (1) Status assessment: By comparing the real-time data with the preset threshold, it is determined whether the tower is tilted or subjected to abnormal stress.

[0101] (2) Trend analysis: Analyze historical data, predict possible risk trends of towers, and issue early warning signals.

[0102] (3) Intelligent alarm: Combined with preset alarm rules (such as the tilt angle exceeds a certain threshold, the vibration amplitude is too large, etc.), an alarm is issued to prompt relevant personnel to take measures.

[0103] The edge computing module is used to quickly analyze the collected data and provide preliminary tower status assessment and real-time alarm capabilities. The edge computing module includes an embedded microprocessor, an algorithm module, and a short-term storage unit. The embedded microprocessor runs the Kalman filter algorithm to smooth the tilt angle data; the algorithm module integrates the random forest algorithm to classify the tower status into three categories: "normal", "slightly tilted", and "severely tilted". When the classification model detects a severe tilt state, it immediately triggers a local alarm signal and sends an early warning message to the central system; the short-term storage unit is used to store the most recent monitoring data to facilitate retrospective analysis of abnormal changes.

[0104] The data transmission and synchronization module is used for data communication between local devices and the central processing system to ensure the real-time and integrity of the data. The data transmission and synchronization module includes a data slicing and compression module and a wireless communication module; the data slicing and compression module is used to divide the data collected by the sensor into small blocks and compress them to optimize the transmission efficiency, and use the network time protocol (NTP) to synchronize the time of multiple devices to ensure the consistency and comparability of the data. The wireless communication module includes a LoRa module, a 4G / 5G module and a WiFi module. The LoRa module is used for low power consumption and long distance situations, and is suitable for tower monitoring in remote areas. The 4G / 5G module is used for real-time transmission of high-frequency and high-capacity data to ensure that the monitoring platform receives information in a timely manner. The WiFi module is suitable for tower group monitoring in a local area network environment.

[0105] The central processing and intelligent analysis module is used to conduct in-depth analysis of the large-scale data collected on the central server; use the long short-term memory network to analyze the time series data, predict the future tower tilt trend and assess the risk; and combine environmental data, vibration data and tilt angle changes to comprehensively assess the health status of the tower; accurately locate the fault area based on the location distribution of tilt angle and vibration anomalies.

[0106] The central processing and intelligent analysis module includes a central processing unit, a graphics processing unit, a database module, and a decision module. The central processing unit (CPU) performs large-scale data processing, intelligent algorithm analysis, and global risk assessment. The graphics processing unit (GPU) runs the long short-term memory network (LSTM) prediction model to perform time series analysis on the tilt trend of the tower and predict future risk status. The database module stores historical data for model optimization and trend comparison analysis. The decision module combines real-time data and prediction results to generate warning signals and locate fault areas.

[0107] The central processing unit (CPU) receives data from various sensors, executes corresponding algorithms, processes and fuses data, and then determines the stability of the tower and triggers an early warning. After receiving the sensor data, the CPU analyzes data such as the tilt angle, tower vibration, and environmental factors to more comprehensively analyze the health of the tower. The CPU compares the real-time collected data with the set safety threshold to determine whether the tower is in an abnormal state. If a parameter (such as the tilt angle, vibration amplitude, etc.) exceeds the threshold, the system will consider the tower to be in a dangerous state. The CPU can also analyze historical data to identify the changing trend of the tower's stability. Through the time series analysis of the data, it is predicted whether the tower may tilt or collapse in the next few hours or days, and early warning measures can be taken in advance. The random forest algorithm and LSTM time series prediction model are used to predict the tilt angle and collapse trend of the tower respectively to determine whether the tower is in a dangerous state. The visualization and early warning module is used to display the real-time status and prediction results of the tower, and send early warning information through various alarm methods.

[0108] The visualization and early warning module includes an audible and visual alarm, a remote notification system, and a visualization terminal. When the tower is tilted or vibrates abnormally, the audible and visual alarm emits an audible and visual signal to attract the attention of the staff. The remote notification system sends early warning information to maintenance personnel via SMS, APP push, and email, including a detailed fault analysis report. The visualization terminal intuitively displays the real-time status, data trends, and risk classification information of the tower on the monitoring platform, and uses a graphical interface to display sensor data trend charts, tower status charts, and output results of the prediction model. The tower status is marked with different colors according to the risk level (normal, warning, dangerous) for quick judgment. Maintenance personnel can query historical data and algorithm analysis results through the system interface to support quick decision-making.

