Safety control method and system for fan maintenance operation area

Abstract

The invention discloses a safety management and control method and system for a fan maintenance operation area, and the method comprises the steps: collecting the data of the fan maintenance operation area, and building a risk analysis model for processing the collected data; the risk level of the operation area is judged according to the analysis result of the risk analysis model, the risk level is dynamically adjusted in combination with real-time data, and an early warning signal is sent out; intelligent equipment is used for continuously monitoring the equipment state and the physiological state of an operator, the state of the operator is judged through a reward mechanism, and safety management and control of a draught fan overhaul operation area are achieved. According to the invention, through data acquisition and a risk analysis model of a deep neural network, accurate monitoring of a working environment is ensured; secondly, the risk level is dynamically adjusted, early warning is given out in time, and the early warning accuracy and response speed are remarkably improved; and finally, through monitoring of intelligent equipment and introduction of a reward mechanism, health and safety of operators are guaranteed, and potential safety hazards caused by sudden environment change or poor personnel state are effectively reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of fan maintenance, and in particular to a safety control method and system for a fan maintenance operation area. Background Art

[0002] With the growth of global energy demand and the improvement of environmental protection awareness, wind power generation has been widely used as a renewable energy source. As the core equipment of wind power generation, the reliability and safety of wind turbines are crucial to the stable operation of wind power generation systems. In recent years, the scale and height of wind turbines have continued to increase, especially large wind turbines in offshore wind farms, which makes wind turbine maintenance operations more complicated and dangerous. Traditional wind turbine maintenance mainly relies on manual inspection and regular maintenance, but with the increase in the number of wind turbines and the increase in the capacity of single units, traditional maintenance methods appear to be inefficient and high-risk. Therefore, how to improve the efficiency of wind turbine maintenance operations while ensuring the safety of operators has become an important research direction in the current technical field.

[0003] Most of the existing wind turbine maintenance safety technologies focus on fault diagnosis and predictive maintenance of the equipment itself, while paying less attention to the safety management and control of operators. Traditional methods mainly judge risks through manual experience, lack scientific risk analysis and real-time monitoring methods, which makes safety accidents easy to occur in high-risk operation areas. In addition, the existing technologies mostly use static analysis for risk assessment of the operation area, and fail to fully consider the dynamic changes of real-time environmental data, resulting in delayed or inaccurate early warning signals. At the same time, most of the existing intelligent equipment only has basic physiological parameter monitoring functions, which cannot be fully combined with the risk analysis model, and thus cannot effectively guarantee the safety of operators when dealing with complex operating environments. Therefore, the existing technology has obvious deficiencies in the safety management and control of wind turbine maintenance operations, and there is an urgent need for a safety management and control method that can be dynamically adjusted and warned in real time to improve the safety and efficiency of wind turbine maintenance operations.

[0004] The existing safety control technology for wind turbine maintenance operations has problems such as inaccurate risk assessment, insufficient dynamic warning, and single intelligent equipment functions, and the content of the present invention solves these problems. The present invention belongs to the field of wind power generation, and specifically relates to a safety control method and system for wind turbine maintenance operation areas. Through the combination of risk analysis models and intelligent equipment, real-time monitoring and dynamic warning of the physiological state of operators and the risk level of the operation area are achieved, which significantly improves the safety of wind turbine maintenance operations. Summary of the invention

[0005] In view of the above-mentioned problems, the present invention is proposed.

[0006] Therefore, the technical problem solved by the present invention is that the prior art has problems such as inaccurate risk assessment, insufficient dynamic warning and single function of intelligent equipment.

[0007] To solve the above technical problems, the present invention provides the following technical solutions: a method for safety management and control of a fan maintenance operation area, comprising: collecting data of the fan maintenance operation area, establishing a risk analysis model to process the collected data; determining the risk level of the operation area according to the analysis results of the risk analysis model, dynamically adjusting the risk level in combination with real-time data and issuing an early warning signal; using intelligent equipment to continuously monitor the equipment status and physiological status of the operating personnel, judging the operating personnel status through a reward mechanism, and realizing safety management and control of the fan maintenance operation area.

[0008] As a preferred solution of the safety control method of the wind turbine maintenance operation area described in the present invention, the data of the wind turbine maintenance operation area includes real-time collection of wind speed, temperature and humidity, vibration and noise data using sensors installed around the wind turbine and in the operation area; the operation area is divided into a tower body area, a cabin area and a blade area according to different maintenance work, and the data of each operation area is collected independently and analyzed using a risk analysis model.

