A Fan Fault Emergency Prediction and Control Method Based on Data-Driven Strategy

Abstract

The present invention discloses a fan fault emergency prediction control method based on a data-driven strategy, which relates to the field of intelligent fan control algorithms. The present invention collects internal state data and external environment data of the fan in normal and emergency states through a historical data set, and uses these data to construct and train a deep learning model, so as to achieve accurate prediction and intelligent control of the fan state. Moreover, the intelligent control method proposed by the present invention can not only improve the operating efficiency of the fan, reduce the maintenance cost, but also significantly enhance the safety and reliability of the system.

Description

Technical Field

[0001] The present invention relates to the field of wind turbine control, and specifically to a wind turbine fault emergency prediction control method based on a data-driven strategy. Background Art

[0002] In the field of modern industrial automation, as one of the key devices, the stability and reliability of wind turbines are crucial for the safe operation of the entire system. With the development of technology, traditional wind turbine control methods have gradually been unable to meet the growing demands for intelligence and automation. The existing technologies mainly rely on simple sensors and fixed control logics, which often lack flexibility and adaptability and are difficult to cope with complex and changing working environments. For example, traditional wind turbine control systems usually only consider the operating state of the wind turbine and ignore external environmental factors and potential emergency situations such as sudden events like fires. This limitation results in slow responses of wind turbines to sudden events and may even lead to more serious safety accidents.

[0003] Another defect of the existing technologies lies in the singularity of their control strategies and the lack of prediction capabilities. Most wind turbine control systems adopt rule-based control strategies, which often cannot adapt to the changing demands of wind turbines in different operating states. In addition, these systems lack the ability to predict the future state of wind turbines and cannot make corresponding adjustments and optimizations in advance. In practical applications, wind turbines may enter a fault or emergency state due to changes in the external environment or abnormalities in the internal state, while the existing technologies often cannot identify these states in time and make effective responses, which not only affects the operating efficiency of wind turbines but also increases the maintenance cost and safety risks. Summary of the Invention

[0004] The present invention provides a wind turbine fault emergency prediction control method based on a data-driven strategy, aiming to provide a more efficient and accurate control algorithm to improve the intelligence and precision of wind turbine control. Among them:

[0005] A wind turbine fault emergency prediction control method based on a data-driven strategy includes the following steps:

[0006] S1. Collect the internal state data and external environment data of the wind turbine in normal and emergency states through a historical data set. The internal state data includes current and voltage data, amplitude and rotational speed data, and the external environment data includes external component state data, temperature data, and atmospheric data;

[0007] S2. Use the internal status data and external environment data collected under the normal and emergency statuses of the fan as sample data to construct and train a deep learning model, and output prediction results and control decisions through the deep learning model. The prediction results include the emergency status of fan faults and the normal status of the fan, and the control decisions include fan adjustment parameters and fan working mode selection for the fan status;

[0008] S3. Use the constructed deep learning model to input new data, and the control terminal controls and adjusts the fan according to the output prediction results and control decisions;

[0009] Among them, the step S2 specifically includes the following sub-steps:

[0010] S201. Define the state space according to the sample data, define the action space of the fan control actions regarding the sample data, and design the reward function;

[0011] S202. Construct a normal policy network and an emergency policy network through a recurrent neural network;

[0012] S203. Define the loss functions of the recurrent neural network model, which respectively include the policy loss function, the value loss function, and the emergency loss function, and calculate the total loss function according to the defined policy loss function, value loss function, and emergency loss function;

[0013] S204. Establish and train the model, and input the sensing data collected by the fan front-end data acquisition unit into the trained model to output the fan control decision and the fan working mode control decision for the fan operation.

[0014] Furthermore, in the step S201, the definition of the state space is specifically as follows:

[0015] ;

[0016] Among them, the represents the state space, the is a composite state vector including sensing data and an emergency status identifier, the represents the sensing data, and the represents the emergency status identifier;

[0017] The definition of the action space is specifically as follows:

[0018] ;

[0019] Among them, the represents the action space, the represents the control actions of the fan, and the control actions at least include fan control parameter adjustment and fan working mode switching.

