Flexible fresh air control method and system for intelligent stadium

Abstract

The invention discloses a flexible fresh air control method and system for an intelligent stadium, and the system comprises a data collection module which is used for detecting the visitor flow rate, the age and the somatosensory comfort degree, and collecting the operation parameters of a fan motor and environment parameters; the big data analysis and deep learning module comprises a somatosensory comfort model, a motor operation optimization model and an age and physiological difference adjustment model, and is used for dynamically outputting control parameters of the fresh air system and the air conditioning system according to the data acquired by the data acquisition module; and the control execution module is used for adjusting a fresh air system, an air conditioning system and motor frequency conversion speed regulation and load distribution of the stadium according to the control parameters output by the big data analysis and deep learning module. The invention relates to the technical field of artificial intelligence, and can solve the problems of low intelligent degree and low control refinement degree of a fan control system in the prior art.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a flexible fresh air control method and system for an intelligent sports stadium. Background Art

[0002] As a power device that can adjust airflow, fans play a vital role in many fields. In the construction field, fans are the core of air conditioning and ventilation systems, providing a healthy and safe indoor environment for large buildings and public buildings.

[0003] By studying the control technology of different types of ventilation fans, the control accuracy of the fans can be improved to achieve a more efficient and stable ventilation effect. For example, by using variable frequency control technology, the fan speed can be adjusted according to actual needs to achieve the purpose of energy saving and consumption reduction. At the same time, the study of advanced sensor technology and automatic control algorithms can monitor and adjust the operating status of the fan in real time to ensure that the air volume always meets the needs of specific scenarios. In addition, by optimizing the control system of the fan, the reliability and stability of the fan can be improved, the maintenance cost can be reduced, and a more reliable ventilation guarantee can be provided for the development of various fields.

[0004] Chinese invention patent application CN116066938A discloses a dual fan control system applied to a mechanical ventilation system, comprising: a temperature sensor, which is arranged in a ventilation room to detect the temperature value in the ventilation room; a pressure difference sensor, which is arranged in the ventilation room where the temperature sensor is located and an adjacent room or the outside world to detect the pressure difference between the ventilation room where the temperature sensor is located and the adjacent room, or the outside world; a fan frequency conversion controller, which is used to control the ventilation volume of the supply fan and the exhaust fan based on the temperature value or the pressure difference. The supply frequency conversion fan adopts room temperature control, and can adjust the fan air volume according to the room load change from time to time to achieve energy saving; the exhaust frequency conversion fan adopts the pressure difference between the inside and outside of the room, and the exhaust fan can well control the pressure in the ventilation room while following the air volume adjustment of the supply fan to achieve energy saving. The dual fan control system only adjusts the fan air volume to achieve energy saving through the artificially set temperature and pressure difference, and the degree of intelligence and control refinement is low.

[0005] Chinese utility model patent CN219222770U discloses an integrated ventilation control system based on hospital carbon dioxide concentration monitoring, including: CO2 concentration control host, CO2 concentration controller, fan frequency conversion controller, evenly distributed CO2 concentration monitor, on-site manual start button, ventilation fan, time controller, centralized control room start button, mobile phone controller, sound and light alarm. The ventilation control system needs to be manually started according to the monitoring situation or the fan operation status is controlled by the time controller setting, and the degree of intelligence and control refinement is low.

[0006] Therefore, it is necessary to provide a flexible fresh air control method and system for smart sports stadiums, which can solve the problems of low intelligence and control refinement of fan control systems in the prior art. Summary of the invention

[0007] The object of the present invention is to provide a flexible fresh air control method and system for an intelligent sports stadium, which can solve the problems of low intelligence and control refinement of fan control systems in the prior art.

[0008] The present invention is achieved in that:

[0009] A smart sports stadium flexible fresh air control system, comprising:

[0010] Data acquisition module, used to detect human flow, age and body comfort, and collect fan motor operating parameters and environmental parameters;

[0011] Big data analysis and deep learning module, including somatosensory comfort model, motor operation optimization model, age and physiological difference adjustment model, used to dynamically output the control parameters of the fresh air system and air conditioning system according to the data collected by the data acquisition module;

[0012] Generate an adjustment strategy and pass it to the control execution module, which is used to adjust the stadium's fresh air system, air-conditioning system, and its motor variable frequency speed regulation and load distribution according to the control parameters output by the big data analysis and deep learning module.