[0109] The power management module is used to provide a stable power supply to ensure long-term stable operation of the system. The power management module includes a solar power supply system, a backup battery module and a power monitoring module; the solar power supply system uses the outdoor conditions of the tower to provide clean energy; the backup battery module provides backup power when there is insufficient light to ensure continuous operation of the system; the power monitoring module monitors the power supply status in real time to prevent system interruption due to power failure.

[0110] The main power supply of the present invention is supplied by solar panels, and the backup power supply is supplied by lead-acid batteries. By combining the solar panels with batteries, a supplementary power supply of renewable energy can be provided to ensure that the system can operate even without an external power supply. An intelligent charging management system is used to adjust the charging current and voltage of the solar panel according to the battery power, temperature and other conditions to avoid overcharging or over-discharging and extend the service life of the battery. In addition, it also has a battery power monitoring function to monitor the battery power status in real time. When the backup power supply is too low, a warning is issued in advance to remind maintenance personnel to replace it. The power module is equipped with a status indicator, an LCD display or other methods to display the working status of the power supply in real time, such as battery power, output voltage, current, etc., to facilitate maintenance and inspection by staff.

[0111] The model training and optimization module uses the latest data collected in real time to incrementally train the deep learning model and optimize the model performance; evaluates the prediction accuracy of the new model through cross-validation and test sets to ensure the applicability of the optimized model; and automatically adjusts the parameters of the Kalman filter and random forest model based on historical data trends. The early warning mechanism is set based on a multi-level alarm strategy, which is divided into first-level warning, second-level warning and third-level warning, which are defined as follows:

[0112] Level 1 warning: The tilt angle exceeds the set threshold but does not reach the dangerous value, prompting maintenance.

[0113] Level 2 warning: The tilt angle or acceleration is close to dangerous values, and emergency inspection is recommended.

[0114] Level 3 warning: The risk of tilting or collapse is extremely high. It is recommended to stop the relevant power transmission lines immediately.

[0115] Through simulation experiments and actual data collection, we built a sample data set, trained the algorithm model using the PyTorch deep learning framework, and improved the model performance through hyperparameter tuning and cross-validation. We verified the reliability and accuracy of the device through simulation tests and actual field tests, and optimized the hardware design and algorithm parameters.

[0116] The system management and maintenance module is used for the management and maintenance functions of the entire early warning system.

[0117] The system management and maintenance module includes log management, remote update, health diagnosis and permission management. The log management part is used to record the system operation status, data transmission and fault occurrence time, providing a basis for problem tracking. The remote update part updates the system software and algorithm models online through the network to improve system functions. The health diagnosis part is used to perform health checks on hardware devices and software modules to promptly discover and fix potential problems. The permission management part is used to manage user access rights in a hierarchical manner to ensure data security and the standardization of system operations.

[0118] The sensor module of the present invention collects the inclination, vibration and environmental parameters of the tower in real time, and the data is preliminarily digitized and denoised through the data acquisition module. The edge computing module quickly analyzes the data, detects obvious anomalies and triggers local alarms in real time. The data is transmitted to the central processing unit through the wireless communication module. The central unit combines the Kalman filter algorithm and the LSTM prediction model to analyze the data trend and predict the future risk status. When a potential risk is detected, the early warning and display module transmits the alarm information to relevant personnel in a variety of ways, and provides detailed visual data and analysis results through the monitoring platform. The power management module ensures stable power supply for the system and supports real-time monitoring around the clock. The architecture that combines edge computing with cloud-based intelligent algorithms not only improves real-time performance but also reduces the communication burden. A variety of sensors and Kalman filter algorithms are integrated to ensure the accuracy and robustness of the data. The LSTM model predicts the inclination trend of the tower, realizes early warning and positioning of faults, and reduces safety hazards.

[0119] Correspondingly, based on the above-mentioned tower tilting and collapse warning system, the present invention also proposes a tower tilting and collapse warning method, which specifically includes:

[0120] S1. Data acquisition: The sensor module is used to monitor the tower status in real time, including the tilt angle, vibration intensity and environmental parameters. The data is transmitted to the edge computing module through a multi-channel interface.

[0121] S2. Data preprocessing: De-noise, calibrate and format the collected raw data to ensure data quality and consistency; and use the Kalman filter algorithm to smooth the tilt angle data to reduce measurement errors and noise effects.