[0020] Further, in step S202, constructing the normal policy network specifically includes: according to the characteristics of the wind turbine status data changing over time, constructing the normal policy network through a recurrent neural network; wherein, when the control actions of the wind turbine are continuous, a linear activation function is used as the output layer, and the output vector of the output layer represents a point in the action space; when the control actions of the wind turbine are discrete, the probability of each action being selected is calculated through the softmax function, specifically:

[0021] ;

[0022] wherein, the represents the normal policy network, the represents the action, the represents the bias of the action , the represents the bias of the action , the represents the weight of the action , the represents the weight of the action , the represents the transpose, the represents the feature representation of the state , the represents the size of the action space, the represents the probability of taking the action under the state .

[0023] Further, in step S202, constructing the emergency policy network specifically includes: according to the characteristics of the emergency state data changing over time, processing the time series data of the emergency signal through a recurrent neural network; the input layer receives the emergency state identifier , the hidden layer processes the input data through time steps, and the output layer generates a decision signal; wherein, the output layer of the emergency policy network converts the output of the recurrent neural network through the sigmoid function to generate a value between 0 and 1, representing the probability of taking an emergency action.

[0024] Further, in step S203, the policy loss function is specifically expressed as:

[0025] ;

[0026] wherein:

[0027] ;

[0028] the represents the policy loss function, the represents the expectation, the represents the probability of taking action under the old policy in state . The represents the probability of taking action under the new policy in state . The represents the new set of parameters. The represents the ratio of the probabilities between the new policy and the old policy. The represents a preset small constant used to control the amplitude of policy update. The represents the advantage function of the old policy, which is used to represent the difference between the cumulative return obtained by taking action in state and the average cumulative return under the old policy.

[0029] Furthermore, in step S203, the value loss function is specifically expressed as:

[0030] ;

[0031] wherein, the represents the value loss function, the represents the expectation, the represents the value network, the represents the time step the immediate return obtained after taking the action. The represents the time step state of represents the discount factor.

[0032] Furthermore, in step S203, the contingency policy loss function is specifically expressed as:

[0033] ;

[0034] wherein, the represents the expectation, the represents the contingency policy loss function; the represents the probability of taking a contingency action given the contingency state identifier . The represents the contingency state identifier.

[0035] Furthermore, in step S203, the total loss function is specifically expressed as:

[0036] ;

[0037] wherein, the represents the total loss function, the represents The weight of denotes the weight of denotes the policy loss function, and denotes the value loss function, and denotes the emergency policy loss function.

[0038] Preferably, a fan fault emergency prediction and control system based on a data-driven policy is proposed. The system is implemented based on the fan fault emergency prediction and control method based on a data-driven policy described in any one of the above, and includes:

[0039] A data acquisition module, configured to collect internal state data and external environment data of the fan in the normal state and the emergency state through a historical data set. The internal state data includes current and voltage data, amplitude and rotation speed data, and the external environment data includes external component state data, temperature data, and atmospheric data;

[0040] A model construction module, configured to construct and train a deep learning model by using the internal state data and external environment data of the fan in the normal state and the emergency state collected as sample data, and output a prediction result and a control decision through the deep learning model. The prediction result includes the fan fault emergency state and the fan normal state, and the control decision includes fan adjustment parameters and fan working mode selection of the fan state;

[0041] A control output module, configured to input new data through the constructed deep learning model, and the control end controls and adjusts the fan according to the output prediction result and control decision.

[0042] Further, the model construction module specifically includes:

[0043] A data definition unit, configured to define a state space according to the sample data, define an action space of the fan control action with respect to the sample data, and design a reward function;

[0044] A network construction unit, configured to construct a normal policy network and an emergency policy network through a recurrent neural network;

[0045] Define the loss function of the recurrent neural network model. The loss function respectively includes a policy loss function, a value loss function, and an emergency loss function, and calculate the total loss function according to the defined policy loss function, value loss function, and emergency loss function;

[0046] A model training unit, configured to establish and train the model, and input the sensing data collected by the fan front-end data acquisition unit into the trained model, and output the fan control decision and the fan working mode control decision of the fan operation.