[0009] As a preferred solution of the safety control method for the wind turbine maintenance operation area described in the present invention, the risk analysis model includes: according to the data of the wind turbine maintenance operation area, based on the deep neural network DNN, a risk analysis model is established for risk identification. The structure of the risk analysis model is an input layer, a first hidden layer, an adaptive activation function layer, a feature interaction layer, a dynamic time convolution layer and an output layer. The input layer receives the input data and directly passes it to the hidden layer. In the hidden layer, the input data is preliminarily weighted and summed, and a bias term is added; in the adaptive activation function layer, the activation function is constructed by combining Sigmoid and hyperbolic tangent functions, and the adaptive parameter θ is added to flexibly adjust the response degree according to the change of the input data. The linearly transformed data is activated by the activation function, which is expressed as:

[0010]

[0011] Among them, σ(z,θ) represents the activation function; tanh represents the hyperbolic tangent function; z represents the input value of the activation function; α represents the adjustment coefficient; n represents the total number of input data; H (m) represents the output of the mth layer of the adaptive activation function layer; x i Represents different features in the input data; in traditional DNN, input features are usually assumed to be independent of each other, but there are complex interactions between the data in the wind turbine maintenance operation area. Therefore, a feature interaction layer is added to represent the interaction between input features through multiplication to capture the nonlinear relationship between the features, which can be expressed as:

[0012]

[0013] Among them, F int (x i ,x j ) represents the feature interaction term; k represents the index variable; λ k represents the weight of the feature interaction term; represents the i-th feature in the input data; the dynamic time convolution layer captures the changes in the data in the time dimension through the time convolution kernel in response to sudden changes in real-time data, and enhances the risk analysis model's ability to capture sudden changes, which is expressed as:

[0014]

[0015] in, represents the output of the kth layer of the dynamic temporal convolutional layer; κ t represents the time convolution kernel; t represents the time index; T represents the time step; the output layer outputs the final prediction result, which is expressed as:

[0016]

[0017] in, represents the final risk prediction value; W (k) represents the weight matrix of the kth layer; b k Represents the bias term of the kth layer.

[0018] As a preferred solution of the safety management method of the wind turbine maintenance operation area described in the present invention, the risk level of the operation area includes: The risk level of the operation area is determined by comparing it with the preset risk threshold. The risk level is divided into three levels: low risk, medium risk and high risk. The risk level of the operation area is visualized and sent to the mobile terminal of the maintenance operator. There are no restrictions on the low-risk operation area, and maintenance operations are allowed. The medium-risk operation area is analyzed every five minutes through the risk analysis model. If it is three consecutive times, There is an upward trend or If the high-risk level threshold is reached at 90%, an early warning signal will be issued to remind maintenance workers to evacuate; maintenance work is not allowed in high-risk operation areas.

[0019] As a preferred solution of the safety control method for the wind turbine maintenance operation area described in the present invention, the intelligent equipment includes two parts: a safety helmet and a safety belt. A physiological state sensor and a posture sensor are embedded in the safety helmet. The safety belt is a safety equipment covering the whole body including shoulder straps, chest straps, waist belts and leg straps. Tensioning sensors are embedded in the shoulder straps, chest straps and waist belts, and position sensors are embedded in the leg straps and waist belts. There is a groove on the plug and socket parts of each buckle of the safety helmet and the safety belt, and a micro switch is embedded. When the plug is inserted into the socket and fully buckled, the trigger rod of the micro switch is pressed, and a signal path is formed, so it is judged that the buckle is fully buckled.

[0020] As a preferred scheme of the safety control method for the wind turbine maintenance operation area described in the present invention, wherein: the equipment status includes, before performing the maintenance operation, after the maintenance worker wears the smart equipment, the smart equipment performs self-inspection to determine the equipment status, and the safety helmet collects the angle information of the three axes of pitch angle, roll angle and yaw angle through the attitude sensor. The preset standard wearing angle range is -5° to 15° for pitch angle, -10° to 10° for roll angle, and -15° to 15° for yaw angle. The angles of the three axes are compared and judged separately. When the angles of the three axes are all within the preset standard wearing angle range and the buckle of the safety helmet is fully fastened, it is judged that the safety helmet is worn correctly; the tensioning sensor of the safety belt measures the tensioning degree of the shoulder strap, chest strap and waist belt, and makes a tensioning judgment based on the tension value. If the detected tension value is within the preset range and the buckle is fully fastened, it is judged that the safety belt is worn correctly.

[0021] As a preferred solution of the safety control method for the wind turbine maintenance operation area of ​​the present invention, the physiological state includes: according to the physiological state of the operator, a dynamic early warning is performed through a reward mechanism combined with the risk level and the operation time, and the state space is defined as the real-time physiological state data of the operator, and the output of the risk analysis model and operation time, the actions include issuing warnings and continuing monitoring; the Q value of each action in different states is initialized to 0, the initial learning rate α is set to 0.1, and the discount factor γ is set to 0.9; the reward mechanism is divided into two parts: a risk-based reward mechanism and a work time-based reward mechanism. The risk-based reward mechanism is only activated when working in a medium-risk operation area, and whether to trigger a warning action is determined based on the physiological state data, which is expressed as:

[0022]

[0023] Among them, R risk represents the risk reward value; λ represents the risk coefficient; n represents the total number of types of physiological state data; w i represents the weight of physiological state data i; f i (s i) represents the abnormal measurement function of the ith physiological state parameter; when the maintenance operator is working in a low-risk operation area, R risk =0; the reward mechanism based on the working time predefines a safe working time window T safe , the probability of early warning triggering when the physiological state is abnormal within the safe operation time window is low, which is expressed as:

[0024]

[0025] Among them, R time represents the time reward value; β represents the time reward coefficient; t represents the operation time; τ represents the time growth factor; combining the two reward mechanisms, the optimal action is selected by comprehensive Q value calculation, which is expressed as:

[0026] Q(s,a)←Q(s,a)+α×((R risk +R time )+γ×max a Q(s′,a′)-Q(s,a))

[0027] Among them, Q(s,a) represents the expected return after taking action a in state s; max a Q(s′,a′) represents the maximum expected benefit of selecting the optimal action a′ in the next state s′; each time an action is selected, an ε-greedy strategy is adopted to randomly select an action with an ε probability and select the action with the maximum current Q value with a probability of 1-ε; a low threshold Q of the comprehensive Q value is set l and high threshold Q h , when Q(s,a) l When Q l ≤Q(s,a)≤Q h When Q(s,a)>Q h When a high-level warning occurs, it will issue an alert and notify the monitoring center, reminding the operators to stop the operation immediately and evacuate the fan maintenance area.

[0028] ​In the second aspect, the present invention also provides a safety management and control system for a wind turbine maintenance operation area, including a risk analysis module, which collects environmental data of the wind turbine maintenance operation area through sensors, divides the operation area into a tower main body area, a cabin area and a blade area, and establishes a risk analysis model to process the data of the three areas respectively; a risk level module, which determines the risk level of the operation area according to the analysis results of the risk analysis model, takes different measures for areas with different risk levels, dynamically adjusts the risk level in combination with real-time data, and issues an early warning signal; a personnel monitoring module, which uses intelligent equipment to continuously monitor the equipment status and physiological status of the operating personnel, judges the operating personnel status through a reward mechanism, and realizes safety management and control of the wind turbine maintenance operation area.

[0029] In a third aspect, the present invention further provides a computing device, comprising: a memory and a processor;

[0030] The memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions. When the computer executable instructions are executed by the processor, the steps of the safety management and control method of the wind turbine maintenance operation area are implemented.

[0031] In a fourth aspect, the present invention further provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, can implement the steps of the method for safety management and control of the wind turbine maintenance operation area.

[0032] Beneficial effects of the present invention: The method of the present invention ensures accurate monitoring of the working environment through data collection and risk analysis models of deep neural networks; secondly, it dynamically adjusts the risk level and issues early warnings in a timely manner, significantly improving the accuracy and response speed of early warnings; finally, through the introduction of monitoring and reward mechanisms for intelligent equipment, the health and safety of operating personnel are guaranteed, effectively reducing safety hazards caused by sudden environmental changes or poor personnel conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0034] Figure 1 An overall flow chart of a safety management and control method for a wind turbine maintenance operation area provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0035] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0036] Example 1, reference Figure 1 , is an embodiment of the present invention, and provides a safety management and control method for a wind turbine maintenance operation area, comprising:

[0037] S1: Collect data from the fan maintenance operation area and establish a risk analysis model to process the collected data.

[0038] A wind turbine refers to a wind turbine, which is a device that converts wind energy into electrical energy. A wind turbine consists of blades, a nacelle, a tower and a base. The blades usually have three pieces and are made of lightweight but strong materials. The wind drives the blades to rotate. The nacelle is located at the top of the tower and contains core components such as the generator, gearbox, and control system. The tower is a structure that supports the wind turbine and is usually tens to hundreds of meters high to ensure that the blades can be exposed to stronger winds. The base provides stability for the wind turbine and is usually made of concrete.

[0039] The main contents of wind turbine maintenance work are blade inspection and maintenance, equipment maintenance inside the cabin and internal inspection of the tower. During the operation, it is necessary to use climbing equipment (such as lifting platforms, rope systems) to safely enter the tower or blade high-altitude operation area. Therefore, it is very necessary to conduct a risk assessment before the operation, identify the dangerous points in the operation area and formulate safety measures, and equip necessary personal protective equipment (such as safety belts, helmets, gloves, etc.) to ensure the safety of the operators.

[0040] Furthermore, before the wind turbine maintenance work is carried out, the system first activates the multi-layer perception mechanism, using multiple sensors installed around the wind turbine and in the operating area to collect key environmental parameters in real time. The sensors are distributed on the ground, inside the tower and at high altitudes to ensure full coverage of the entire operating area.

[0041] Key environmental parameters include wind speed, temperature and humidity, vibration and noise data. Wind speed is the speed at which wind moves on a horizontal plane per unit time. Excessively high wind speed may affect the stability of operators and even cause equipment to lose control.

[0042] Temperature data indicates the heat of the air, and humidity indicates the content of water vapor in the air. High temperature may cause heatstroke to personnel or overheating of equipment, while high humidity may cause equipment to become slippery and increase the risk of corrosion.

[0043] Vibration data indicates the amplitude and frequency of movement of wind turbine towers, blades or other mechanical parts. Excessive vibration may indicate structural problems, such as loose bolts or unbalanced blades, which may cause safety accidents.