[0122] When cleaning and fusing data, the Kalman filter algorithm is used to comprehensively organize the data from multiple sensors and filter out noise and outliers to improve the smoothness of the data. Specifically, it includes:

[0123] Set the system state to x k , represents the actual tilt angle of the tower at time k; the sensor measurement value z k For state x kThe observed value of; then the state transition model is expressed as:

[0124] x k =Fx k-1 +ω k (1);

[0125] Where F is the state transfer matrix, which represents the change of the system over time. It is set to 1, assuming that the angle changes smoothly; x k-1 represents the actual tilt angle of the tower at time k-1; ω k is process noise, which follows a Gaussian distribution It represents the process noise of the system in each time step, that is, the dynamic change of the system is not completely predictable, and the change of the error follows a Gaussian distribution with a mean of 0 and a covariance matrix of Q.

[0126] The mean is 0: It means that the change in the system state is a random disturbance with zero mean, that is, the expected value of the noise is 0.

[0127] Covariance matrix Q: represents the relationship between the size of noise and different dimensions. In Kalman filtering, Q is used to describe the volatility of process noise (uncertainty of system dynamics). If Q is large, it means that the uncertainty of system dynamics is high; if Q is small, it means that the system dynamics is relatively certain and the noise is small.

[0128] The observation model is expressed as:

[0129] z k =Hx k +ν k (2);

[0130] In the formula, z k is the actual measurement value at the current moment; H is the observation matrix, which represents the relationship between the sensor measurement value and the true state; ν k is the measurement noise, which follows a Gaussian distribution

[0131]

[0132] represents a Gaussian distribution with mean zero and covariance matrix R.

[0133] The mean is 0: The expected value of noise is 0, which means that the average value of the measured noise has no bias towards the actual measured value, and the noise is not systematically higher or lower. In simple terms, the "average" of the measurement error is zero, and there is no bias.

[0134] Covariance matrix R: describes the distribution range of noise and the correlation between different measurement dimensions. It indicates the correlation between different measurement dimensions (measurements of multiple sensors). The diagonal elements of the matrix represent the noise variance in each dimension, while the non-diagonal elements represent the covariance between different dimensions.

[0135] The state prediction and covariance prediction formulas of the Kalman filter algorithm in the prediction stage are expressed as:

[0136]

[0137] In the formula, represents the predicted state (estimated value) at the current moment, which is calculated based on the estimated value at the previous moment; F is the state transition matrix, which describes the state change of the system from time t-1 to time t; Indicates the estimated state at the previous moment; is the covariance matrix predicted at the current moment; P k-1 is the estimation error covariance matrix of the previous moment, which represents the uncertainty of the system state estimation; F ┬ Represents the transpose of the state transition matrix (StateTransition Matrix) F;

[0138] The Kalman gain, state update and covariance update formulas in the update phase are expressed as:

[0139]

[0140] In the formula, K k represents the Kalman gain; H ┬ is the transpose of the measurement matrix H; is the updated state at the current moment; is the predicted state obtained in the prediction phase; is the measurement value prediction calculated based on the predicted state, where H is the observation matrix, representing the mapping from state space to measurement space; P k is the updated covariance matrix at the current moment; I is the identity matrix.

[0141] Set the acceleration sensor's measurement value to z acc , the noise covariance is R acc ; The angle sensor's measurement value is z angle , the noise covariance is R angle ; Through weighted average fusion, the comprehensive tilt angle estimation value is obtained

[0142]

[0143] The Kalman filter algorithm effectively filters out high-frequency noise through model prediction and measurement update; for detected abnormal values, it uses the measurement residual The statistical properties of z k is the actual measurement value (obtained by the sensor) at the kth moment. Set the threshold:

[0144]

[0145] In the formula, α is a constant used to adjust the size of the tolerated residual. This constant is adjusted according to the noise level in the application scenario. After the outliers are removed, the predicted value is used instead. Make a substitution; is the predicted state covariance matrix at the kth moment, which indicates the uncertainty of the filter's prediction of the current state. The larger the covariance matrix, the more uncertain the predicted state is, and the greater the tolerance for the measurement residual (i.e., a larger residual is allowed). Conversely, the smaller the covariance matrix, the smaller the tolerated residual, requiring more precise measurement.

[0146] Formula (9) This condition is used to determine the measurement residual |e k |Is it greater than a certain threshold, the threshold is determined by α and covariance Determined by both; if the condition is met: it means that the current measurement residual is large, there may be outliers or measurement errors, then the measurement value is considered abnormal and ignored or further processed; if the condition is not met, it means that the current measurement residual is within a reasonable range, and the filter continues to update.

