Intelligent safety early warning method for wind power hoisting operation

Abstract

The invention relates to the technical field of wind power generation, discloses an intelligent safety early warning method for wind power hoisting operation, and aims to solve the technical problems of insufficient perception and lack of data fusion and intelligent analysis decision in existing operation safety management. The method is characterized by comprising the following steps: constructing an intelligent terminal integrated with a UWB / IMU / safety belt / environment sensor; deploying a positioning base station network and a signal relay system; a multi-source data fusion intelligent analysis platform is established, and high-precision positioning, behavior recognition and deep learning risk prediction are achieved; and a closed-loop intelligent decision-making and execution system is constructed, and graded early warning, electronic fence, environment linkage and one-key help calling are realized. According to the method, a traditional experience driving mode is innovated into a data driving mode, the safety level and the operation efficiency of wind power hoisting operation are remarkably improved, and the defects of insufficient perception, information isolation and the like are overcome.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of wind power generation, in particular to an intelligent safety warning method for wind power hoisting operation. BACKGROUND

[0002] Energy shortage and environmental pollution problems are increasingly prominent, and have become a major challenge to the sustainable development of global economic society. Under this background, countries and regions around the world have paid great attention to the transformation and upgrading of energy structure, and the full use of clean energy and renewable energy is regarded as a key path to cope with the above challenges. Among them, wind resources, as a kind of inexhaustible and inexhaustible natural clean energy, its effective development and utilization has significantly promoted the vigorous development of wind power industry. The construction of wind power project has the characteristics of fast effect and short cycle, combined with the continuous maturity of the manufacturing level of related equipment, which creates a broad space for the industrialization process of wind power market.

[0003] Construction safety is of great importance in all walks of life, especially for personnel safety, which is the basis and guarantee for the healthy development of the industry. Only under the premise of fully guaranteeing the safety of operation, can various construction tasks be successfully completed, and the realization of production benefit be ensured. In the construction stage of wind power industry, due to its inherent complexity and high technical requirements, there are a large number of high-risk operation links, such as climbing operation, hoisting of large parts, etc. These operations not only put forward high requirements on the safety consciousness of construction personnel, but also have strict specifications for safety protection equipment and site management level. In order to ensure the safety of life and property of personnel in wind power industry construction operation, and further improve the intelligent level of safety management of wind power plant, the industry urgently needs to build an intelligent system that can realize the precise management and real-time warning of construction personnel. Therefore, the traditional safety management mode of wind power construction mainly relies on the patrol, visual observation, intercom communication and experience-based judgment of on-site management personnel. In order to improve the management efficiency, some wind power construction sites introduce basic intelligent auxiliary means, such as realizing remote visualization of operation area by deploying fixed video monitoring system, or using general GPS positioning technology to track the approximate position of construction personnel, combined with environmental sensors such as anemometer and thermometer to monitor the meteorological conditions. These early attempts have alleviated the pressure of pure manual management to a certain extent, providing preliminary digital information support for managers, so that some obvious violations or extreme environmental conditions can be discovered and trigger simple threshold-based alarm. Through these technologies, the safety management of wind power construction has made a solid first step towards dataization and remoteization, laying a foundation for further intelligent exploration.

[0004] In view of this, how to build a set of wind power hoisting operation of construction personnel to achieve high-precision, multi-dimensional, real-time situation awareness, and effectively integrate multi-source data to support intelligent risk assessment and decision-making, and ultimately realize the transformation from passive response to active early warning of intelligent safety management system has become the key challenge and technical problem to be solved for the current technical personnel in the field. SUMMARY

[0005] The present application aims to solve the technical contradiction of the serious lack of high-precision, multi-dimensional, real-time situation awareness in the existing wind power hoisting operation safety management mode, which leads to data fusion obstacles and lack of intelligent analysis and decision-making capabilities. In order to achieve the above invention purpose, the present application provides an intelligent safety warning method for wind power hoisting operation. This method realizes the comprehensive and accurate acquisition of individual identity, high-precision position, real-time posture, physiological state of construction personnel and micro-environment parameters of operation area by building a multi-level, multi-modal perception network, and further deeply fuses and intelligently analyzes the above multi-source heterogeneous data. Through advanced deep learning models, the potential risks are predictively evaluated, and finally a closed-loop intelligent decision and execution mechanism is formed, so as to innovate the traditional "experience-driven" safety management mode into a "data-driven" intelligent prevention mode, and significantly improve the overall safety level and operation efficiency of wind power hoisting operation.

[0006] In order to solve the above technical problems, the present application provides the following technical solutions: An intelligent safety warning method for wind power hoisting operation, comprising: S1: Real-time acquisition of multiple physical parameters and micro-environment parameters of construction personnel through at least one intelligent sensing terminal of construction personnel, wherein the intelligent sensing terminal is provided with an ultra-wideband positioning module, an inertial measurement unit, an air pressure and temperature sensor module and a safety belt wearing state sensor; S2: The ultra-wideband positioning module cooperates with the positioning base station network to assist in acquiring the high-precision position information of the construction personnel in combination with the accurate coordinates of multiple ultra-wideband positioning base stations in the positioning base station network and the preliminary positioning results of the intelligent sensing terminal; S3: The data collected and preliminarily processed by the intelligent sensing terminal, the positioning auxiliary data of the positioning base station network and the environmental monitoring data of the on-site environmental monitoring equipment are uploaded to the central server in real time through the wireless communication unit or the signal relay system; S4: In the central server, a multi-source heterogeneous data fusion and intelligent analysis platform is run to perform time stamp alignment and data cleaning preprocessing operation on the received multi-source data, and to perform high-precision personnel position and trajectory reconstruction, personnel posture and behavior recognition, micro-environment parameter and personnel activity association, and construction site space situation visualization; S5: Based on the high-precision personnel position data, behavior recognition results and environmental parameters output by S4, the current operation state and potential safety risks are real-time evaluated and predicted through the hoisting risk intelligent evaluation and prediction model in the multi-source heterogeneous data fusion and intelligent analysis platform. The hoisting risk intelligent evaluation and prediction model adopts a multi-modal fusion architecture of deep learning; S6: According to the risk evaluation results output by the hoisting risk intelligent evaluation and prediction model in S5, a hierarchical response strategy is adopted through a closed-loop intelligent decision and execution system to perform hierarchical risk warning and notification, electronic fence management, height and behavior control, environmental linkage safety response, one-key help and emergency rescue, and operation trajectory and event backtracking.

[0007] Preferably, the construction personnel intelligent perception terminal comprises: the sensor unit configured to acquire multiple physical parameters and environmental parameters of the construction personnel in real time; the data acquisition and processing unit configured to preprocess, filter, fuse and preliminarily calculate the raw data collected by the sensor unit; the human-computer interaction unit configured to provide real-time information feedback and emergency call function to the construction personnel; the wireless communication unit configured to upload the processed data to a remote server in real time and receive instructions from the server; and the power supply unit configured to provide stable power supply for each module of the intelligent perception terminal.

[0008] Preferably, the sensor unit specifically comprises: The ultra-wideband (UWB) positioning module uses a pulse signal with extremely narrow time domain width for distance measurement, the working frequency range is 3.1 GHz to 10.6 GHz, the pulse repetition frequency can reach 100 MHz, the distance measurement accuracy can reach 10 centimeter level, and it has strong anti-interference ability in complex multipath environment, to send nanosecond-level pulse signals to the surrounding environment at a preset period; The inertial measurement unit (IMU) is a nine-axis micro-electromechanical system (MEMS) sensor, which integrates a three-axis accelerometer, a three-axis gyroscope and a three-axis magnetometer. The acceleration measurement range is ±16g, the angular velocity measurement range is ±2000° / s, and the internal attitude fusion algorithm is used to real-time solve the pitch angle, roll angle and heading angle of the construction personnel. The attitude solution accuracy can reach ±1° in a dynamic environment; The air pressure and temperature sensor module is composed of a high-precision digital air pressure sensor and a thermistor. The air pressure measurement range is 300 hPa to 1100 hPa, the resolution can reach 0.01 hPa, the temperature measurement range is -40℃ to 85℃, and the accuracy is ±0.5℃; and The safety belt wearing state sensor is configured as a miniature mechanical sensor or a contact switch, is installed at a key connection point of the safety belt, has a response time less than 100 milliseconds, outputs a specific signal when the safety belt is normally locked, and outputs another signal when the safety belt is off, loose or abnormally stressed.