[0047] The beneficial effects of the present invention are as follows:

[0048] (1) By collecting the internal state data and external environment data of the fan in the normal state and emergency state through the historical data set, the present invention constructs and trains a deep learning model by using these data, so as to realize the accurate prediction and intelligent control of the fan state;

[0049] (2) The present invention processes time series data through the normal policy network and emergency policy network constructed by the recurrent neural network, captures the dynamic relationship between the fan state and environmental factors, so as to provide a more accurate control strategy; By defining the policy loss function, value loss function and emergency loss function, the prediction and control capabilities of the model are optimized to ensure that the fan can maintain the best performance in various working states; It can dynamically adjust the operation parameters and working modes of the fan according to real-time sensing data, improve the response speed of the system and the ability to handle emergencies; The intelligent control method proposed by the present invention can not only improve the operation efficiency of the fan, reduce the maintenance cost, but also significantly improve the safety and reliability of the system. Description of the Drawings

[0050] Figure 1 It is a method flow chart of a fan fault emergency prediction control method based on a data-driven strategy proposed by the present invention. Detailed Embodiments

[0051] The technical solutions of the present invention will be further described in detail below with reference to the drawings, but the protection scope of the present invention is not limited to the following.

[0052] In order to make the purpose, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail in conjunction with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention, that is, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Usually, the components of the embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations.

[0053] Therefore, the detailed description of the embodiments of the present invention provided in the drawings below is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work fall within the protection scope of the present invention. It should be noted that relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations.

[0054] Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, article or apparatus comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or apparatus. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, article or apparatus comprising said element.

[0055] The features and performance of the present invention will be further described in detail below in conjunction with embodiments.

[0056] A fan fault emergency prediction control method based on a data-driven strategy, as Figure 1 , includes the following steps:

[0057] S1. Collect the internal state data and external environment data of the fan in the normal state and the emergency state through the historical data set. The internal state data includes current and voltage data, amplitude and rotational speed data, and the external environment data includes external component state data, temperature data, and atmospheric data;

[0058] S2. Use the collected internal state data and external environment data of the fan in the normal state and the emergency state as sample data to construct and train a deep learning model. Output the prediction results and control decisions through the deep learning model. The prediction results include the fan fault emergency state and the fan normal state, and the control decisions include the fan adjustment parameters and the fan operating mode selection of the fan state;

[0059] S3. Use the constructed deep learning model, input new data, and the control terminal performs control adjustment on the fan according to the output prediction results and control decisions;

[0060] Among them, the step S2 specifically includes the following sub-steps:

[0061] S201. Define the state space according to the sample data, define the action space of the fan control action regarding the sample data, and design the reward function;

[0062] S202. Construct a normal policy network and an emergency policy network through a recurrent neural network;

[0063] S203. Define the loss function of the recurrent neural network model. The loss function includes a policy loss function, a value loss function, and an emergency loss function respectively. Calculate the total loss function according to the defined policy loss function, value loss function, and emergency loss function;

[0064] S204. Model establishment and training, and input the sensing data collected by the front-end data acquisition unit of the fan into the trained model to output the fan control decision and the fan operation mode control decision for the operation of the fan.

[0065] Furthermore, according to the specific implementation principle of the method proposed in this embodiment above, its specific composition structure is proposed: including an intelligent fan (including a fan, a data acquisition unit for collecting front-end data of the fan, and an FPGA fan signal processing unit for processing fan data), a rear-end data calculation unit of the fan, an atmospheric environment sensor, and a fan frequency converter implementation. A set of rear-end data calculation units of the fan can control multiple (multiple groups) intelligent fans. The front-end data acquisition unit of the fan is a variety of sensors for monitoring the operating state of the fan; the atmospheric environment sensor group is used to monitor air quality to cooperate with the realization of intelligent control of the fan operation; the FPGA fan signal processing unit collects, performs spectrum analysis and processing on the sensors in the front-end data acquisition unit of the fan, and transmits the relevant processed data to the rear-end data calculation unit of the fan or other relevant gateway devices through various interface forms, depending on the design requirements of the actual project. The rear-end data calculation unit of the fan controls the operation of the fan through the frequency converter. The entire ventilation system supports both manual and automatic control, and can not only achieve local control of the fan, but also achieve remote control of the fan. In addition, it should be noted that this embodiment can be applied to building environments, traffic tunnel transportation, or industrial processes, and the type of fan can be changed according to actual needs. Exemplarily, this embodiment uses a jet fan.