[0044] Noise data indicates the sound intensity in the operating area, with a particular focus on mechanical noise and environmental noise. Sudden or abnormal noise is often a precursor to equipment failure, such as bearing damage or blade striking.

[0045] Furthermore, according to the data of the wind turbine maintenance operation area, a risk analysis model is established based on the deep neural network DNN for risk identification. The structure of the risk analysis model is input layer, the first hidden layer, adaptive activation function layer, feature interaction layer, dynamic time convolution layer and output layer. The input layer receives the input data and passes it directly to the hidden layer. In the hidden layer, the input data is preliminarily weighted summed and a bias term is added.

[0046] In the adaptive activation function layer, the activation function is constructed by combining Sigmoid and hyperbolic tangent functions. The adaptive parameter θ is added to flexibly adjust the response degree according to the changes in the input data. The linearly transformed data is activated by the activation function, which is expressed as:

[0047]

[0048] Among them, σ(z,θ) represents the activation function; tanh represents the hyperbolic tangent function; z represents the input value of the activation function; α represents the adjustment coefficient; n represents the total number of input data; H (m) represents the output of the mth layer of the adaptive activation function layer; x i Represents different features in the input data.

[0049] Through the adaptive activation function, the model can dynamically adjust the response strength, adapt to the changes in different data characteristics, and improve the ability to capture nonlinear characteristics, so that the model can still maintain high-precision prediction in the complex environment of wind turbine maintenance.

[0050] In traditional DNN, input features are usually assumed to be independent of each other. However, there are complex interactions between the data in the wind turbine maintenance operation area (for example, there may be a certain correlation between wind speed and vibration). Therefore, a feature interaction layer is added to express the interaction between input features through product, capturing the nonlinear relationship between the features, which can be expressed as:

[0051]

[0052] Among them, F int (xi,xj) represents the feature interaction term; k represents the index variable; λ k represents the weight of the feature interaction term; Represents the i-th feature in the input data.

[0053] The dynamic time convolution layer targets sudden changes in real-time data by capturing changes in the time dimension of data through the time convolution kernel, thus enhancing the risk analysis model’s ability to capture sudden changes, which can be expressed as:

[0054]

[0055] in, represents the output of the kth layer of the dynamic temporal convolutional layer; κ t represents the temporal convolution kernel; t represents the time index; T represents the time step.

[0056] The output layer outputs the final prediction result, expressed as:

[0057]

[0058] in, represents the final risk prediction value; W (k) represents the weight matrix of the kth layer; b k Represents the bias term of the kth layer.

[0059] S2: Determine the risk level of the operating area based on the analysis results of the risk analysis model, dynamically adjust the risk level based on real-time data and issue an early warning signal.

[0060] The classification of risk levels (low risk, medium risk and high risk) directly affects the safety of maintenance operations. The definition of risk levels is based on the comparison between the output of the risk analysis model and the preset risk threshold. The preset risk threshold is used to distinguish the boundaries of different risk levels. It is set based on a large amount of historical data and actual operating experience, and can accurately reflect the safety of the operating environment.

[0061] The risk level of the operating area includes the output of the risk analysis model The risk level of the operating area is determined by comparing it with the preset risk threshold. The risk level is divided into three levels: low risk, medium risk and high risk. The risk level of the operating area is visualized and sent to the mobile terminal of the maintenance personnel.

[0062] There are no restrictions on low-risk operation areas, and maintenance operations are allowed; medium-risk operation areas are analyzed every five minutes using the risk analysis model. If three consecutive There is an upward trend or If the high-risk level threshold is reached at 90%, an early warning signal will be issued to remind maintenance workers to evacuate; maintenance work is not allowed in high-risk operation areas.

[0063] The risk level is visualized and sent to the operator's mobile terminal to ensure that the operator can understand the current risk status intuitively and in a timely manner. Compared with the traditional risk notification method, the visualization method of the present invention makes the risk information more intuitive and easy to understand, reducing the possibility of safety accidents caused by misunderstandings or information lags. In addition, the dynamic adjustment and continuous monitoring function of the real-time risk level enables the operator to get accurate safety guidance at the first time, significantly improving the safety and efficiency of wind turbine maintenance operations.

[0064] S3: Use intelligent equipment to continuously monitor the equipment status and physiological status of operators, judge the status of operators through a reward mechanism, and achieve safe management and control of the wind turbine maintenance operation area.

[0065] Smart equipment refers to personal protective equipment (PPE) with built-in sensors and communication modules that can monitor the physiological and working status of workers in real time and communicate with the central control system to ensure safety.

[0066] Smart equipment consists of a helmet and a safety belt. The helmet is embedded with physiological state sensors and posture sensors, including heart rate monitoring, GPS positioning, gyroscopes, cameras and wireless communication modules. It can monitor the position and posture of workers in real time, record the working environment, detect head impacts, or provide real-time communication functions.

[0067] A safety belt is a full-body safety equipment that includes shoulder straps, chest straps, waist straps and leg straps, covering the upper and lower body to ensure that workers are fully protected when working at heights. This full-body safety belt has various built-in sensors that can monitor the worker's posture, tension, suspension status, etc., and issue alarms or automatically record data in abnormal situations. Tension sensors are embedded in the shoulder straps, chest strap and waist strap of the safety belt, and position sensors are embedded in the leg straps and waist strap.