[0009] Preferably, the data acquisition and processing unit is composed of a low-power microcontroller (MCU) and its supporting circuit, integrates an analog-to-digital converter (ADC), a digital signal processor (DSP) and a memory, the MCU runs embedded firmware, is responsible for synchronous acquisition of sensor data, time stamp marking, denoising processing, data formatting and preliminary algorithm calculation, including initial calculation of a positioning algorithm based on a UWB signal time difference of arrival (TDOA), an IMU attitude solving algorithm and an air pressure height conversion algorithm, and has a power management function to monitor the power state of the power supply unit. The man-machine interaction unit includes a low-power OLED display screen for displaying personnel ID, current position overview, height, battery power and warning information, and a waterproof and mistaken touch-proof emergency call button, after the button is pressed, the wireless communication unit immediately sends a high-priority distress signal. The wireless communication unit adopts LoRa wireless communication technology, has a working frequency of 433 MHz or 868 MHz / 915 MHz, has the characteristics of adjustable spreading factor (SF), adjustable bandwidth (BW) and adjustable coding rate (CR), a transmission distance of several kilometers, excellent anti-interference performance and low power consumption, and supports point-to-point communication and star network networking.

[0010] Preferably, the positioning base station network is composed of a plurality of fixedly deployed ultra-wideband (UWB) positioning base stations, the UWB positioning base stations and the UWB module of the intelligent sensing terminal adopt the same chip set and communication protocol, each positioning base station includes a UWB positioning module, a microcontroller (MCU), a wireless communication unit and a power supply unit, the positioning base station receives UWB pulse signals from the intelligent sensing terminal, accurately measures the time of arrival (ToA) of the signals, sends the ToA information together with the accurate coordinates of the base station itself to the intelligent sensing terminal of the construction personnel or the signal relay system through the wireless communication unit (LoRa module) of the positioning base station. The signal relay system is composed of a plurality of signal repeaters which are data forwarding nodes in the LoRa network, and the structure of the signal repeaters includes a microcontroller (MCU), a wireless communication unit and a power supply unit, the signal repeaters are deployed at different heights or signal shielding areas inside the wind turbine tower, for receiving LoRa signals from the intelligent sensing terminal or the positioning base station, and forwarding them to the LoRa gateway or the server, the microcontroller in the signal repeater has a signal strength monitoring function and can dynamically adjust the forwarding strategy.

[0011] Preferably, the multi-source heterogeneous data fusion and intelligent analysis platform is deployed on a central server, and after performing data preprocessing, the following function modules are further executed: The high-precision personnel position and trajectory reconstruction module combines the original ranging data from the UWB positioning module, the accurate coordinates of the positioning base station and the preliminary positioning results of the intelligent sensing terminal, and the high-precision position reconstruction module adopts a TDOA algorithm, and the positioning formula is: ; Where (x,y) is the personnel coordinate, (x i ,y i ) and (x j ,y i ) are base station coordinates, and c is the speed of light; The personnel posture and behavior recognition module deeply analyzes the acceleration, angular velocity and magnetometer data from the inertial measurement unit (IMU) of the intelligent sensing terminal, identifies the real-time posture and abnormal behavior of the construction personnel through a pre-trained machine learning or deep learning model, and the personnel body inclination calculation formula is: Where ax is the left-right direction acceleration and az is the vertical direction acceleration; The micro-environment parameter and personnel activity association module associates the real-time air pressure and temperature data from the air pressure and temperature sensor module with the macro-environment data from the on-site environment monitoring equipment through a spatial matching algorithm and a time series, and accurately binds them to the real-time position of a specific construction personnel, and the height change calculation formula is: Where P1 and P2 are air pressure values at different times, and K is a conversion constant; and The construction site space situation visualization module uses digital twinning technology to present the overall space situation of the wind power hoisting operation site in a three-dimensional visual manner, accurately displays the real-time position, height, posture and motion trajectory of all construction personnel, and the real-time position, posture, load parameter, movement path and safety distance of the hoisting equipment, and labels the fixed obstacles, safety channels, dangerous areas and micro-environment parameters in the work area; The distance D between the personnel and the equipment is calculated by the following formula: wherein R is the current working radius of the equipment, L is the maximum outer dimension of the hoisted object, and S is the set minimum safety margin.

[0012] Preferably, the hoisting risk intelligent evaluation and prediction model adopts a fusion architecture of an improved ResNet50 and a Bi-directional Long Short-Term Memory (BiLSTM) network, and includes: Input data: the model receives multi-modal data input, including a high-definition video stream frame sequence of a construction site, real-time data uploaded by the intelligent sensing terminal, operating parameters of the hoisting equipment, and environmental monitoring data; Feature extraction: the model extracts visual features in the video frames through an improved convolutional neural network (CNN) branch. The CNN branch extracts low-level features of personnel, hoisted objects, and obstacles in the image through multiple 3x3 convolution layers, and uses a spatial attention mechanism to strengthen the feature weights of high-risk areas to generate a distance feature matrix of 128x128 dimensions. At the same time, a BiLSTM branch processes time series data. The BiLSTM branch adopts a double-layer structure, each layer containing 128 neurons and embedding a gated recurrent unit (GRU), and finally outputs a dynamic feature vector of 64 dimensions; The attention weight calculation formula is: wherein F is a feature map, and W and b are learnable parameters; Data fusion and risk prediction: the distance feature matrix extracted by the CNN branch and the dynamic feature vector output by the BiLSTM branch are spliced and fused through a fully connected layer. The fused feature vector is input into a Softmax activation function layer, and the probability distribution of the current working state being in the “safe”, “warning”, or “dangerous” three safety level labels is output; The risk level probability calculation formula is: wherein X is the fused feature vector; Before being put into use, the model undergoes initial training on a large-scale multi-scenario and multi-modal data set. The training process uses a Focal Loss function to balance the sample class distribution and iteratively optimizes parameters through an Adam optimizer. The model training uses a Focal Loss function to balance the sample class distribution. The loss function calculation formula is: wherein Pt is the probability that the model predicts the true class, and γ is the focus parameter. During the system running, the platform periodically collects new pre-warning events and dangerous event data to form an incremental data set, and performs incremental training on the model. During the incremental training, the first 80% of the parameters of the CNN and BiLSTM are frozen, and only the fully connected layer and attention weight are fine-tuned; for the offshore wind power scene, the model further introduces a Transformer module, which captures the complex correlation between long-time wind speed and swinging objects through an eight-head self-attention mechanism, and adds tidal height as a feature dimension. The optimization effect of the model is evaluated by a precision rate of no less than 95% and a recall rate of no less than 98%.

[0013] The safety state score S of the hoisting risk intelligent assessment and prediction model is calculated by the following formula: S = 0.4P + 0.4B + 0.2E; wherein P is the position safety value, ranging from 0 to 1, determined by the distance between the personnel and the nearest electronic fence danger zone; B is the behavior safety value, ranging from 0 to 1, determined by the position of the safety belt and the stability of the posture; and E is the environmental safety value, ranging from 0 to 1, determined by the environmental wind speed.