[0066] Among them, the fan monitoring and sensing unit can accurately sense the safety and operating state of the fan, and can upload the parameters to the rear-end data calculation unit of the fan in real time. This sensing unit integrates 4-channel safety suspension structure member monitoring sensors, 1-channel motor current / voltage sensor, 1-channel temperature sensor, 1-channel vibration sensor, and 1-channel rotational speed sensor, etc., to monitor the state of the safety suspension structure member, current and voltage values, the operating temperature of the motor, the amplitude of the impeller shaft, and the rotational speed of the impeller, etc. data, and can upload the parameters to the rear-end data calculation unit of the fan in real time.

[0067] Furthermore, the rear-end data calculation unit of the fan adopts an ARM architecture 1GHz MCU, a Linux kernel system, 512MB DDR3, 8GB eMMC, with a microSD slot x1, a 10 / 100M Ethernet port, 1xRS-232 / 422, 2xRS-485-2w and other characteristics, providing an operating environment for the overall intelligent fan control strategy.

[0068] Furthermore, the atmospheric environment sensor is used to monitor air pollution indicators such as carbon monoxide (CO), VI (smoke concentration), SO2, NO2, O3, PM10, and PM2.5.

[0069] Further, the fan frequency converter is mainly controlled by the data calculation unit at the rear end of the fan. A variable frequency speed control device is adopted to change the speed of the fan, thereby changing the air volume of the fan to meet the air volume requirement, so as to achieve the most energy-saving operation and the highest comprehensive benefit.

[0070] Further, in the step S201, the definition of the state space is specifically as follows:

[0071] ;

[0072] Among them, the represents the state space, the is a composite state vector including sensing data and an emergency state identifier, the represents the sensing data, and the represents the emergency state identifier;

[0073] The definition of the action space is specifically as follows:

[0074] ;

[0075] Among them, the represents the action space, the represents the control action of the fan, and the control action at least includes the adjustment of the fan control parameters and the switching of the fan working mode.

[0076] Further, in the step S202, the construction of the conventional policy network is specifically as follows: According to the characteristics of the fan state data changing with time, a conventional policy network is constructed through a recurrent neural network; among them, when the control actions of the fan are continuous, a linear activation function is used as the output layer, and the output vector of the output layer represents a point in the action space; when the control actions of the fan are discrete, the probability of each action being selected is calculated through the softmax function, specifically as follows:

[0077] ;

[0078] Among them, the represents the conventional policy network, the represents the action, the represents the bias of the action , the represents the bias of the action , the represents the weight of the action , the represents the weight of the action , the represents the transpose, the represents the feature representation of the state, and the represents the size of the action space. Indicates the probability of taking an action in the state .

[0079] Furthermore, in step S202, constructing the emergency policy network specifically includes: according to the characteristics of the change of emergency state data over time, processing the time series data of the emergency signal through a recurrent neural network; the input layer receives the emergency state identifier , the hidden layer processes the input data through time steps, and the output layer generates a decision signal; among them, the output layer of the emergency policy network converts the output of the recurrent neural network through the sigmoid function to generate a value between 0 and 1, indicating the probability of taking an emergency action.

[0080] It should be noted that in step S202, the output decision of the fire emergency policy network specifically is: according to the binary decision to decide whether to switch the working mode, that is, whether to switch to the emergency power frequency circuit; among them, the represents the fire emergency policy network, the represents the fire state identifier, the represents the bias, the represents the probability of taking an action in the emergency action state , the represents the weight, the represents the transpose, the represents the sigmoid function.