[0013] The data acquisition module includes: infrared sensors installed at the entrance and exit of the stadium for crowd detection, several AI cameras and several WiFi / Bluetooth detectors installed in the stadium for crowd detection, a receiving module installed in the stadium for collecting data from smart wearable devices, several environmental sensors installed in the stadium for environmental information detection, and voltage sensors, current sensors and wind speed sensors for collecting fan motor operating parameters.

[0014] A control method for a flexible fresh air control system of an intelligent sports stadium comprises the following steps:

[0015] Step 1: The data acquisition module detects the flow of people in the stadium through infrared sensors and AI cameras;

[0016] Step 2: The data acquisition module detects the signals emitted by smart devices in the stadium through WiFi / Bluetooth detectors, and estimates the crowd density and flow direction in the area where the WiFi / Bluetooth detectors are located;

[0017] Step 3: The data collection module uses facial recognition technology to identify the age group of users in the stadium based on the images collected by the AI ​​camera;

[0018] Step 4: The data acquisition module collects environmental information of the area where the environmental sensor is located through the environmental sensor;

[0019] Step 5: The big data analysis and deep learning module determines the user's comfort level based on the crowd density, crowd flow direction, user age group, and environmental information in each area, and dynamically adjusts the environmental conditions;

[0020] Step 6: Generate an adjustment strategy and pass it to the control execution module to adjust the air flow rate of the fresh air system and the air temperature of the air conditioning system in the stadium according to the environmental conditions of the big data analysis and deep learning module.

[0021] In the step 1, the flow of people is detected first, and the flow of people in the stadium is monitored by infrared sensors, AI cameras combined with AI analysis and WiFi / Bluetooth detectors; infrared sensors are installed at the entrance and exit of the stadium, and the flow of people is monitored in real time by detecting the infrared radiation released by the human body, providing an accurate number of people entering and leaving, thereby calculating the number of people in the stadium;

[0022] The AI ​​camera combines AI technology to detect and count dynamic crowds by capturing images inside the stadium and using YOLO and convolutional neural network algorithms. At the same time, the interior of the stadium is divided into zones and AI cameras are installed, with each AI camera capturing images within the corresponding area.

[0023] In step 3, the image captured by the AI ​​camera is combined with facial recognition technology to capture the person in the image, and the age group of the user is identified by the AgeNet algorithm after the facial image is extracted, and the age group is divided into children, youth, middle-aged, and elderly;

[0024] For situations where users of different age groups, or children and the elderly coexist in the same area, the system makes comprehensive adjustments based on ambient temperature, air flow rate and the user's real-time physiological state, so that the comfort of users of different age groups can be taken into account.

[0025] In step 4, the environmental information includes ambient temperature, ambient humidity, and air quality data; wherein the control quality data includes PM2.5 concentration and CO2 concentration.

[0026] In step 5, the somatosensory comfort model of the big data analysis and deep learning module is based on the temperature, humidity, and air quality data obtained from the environmental sensor and the user's physiological state data obtained from the data acquisition module, and is comprehensively analyzed through a convolutional neural network and a long short-term memory network;

[0027] The convolutional neural network of the somatosensory comfort model is used to process environmental data from different areas, including the distribution of temperature, humidity, and air quality, and extract local spatial features from them to identify the comfort requirements of different areas. The long short-term memory network of the somatosensory comfort model is used to process the user's time series physiological data, including physiological signals of changes in heart rate and sweating, and capture the time dependency of these physiological signals.

[0028] The somatosensory comfort model is based on the output of convolutional neural networks and long short-term memory networks, and is fine-tuned according to historical data and real-time data to automatically adjust the temperature and humidity in sports stadiums. The somatosensory comfort model adjusts model parameters through reinforcement learning based on user complaints, changes in user expressions captured by AI cameras, and feedback on body temperature changes obtained by WiFi / Bluetooth detectors to improve the accuracy of comfort prediction.

[0029] In step 5, if there are a large number of children and a small number of elderly people in one area, the system adopts a multi-regional control strategy according to the temperature requirements of different users, and physically divides the area into multiple small areas; for the small area where children are more active, the temperature is lowered and the ventilation is enhanced; and for the small area where there are relatively more elderly people, the temperature is increased and the air flow rate is slowed down;

[0030] The age and physiological difference adjustment model of the big data analysis and deep learning module uses multimodal deep learning and generative adversarial networks to process the somatosensory differences of users of different age groups. The age and physiological difference adjustment model sets weights according to the age distribution of the population, with a weighting coefficient of 0.4 for children and 0.6 for the elderly. Multimodal learning automatically adjusts and outputs control parameters by integrating the user's age, physiological state, and environmental data.