[0014] Preferably, the hierarchical risk warning and notification in the closed-loop intelligent decision and execution system comprises: Level I warning: when the system determines that there is a potential risk, such as the horizontal distance between personnel and equipment is less than 3 meters but greater than 1.5 meters, or the vertical clearance is less than 2 meters but the personnel are not in the moving path of the equipment, the platform pops up a risk prompt window through the remote monitoring interface, and notifies the relevant management personnel through SMS or APP, while the construction personnel intelligent perception terminal issues a slight vibration prompt or a low-frequency voice alarm; Level II warning: when the risk level is further increased, such as personnel invading the range of 5 meters below the hoisting object, or the real-time wind speed exceeding 15 m / s for 30 seconds, the platform immediately triggers an audible and visual alarm, and issues a warning message to all personnel on site through a voice broadcast system, while the construction personnel intelligent perception terminal issues a moderate vibration and plays a pre-recorded voice, and the platform automatically sends evacuation instructions to the foreman intercom and sends speed reduction or pause operation instructions to the hoisting equipment operator; Level III: When the system identifies a serious dangerous situation, such as the horizontal distance between personnel and equipment being less than 1.5 meters, or multiple people being in a high-altitude violation state at the same time, or the wind speed continuously exceeding 18 m / s, the platform immediately triggers the highest level of emergency sound and light alarm, issues mandatory evacuation instructions through the automatic voice system, sends emergency braking instructions to the hoisting equipment control system to make the hoisting equipment stop running immediately, the construction personnel intelligent sensing terminal issues a strong vibration and plays an emergency voice, and starts the strobe red light outside the tower drum, the platform locks the elevator, sends a list of responsible persons (including real-time positioning link) to the emergency management department through SMS, and starts the emergency broadcast system to play the preset emergency plan, Preferably, the closed-loop intelligent decision-making and execution system further comprises: Electronic fence management: The platform pre-defines a "red-yellow-green" three-level electronic fence area in three-dimensional space, the red area is a high-risk area, the yellow area is an operation edge buffer area or a potential danger area, and the green area is a safe passage or rest area. When the real-time position of the construction personnel intrudes into the red area or crosses the fence boundary without authorization, the system immediately triggers the corresponding early warning and response measures according to the preset risk level; Height and behavior control: When the real-time altitude of the construction personnel exceeds the safe operation platform range, or when the safety belt wearing state sensor detects that the safety belt hook is not locked, the load bearing point displacement exceeds 0.5 meters, etc. The system automatically triggers a warning and pushes it to the safety management personnel, and at the same time, the personnel are prompted by voice through the intelligent sensing terminal; Environmental linkage safety response: When the real-time wind speed detected by the on-site environmental monitoring equipment exceeds the preset safety threshold, or the lightning warning sensor detects that lightning is approaching, the platform automatically issues an "stop hoisting operation" instruction to the site, and confirms whether all construction personnel have evacuated to the safe area through high-precision positioning of personnel; One-key help and emergency rescue: When the construction personnel encounter an emergency, they can press the one-key help button on the intelligent sensing terminal. After the system receives the help signal, it immediately displays the accurate position, real-time height, personal identity information and current physiological state of the personnel on the monitoring platform, and quickly calculates and displays the position of the nearest rescue force through the map service; Operation trajectory and event backtracking: The platform stores and manages all sensor data, personnel activity trajectory, equipment operation parameters, environmental change curve and system warning records for a long time. When a safety event or accident occurs, the management personnel can replay all related data before and after the accident through the platform.

[0015] Preferably, the method further comprises: System networking and wireless communication protocol: the system adopts LoRa wireless communication technology to build a star-shaped networking structure, the intelligent sensing terminal as the end node, the signal repeater or LoRa gateway as the center node, the communication system realizes the data transmission and reception between multiple terminal devices through the polling mode, ensures the order and reliability of data transmission, and each intelligent sensing terminal is configured with a unique identifier (ID) which is not repeated. Communication path selection: the control unit of the intelligent sensing terminal can automatically judge whether the personnel is located inside or outside the wind turbine tower, the judgment mechanism includes installing an access detection device at the entrance of the wind turbine tower, or the platform pre-labels the accurate three-dimensional coordinate area of the wind turbine tower, when the real-time position calculated by the UWB positioning module of the personnel coincides with the tower coordinate area, the system determines that the personnel has entered the tower; when the personnel is located inside the tower, the LoRa module of the intelligent sensing terminal preferentially establishes a connection with the nearest signal repeater, and forwards the data to the LoRa gateway through the repeater; when the personnel is located in the unobstructed area outside the tower, the LoRa module of the intelligent sensing terminal directly establishes a connection with the LoRa gateway. The LoRa gateway serves as the wide area network interface of the entire LoRa network, is responsible for uploading all data collected from the LoRa network to the remote central server through the cellular network or wired network, and receives instructions from the server, and the LoRa gateway performs identity authentication and encryption handshake before establishing a connection with the server, and continuously attempts to connect until success.

[0016] The present application constructs an intelligent safety warning system integrating high-precision sensing, multi-source data fusion, intelligent evaluation and prediction, closed-loop decision and execution, realizes the all-round, real-time and active management of the safety of the wind power hoisting operation personnel, and overcomes the inherent defects of the traditional technology in the complex operation environment, such as insufficient sensing accuracy, information isolation and risk judgment lag, and significantly improves the intelligent level of safety management of wind power hoisting operation. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 The present application is an intelligent safety warning method for wind power hoisting operation, and the overall system architecture schematic diagram is shown in the figure. Figure 2 The present application is an intelligent safety warning method for wind power hoisting operation, and the overall system architecture schematic diagram is shown in the figure. Figure 1 The structure schematic diagram of the construction personnel intelligent sensing terminal shown in the figure. Figure 3 The functional block diagram of the multi-source heterogeneous data fusion and intelligent analysis platform of the present application is shown in the figure. Figure 4 The core architecture schematic diagram of the hoisting risk intelligent evaluation and prediction model of the present application is shown in the figure. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] This invention provides an intelligent safety early warning method for wind power installation operations. To enable those skilled in the art to more clearly understand this invention, the specific embodiments of the invention will be described in detail below with reference to the accompanying drawings. It should be understood that the described embodiments are only for illustrating the invention and are not intended to limit the scope of the invention. Those skilled in the art can make various modifications or equivalent substitutions to the technical features without departing from the spirit and scope of the invention.

[0020] Reference Figure 1 The schematic diagram of the overall system architecture of this invention illustrates that the core of the intelligent safety early warning method for wind power installation operations lies in constructing a comprehensive platform integrating high-precision sensing, multi-source data fusion, intelligent assessment and prediction, and closed-loop decision-making and execution. This system architecture encompasses intelligent sensing terminals for construction personnel, a positioning base station network, a signal relay system, a LoRa gateway, and a multi-source heterogeneous data fusion and intelligent analysis platform and a closed-loop intelligent decision-making and execution system deployed on a central server. On-site environmental monitoring equipment, such as high-precision anemometers, lightning warning sensors, and high-definition video surveillance systems, are also integrated into this platform as data input sources.

[0021] In one specific embodiment, the intelligent sensing terminal for construction workers is designed as a highly integrated and compact wearable device, aiming to achieve comprehensive and accurate acquisition of individual worker identity, high-precision location, real-time posture, physiological state (if related sensors are integrated), and micro-environmental parameters of the work area. Figure 2 As shown, the main structure of the intelligent sensing terminal integrates a sensor unit, a data acquisition and processing unit, a human-computer interaction unit, a wireless communication unit, and a power supply unit.