[0081] Furthermore, as a preferred implementation of the above embodiment, it also includes compounding the fire emergency policy network and the conventional fan control policy network, and deploying the policy using the compound policy network. Specifically: the final state vector . In addition, it is also necessary to design the reward function. Exemplarily, the designed reward function is: , where the represents the energy consumption, the represents the system safety index, the represents the equipment life loss, the represents the weight of the equipment life, the represents the weight of the system safety, the represents the weight of the energy consumption.

[0082] Furthermore, in step S203, the policy loss function is specifically expressed as:

[0083] ;

[0084] Among them:

[0085] ;

[0086] The represents the policy loss function, and the represents the expectation. The represents the probability of taking action under the old policy in state . The represents the probability of taking action under the new policy in state . The represents the new set of parameters. The represents the ratio of the probabilities between the new policy and the old policy. The represents a preset small constant used to control the magnitude of policy update. The represents the advantage function of the old policy, which represents the difference between the cumulative return obtained by taking action in state and the average cumulative return under the old policy. Exemplarily, assume that in wind turbine control, the new policy recommends operating the wind turbine at a higher speed for energy conservation, while the old policy recommends a lower speed.

[0087] Furthermore, in step S203, the value loss function is specifically expressed as:

[0088] ;

[0089] wherein, the represents the value loss function, the represents the expectation, the represents the value network, the represents the time step the immediate return obtained after taking action, the represents the time step state of , and the

[0090] represents the discount factor. Exemplarily, if the wind turbine operates in a certain state and the value network predicts its return to be 10, while the actual return obtained (including energy conservation and system safety) is 12, then the TD error is 2, and the value loss will encourage the value network to reduce this prediction error.

[0091] ;

[0092] wherein, the represents the expectation, the represents the emergency policy loss function; the represents the emergency state identifier given In the case of, the probability of taking emergency actions, the represents an emergency status identifier. Exemplarily, when a fire occurs, if the emergency strategy network correctly decides to switch to the emergency power frequency circuit, then is 1, and the model will receive positive feedback for the correct decision. Conversely, if the decision is incorrect, negative feedback will be received.

[0093] Furthermore, in the step S203, the total loss function is specifically expressed as:

[0094] ;

[0095] wherein, the represents the total loss function, the represents the weight of, the represents the weight of, the represents the policy loss function, the represents the value loss function, the represents the emergency strategy loss function.

[0096] Furthermore, the above embodiments can optimize the response speed and decision-making quality of the model through the emergency strategy loss function for emergency situations, ensuring a rapid switch to the emergency power frequency circuit when an emergency occurs. In addition, the mean squared error loss function of the value network in the above embodiments ensures an accurate prediction of future rewards, which helps the fan control strategy to also consider other value parameters when considering energy conservation. Compared with the prior art, the above embodiments can enable the model to simultaneously consider multiple objectives such as energy conservation, efficiency, safety, and emergency response through the combination of three loss functions, achieving multi-objective optimization.

[0097] Furthermore, as a preferred implementation of the above embodiments, when applied to the tunnel traffic field, the fan backend data calculation unit is also connected to an external network and an atmospheric environment sensor group, and can receive traffic data such as vehicle flow according to the external network and receive atmospheric environment data according to the atmospheric environment sensor group. The atmospheric environment data includes, but is not limited to, CO, SO2, NO2, and O3. The above environmental data and traffic data can both be used as state vectors of the state space to make decisions in combination with a control algorithm.

[0098] The above embodiments construct a composite strategy network to handle conventional fan control and fire emergency mode control. For the existing DRL solutions, the present embodiments can handle both conventional control and emergency control, and introduce hierarchical decision-making, where the conventional fan control strategy and the fire emergency strategy are dynamically selected according to the environmental state (such as a fire signal), increasing the flexibility of the system.