[0068] There is a groove in the plug and socket part of each buckle of the safety helmet and safety belt, in which a micro switch is embedded. When the plug is inserted into the socket and fully buckled, the trigger rod of the micro switch is pressed, and a signal path is formed, which determines that the buckle is fully buckled.

[0069] Furthermore, before performing maintenance work, after the maintenance workers wear the smart equipment, the smart equipment performs self-inspection to determine the equipment status. The safety helmet collects angle information of the three axes: pitch angle, roll angle and yaw angle through the attitude sensor. The preset standard wearing angle range is -5° to 15° for pitch angle, -10° to 10° for roll angle, and -15° to 15° for yaw angle. The angles of the three axes are compared and judged separately. When the angles of the three axes are all within the preset standard wearing angle range and the buckle of the safety helmet is fully fastened, it is judged that the safety helmet is worn correctly.

[0070] The seat belt tension sensor measures the tension of the shoulder strap, chest strap and waist belt, and makes a tension judgment based on the tension value. If the detected tension value is within the preset range and the buckle is fully fastened, the seat belt is judged to be worn correctly.

[0071] Furthermore, during the maintenance process, intelligent equipment will obtain the operator's physiological data in real time from multiple sensors (such as heart rate monitoring, blood oxygen saturation, body temperature, respiratory rate, GSR, etc.), such as heart rate, blood oxygen saturation, body temperature, respiratory rate, galvanic skin response (GSR), etc., and through the reward mechanism, dynamic early warning will be carried out in combination with risk level and operation time. The system can regard actions under different physiological states (such as whether to issue a warning or not) as multiple slot machine arms, and update the priority of each action through feedback after each operation (such as improvement or deterioration of physiological state, operator response, etc.). The weight of risk level and operation time can directly affect the size of the reward, thereby giving priority to early warning actions under high-risk or long-term operations.

[0072] Define the state space as the real-time physiological state data of the operator and the output of the risk analysis model and operation time. The actions include issuing warnings and continuing monitoring. The Q value of each action in different states is initialized to 0, and the initial learning rate α is set to 0.1, and the discount factor γ is set to 0.9.

[0073] The reward mechanism is divided into two parts: risk-based reward mechanism and work time-based reward mechanism. The risk-based reward mechanism is only activated when working in a medium-risk work area. It determines whether to trigger the warning action based on the physiological status data, which is expressed as:

[0074]

[0075] Among them, R risk represents the risk reward value; λ represents the risk coefficient; n represents the total number of types of physiological state data; w i represents the weight of physiological state data i; f i (s i ) represents the abnormal measurement function of the ith physiological state parameter; when the maintenance operator is working in a low-risk operation area, R risk is 0; when When the person is in a medium-risk area and has an abnormal physiological state (such as excessive heart rate, low blood oxygen, etc.), the system increases the Q value through a reward function to increase the probability of selecting a warning action.

[0076] It should be noted that when the risk analysis model outputs When the risk is low, the system considers the working environment to be relatively safe, so the reward mechanism of the risk analysis model will not be activated. In this case, R riskIt is directly set to 0, which means that the risk factor does not affect the current Q value update, and the system mainly relies on other factors (such as operation time, real-time physiological status, etc.) to make judgments.

[0077] In high-risk situations, which usually means that the operation cannot be carried out, the system will directly prohibit the operation or stop the warning-related calculations. Therefore, the probability of triggering the warning will be increased only when working in a medium-risk area through a risk-based reward mechanism.

[0078] The reward mechanism based on working time predefines a safe working time window T safe , the probability of early warning triggering when the physiological state is abnormal within the safe operation time window is low, which is expressed as:

[0079]

[0080] Among them, R time represents the time reward value; β represents the time reward coefficient; t represents the operation time; τ represents the time growth factor; when the operation time exceeds the set safety time threshold (such as 2 hours), the system begins to apply the time reward function. As the operation time increases, the Q value of triggering the warning action will increase exponentially, which significantly increases the probability of triggering the warning. This ensures that when the operation time is too long, the operator can be reminded and rest in time.

[0081] Combining the two reward mechanisms, the optimal action is selected by comprehensive Q value calculation, expressed as:

[0082] Q(s,a)←Q(s,a)+α×((R risk +R time )+γ×max a Q(s′,a′)-Q(s,a))

[0083] Among them, Q(s,a) represents the expected return after taking action a in state s; max a Q(s′,a′) represents the maximum expected benefit of selecting the optimal action a′ in the next state s′.

[0084] Each time an action is selected, the ε-greedy strategy is used to randomly select an action with an ε probability and select the action with the largest current Q value with a probability of 1-ε. The ε-greedy strategy is used to select actions in the state space to ensure that the system maintains a balance between exploring new strategies and using existing optimal strategies. The specific steps are:

[0085] The initial value of ε is set to 0.1, indicating that the system has a 10% probability of randomly selecting actions to explore new decision paths. As the system learns more deeply, the value of ε is gradually reduced to close to 0, enhancing the system's reliance on the learned optimal strategy.