[0147] like Figure 1 As shown in the figure, the Kalman filter data preprocessing time series diagram is generated by combining the collected data to show the true value, the original measurement value and the filtering / fusion result. The Kalman filter predicts and updates the tower tilt angle data through a mathematical model, which can effectively filter out noise and outliers. At the same time, multi-sensor data fusion can combine the advantages of different sensors and significantly improve the accuracy and stability of tilt angle estimation.

[0148] S3. Data analysis: Use the edge computing module to run the random forest classification model to quickly classify the tower status into "normal", "slightly tilted" and "severely tilted". When the status is classified as "severely tilted", a local alarm is triggered and the data is uploaded to the central system.

[0149] S4. Data transmission: The sensor data and preliminary analysis results are transmitted to the central server through the wireless communication module to ensure integrity verification and time synchronization during data transmission.

[0150] S5. In-depth data analysis: The central server uses the long short-term memory network (LSTM) prediction model to analyze the time series data of the tower to predict the tilt trend and future risks. It also conducts a multi-dimensional assessment based on environmental parameters and historical data to generate a tower health status report.

[0151] The tilt detection of the tower specifically includes:

[0152] Set the safety threshold of the tower inclination angle and acceleration, and trigger an alarm when the monitoring value exceeds the threshold. Use the random forest algorithm to classify according to historical data to determine whether the tower is in a dangerous state. The flowchart of the random forest algorithm is as follows: Figure 2 As shown, the collected data visualization test results are as follows Figure 3 The specific steps are as follows:

[0153] 1) The original training set is N, and the bootstrap sampling method is used to randomly select m new bootstrap sample sets with replacement, and then construct m classification trees;

[0154] 2) If there are D features in the feature space, then in each round of decision tree generation, d features are randomly selected from the D features to form a new feature set. By using the new feature set to generate a decision tree, a total of m independent decision trees are generated in m rounds;

[0155] 3) The generated multiple trees are combined into a random forest. The importance of several independent decision trees is equal, and there is no need to consider their weights.

[0156] When analyzing the collapse of a tower, the inclination change of the tower is usually affected by factors such as the external environment, structural deformation, and equipment aging. In order to warn of possible inclination or collapse, historical sensor data is analyzed to predict trends in the future. The Long Short-Term Memory (LSTM) prediction model can capture long-term dependencies in the data and complete the task of accurately predicting the inclination data of the tower. LSTM (Long Short-Term Memory) is a special recurrent neural network (RNN) designed to solve the gradient vanishing and gradient exploding problems in traditional RNNs. LSTM introduces a gating mechanism to determine the input information and forgotten information at each time step, thereby better capturing the temporal dependencies over long time spans.

[0157] The structure of the LSTM unit includes the forget gate, input gate, unit state update and output gate. They represent three different information flow control methods. The internal structure diagram is as follows Figure 4 shown.

[0158] The Forget Gate can control which information needs to be discarded from the cell state. The specific formula is as follows:

[0159] f t =σ(W f ·[h t-1 ,x t ]+b f ) (10);

[0160] In the formula, f t is the output of the forget gate, W f and b f are weights and biases, σ is the activation function sigmoid, whose output range is 0 to 1, which determines the degree of "gating" of information. The closer the sigmoid output is to 1, the more input information there is; the closer it is to 0, the less input information there is. t-1 is the hidden state of the previous time step, x t is the input for the current time step.

[0161] The input gate is used to determine which information needs to be updated to the cell state. The specific formula is as follows:

[0162] i t =σ(W i ·[h t-1 ,x t ]+b i ) (11);

[0163]

[0164] In the formula, i t is the output of the input gate, indicating the degree of selection of the input information at the current moment. The value is between 0 and 1. The closer it is to 1, the more the input information at the current moment is updated to the unit state. The closer it is to 0, the less influence the input information has on the current state. i is the weight matrix of the input gate, which is used to transform the previous hidden state h t-1 and the current input x t Mapped to the space of the input gate; [h t-1 ,x t ] means to change the hidden state h of the previous moment t-1 and the current input x t concatenate into a vector as input to the input gate; b i is the bias term of the input gate; is the candidate unit state, which represents the potential update value of the current time step; tanh is the hyperbolic tangent activation function, whose output range is [-1,1], which is used to generate nonlinear transformations of candidate unit states; W Cis the weight matrix of the candidate unit state, which transforms the hidden state h at the previous moment t-1 and the current input x t Mapped into the space of candidate unit states; b C is the bias term of the candidate unit state.

[0165] The specific formula for updating the unit status is as follows:

[0166]

[0167] In the formula, C t is the cell state at the current time step, C t-1 is the cell state at the previous time step.