[0099] Specifically, through the above embodiments, the intelligent fan has the following advantages:

[0100] Energy conservation and environmental protection: Applying the variable frequency commissioning technology to the traditional fan control system can achieve speed regulation in stages, effectively reduce power consumption and improve the energy-saving effect. At the same time, it can reduce the fan noise and extend the service life of the fan.

[0101] Optimized control method: The traditional ventilation system generally adopts the threshold control method, and uses distributed CO / VI sensors to detect the concentration values of carbon monoxide and smoke. When the values reach a certain level, the fan is controlled to operate. This method is prone to cause frequent start and stop of the fan, affecting the service life of the fan. The intelligent fan control system can collect a variety of environmental parameters, combine with the fan variable frequency speed regulation technology, and adopt an optimized control algorithm to control the operation of the fan.

[0102] Telemetry function: The traditional ventilation system can only monitor the operation status of the fan on the central control and management platform; the intelligent fan control system can perform remote monitoring in real time on the mobile APP side.

[0103] Deep intelligent monitoring function: The traditional ventilation system can only monitor the basic operation status of the fan on the central control and management platform, such as forward and reverse rotation, whether there is a fault, etc. The intelligent fan control system deploys sensors such as detection acceleration, rotation speed, vibration, voltage, current and temperature on the fan to monitor the safety of the fan components, and at the same time monitors the failure status of the motor and blades. By collecting these sensor data, it can deeply judge the current operation status of the fan, and can predict and warn of abnormal status to ensure the safe operation of the fan control system.

[0104] Preventive maintenance and equipment health monitoring function:

[0105] The maintenance work of the traditional ventilation system adopts the methods of manual inspection and passive maintenance;

[0106] The intelligent fan control system can perform preventive maintenance and health analysis, monitor the whole life cycle of the fan, and can accurately locate the maintenance and repair parts, improving the fine level of operation and management.

[0107] Furthermore, a fan fault emergency prediction control system based on a data-driven strategy is proposed. The system is implemented based on the fan fault emergency prediction control method based on a data-driven strategy described in the above embodiments, and includes:

[0108] A data acquisition module, used to collect the internal state data and external environment data of the fan in the normal state and emergency state through the historical data set. The internal state data includes current and voltage data, amplitude and rotation speed data, and the external environment data includes external component state data, temperature data and atmospheric data;

[0109] A model construction module, which is used to construct and train a deep learning model by using the internal state data and external environment data of the fan in the normal state and emergency state collected as sample data, and output a prediction result and a control decision through the deep learning model. The prediction result includes the emergency state of the fan failure and the normal state of the fan, and the control decision includes the fan adjustment parameters and the fan working mode selection of the fan state;

[0110] A control output module, which is used to input new data through the constructed deep learning model, and the control end controls and adjusts the fan according to the output prediction result and control decision.

[0111] Furthermore, the model construction module specifically includes:

[0112] A data definition unit, which is used to define the state space according to the sample data, define the action space of the fan control action regarding the sample data, and design the reward function;

[0113] A network construction unit, which is used to construct a normal policy network and an emergency policy network through a recurrent neural network;

[0114] Define the loss function of the recurrent neural network model. The loss function respectively includes a policy loss function, a value loss function, and an emergency loss function, and calculate the total loss function according to the defined policy loss function, value loss function, and emergency loss function;

[0115] A model training unit, which is used for model establishment and training, and input the sensing data collected by the fan front-end data collection unit into the trained model, and output the fan control decision and the fan working mode control decision of the fan operation.

[0116] Further, as a preferred implementation of this embodiment, an implementation of an intelligent fan control system applied to tunnel ventilation is proposed, which is as follows: Under normal conditions (i.e., conventional fan control), the intelligent fan control system is controlled by the fan backend data calculation unit. The fan backend data calculation unit is deployed in the fan control box. The fan backend data calculation unit is connected to the FPGA fan signal processing unit, the atmospheric environment sensor group, etc. through interfaces such as RS485 (multiple interface types are available), and communicates with the FPGA fan signal processing unit through the ModbusRTU bus communication protocol. Based on the real-time collection of various environmental parameters and traffic volume information, the effective control of the fan working process and operating speed is achieved through a control algorithm (PPO) to improve the ventilation efficiency of the tunnel, enhance the comfort of tunnel passage, and achieve the purpose of saving energy. At the same time, the fan backend data calculation unit also obtains the relevant operating state parameters of the intelligent fan from the FPGA fan signal processing unit, and judges and adjusts the fan control according to the operating state parameters and reports the working state information of relevant equipment to the control center. The on-site wiring is simple and convenient, and remote fault diagnosis can be realized by matching independent addresses.