[0086] Generate a random number r in the range of [0,1]. If r<ε, the system randomly selects an action for exploration, which can be early warning, continued monitoring, and other actions. Random selection helps to discover potential better strategies.

[0087] If r ≥ ε, the system selects the action with the highest current Q value and uses it. This is the optimal decision based on the existing data, aiming to maximize the expected benefit of the current state.

[0088] After selecting an action, the system updates the Q value based on the current state and the execution result of the selected action (such as changes in the operator's physiological state, whether an early warning is triggered, etc.). The Q value update formula is as described above.

[0089] As the system learns, the ε value is gradually reduced, so that the system has more exploration behavior in the early stage and relies more on the learned optimal strategy in the later stage. The adjustment formula is expressed as:

[0090] ∈=∈ min +(∈ max -∈ min )×exp(-κ×n)

[0091] Among them, ∈ max represents the maximum value of ∈; ∈ min represents the minimum value of ∈; κ represents the parameter that controls the decreasing rate of the ∈ value; n represents the current number of learning times or rounds.

[0092] After each Q-value update, the system further adjusts the reward function and Q-value update parameters based on the operator's actual feedback (such as recovery or deterioration of physiological state) to optimize the strategy.

[0093] Set the lower threshold Q of the comprehensive Q value l and high threshold Q h , when Q(s,a) l When the alarm is raised, no warning is given and detection continues.

[0094] When Q l ≤Q(s,a)≤Q h When , a primary warning is issued to remind the operators to take a rest, but the operation time t is not reset.

[0095] When Q(s,a)>Q h When a high-level warning occurs, it will issue an alert and notify the monitoring center, reminding the operators to stop the operation immediately and evacuate the fan maintenance area.

[0096] ​The present embodiment also provides a safety management and control system for a wind turbine maintenance operation area, including: a risk analysis module, which collects environmental data of the wind turbine maintenance operation area through sensors, divides the operation area into a tower main body area, a cabin area and a blade area, and establishes a risk analysis model to process the data of the three areas respectively; a risk level module, which determines the risk level of the operation area according to the analysis results of the risk analysis model, takes different measures for areas with different risk levels, dynamically adjusts the risk level in combination with real-time data, and issues an early warning signal; a personnel monitoring module, which uses intelligent equipment to continuously monitor the equipment status and physiological status of the operating personnel, judges the operating personnel status through a reward mechanism, and realizes safety management and control of the wind turbine maintenance operation area.

[0097] This embodiment also provides a computing device, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the method for safety management and control of the wind turbine maintenance operation area as proposed in the above embodiment.

[0098] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the safety management and control method of the wind turbine maintenance operation area proposed in the above embodiment is implemented.

[0099] The storage medium proposed in this embodiment and the safety control method of the wind turbine maintenance operation area proposed in the above embodiment belong to the same inventive concept. The technical details not described in detail in this embodiment can be referred to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0100] Through the above description of the implementation method, the technicians in the relevant field can clearly understand that the present invention can be implemented by means of software and necessary general hardware, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ReadOnly, Memory, ROM), random access memory (RandomAccessMemory, RAM), flash memory (FLASH), hard disk or optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform the methods of various embodiments of the present invention.

[0101] Example 2, the following is an embodiment of the present invention, which provides a safety management and control method for a wind turbine maintenance operation area. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through simulation experiments.

[0102] The purpose of this embodiment is to verify the application effect of the risk analysis model based on deep neural network (DNN) and intelligent equipment combined with reward mechanism in wind turbine maintenance operations. The experimental data includes environmental parameters (wind speed, temperature and humidity, vibration, noise) and the physiological status data of the operators (heart rate, blood oxygen saturation, etc.). To ensure the controllability of the experimental data, the changes in environmental parameters will be simulated in the form of human input to simulate the real operation scene, rather than collected in real time. During the experiment, the experiment simulates the changes in the wind turbine maintenance operation environment to evaluate the early warning triggering and response effects of the system.

[0103] By manually inputting environmental data, each simulation scenario tests five serial data points. When the system is at a medium risk level, it will perform risk analysis again after each data point. If three consecutive data points show an increase in risk, the system will issue an early warning. The data of one simulation experiment is shown in Table 1.

[0104] Table 1 Environmental parameters and risk level data table

[0105]

[0106] It can be seen from the data in Table 1 that the wind speed between test No. 2 and No. 3 increased from 13m / s to 15m / s, and the vibration frequency also increased from 1.0Hz to 1.3Hz accordingly. This synchronous increase reflects the synergistic effect of environmental changes. The increase in wind speed may lead to enhanced structural vibration, thereby causing the system to adjust the risk level from low risk to medium risk.

[0107] In sequence number 5, wind speed, vibration and noise all increased significantly. The risk analysis model accurately identified this change and raised the risk level to high risk. This adjustment not only illustrates the sensitivity of the DNN model to changes in multi-dimensional environmental parameters, but also proves the accuracy and real-time performance of the model in complex environments.