[0168] The output gate is used to control the output of the current time step. The specific formula is as follows:

[0169] o t =σ(W o ·[h t-1 ,x t ]+b o ) (14);

[0170] h t =o t *tanh(C t ) (15);

[0171] In the formula, o t is the output of the output gate, controlling the hidden state h t What information needs to be output in the hidden state? Its value is also between 0 and 1. The closer it is to 1, the more information in the hidden state will be output. o is the weight matrix of the output gate, and the hidden state h of the previous moment is t-1 and the current input x t Mapped into the space of the output gate; b o is the bias term of the output gate; h t is the hidden state at the current moment, which represents the current output of the model and is usually passed to the next moment or used as the final output; tanh(C t ) means that the current cell state Ct is activated by the hyperbolic tangent function to ensure that the hidden state h t In the range [-1,1], avoid numerical overflow.

[0172] Through the above gating mechanism, the LSTM network can effectively transfer and update information at each time step, thereby capturing long-term dependencies in the time series. For the prediction of the tilt or collapse of the substation tower, LSTM can predict the future tilt trend through historical sensor data. Assuming that the tower changes from 90° (upright) to 0° (completely collapsed), the mean square error (MSE) loss function is used for training, and the Adam optimizer is used to optimize the model. Figure 5 This is the loss curve of the LSTM network during the training and testing of the collected sensor data. It can be seen that the training and testing loss curves are gradually converging. Figure 6 It shows the model's prediction of real data. The training and test prediction curves are significantly different from the real data, and the loss curve converges smoothly from high to low.

[0173] S6. Risk assessment and fault location: Combined with the in-depth analysis results, the specific fault location of the tower is determined through the fault location algorithm, and the status is displayed according to the assessed risk level: normal, warning or dangerous.

[0174] S7. Warning and notification: The system triggers the warning mechanism according to the risk level, including:

[0175] Normal status: data recording, no notification required;

[0176] Warning status: Send yellow warning information to maintenance personnel and recommend inspection;

[0177] Dangerous state: Send red emergency warning information, notify relevant personnel through SMS and APP push, and trigger sound and light alarm equipment;

[0178] S8. Visual display: The tower status is displayed in real time on the monitoring platform, including the tilt angle trend graph, vibration intensity change graph and risk assessment results, and the predicted tower tilt trend and specific fault location are displayed to facilitate quick decision-making.

[0179] S9. Model optimization: Based on the latest data collected in real time, incremental training is performed on the LSTM prediction model and the random forest classification model to improve prediction accuracy; Kalman filter parameters and other algorithm configurations are automatically adjusted to adapt to environmental changes.

[0180] S10. System management and maintenance: The system automatically monitors the operating status of hardware and the health of software modules, and promptly alerts when abnormalities are found; maintenance personnel perform software updates, log viewing and data management through remote operations to ensure that the system continues to operate efficiently.

[0181] like Figure 1-7As shown, the tower tilt and collapse warning method provided by the present invention forms a closed loop from data collection to in-depth analysis, and then to risk warning and model optimization. It can efficiently and accurately realize real-time monitoring of the substation tower status and tilt and collapse warning, which significantly improves the operation safety and management efficiency of the substation.

[0182] Obviously, the embodiments described above are only some embodiments of the present application, rather than all embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application is described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions recorded in the aforementioned specific implementation methods, or to perform equivalent replacement of some of the technical features therein. Any equivalent structure made using the contents of the specification and drawings of this application, directly or indirectly used in other related technical fields, is similarly within the scope of patent protection of this application.

Claims

1. A tower tilt collapse early warning system, characterized in that: It includes sensor module, data acquisition and preprocessing module, edge computing module, data transmission and synchronization module, central processing and intelligent analysis module, visualization and early warning module, power management module, model training and optimization module, and system management and maintenance module; The sensor module is used to collect physical parameters of the tower state in real time, including tilt angle, vibration amplitude and external environmental conditions, to provide a data basis for subsequent data analysis; The data acquisition and preprocessing module is used to integrate, clean and preliminarily analyze the raw data collected by the sensor module to provide high-quality data input for subsequent data processing; The edge computing module is used to quickly analyze the collected data and provide preliminary tower status assessment and real-time alarm capabilities; The data transmission and synchronization module is used for data communication between the local device and the central processing system to ensure the real-time and integrity of the data; The central processing and intelligent analysis module is used to conduct in-depth analysis of the large-scale data collected on the central server; use the long short-term memory network to analyze the time series data, predict the future tower tilt trend and assess the risk; and combine environmental data, vibration data and tilt angle changes to comprehensively assess the health status of the tower; accurately locate the fault area based on the location distribution of tilt angle and vibration anomalies; The visualization and early warning module is used to display the real-time status and prediction results of the tower and send early warning information through various alarm methods; The power management module is used to provide a stable power supply to ensure long-term stable operation of the system; The model training and optimization module uses the latest data collected in real time to perform incremental training on the deep learning model to optimize model performance; the prediction accuracy of the new model is evaluated through cross-validation and test sets to ensure the applicability of the optimized model; The parameters of the Kalman filter and random forest model are automatically adjusted based on historical data trends; The system management and maintenance module is used for the management and maintenance functions of the entire early warning system.