[0022] Specifically, the sensor unit is responsible for real-time sensing and acquiring various physical parameters of the construction personnel and their surrounding environment. The core components of this unit include: an ultra-wideband (UWB) positioning module, an inertial measurement unit (IMU), an air pressure and temperature sensor module, and a safety belt wearing state sensor. The ultra-wideband (UWB) positioning module uses nanosecond-level pulse signals with extremely narrow time-domain width for distance measurement, with a working frequency range set at 3.1 GHz to 10.6 GHz and a pulse repetition frequency up to 100 MHz. This module calculates distance by measuring signal round-trip time (ToF) or time difference of arrival (TDoA), and its ranging accuracy has been verified to be up to 10 centimeters in actual tests. In complex wind power hoisting operation environments, especially in the presence of a large number of metal structures and multipath reflections, the UWB positioning module exhibits excellent anti-interference capability thanks to its high-resolution pulse and spread spectrum technology. After starting, the UWB positioning module automatically sends nanosecond-level pulse signals to the surrounding environment at a preset period (e.g., once per second) to ensure the real-time nature of positioning data.

[0023] The inertial measurement unit (IMU) uses a nine-axis micro-electromechanical system (MEMS) sensor, which internally integrates a three-axis accelerometer, a three-axis gyroscope, and a three-axis magnetometer. The measurement range of the accelerometer is ±16g, and the measurement range of the angular velocity meter is ±2000° / s. Through the internal running of advanced attitude fusion algorithms, such as the Extended Kalman Filter (EKF) algorithm, the IMU can calculate the pitch angle (Pitch), roll angle (Roll), and yaw angle (Yaw) of the construction personnel in real time. In a dynamic environment, its attitude calculation accuracy can be stabilized within ±1°. These attitude data, combined with the linear motion data of the accelerometer, provide key inputs for accurately determining the motion mode (such as stationary, walking, running, climbing) and abnormal state (such as falling, violent shaking, long-term stationary) of the personnel.

[0024] Further, the air pressure and temperature sensor module is composed of a high-precision digital air pressure sensor and a thermistor. The air pressure sensor has a measurement range of 300 hPa to 1100 hPa and a resolution of 0.01 hPa, which can be used to accurately calculate the altitude of the construction personnel. The thermistor has a temperature measurement range of -40°C to 85°C and an accuracy of ±0.5°C. By real-time monitoring of air pressure data and comparison with the preset reference air pressure value, the fine positioning of the construction personnel in the vertical dimension can be realized, which is particularly critical for high-altitude work. The safety belt wearing state sensor is configured as a miniature mechanical sensor or a contact switch, which is cleverly installed at the key connection point of the safety belt, such as the inside of the lock or the load-bearing webbing. When the safety belt is in the normal locking state, the sensor outputs a specific electrical signal; when the safety belt is off, loose, or detects abnormal stress (such as instantaneous impact or displacement beyond the safe range), it outputs another clear signal. The response time of the sensor is strictly controlled to be less than 100 milliseconds to ensure immediate detection of abnormal states of the safety belt.

[0025] The data acquisition and processing unit, as the "brain" of the intelligent sensing terminal, is composed of a low-power microcontroller (MCU) and its supporting circuit. The MCU has a high-performance analog-to-digital converter (ADC), a digital signal processor (DSP), and a non-volatile memory integrated inside. The MCU runs a carefully designed embedded firmware that is responsible for the synchronous acquisition, accurate timestamping, denoising, and formatting of the raw data collected by the sensor unit, as well as preliminary algorithmic calculations. For example, it is responsible for performing initial calculations based on the UWB signal time difference of arrival (TDoA) positioning algorithm, processing the IMU raw data with attitude solving algorithms, and converting the air pressure data into altitude. The MCU also has a perfect power management function, which can monitor the power state of the power supply unit in real time, and automatically trigger a low-power mode or send a low-power warning signal when the power is detected to be below the preset threshold.

[0026] The human-machine interaction unit provides real-time information feedback and emergency call functions for construction personnel, which includes a low-power OLED display screen and an emergency call button. The OLED display screen is used to clearly display the personnel ID, overview information of the current position (such as the distance from the base station), real-time altitude, battery remaining power, and various warning information issued by the system. The emergency call button is designed as a waterproof, anti-misoperation high-reliability physical button. When the construction personnel press the button in an emergency, the wireless communication unit will immediately send a high-priority distress signal to the central server, and a confirmation prompt will be given on the display screen.

[0027] The wireless communication unit is responsible for uploading the data processed by the data acquisition and processing unit to the remote central server in real time, and receiving instructions from the server. This unit uses LoRa wireless communication technology, and its working frequency can be configured as 433MHz or 868MHz / 915MHz according to regional regulations. The LoRa technology can provide excellent anti-interference performance and extremely low power consumption while ensuring data transmission distance (up to several kilometers) due to its unique adjustable spreading factor (SF), adjustable bandwidth (BW) and adjustable coding rate (CR). It is very suitable for wind farms with vast and complex signal environments, and supports point-to-point communication and efficient star network networking.

[0028] The power supply unit is composed of a high energy density lithium polymer battery and an intelligent power management module, which provides stable and durable power supply for each module of the intelligent sensing terminal. The power supply unit can support multiple charging methods, including USB-C interface fast charging and wireless induction charging, to ensure convenient power supply in different operating scenarios.

[0029] In the second aspect, the application provides a positioning base station network and signal relay system for wind power hoisting operations. The positioning base station network is composed of multiple fixedly deployed ultra-wideband (UWB) positioning base stations. The UWB positioning base stations and the UWB modules in the intelligent sensing terminal use the same chip set and communication protocol, ensuring the compatibility and interoperability of UWB signals within the system. Each positioning base station includes a UWB positioning module, a microcontroller (MCU), a wireless communication unit, and a power supply unit. The positioning base station accurately receives the UWB pulse signal from the intelligent sensing terminal and measures the time of arrival (ToA) of the signal, and sends the ToA information together with the accurate coordinate information of the base station itself to the construction personnel intelligent sensing terminal or the signal relay system through the wireless communication unit (LoRa module). The microcontroller is responsible for the accurate control of the UWB module, the processing of raw data, and the seamless interface with the wireless communication unit. The wireless communication unit also uses the same LoRa module as the intelligent sensing terminal to ensure the compatibility and interoperability of data transmission.

[0030] To effectively cope with the serious shielding and attenuation effects of wireless signals by large metal structures such as wind turbine towers, the application further configures multiple signal repeaters. Signal repeaters act as key data forwarding nodes in the LoRa network, and their structure includes a microcontroller (MCU), a wireless communication unit, and a power supply unit. These signal repeaters are strategically deployed at different heights inside the wind turbine tower or in areas with severe signal obstruction. Their core function is to receive LoRa signals from intelligent sensing terminals or positioning base stations and reliably forward them to LoRa gateways or directly transmit them to the central server, thereby establishing stable wireless communication links in complex electromagnetic environments. The microcontroller in the signal repeater has advanced signal strength monitoring capabilities, allowing it to dynamically adjust the forwarding strategy based on real-time signal quality, such as selecting the best path or adjusting the transmission power, to optimize the stability and real-time performance of data transmission. The wireless communication unit also uses high-reliability LoRa modules and supports advanced functions such as relay networking and signal detection to adapt to the complex and variable electromagnetic environment of wind farms.

[0031] The positioning base stations and signal repeaters use independent solar panels combined with large-capacity energy storage batteries to form a composite power supply unit, ensuring continuous and stable operation in remote or off-grid sites. In a typical deployment scenario, positioning base stations are deployed around the hoisting operation area to form a UWB positioning field with a coverage range of hundreds of meters. The number and specific location of the base stations are fine-tuned based on the size of the operation area and the required positioning accuracy. For example, in a circular operation area with a radius of 200 meters, at least four UWB positioning base stations are deployed to achieve high-precision three-dimensional positioning. For higher accuracy requirements, the number of base stations can be increased to six or more. The positioning base station network is synchronized in time through the LoRa network to ensure the accuracy of the TDoA positioning algorithm for time difference measurement, thereby ensuring the accuracy of the overall positioning system.