[0117] In addition, the control loop of the intelligent fan control system is designed as follows:

[0118] Install the frequency converter in the power distribution cabinet, connect the cable to the front-stage circuit breaker, and lead the lower end of the circuit breaker to the input end of the frequency converter and the upper end of the industrial frequency forward and reverse contactor respectively;

[0119] Short-circuit the output side of the frequency converter with the motor industrial frequency circuit and lead it to the corresponding fan end through a cable;

[0120] The control system has automatic and manual modes, and is provided with a manual operation button.

[0121] Under normal conditions, it is in the automatic control mode. After the fan backend data calculation unit obtains data information from the FPGA fan signal processing unit and the atmospheric environment sensor, it controls the operation of the fan;

[0122] When in the fire condition, the control signal disconnects the output end of the frequency converter and switches to the emergency industrial frequency circuit to control the fan.

[0123] The above is only a preferred implementation manner of the present invention. It should be understood that the present invention is not limited to the form disclosed herein, should not be regarded as excluding other embodiments, but can be used in various other combinations, modifications and environments, and can be changed within the scope of the concept described herein through the above teachings or the technology or knowledge in the relevant field. And the changes and modifications made by those skilled in the art without departing from the spirit and scope of the present invention should all be within the protection scope of the appended claims of the present invention.

Claims

1. A fan fault emergency prediction and control method based on a data-driven strategy, characterized in that Including: S1. Collect the internal state data and external environment data of the fan under normal and emergency states through the historical data set. The internal state data includes current and voltage data, amplitude and rotation speed data, and the external environment data includes external component state data, temperature data, and atmospheric data; S2. Use the collected internal state data and external environment data of the fan under normal and emergency states as sample data to construct and train a deep learning model. Output the prediction result and control decision through the deep learning model. The prediction result includes the fan failure emergency state and the fan normal state, and the control decision includes the fan adjustment parameter and the fan working mode to select the fan state; S3. Use the constructed deep learning model, input new data, and the control end controls and adjusts the fan according to the output prediction result and control decision; Among them, the step S2 specifically includes the following sub-steps: S201. Define the state space according to the sample data, define the action space of the fan control action regarding the sample data, and design the reward function; S202. Construct a normal policy network and an emergency policy network through a recurrent neural network; S203. Define the loss function of the recurrent neural network model. The loss function respectively includes a policy loss function, a value loss function, and an emergency policy loss function, and calculate the total loss function according to the defined policy loss function, value loss function, and emergency policy loss function; S204. Establish and train the model, and input the sensing data collected by the fan front-end data acquisition unit into the trained model to output the fan control decision and the fan working mode control decision for the fan to work; In the step S201, the definition of the state space is specifically: ； Among them, the represents the state space, and the composite state vector including sensing data and emergency status identifier, where the represents the sensing data, and the represents the emergency status identifier; The definition of the action space is specifically: ； Among them, the represents the action space, and the represents the control actions of the fan, and the control actions at least include the adjustment of the fan control parameters and the switching of the fan operating modes; In the step S202, specifically constructing the emergency strategy network: according to the characteristics of the emergency status data changing over time, the time series data of the emergency signal is processed by a recurrent neural network; the input layer receives the emergency status identifier , the hidden layer processes the input data through time steps, and the output layer generates a decision signal; among them, the output layer of the emergency strategy network converts the output of the recurrent neural network through the sigmoid function to generate a value between 0 and 1, indicating the probability of taking an emergency action; In the step S203, the emergency policy loss function is specifically expressed as: ； Among them, the represents expectation, the represents the emergency strategy loss function; the represents the probability of taking an emergency action under the given emergency state identifier , and the represents the emergency state identifier; The emergency strategy network includes a fire emergency strategy network, and the output decision of the fire emergency strategy network is specifically: according to the binary decision: Decide whether to switch the working mode, that is, whether to switch to the emergency power frequency circuit; among them, the represents the fire emergency strategy network, the represents the fire status identifier, the represents the bias, the represents the probability of taking actions in the emergency action state the represents the weight, the represents the transpose, the represents the sigmoid function.