2. The tower tilt and collapse early warning system according to claim 1 is characterized in that: The sensor module includes an acceleration sensor, an angle sensor, a vibration sensor and an environmental sensor; the acceleration sensor uses three-axis acceleration to measure the inclination angle of the tower and calculates the change in the tower posture in real time; the angle sensor is used to detect the rotation rate of the tower and cooperates with the acceleration sensor to provide more accurate inclination angle change data; the vibration sensor is used to monitor the vibration frequency and amplitude of the tower to determine whether the tower is in an abnormal state due to external force; the environmental sensor includes a temperature and humidity sensor and an anemometer to record the impact of external environmental changes on the tower state and assist in the analysis of the cause of the inclination; The data acquisition and preprocessing module includes an analog-to-digital converter, a multi-channel data collector and a filtering module; the analog-to-digital converter is used to convert analog sensor signals into digital signals; the multi-channel data collector is used to synchronize the data streams of multiple sensors to ensure the consistency of acquisition timing; the filtering module is used to denoise the data to eliminate the influence of external electromagnetic interference and environmental noise.

3. The tower tilt and collapse early warning system according to claim 1 is characterized in that: The edge computing module includes an embedded microprocessor, an algorithm module and a short-term storage unit. The embedded microprocessor runs a Kalman filter algorithm to smooth the tilt angle data. The algorithm module integrates a random forest algorithm to classify the tower status into three categories: "normal", "slightly tilted" and "severely tilted". When the classification model detects a severe tilt state, it immediately triggers a local alarm signal and sends an early warning message to the central system. The short-term storage unit is used to store the most recent monitoring data to facilitate retrospective analysis of abnormal changes. The data transmission and synchronization module includes a data segmentation and compression module and a wireless communication module; the data segmentation and compression module is used to divide the data collected by the sensor into small blocks and compress them to optimize the transmission efficiency, and use the network time protocol to synchronize the time of multiple devices to ensure the consistency and comparability of the data; the wireless communication module includes a LoRa module, a 4G / 5G module and a WiFi module, the LoRa module is used for low-power consumption and long-distance transmission; the 4G / 5G module is used for real-time transmission of high-frequency and high-capacity data; the WiFi module is used for monitoring tower groups in a local area network environment.

4. The tower tilt and collapse early warning system according to claim 1 is characterized in that: The central processing and intelligent analysis module includes a central processing unit, a graphics processing unit, a database module and a decision module; the graphics processing unit runs a long short-term memory network LSTM prediction model to perform time series analysis on the tilt trend of the tower and predict future risk status; The database module stores historical data for model optimization and trend comparison analysis; the decision module combines real-time data and prediction results to generate early warning signals and locate fault areas.

5. The tower tilt and collapse early warning system according to claim 1 is characterized in that: The visualization and early warning module includes an audible and visual alarm, a remote notification system and a visualization terminal; the audible and visual alarm emits audible and visual signals to attract the attention of the staff when the tower is tilted or vibrates abnormally; the remote notification system sends early warning information to maintenance personnel through SMS, APP push and email, including a detailed fault analysis report; the visualization terminal intuitively displays the real-time status, data trend and risk classification information of the tower on the monitoring platform, and uses a graphical interface to display the sensor data trend chart, tower status chart and the output results of the prediction model; the tower status is marked with different colors according to the risk level to facilitate quick judgment; maintenance personnel can query historical data and algorithm analysis results through the system interface to support quick decision-making; The power management module includes a solar power supply system, a backup battery module and a power monitoring module; the solar power supply system utilizes the outdoor conditions of the tower to provide clean energy; the backup battery module provides backup power when there is insufficient light to ensure continuous operation of the system; the power monitoring module monitors the power supply status in real time to prevent system interruption due to power failure.