[0032] In the third aspect, the application provides a multi-source heterogeneous data fusion and intelligent analysis platform. The platform is deployed on the central server and is a high-performance and highly available computing cluster, responsible for receiving and aggregating massive amounts of real-time data from intelligent sensing terminals, positioning base station networks, signal relay systems, and other on-site environmental monitoring devices. Environmental monitoring devices include high-precision anemometers, lightning warning sensors, and high-definition video monitoring systems, which provide macro-environment and visual situation information. The platform first performs strict preprocessing operations on the received multi-source heterogeneous data, including timestamp alignment, data cleaning (removing noise and redundant information), missing value filling (using interpolation or machine learning methods), and outlier detection (based on statistics or model prediction), to ensure the quality and reliability of the data for subsequent analysis.

[0033] A high-precision personnel position and trajectory reconstruction module combines raw ranging data from the UWB positioning module, accurate coordinates of the positioning base stations, and preliminary positioning results of the intelligent sensing terminal. The high-precision position reconstruction module adopts a TDOA algorithm, and the positioning formula is: ; where (x, y) is the personnel coordinate, (x i ,y i ) and (x j ,y i ) are base station coordinates, and c is the speed of light.

[0034] Subsequently, the platform performs the following key function modules, as shown in Figure 3 : A high-precision personnel position and trajectory reconstruction module combines raw ranging data (ToA / TDoA information) from the UWB positioning module, accurate coordinates of the positioning base stations, and preliminary positioning results of the intelligent sensing terminal. Advanced multi-point positioning algorithms, such as a multi-lateration positioning algorithm based on the least squares method or a positioning fusion algorithm based on extended Kalman filtering (EKF), are used to achieve centimeter-level or even sub-meter-level high-precision three-dimensional position calculation of construction personnel on the server side. At the same time, through time series analysis and motion model prediction of position data, the fine motion trajectory of construction personnel in three-dimensional space is reconstructed in real time and continuously.

[0035] A personnel posture and behavior recognition module deeply analyzes acceleration, angular velocity, and magnetometer raw data from the inertial measurement unit (IMU) of the intelligent sensing terminal.

[0036] The personnel body inclination calculation formula is: where ax is the left-right direction acceleration, and az is the vertical direction acceleration.

[0037] Through a pre-trained machine learning or deep learning model, such as a model based on a support vector machine (SVM) or a long short-term memory network (LSTM), various postures (such as standing, walking, running, climbing, bending, and crouching) and abnormal behaviors (such as falling, long-time stillness, violent shaking, and sudden acceleration / deceleration) of construction personnel are recognized in real time. Combined with the data of the safety belt wearing state sensor, it is accurately judged whether the safety belt is correctly worn, including whether it is locked, whether it is loose, and whether the stress point is abnormal.

[0038] Micro-environment parameter and personnel activity association module: This module associates real-time air pressure and temperature data from the air pressure and temperature sensor module, as well as macro-environmental data from external high-precision wind speed and direction instruments, lightning warning sensors, visibility sensors, and other environmental monitoring equipment, through spatial matching algorithms (such as GeoHash or R-tree-based spatial indexing) and time series association, and accurately binds to the real-time location of a specific construction personnel, forming a real-time perception map of the micro-environment in which the personnel are located. This helps to understand the impact of specific environmental factors on individual behavior and risk.

[0039] wherein the height change amount calculation formula is: P1 and P2 are air pressure values at different times, and K is a conversion constant.

[0040] Construction site space situation visualization module: The platform uses advanced digital twin technology to present the overall space situation of the wind power hoisting operation site in real time in a high-fidelity three-dimensional visualization manner. The visualization interface accurately displays the real-time positions, altitudes, postures, and motion trajectories of all construction personnel, as well as the real-time positions, key attitude parameters (such as boom angle, telescopic length), load parameters, pre-set movement paths, and safety distances from personnel of hoisting equipment (such as main cranes, auxiliary cranes, hoist arms, and hoists). The distance D between personnel and equipment is calculated by the formula: wherein R is the current working radius of the equipment, L is the maximum outer dimension of the hoisted object, and S is the set minimum safety margin.

[0041] The interface also accurately labels fixed obstacles, planned safety passages, pre-set dangerous areas (marked by electronic fences), and real-time micro-environment parameters (such as local wind speed, temperature, and visibility) in the work area.

[0042] Hoisting risk intelligent assessment and prediction model: This is the core intelligent module of the invention, which is used to analyze the current operation state and predict potential safety risks in the future. The model uses an improved ResNet50 and Bi-directional Long Short-Term Memory (BiLSTM) network fusion architecture, as shown in Figure 4 .

[0043] Input data: The model receives multi-modal data input, including: 1) Frame sequence of high-definition video stream of construction site (resolution 1920x1080) for extracting spatial features of personnel, hoisting objects, equipment, and obstacles; 2) Real-time data uploaded by intelligent sensing terminals, including accurate position of personnel (centimeter level), altitude, posture, and safety belt status; 3) Operating parameters of hoisting equipment, including crane boom angle (0°-90°), telescopic length, load (0-50 tons), operating speed, and rotation angle; 4) Environmental monitoring data, including real-time wind speed (1Hz sampling, 0-20m / s), wind direction, temperature (-20℃-40℃), humidity, visibility, lightning index, and tidal height (for offshore wind power scenarios, range 0-10 meters).

[0044] Feature extraction: The model extracts visual features in video frames through an improved convolutional neural network (CNN) branch based on ResNet50 architecture. Specifically, the CNN branch first extracts low-level features such as edges and textures of personnel, hoisting objects, and obstacles in the image through multiple 3x3 convolution layers. Then, it uses a spatial attention mechanism to strengthen the feature weights of high-risk areas such as the area below the hoisting object, the equipment operating track, and the personnel-intensive area, to generate a 128x128-dimensional distance feature matrix that accurately represents the relative position relationship and potential collision risk between personnel, hoisting objects, and obstacles.

[0045] The attention weight calculation formula is: where F is the feature map, and W and b are learnable parameters.

[0046] The time series data is processed through a BiLSTM branch. The BiLSTM branch uses a double-layer structure, with 128 neurons in each layer. The forward LSTM is used to capture the motion trend of equipment (such as the rotation angle velocity and lifting speed of the boom), and the backward LSTM is used to analyze the dynamic changes of environmental parameters (such as the gradient of wind speed and the fluctuation of temperature). A gated recurrent unit (GRU) is embedded to effectively solve the gradient vanishing problem that may occur in long sequence (such as video sequences of more than 1000 frames or sensor data sequences of several hours) training. The BiLSTM branch finally outputs a 64-dimensional dynamic feature vector representing the patterns and trends in the time series data.

[0047] Data fusion and risk prediction: The 128x128 distance feature matrix extracted by the CNN branch is flattened and processed through a fully connected layer, and then concatenated and fused with the 64-dimensional dynamic feature vector output by the BiLSTM branch through a fully connected layer. The fused feature vector is input into a Softmax activation function layer, which outputs the probability distribution of the current operating state being in "safe", "warning", or "dangerous" three safety level labels.

[0048] The risk level probability calculation formula is: where X is the fused feature vector.

[0049] The safety state score S of the lifting risk intelligent assessment and prediction model is calculated by the following formula: S = 0.4P + 0.4B + 0.2E; wherein P is a position safety value, ranging between 0-1, determined by the distance of the personnel from the nearest electronic fence danger zone; B is a behavior safety value, ranging between 0-1, determined by the position of the safety belt and the stability of the posture, E is an environmental safety value, ranging between 0-1, determined by the environmental wind speed. For example, [0.95, 0.04, 0.01] indicates a 95% probability of safety.

[0050] Before being put into use, the model has undergone initial training on a large-scale, multi-scenario, multi-modal data set, and the training data contains normal operation data and typical accident case data of different types of wind farms (such as mountainous, plain, and offshore). The model training uses the Focal Loss function to balance the sample class distribution, and the loss function calculation formula is: where Pt is the probability that the model predicts the real category, and γ is the focus parameter.