[0108] The monitoring experiment of the physiological state of the operators continued. The operators wore smart helmets and safety belts to simulate their state when working at high altitudes. The system monitored the physiological state of the operators such as heart rate and blood oxygen saturation in real time through smart equipment, and combined with the environmental risk level and working time, the system used a reward mechanism to dynamically judge the necessity of early warning. The experimental data is shown in Table 2.

[0109]

[0110] Between test sequence 2 and sequence 3, the heart rate dropped from 80bpm to 77bpm, while the breathing rate increased from 18 times / minute to 19 times / minute. During this period, the simulated personnel did not perform any activities, simulating the rest process after the primary warning. In actual high-altitude work situations, this fluctuation may be caused by the operator taking a short rest or adapting to environmental changes. In this case, the system maintained the primary warning state, indicating that the system can recognize and adapt to natural fluctuations in physiological parameters.

[0111] As the working time increases, the body temperature and galvanic skin response (GSR) show a gradual increase trend, especially in test sequence 5, the body temperature rises to 37.0℃, and the GSR reaches 17μS. The system triggers a high-level warning in this state. This shows that the system can effectively combine physiological status data and working time to determine the physiological stress of the operator and prevent safety problems caused by excessive fatigue or other stress factors.

[0112] In summary, the DNN model of the present invention can effectively handle complex changes in the real environment, especially under nonlinear and multivariable conditions, and accurately and dynamically adjust the risk level. This dynamic response capability significantly improves the safety of wind turbine maintenance operations and ensures that operators are warned at the first time when risks increase.

[0113] By introducing a reward mechanism and Q-value update strategy, accurate monitoring and early warning of the physiological state of operators are achieved, significantly improving operational safety. The system can not only handle the natural fluctuations of physiological data, but also accurately trigger early warnings at critical moments to prevent potential safety accidents.

[0114] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for safety control of a fan maintenance operation area, characterized in that: include: Collect data from the fan maintenance operation area and establish a risk analysis model to process the collected data; Determine the risk level of the operating area based on the analysis results of the risk analysis model, dynamically adjust the risk level based on real-time data and issue early warning signals; Intelligent equipment is used to continuously monitor the equipment status and physiological status of operators, and the status of operators is judged through a reward mechanism to achieve safe management and control of the wind turbine maintenance operation area.

2. The method for safety control of a wind turbine maintenance operation area according to claim 1, characterized in that: The data of the wind turbine maintenance operation area includes real-time collection of wind speed, temperature and humidity, vibration and noise data using sensors installed around the wind turbine and in the operation area; The operation area is divided into the tower main area, the nacelle area and the blade area according to the different maintenance work. The data of each operation area is collected independently and analyzed using the risk analysis model.

3. The method for safety control of a wind turbine maintenance operation area according to claim 2, characterized in that: The risk analysis model includes, based on the data of the wind turbine maintenance operation area, establishing a risk analysis model for risk identification based on the deep neural network DNN, the structure of the risk analysis model is an input layer, a first hidden layer, an adaptive activation function layer, a feature interaction layer, a dynamic time convolution layer and an output layer, the input layer receives input data and directly passes it to the hidden layer, the input data is preliminarily weighted summed in the hidden layer, and a bias term is added; In the adaptive activation function layer, the activation function is constructed by combining Sigmoid and hyperbolic tangent functions. The adaptive parameter θ is added to flexibly adjust the response degree according to the changes in the input data. The linearly transformed data is activated by the activation function, which is expressed as: Among them, σ(z,θ) represents the activation function; tanh represents the hyperbolic tangent function; z represents the input value of the activation function; α represents the adjustment coefficient; n represents the total number of input data; H (m) represents the output of the mth layer of the adaptive activation function layer; x i Represent different features in the input data; In traditional DNN, input features are usually assumed to be independent of each other, but there are complex interactions between the data in the wind turbine maintenance operation area. Therefore, a feature interaction layer is added to express the interaction between input features through multiplication to capture the nonlinear relationship between the features, which can be expressed as: Among them, F int (x i ,x j ) represents the feature interaction term; k represents the index variable; λ k represents the weight of the feature interaction term; Represents the i-th feature in the input data; The dynamic time convolution layer targets sudden changes in real-time data by capturing changes in the time dimension of data through the time convolution kernel, thus enhancing the risk analysis model’s ability to capture sudden changes, which can be expressed as: in, represents the output of the kth layer of the dynamic temporal convolutional layer; κ t represents the time convolution kernel; t represents the time index; T represents the time step; The output layer outputs the final prediction result, expressed as: in, represents the final risk prediction value; W (k) represents the weight matrix of the kth layer; b k Represents the bias term of the kth layer.