2. The emergency control method for a fan based on a data-driven strategy according to claim 1, characterized in that In the step S202, the construction of the normal policy network is specifically: According to the characteristics of the fan state data changing with time, construct a normal policy network through a recurrent neural network; among them, when the control action of the fan is continuous, use a linear activation function as the output layer, and the output vector of the output layer represents a point in the action space; when the control action of the fan is discrete, calculate the probability of each action being selected through the softmax function, specifically: ； Among them, the represents the conventional policy network, the represents the action, the represents the bias of the action , and the represents the bias of the action . The represents the weight of the action , and the represents the weight of the action . The represents the transpose, the represents the feature representation of the state , the represents the size of the action space, and the represents the probability of taking the action in the state .

3. The emergency control method for a fan based on a data-driven strategy according to claim 1, wherein, In the step S203, the policy loss function is specifically expressed as: ； Wherein: ； The represents the policy loss function, the represents the expectation, the represents the probability of taking action under the old policy in state , the represents the probability of taking action under the new policy in state , the represents the new set of parameters, the represents the ratio of the probabilities between the new policy and the old policy, the represents a preset small constant used to control the magnitude of policy update, the represents the advantage function of the old policy, which represents the difference between the cumulative return obtained by taking action in state and the average cumulative return under the old policy.

4. The emergency control method for a fan based on a data-driven strategy according to claim 1, characterized in that In the step S203, the value loss function is specifically expressed as: ； Among them, the represents a value loss function, the represents an expectation, the represents a value network, the represents a time step is the immediate reward obtained after taking an action, the represents a time step is the state of and represents a discount factor.

5. The emergency control method for a fan based on a data-driven strategy according to claim 1, characterized in that, In the step S203, the total loss function is specifically expressed as: ； Among them, the represents the total loss function, the represents the weight of represents the weight of represents the policy loss function, the represents the value loss function, the represents the emergency policy loss function.

6. A fan fault emergency prediction and control system based on a data-driven strategy, the system is implemented based on a fan fault emergency prediction and control method according to any one of claims 1-5, characterized in that, Including: A data acquisition module for collecting the internal state data and external environment data of the fan under normal and emergency states through the historical data set. The internal state data includes current and voltage data, amplitude and rotation speed data, and the external environment data includes external component state data, temperature data, and atmospheric data; The model construction module is used to construct and train a deep learning model with the internal state data and external environment data collected under the normal state and emergency state of the fan as sample data, and output a prediction result and a control decision through the deep learning model. The prediction result includes the fan fault emergency state and the fan normal state, and the control decision includes the fan adjustment parameters and the fan working mode selection for the fan state; The control output module is used to input new data through the constructed deep learning model, and the control end controls and adjusts the fan according to the output prediction result and control decision.

7. The emergency prediction and control system for fan faults based on a data-driven strategy according to claim 6, characterized in that The model construction module specifically includes: The data definition unit is used to define the state space according to the sample data, define the action space of the fan control action regarding the sample data, and design the reward function; The network construction unit is used to construct a normal policy network and an emergency policy network through a recurrent neural network; Define the loss functions of the recurrent neural network model, where the loss functions respectively include a policy loss function, a value loss function, and an emergency policy loss function, and calculate the total loss function according to the defined policy loss function, value loss function, and emergency policy loss function; The model training unit is used to establish and train the model, input the sensing data collected by the fan front-end data acquisition unit into the trained model, and output the fan control decision and the fan working mode control decision for the fan operation.

Images