[0051] The training data set covers device parameters such as crane boom angle 0°-90°, load change 0-50 tons, and environmental data such as wind speed 0-20 m / s sampled at 1 Hz, temperature -20℃-40℃, and contains personnel real-time coordinate accuracy ±30 cm data. The accident case includes video frames and synchronous sensor data of scenes such as collision of hoisted objects and personnel entering dangerous areas. The training process uses the Focal Loss function (γ=2) to balance the sample class distribution, especially setting the weight of dangerous samples to 5 times that of safe samples to solve the sample imbalance problem, and the Adam optimizer is used for parameter iterative optimization, with the learning rate initially set to 0.001. During system operation, the platform periodically collects new pre-warning event and dangerous event data (e.g., daily or weekly), forms an incremental data set, and performs incremental training on the model every month. During incremental training, the first 80% of the parameters of the CNN and BiLSTM are frozen, only the fully connected layer and attention weight are fine-tuned, and the small batch gradient descent method is used with a learning rate of 0.0001 to avoid model forgetting historical knowledge and enable the model to adapt to new risk situations and changes in operating scenarios in a timely manner. For special scenarios such as offshore wind power, the model further introduces a Transformer module to capture complex correlation features between long-term (e.g., 1 hour) wind speed and hoisted object swings through an eight-head self-attention mechanism, and adds tidal height as a feature dimension to improve the recognition and prediction ability of specific environmental risks. The model optimization effect is evaluated by precision (Precision) and recall (Recall) confusion matrix indicators to ensure that the precision is not less than 95% and the recall is not less than 98% when identifying dangerous events.

[0052] In a fourth aspect, the present application provides a closed-loop intelligent decision-making and execution system. Based on the risk assessment results output by the multi-source heterogeneous data fusion and intelligent analysis platform, the system adopts a hierarchical response strategy to achieve a fundamental change from traditional passive response to active prevention. The intelligent decision-making and execution system includes the following core functions: Hierarchical risk warning and notification: based on the output results of the hoisting risk intelligent assessment and prediction model, the risk is divided into level I (minor warning), level II (moderate warning), and level III (serious danger).

[0053] Level I warning: when the system determines that there is a potential risk, such as the horizontal distance between the construction personnel and the hoisting equipment is less than 3 meters but still greater than 1.5 meters, or the vertical clearance is less than 2 meters but the personnel are not yet in the predicted path of the equipment movement, the platform immediately pops up an eye-catching risk prompt window on the remote monitoring interface and pushes notifications to all relevant management personnel (e.g., team leaders, safety officers) through SMS or APP. At the same time, the construction personnel intelligent perception terminal issues a slight vibration prompt or plays a low-frequency voice alarm (e.g., "Please pay attention to the surrounding environment").

[0054] Level II warning: When the risk level is further increased, for example, the construction personnel intrudes into the area 5 meters below the hoisted object within a radius of 5 meters, or the real-time wind speed exceeds 15 m / s for 30 seconds (which is close to the upper limit of safe operation), the platform immediately triggers the high-decibel sound and light alarm on the scene, and sends clear pre-warning information to all personnel on the scene through the high-power voice broadcast system. At the same time, the construction personnel intelligent sensing terminal sends a moderate vibration and plays a pre-recorded voice (such as "Please keep away from the dangerous area and pay attention to avoid") in a loop, the platform automatically sends evacuation instructions to the foreman's intercom, and sends detailed deceleration or suspension operation instructions to the hoisting equipment operator, to ensure that the equipment action is controllable.

[0055] Level III danger: When the system identifies a serious dangerous situation, for example, the horizontal distance between the construction personnel and the hoisting equipment is less than 1.5 meters (which has entered the high-risk collision area), or multiple personnel are in a high-altitude violation state (such as not wearing a safety rope), or the wind speed continuously exceeds 18 m / s (which has exceeded the safe operation threshold of the equipment), the platform immediately triggers the highest level of emergency sound and light alarm, and sends a mandatory evacuation instruction through the automatic voice system, clearly indicating that all personnel immediately evacuate to the designated safe area. At the same time, send an emergency braking instruction to the control system of the hoisting equipment to make the hoisting equipment immediately stop all running actions. The construction personnel intelligent sensing terminal sends a strong vibration and plays an emergency voice (such as "Immediately evacuate, there is a risk of falling or collision") in a loop, and starts the high-brightness flashing red light outside the wind turbine tower, which alerts the distance. The platform also locks the relevant elevators and sends the list of responsible persons (including real-time positioning link) to the emergency management department through SMS or a dedicated communication link, and starts the emergency broadcast system to play the preset emergency plan, guiding the on-site personnel to take emergency measures.

[0056] Electronic fence management: The platform pre-defines "red-yellow-green" three-level electronic fence areas in three-dimensional space according to the characteristics and safety regulations of wind power hoisting operations. The red area is a high-risk area, such as the hoist arm below the operation area, the heavy object running track area, the high-voltage live area, or the equipment rotating blind area; the yellow area is the operation edge buffer area or the potential dangerous area, such as the 3-5 meter range of the hoisting equipment edge; the green area is the safe passage or the designated rest area. When the real-time position of the construction personnel (provided by the high-precision personnel position and trajectory reconstruction module) intrudes into the red area or crosses the fence boundary without authorization, the system immediately triggers the corresponding pre-warning and response measures according to the preset risk level, and the response logic is closely linked with the hierarchical warning mechanism.

[0057] Height and behavior control: When the real-time altitude of the construction personnel (calculated by the combination of barometric pressure and temperature sensor modules and the UWB positioning module) exceeds the safe operation platform range, such as when the personnel are not in a protected basket on a high-altitude operation platform and the height exceeds 80 meters, or when the safety belt wearing state sensor detects that the safety belt hook is not locked, the load point displacement exceeds 0.5 meters (indicating that the safety belt may be loose or subjected to abnormal impact), or other abnormal conditions, the system automatically triggers a warning and pushes detailed information (personnel ID, location, abnormal type) to the mobile terminal of the safety management personnel. At the same time, the personnel are prompted by the intelligent sensing terminal (such as "please check the safety belt state") to guide them to correct unsafe behavior.

[0058] Environmental linkage safety response: When the environmental monitoring system detects that the real-time wind speed exceeds the preset safety threshold (such as a sustained 5-minute average wind speed of 10.8 m / s or a gust exceeding 15 m / s), or detects that lightning is approaching (such as detecting lightning strikes within a radius of 20 kilometers) through a lightning warning sensor, the platform automatically issues an "stop hoisting operation" instruction to the field and confirms whether all construction personnel have evacuated to the designated safe area through high-precision personnel positioning. If it is found that there are still personnel remaining, they will be given priority warning and voice guidance, and the specific location of the remaining personnel will be pushed to the emergency team.

[0059] One-key call for help and emergency rescue: When the construction personnel encounter an emergency situation, they can press the one-key call for help button on the intelligent sensing terminal. The system receives the distress signal and immediately displays the accurate position (combined with UWB positioning), real-time height, personal identity information, and current physiological state (such as heart rate, body temperature, if the intelligent sensing terminal integrates relevant physiological sensors) of the personnel on the central monitoring platform. The system quickly calculates and displays the location of the nearest rescue force (such as an ambulance or safety team) through map services, so that the remote command center can quickly dispatch and guide rescue operations and provide accurate route navigation.

[0060] Operation trajectory and event backtracking: The platform stores and manages all sensor data, personnel activity trajectories, equipment operation parameters, environmental change curves, and system warning records for a long time with high reliability. When a safety event or accident occurs, management personnel can replay all relevant data before and after the accident through the platform, including detailed motion trajectories of personnel, equipment operation records, environmental parameter changes, warning information triggering time and content, etc., to quickly and accurately analyze the cause of the accident and define responsibilities, providing detailed data support for subsequent safety improvement measures and prevention strategies.