[0031] In step 5, the motor operation optimization model of the big data analysis and deep learning module processes the time series data of the fan motor operation parameters through the time series prediction network, and predicts the future load demand based on the historical data;

[0032] The motor operation optimization model of the big data analysis and deep learning module is based on the data collection module's data collection of human flow. The camera detects the changes in human flow in each area through video monitoring and the YOLO algorithm. The infrared sensor detects the number of people entering and leaving the stadium, and the WiFi / Bluetooth detector analyzes the signal strength and density changes of smart devices in the stadium to estimate the future changes in human flow and density, and organizes them into multi-dimensional time series according to timestamps.

[0033] The time series prediction network of the motor operation optimization model is trained using historical passenger flow data. The time series prediction network of the motor operation optimization model predicts the passenger flow in a certain period of time in the future by analyzing the trend of passenger flow changes in the past. The input data of the time series prediction network of the motor operation optimization model: the input includes historical passenger flow data of multiple time steps, holiday information, large-scale event announcements, and additional features of weather changes. The output result of the time series prediction network of the motor operation optimization model: the predicted value of passenger flow in each time period in the future.

[0034] When it is predicted that the flow of people will increase rapidly, the fan speed will be gradually increased to ensure effective air circulation, and the air conditioning system will be controlled to gradually lower the temperature.

[0035] In the step 6, in terms of the regulation of the fresh air system, an adjustment strategy is generated and transmitted to the control execution module to adjust the fan speed and air volume of the fresh air system in real time according to the flow of people, air quality and physiological status data; when the flow of people is large, an adjustment strategy is generated and transmitted to the control execution module to automatically increase the wind speed to enhance air circulation; in terms of the regulation of the air-conditioning system, an adjustment strategy is generated and transmitted to the control execution module to automatically adjust the set values ​​of temperature and humidity based on the somatosensory comfort model and age requirements; an adjustment strategy is generated and transmitted to the control execution module to dynamically adjust the air-conditioning output of each area according to the activity volume and crowd distribution in different areas of the stadium.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] The present invention is provided with a data acquisition module, a big data analysis and deep learning module, generates an adjustment strategy and transmits it to the control execution module. The data acquisition module collects data such as pedestrian flow, age recognition, physical comfort, fan operation status, temperature and humidity, and control quality, and intelligently distributes the load of the fresh air system and the air-conditioning system according to the needs of the stadium through the physical comfort model, the age and physiological difference adjustment model, and the fan operation optimization model to avoid overloading or inefficient operation of a single device, so as to achieve more refined and intelligent environmental adjustment. It can be applicable to various public places such as comprehensive stadiums, and can provide reliable protection for the safe and efficient operation of each place. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a structural schematic diagram of the flexible fresh air control system of the intelligent sports stadium of the present invention.

[0039] In the figure, 1 is the data acquisition module, 2 is the big data analysis and deep learning module, and 3 generates the adjustment strategy and transmits it to the control execution module. DETAILED DESCRIPTION

[0040] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0041] Please see attached Figure 1 , a smart sports stadium flexible fresh air control system, comprising:

[0042] Data acquisition module 1, used to detect the flow of people, detect age and physical comfort, and collect fan motor operating parameters and environmental parameters;

[0043] Big data analysis and deep learning module 2, including a somatosensory comfort model, a motor operation optimization model, and an age and physiological difference adjustment model, is used to dynamically output control parameters of the fresh air system and the air conditioning system based on the data collected by the data collection module 1;

[0044] Generate an adjustment strategy and pass it to the control execution module 3, which is used to adjust the fresh air system, air-conditioning system and its motor variable frequency speed regulation and load distribution of the stadium according to the control parameters output by the big data analysis and deep learning module 2.

[0045] The data acquisition module 1 includes: infrared sensors installed at the entrance and exit of the stadium and used for crowd detection, several AI cameras and several WiFi / Bluetooth detectors installed in the stadium and used for crowd detection, a receiving module installed in the stadium and used to collect data from smart wearable devices, several environmental sensors installed in the stadium and used for environmental information detection, and voltage sensors, current sensors and wind speed sensors for collecting fan motor operating parameters.