6. The tower tilt and collapse early warning system according to claim 1, characterized in that: The system management and maintenance module includes a log management part, a remote update part, a health diagnosis part and a rights management part; The log management part is used to record the system operation status, data transmission status and fault occurrence time, providing a basis for problem tracking; The remote update part updates the system software and algorithm models online through the network to improve system functions; the health diagnosis part is used to perform health checks on hardware devices and software modules to promptly discover and repair potential problems; the permission management part is used to hierarchically manage user access rights to ensure data security and the standardization of system operations.

7. A tower tilt and collapse early warning method implemented based on the tower tilt and collapse early warning system according to claim 1, characterized in that: The method is implemented based on the following steps: S1. Data collection: Use the sensor module to monitor the tower status in real time, including tilt angle, vibration intensity and environmental parameters. The data is transmitted to the edge computing module through a multi-channel interface. S2. Data preprocessing: De-noise, calibrate and format the collected raw data to ensure data quality and consistency; and use the Kalman filter algorithm to smooth the tilt angle data to reduce measurement errors and noise effects; S3. Data analysis: Use the edge computing module to run the random forest classification model to quickly classify the tower status into "normal", "slightly tilted" and "severely tilted". When the status is classified as "severely tilted", a local alarm is triggered and the data is uploaded to the central system. S4. Data transmission: The sensor data and preliminary analysis results are transmitted to the central server through the wireless communication module to ensure integrity verification and time synchronization during data transmission; S5. In-depth data analysis: The central server uses the long short-term memory network (LSTM) prediction model to analyze the time series data of the tower to predict the tilt trend and future risks. It also conducts a multi-dimensional assessment based on environmental parameters and historical data to generate a tower health status report. S6. Risk assessment and fault location: Combined with the in-depth analysis results, the specific fault location of the tower is determined through the fault location algorithm, and the status is displayed according to the assessed risk level: normal, warning or dangerous; S7. Warning and notification: The system triggers the warning mechanism according to the risk level, including: Normal status: data recording, no notification required; Warning status: Send yellow warning information to maintenance personnel and recommend inspection; Dangerous state: Send red emergency warning information, notify relevant personnel through SMS and APP push, and trigger sound and light alarm equipment; S8. Visual display: The tower status is displayed in real time on the monitoring platform, including the tilt angle trend graph, vibration intensity change graph and risk assessment results, and the predicted tower tilt trend and specific fault location are displayed to facilitate quick decision-making; S9. Model optimization: Based on the latest data collected in real time, incremental training is performed on the LSTM prediction model and the random forest classification model to improve prediction accuracy; Kalman filter parameters and other algorithm configurations are automatically adjusted to adapt to environmental changes; S10. System management and maintenance: The system automatically monitors the operating status of hardware and the health of software modules, and promptly alerts when abnormalities are found; maintenance personnel perform software updates, log viewing and data management through remote operations to ensure that the system continues to operate efficiently.

8. The tower tilt and collapse early warning method according to claim 7, characterized in that: When cleaning and fusing the data in step S2, the Kalman filter algorithm is used to comprehensively organize the data of multiple sensors, and the noise and outliers are filtered to improve the smoothness of the data; specifically, it includes: Set the system state to x k , represents the actual tilt angle of the tower at time k; the sensor measurement value z k For state x k The observed value of; then the state transition model is expressed as: x k =Fx k-1 +ω k (1); Where F is the state transfer matrix, which represents the change of the system over time. It is set to 1, assuming that the angle changes smoothly; x k-1 represents the actual tilt angle of the tower at time k-1; ω k is process noise, which follows a Gaussian distribution Represents the process noise of the system in each time step, that is, the dynamic changes of the system are not completely predictable, the change of the error follows a Gaussian distribution with a mean of 0 and a covariance matrix of Q; The observation model is expressed as: z k =Hx k +ν k (2); In the formula, z k is the actual measurement value at the current moment; H is the observation matrix, which represents the relationship between the sensor measurement value and the true state; ν k is the measurement noise, which follows a Gaussian distribution represents a Gaussian distribution with a mean of zero and a covariance matrix of R; The state prediction and covariance prediction formulas of the Kalman filter algorithm in the prediction stage are expressed as: In the formula, Represents the predicted state at the current moment, which is calculated based on the estimated value at the previous moment; Indicates the estimated state at the previous moment; is the covariance matrix predicted at the current moment; P k-1 is the estimation error covariance matrix of the previous moment, which represents the uncertainty of the system state estimation; F ┬ represents the transpose of the state transfer matrix F; The Kalman gain, state update and covariance update formulas in the update phase are expressed as: In the formula, K k represents the Kalman gain; H ┬ is the transpose of the measurement matrix H; is the updated state at the current moment; is the predicted state obtained in the prediction phase; is the measured value prediction calculated based on the predicted state; P k is the updated covariance matrix at the current moment; I is the identity matrix; Set the acceleration sensor's measurement value to z acc , the noise covariance is R acc ; The angle sensor's measurement value is z angle , the noise covariance is R angle ; Through weighted average fusion, the comprehensive tilt angle estimation value is obtained The Kalman filter algorithm effectively filters out high-frequency noise through model prediction and measurement update; for detected abnormal values, it uses the measurement residual Statistical characteristics of , set the threshold: In the formula, α is a constant used to adjust the size of the tolerated residual. This constant is adjusted according to the noise level in the application scenario. After the outliers are removed, the predicted value is used instead. Make a substitution; Formula (9) This condition is used to determine the measurement residual |e k |Is it greater than a certain threshold, the threshold is determined by α and covariance Determined by both; if the condition is met: it means that the current measurement residual is large, there may be outliers or measurement errors, then the measurement value is considered abnormal, ignored or further processed; if the condition is not met, it means that the current measurement residual is within a reasonable range, and the filter continues to update.