[0061] In a fifth aspect, the present application provides a system networking and wireless communication protocol. The system adopts LoRa wireless communication technology to build a star-shaped networking structure, in which the intelligent sensing terminal serves as the end node, and the signal repeater or LoRa gateway serves as the center node. The communication system realizes data transmission and reception among multiple terminal devices through the polling mode, ensuring the orderliness and reliability of data transmission. Specifically, each intelligent sensing terminal is configured with a unique identifier (ID) that is not repeated. When the terminal establishes a connection with the base station or repeater, the base station or repeater records the ID. When multiple terminals simultaneously connect to the base station or repeater, the base station or repeater will send data transmission instructions to the terminals in a predetermined order (e.g., from small to large) according to the terminal ID. The terminal can send the collected data to the base station after receiving the instruction. This mechanism effectively avoids data conflicts and channel congestion.

[0062] The control unit of the intelligent sensing terminal can automatically determine whether the personnel is located inside or outside the wind turbine tower, and select the appropriate communication path accordingly. The judgment mechanism is multi-modal fusion: first, an entry and exit detection device is installed at the entrance of the wind turbine tower, such as an RFID or NFC-based sensor. When the intelligent sensing terminal worn by the construction personnel passes through the detection device, the system determines that the personnel has entered the tower interior and updates their area status in the system. Second, the platform pre-accurately marks the precise three-dimensional coordinate area of the wind turbine tower. When the real-time position calculated by the UWB positioning module of the construction personnel coincides with the tower coordinate area in space, the system determines that the personnel has entered the tower interior, which is the main basis for judgment. When the personnel is located inside the tower, the LoRa module of the intelligent sensing terminal will preferentially establish a connection with the nearest signal repeater and forward the data to the LoRa gateway through the repeater. When the personnel is located outside the tower in an unobstructed area, the LoRa module of the intelligent sensing terminal directly establishes a connection with the LoRa gateway. The LoRa gateway serves as the wide-area network interface of the entire LoRa network, responsible for uploading all data collected from the LoRa network to the remote central server through the cellular network (e.g., 4G / 5G) or wired network, and receiving instructions from the server, realizing bidirectional data transmission. Before establishing a connection with the server, the LoRa gateway will perform strict identity verification and encryption handshake, and continuously attempt to connect until successful, to ensure the security and reliability of data transmission, preventing unauthorized access and data leakage.

[0063] The intelligent safety early warning system is constructed by organic combination of the technical solutions, and realizes all-around, real-time and active management on safety of the wind power hoisting operation construction personnel.

[0064] Although the embodiments of the present application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made thereto without departing from the principles and spirit of the application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A smart safety early warning method for wind power installation operations, characterized in that, include: S1: The construction personnel's various physical parameters and micro-environmental parameters are acquired in real time through at least one intelligent sensing terminal. The intelligent sensing terminal is equipped with an ultra-wideband positioning module, an inertial measurement unit, an air pressure and temperature sensor module, and a safety belt wearing status sensor. S2: By coordinating the ultra-wideband positioning module with the positioning base station network, and combining the precise coordinates of multiple ultra-wideband positioning base stations in the positioning base station network with the preliminary positioning results of the intelligent sensing terminal, high-precision location information of the construction personnel can be obtained. S3: The data collected and preliminarily processed by the intelligent sensing terminal, the positioning auxiliary data of the positioning base station network, and the environmental monitoring data of the on-site environmental monitoring equipment are uploaded to the central server in real time through the wireless communication unit or signal relay system. S4: On the central server, a multi-source heterogeneous data fusion and intelligent analysis platform is run to perform preprocessing operations such as timestamp alignment and data cleaning on the received multi-source data, and to perform high-precision personnel location and trajectory reconstruction, personnel posture and behavior recognition, correlation between micro-environment parameters and personnel activities, and visualization of the spatial situation of the construction site. S5: Based on the high-precision personnel location data, behavior recognition results and environmental parameters output by S4, the hoisting risk intelligent assessment and prediction model in the multi-source heterogeneous data fusion and intelligent analysis platform is used to conduct real-time assessment and prediction of the current operation status and potential safety risks. The hoisting risk intelligent assessment and prediction model adopts a deep learning multimodal fusion architecture. S6: Based on the risk assessment results output by the intelligent assessment and prediction model for hoisting risks in S5, a graded response strategy is adopted through the closed-loop intelligent decision-making and execution system, which includes graded risk warning and notification, electronic fence management, height and behavior control, environmental linkage safety response, one-click emergency call and emergency rescue, and operation trajectory and event backtracking.

2. The intelligent safety early warning method for wind power hoisting operations according to claim 1, characterized in that, The intelligent sensing terminal for construction workers includes: a sensor unit configured to acquire various physical and environmental parameters of the construction workers in real time; a data acquisition and processing unit configured to preprocess, filter, fuse, and perform preliminary calculations on the raw data acquired by the sensor unit; a human-computer interaction unit configured to provide real-time information feedback and emergency call functions to the construction workers; a wireless communication unit configured to upload the processed data to a remote server in real time and receive instructions from the server; and a power supply unit configured to provide a stable power supply to each module of the intelligent sensing terminal.

3. The intelligent safety early warning method for wind power hoisting operations according to claim 2, characterized in that, The sensor unit specifically includes: The ultra-wideband positioning module uses a pulse signal with an extremely narrow time domain width for ranging. The operating frequency range is 3.1 GHz to 10.6 GHz, the pulse repetition frequency is 100 MHz, the ranging accuracy is at the 10 cm level, and it sends nanosecond-level pulse signals to the surrounding environment at a preset period in complex multipath environments. The inertial measurement unit is a nine-axis microelectromechanical system sensor that integrates a three-axis accelerometer, a three-axis gyroscope, and a three-axis magnetometer. The acceleration measurement range is ±16g, and the angular velocity measurement range is ±2000° / s. The pressure and temperature sensor module consists of a high-precision digital pressure sensor and a thermistor. The pressure measurement range is 300 hPa to 1100 hPa with a resolution of 0.01 hPa, and the temperature measurement range is -40℃ to 85℃ with an accuracy of ±0.5℃. The seatbelt wearing status sensor is configured as a miniature force sensor or a contact switch, installed at the critical connection point of the seatbelt, with a response time of less than 100 milliseconds.

4. The intelligent safety early warning method for wind power hoisting operations according to claim 3, characterized in that, The data acquisition and processing unit consists of a low-power microcontroller and its supporting circuitry, integrating an analog-to-digital converter, a digital signal processor, and a memory. The low-power microcontroller runs embedded firmware and is responsible for the synchronous acquisition of sensor data, timestamp marking, noise reduction, data formatting, and preliminary algorithm calculations, including initial calculations for positioning algorithms based on the UWB signal arrival time difference, IMU attitude calculation algorithms, and barometric altitude conversion algorithms. The low-power microcontroller also has power management functions to monitor the power status of the power supply unit. The human-computer interaction unit includes a low-power OLED display screen for displaying personnel ID, current location overview, altitude, battery level, and warning information; The wireless communication unit adopts LoRa wireless communication technology, operates at a frequency of 433MHz or 868MHz / 915MHz, and has adjustable spreading factor, adjustable bandwidth, and adjustable coding rate.

5. The intelligent safety early warning method for wind power hoisting operations according to claim 1, characterized in that, The positioning base station network consists of multiple fixed-deployment ultra-wideband positioning base stations. The UWB positioning base station and the UWB module of the intelligent sensing terminal use the same chipset and communication protocol. Each positioning base station includes a UWB positioning module, a microcontroller, a wireless communication unit and a power supply unit. The ultra-wideband positioning base station receives UWB pulse signals from the intelligent sensing terminal and accurately measures the arrival time of the signals. The signal relay system consists of multiple signal repeaters. Each signal repeater serves as a data forwarding node in the LoRa network. Its structure includes a microcontroller, a wireless communication unit, and a power supply unit. The signal repeaters are deployed at different heights inside the wind turbine tower or in areas where signals are blocked.