4. The method for safety control of a wind turbine maintenance operation area according to claim 3, characterized in that: The risk level of the operation area includes the output of the risk analysis model. Compare with the preset risk threshold to determine the risk level of the operation area. The risk level is divided into three levels: low risk, medium risk, and high risk. The risk level of the operation area is visualized and sent to the mobile terminal of the maintenance operator; There are no restrictions on low-risk work areas, and maintenance work is allowed; The medium-risk operation area is analyzed by the risk analysis model every five minutes. There is an upward trend or If the high-risk level threshold is reached at 90%, an early warning signal will be issued to remind maintenance workers to evacuate; Maintenance work is not allowed in high-risk work areas.

5. The method for safety control of a wind turbine maintenance operation area according to claim 4, characterized in that: The intelligent equipment includes two parts: a safety helmet and a safety belt. A physiological state sensor and a posture sensor are embedded in the safety helmet. The safety belt is a safety equipment covering the whole body including a shoulder strap, a chest strap, a waist belt and a leg strap. Tension sensors are embedded in the shoulder strap, chest strap and waist belt, and position sensors are embedded in the leg strap and waist belt. There is a groove in the plug and socket parts of each buckle of the safety helmet and the safety belt, and a micro switch is embedded. When the plug is inserted into the socket and fully buckled, the trigger rod of the micro switch is pressed, a signal path is formed, and it is judged that the buckle is fully buckled.

6. The method for safety control of a wind turbine maintenance operation area according to claim 5, characterized in that: The equipment status includes that before the maintenance work is carried out, after the maintenance worker wears the smart equipment, the smart equipment performs self-inspection to determine the equipment status, the safety helmet collects the angle information of the three axes of pitch angle, roll angle and yaw angle through the attitude sensor, the preset standard wearing angle range is -5° to 15° for pitch angle, -10° to 10° for roll angle, and -15° to 15° for yaw angle, and the angles of the three axes are compared and judged separately. When the angles of the three axes are all within the preset standard wearing angle range and the buckle of the safety helmet is fully buckled, it is judged that the safety helmet is worn correctly; The seat belt tension sensor measures the tension of the shoulder strap, chest strap and waist belt, and makes a tension judgment based on the tension value. If the detected tension value is within the preset range and the buckle is fully fastened, the seat belt is judged to be worn correctly.

7. The method for safety control of a wind turbine maintenance operation area according to claim 6, characterized in that: The physiological state includes dynamic warning based on the physiological state of the operator through a reward mechanism combined with risk level and operation time, defining the state space as the real-time physiological state data of the operator, and the output of the risk analysis model and operation time, actions include issuing warnings and continuing monitoring; The Q value of each action in different states is initialized to 0, the initial learning rate α is set to 0.1, and the discount factor γ is set to 0.9; The reward mechanism is divided into two parts: risk-based reward mechanism and work time-based reward mechanism. The risk-based reward mechanism is only activated when working in a medium-risk work area. It determines whether to trigger the warning action based on the physiological status data, which is expressed as: Among them, R risk represents the risk reward value; λ represents the risk coefficient; n represents the total number of types of physiological state data; w i represents the weight of physiological state data i; f i (s i ) represents the abnormal measurement function of the ith physiological state parameter; when the maintenance operator is working in a low-risk operation area, R risk is 0; The reward mechanism based on working time predefines a safe working time window T safe , the probability of early warning triggering when the physiological state is abnormal within the safe operation time window is low, which is expressed as: Among them, R time represents the time reward value; β represents the time reward coefficient; t represents the operation time; τ represents the time growth factor; Combining the two reward mechanisms, the optimal action is selected by comprehensive Q value calculation, expressed as: Q(s,a)←Q(s,a)+α×((R risk +R time )+γ×max a Q(s′,a′)-Q(s,a)) Among them, Q(s,a) represents the expected return after taking action a in state s; max a Q(s′,a′) represents the maximum expected benefit of selecting the optimal action a′ in the next state s′; each time an action is selected, an ε-greedy strategy is adopted to randomly select an action with an ε probability and select the action with the largest current Q value with a 1-ε probability; Set the lower threshold Q of the comprehensive Q value l and high threshold Q h , when Q(s,a) l When the alarm is raised, no warning is given and the detection continues;​ When Q l ≤Q(s,a)≤Q h When , a primary warning is issued to remind the operator to take a rest, but the operation time t is not reset; When Q(s,a)>Q h When a high-level warning occurs, it will issue an alert and notify the monitoring center, reminding the operators to stop the operation immediately and evacuate the fan maintenance area.

8. A safety control system for a wind turbine maintenance operation area using the method according to any one of claims 1 to 7, characterized in that: include, The risk analysis module collects environmental data of the wind turbine maintenance operation area through sensors, divides the operation area into the tower main area, the nacelle area and the blade area, and establishes a risk analysis model to process the data of the three areas respectively; The risk level module determines the risk level of the operating area based on the analysis results of the risk analysis model, takes different measures for areas with different risk levels, dynamically adjusts the risk level based on real-time data, and issues early warning signals; The personnel monitoring module uses intelligent equipment to continuously monitor the equipment status and physiological status of operators, judges the status of operators through a reward mechanism, and realizes safe management and control of the wind turbine maintenance operation area.

9. A computing device comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the method described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions, when executed by a processor, implement the steps of the method according to any one of claims 1 to 7.

Images