9. The tower tilt and collapse early warning method according to claim 7, characterized in that: The step S5 of performing inclination detection on the tower specifically includes: Set the safety thresholds of the tower inclination angle and acceleration. When the monitoring value exceeds the threshold, trigger an alarm. Use the random forest algorithm to classify according to historical data to determine whether the tower is in a dangerous state. The specific steps are as follows: 1) The original training set is N, and the bootstrap sampling method is used to randomly select m new bootstrap sample sets with replacement, and then construct m classification trees; 2) If there are D features in the feature space, then in each round of decision tree generation, d features are randomly selected from the D features to form a new feature set. By using the new feature set to generate a decision tree, a total of m independent decision trees are generated in m rounds; 3) The generated multiple trees are combined into a random forest. The importance of several independent decision trees is equal, and there is no need to consider their weights.

10. The tower tilt and collapse early warning method according to claim 7, characterized in that: When performing collapse analysis on the tower in step S5, the long short-term memory network LSTM prediction model is used to capture the long-term dependencies in the data and perform the tower tilt data prediction task, which specifically includes: The forget gate is used to control which information needs to be discarded from the cell state. The specific formula is as follows: f t =σ(W f ·[h t-1 ,x t ]+b f ) (10); In the formula, f t is the output of the forget gate, W f and b f are weights and biases, σ is the activation function sigmoid, h t-1 is the hidden state of the previous time step, x t is the input of the current time step; The input gate is used to determine which information needs to be updated to the cell state. The specific formula is as follows: i t =σ(W i ·[h t-1 ,x t ]+b i ) (11); In the formula, i t is the output of the input gate, indicating the degree of selection of the input information at the current moment. The value is between 0 and 1. The closer it is to 1, the more the input information at the current moment is updated to the unit state. The closer it is to 0, the less influence the input information has on the current state. i is the weight matrix of the input gate, which is used to transform the previous hidden state h t-1 and the current input x t Mapped to the space of the input gate; [h t-1 ,x t ] means to change the hidden state h of the previous moment t-1 and the current input x t concatenate into a vector as input to the input gate; b i is the bias term of the input gate; is the candidate unit state, indicating the potential update value of the current time step; tanh is the hyperbolic tangent activation function, whose output range is [-1,1], which is used to generate nonlinear transformations of candidate unit states; W C is the weight matrix of the candidate unit state, which transforms the hidden state h at the previous moment t-1 and the current input x t Mapped into the space of candidate unit states; b C is the bias term of the candidate unit state; The specific formula for updating the unit status is as follows: In the formula, C t is the cell state at the current time step, C t-1 is the cell state at the previous time step; The output gate is used to control the output of the current time step. The specific formula is as follows: the t =σ(W o ·[h t-1 ,x t ]+b o ) (14); h t =o t *tanh(C t ) (15); In the formula, o t is the output of the output gate, controlling the hidden state h t What information needs to be output in the hidden state? Its value is also between 0 and 1. The closer it is to 1, the more information in the hidden state will be output. o is the weight matrix of the output gate, and the hidden state h of the previous moment is t-1 and the current input x t Mapped into the space of the output gate; b o is the bias term of the output gate; h t is the hidden state at the current moment, which represents the current output of the model and is usually passed to the next moment or used as the final output; tanh(C t ) means that the current cell state Ct is activated by the hyperbolic tangent function to ensure that the hidden state h t In the range [-1,1], avoid numerical overflow.

Images