[0046] The data acquisition module 1 uses various sensors and other equipment to collect parameters and feed them back to the big data analysis and deep learning module 2. The deep learning model analyzes and judges environmental changes, and dynamically adjusts the temperature of the air conditioning system and the speed of the fan of the fresh air system. For example, when the PM2.5 concentration increases, the wind speed and air volume are increased; when the air temperature changes slightly, the wind speed is reduced to reduce energy consumption.

[0047] The sensors and other equipment in the data acquisition module 1 work together to achieve data fusion, which is beneficial to improving the prediction accuracy of the big data analysis and deep learning module 2.

[0048] Please see attached Figure 1 , a flexible fresh air control method for a smart sports stadium, comprising the following steps:

[0049] Step 1: Data acquisition module 1 detects the flow of people in the stadium through infrared sensors and AI cameras.

[0050] Data collection is the basic link of the intelligent sports stadium flexible fresh air control system (hereinafter referred to as the system), which involves real-time monitoring of multi-dimensional data in the sports stadium. These multi-dimensional data provide a basis for subsequent analysis and optimization, and are collected through a variety of sensors and technical means, covering multiple aspects.

[0051] Specifically, the first is crowd detection, which monitors the flow of people in sports stadiums through infrared sensors, AI cameras combined with AI analysis, and WiFi / Bluetooth detectors. Infrared sensors are installed at the entrances and exits of sports stadiums. They can monitor the flow of people in real time by detecting infrared radiation released by the human body, provide accurate numbers of people entering and leaving, and thus calculate the number of people in the sports stadium.

[0052] AI cameras combine AI technology to detect and count dynamic crowds by capturing images inside stadiums and using algorithms such as YOLO and convolutional neural networks, which is particularly suitable for crowded areas. Preferably, the interior of the stadium can be partitioned and AI cameras can be installed, with each AI camera corresponding to the image captured in the corresponding area.

[0053] Step 2: The data acquisition module 1 detects the signals emitted by the smart devices in the stadium through the WiFi / Bluetooth detector, and estimates the crowd density and crowd flow direction in the area where the WiFi / Bluetooth detector is located.

[0054] The WiFi / Bluetooth detector can estimate the crowd density and flow direction of the current area by detecting the signals sent by the user's smart device (such as a mobile phone, smart watch). Preferably, the interior of the stadium can be divided into zones and WiFi / Bluetooth detectors can be installed, and each WiFi / Bluetooth detector corresponds to detecting the smart devices of users in the corresponding zone.

[0055] Relying on the stadium's dedicated App or authorized smart device platform, when entering or signing in through the stadium App, users need to actively authorize the data collection module 1 to access the data of their smart wearable devices. These data are transmitted through the Bluetooth or WiFi connection between the WiFi / Bluetooth detector and the user's smart wearable device to ensure the legality and compliance of data collection. Stadiums need to clearly inform users of the scope of data use and privacy policy to ensure transparency and users' right to know during the data collection process.

[0056] Step 3: The data collection module 1 identifies the age groups of users in the stadium based on the images collected by the AI ​​camera based on facial recognition technology.

[0057] The images collected by the AI ​​camera can be combined with the existing facial recognition technology to capture the characters in the image. After extracting the facial image, the user's age group can be identified through algorithms such as AgeNet. Preferably, the age groups can be divided into children (0-12 years old), youth (13-45 years old), middle-aged (46-60 years old), and elderly (over 60 years old). The age groups can also be segmented according to actual control needs.

[0058] Since users of different age groups have different demands for temperature, humidity and other environmental conditions, for example, children require a temperature comfort level of 21-24°C, while the elderly require a temperature comfort level of 24-26°C. The age group of users in a certain area is used as one of the control conditions for the fresh air system and air conditioning system.

[0059] Taking into account people of different ages: For users of different ages in the same area, such as children and the elderly, the system can make comprehensive adjustments based on the ambient temperature, air flow rate and the user's real-time physiological state (such as heart rate, body temperature). For example, when a large number of children and a small number of elderly people are detected, the system can set the temperature to a relatively middle range (such as 23-25℃) and adjust the temperature by controlling the air flow rate, so that the comfort of users of different ages can be taken into account.