6. The intelligent safety early warning method for wind power hoisting operations according to claim 1, characterized in that, The multi-source heterogeneous data fusion and intelligent analysis platform is deployed on a central server. After performing data preprocessing, it executes the following functional modules: The high-precision personnel location and trajectory reconstruction module, combining the raw ranging data from the UWB positioning module, the precise coordinates of the positioning base station, and the preliminary positioning results from the intelligent sensing terminal, employs the TDOA algorithm. The positioning formula is as follows: ； Where (x, y) are the coordinates of the personnel, (x... i ,y i ) and (x j ,y i All coordinates are base station coordinates, and c is the speed of light; The personnel posture and behavior recognition module performs in-depth analysis on the acceleration, angular velocity, and magnetometer data from the inertial measurement unit of the intelligent sensing terminal. It identifies the real-time posture and abnormal behavior of construction workers through a pre-trained machine learning or deep learning model. The formula for calculating the personnel's body tilt angle is: , where ax is the acceleration in the left-right direction and az is the acceleration in the vertical direction; The microenvironment parameter and personnel activity correlation module associates real-time air pressure and temperature data from the air pressure and temperature sensor module, as well as macro-environmental data from the on-site environmental monitoring equipment, with spatial matching algorithms and time series correlation to accurately bind them to the real-time location of specific construction personnel. The formula for calculating height change is as follows: P1 and P2 are the air pressure values ​​at different times, and K is the conversion constant; The construction site spatial situation visualization module uses digital twin technology to present the overall spatial situation of the wind power hoisting operation site in real time in a three-dimensional visualization manner. It accurately displays the real-time location, height, posture and movement trajectory of all construction personnel, as well as the real-time location, posture, load parameters, movement path and safety distance of hoisting equipment, and marks fixed obstacles, safety passages, dangerous areas and micro-environment parameters in the operation area. The formula for calculating the distance D between personnel and equipment is: Where R is the current working radius of the equipment, L is the maximum external dimensions of the object being hoisted, and S is the set minimum safety margin.

7. The intelligent safety early warning method for wind power hoisting operations according to claim 6, characterized in that, The intelligent assessment and prediction model for hoisting risks adopts an improved ResNet50 and BiLSTM network fusion architecture, including: Input data: The model receives multimodal data input, including high-definition video stream frame sequences from the construction site, real-time data uploaded by the intelligent sensing terminal, operating parameters of the hoisting equipment, and environmental monitoring data; Feature extraction: The model extracts visual features from video frames through an improved convolutional neural network branch. This branch extracts low-level features of people, hanging objects, and obstacles in the image through multiple 3x3 convolutional layers, and uses a spatial attention mechanism to strengthen the feature weights of high-risk areas to generate a 128x128-dimensional distance feature matrix. Simultaneously, a BiLSTM branch processes time-series data. The BiLSTM branch adopts a two-layer structure, with each layer containing 128 neurons and embedding gated recurrent units, ultimately outputting a 64-dimensional dynamic feature vector. The formula for calculating attention weights is: , where F is the feature map, and W and b are learnable parameters; Data fusion and risk prediction: The distance feature matrix extracted by the convolutional neural network branch and the dynamic feature vector output by the BiLSTM branch are concatenated and fused through a fully connected layer. The fused feature vector is input into a Softmax activation function layer, which outputs the probability distribution of the current operation status being labeled as "safe", "warning" or "dangerous". The formula for calculating the probability of risk level is: , where X is the fused feature vector; Before being deployed, the model underwent initial training on a large-scale, multi-scenario, multi-modal dataset. The training process employed the Focal Loss function to balance the sample class distribution and used the Adam optimizer for iterative parameter optimization. The loss function calculation formula is as follows: , where Pt is the probability that the model predicts the true class, and γ is the focusing parameter; The safety status score S of the hoisting risk intelligent assessment and prediction model is calculated using the following formula: S = 0.4P + 0.4B + 0.2E; Wherein, P is the location safety value, ranging from 0 to 1, determined by the distance between the person and the nearest electronic fence danger zone; B is the behavior safety value, ranging from 0 to 1, determined by the position and posture stability of the safety belt; and E is the environmental safety value, ranging from 0 to 1, determined by the environmental wind speed.

8. The intelligent safety early warning method for wind power hoisting operations according to claim 1, characterized in that, The hierarchical risk warning and notification system in the closed-loop intelligent decision-making and execution system includes: Level I warning: When the system determines that there is a potential risk, a risk warning window will pop up on the remote monitoring interface and relevant management personnel will be notified via SMS or APP; Level II Warning: When the risk level increases, an audible and visual alarm will be triggered immediately, and a warning message will be broadcast to all personnel on site via a voice broadcast system; Level III Hazard: When the system identifies a serious hazard, it immediately triggers the highest level of emergency audible and visual alarm, issues a forced evacuation command through the automatic voice system, and simultaneously sends an emergency braking command to the hoisting equipment control system.

9. The intelligent safety early warning method for wind power hoisting operations according to claim 1, characterized in that, The closed-loop intelligent decision-making and execution system also includes: Electronic fence management: The platform pre-defines three-level electronic fence areas in three-dimensional space: red, yellow and green. The red area is a high-risk area, the yellow area is a buffer zone or potential danger zone at the edge of the operation, and the green area is a safe passage or rest area. Height and Behavior Control: When the real-time altitude of the construction personnel exceeds the range of the safe working platform, or when the safety belt wearing status sensor detects abnormal situations such as the safety belt hook not being locked or the load-bearing point displacement exceeding 0.5 meters, the system automatically triggers an early warning and pushes it to the safety management personnel. At the same time, the system provides voice prompts to the personnel through the intelligent sensing terminal. Environmental linkage safety response: When the on-site environmental monitoring equipment detects that the real-time wind speed exceeds the preset safety threshold, or when the lightning warning sensor detects that lightning is approaching, the platform automatically issues a "stop hoisting operation" command to the site, and confirms whether all construction personnel have evacuated to a safe area through high-precision personnel positioning; One-click SOS and Emergency Rescue: When the construction worker encounters an emergency, he / she can press the one-click SOS button on the intelligent sensing terminal. After receiving the SOS signal, the system will immediately display the worker's precise location, real-time altitude, personal identification information and current physiological status on the monitoring platform, and quickly calculate and display the location of the nearest rescue force through map services. Operation trajectory and event retrospective: The platform stores and manages all sensor data, personnel activity trajectories, equipment operating parameters, environmental change curves, and system early warning records for a long period of time. When a safety incident or accident occurs, the platform can replay all relevant data before and after the accident.

10. The intelligent safety early warning method for wind power hoisting operations according to claim 1, characterized in that, The method further includes: System networking and wireless communication protocol: The system adopts LoRa wireless communication technology to build a star network structure. The intelligent sensing terminal is the end node, and the signal repeater or LoRa gateway is the central node. The communication system realizes data transmission and reception between multiple terminal devices through polling to ensure the orderliness and reliability of data transmission. Each intelligent sensing terminal is configured with a unique identifier. Communication path selection: The control unit of the intelligent sensing terminal can automatically determine whether the personnel are inside or outside the wind turbine tower. The determination mechanism includes installing an entry and exit detection device at the entrance of the wind turbine tower, or pre-marking the precise three-dimensional coordinate area of ​​the wind turbine tower. When the real-time position calculated by the construction personnel through the UWB positioning module coincides with the coordinate area of ​​the tower, it is determined that the personnel have entered the tower. The LoRa gateway, serving as the wide area network interface for the entire LoRa network, is responsible for uploading all data collected from the LoRa network to a remote central server via a cellular or wired network, and for receiving instructions from the server. Before establishing a connection with the server, the LoRa gateway performs authentication and an encrypted handshake, and continuously attempts to connect until successful.

Images