[0060] Some smart wearable devices in the prior art (such as smart bracelets and smart watches) are equipped with skin conductance sensors (GSR), which indirectly reflect sweating conditions by detecting changes in skin conductance through the skin conductance sensors of the user's smart wearable devices. When the user sweats, the skin conductivity increases. By collecting the skin conductivity data of the user's smart wearable device, it is determined whether the user is sweating and the degree of sweating, thereby obtaining the user's physiological state.

[0061] If the user's smart wearable device is not equipped with a skin conductance sensor and cannot directly obtain the sweating status, the heart rate change data of the smart wearable device can be collected and combined with the infrared sensor to detect the user's skin temperature to infer the user's sweating status and thus obtain the user's physiological state.

[0062] Step 4: The data acquisition module 1 collects environmental information of the area where the environmental sensor is located through the environmental sensor.

[0063] The environmental information includes ambient temperature, ambient humidity, air quality (including but not limited to PM2.5 concentration, CO2 concentration, etc.) data, etc.

[0064] Step 5: The big data analysis and deep learning module 2 determines the user's comfort level based on the crowd density, crowd flow direction, user age group, and environmental information in each area, and dynamically adjusts the environmental conditions.

[0065] After data collection is completed, the big data analysis and deep learning module 2 establishes a somatic comfort model, a motor operation optimization model, and an age and physiological difference adjustment model, aiming to support intelligent control and real-time adjustment of the fresh air system and air conditioning system to ensure the comfort of the venue environment and the efficiency of equipment operation. Specific to the different network models in deep learning, the following is a detailed description of each model:

[0066] The somatosensory comfort model is based on the temperature, humidity, air quality data (such as CO2 concentration, PM2.5, etc.) obtained from environmental sensors and the user's physiological state data (such as heart rate, sweating, etc.) obtained from the data acquisition module 1, and is comprehensively analyzed through a convolutional neural network (CNN) and a long short-term memory network (LSTM).

[0067] Among them, the CNN of the somatosensory comfort model is used to process environmental data from different areas, such as the distribution of temperature, humidity, and air quality, from which local spatial features can be extracted to help identify the comfort requirements of different areas.

[0068] The LSTM of the somatosensory comfort model is used to process the user's time series physiological data, such as physiological signals of changes in heart rate and sweating, and to capture the time dependency of these physiological signals. The somatosensory comfort model is based on the output of CNN and LSTM, and is fine-tuned according to historical data and real-time data to automatically adjust the temperature and humidity in the stadium. For example, for the elderly, since they are more sensitive to cold, the system will increase the ambient temperature and reduce air flow to maintain a warm and comfortable environment. In the case of a large number of children, the system will increase air circulation and reduce the temperature to cope with the physiological characteristics of children who sweat easily.

[0069] If there are a large number of children and a small number of elderly people in one area, the system can adopt a multi-regional control strategy based on the temperature requirements of different users, and physically divide the area into multiple small areas. For small areas with more children, the temperature can be lowered and ventilation can be increased; while for small areas with relatively more elderly people, the temperature can be appropriately increased and the air flow rate can be slowed down.

[0070] If physical segmentation is difficult, the system will calculate a weighted average temperature setting value based on the real-time population ratio. For example, it will set it to the middle value between the temperature requirements of children and the elderly, and increase the perceived temperature of the elderly by adjusting the local air flow rate.

[0071] The age and physiological difference adjustment model sets weights according to the age distribution of the population. For example, the weight coefficient for children is 0.4 and that for the elderly is 0.6. The reason is that the elderly have weaker body metabolism and cold tolerance, are usually more sensitive to low temperatures, have higher demands for temperature comfort, and need a relatively warm environment. Therefore, in the weighting of adjusting the temperature, the weight coefficient of the elderly will be larger to avoid them feeling uncomfortable due to the cold. Children have active body metabolism, high activity, easy sweating, and high sensitivity to high temperatures. The temperature control needs for children focus more on air circulation and relatively low temperatures. However, considering that children usually have stronger body adjustment capabilities, the weight coefficient for children is slightly smaller. Based on real-time feedback, the age and physiological difference adjustment model optimizes the user's comfort by adaptively adjusting the weights.

[0072] The age and physiological difference adjustment model uses multimodal deep learning and generative adversarial networks (GANs) to process the physical differences of users of different age groups, further enhancing the system's personalized adjustment capabilities. Multimodal learning automatically adjusts and outputs control parameters for controlling the environment in sports stadiums by integrating different data sources such as the user's age, physiological state (such as heart rate, body temperature, sweating), and environmental data (such as ambient temperature, ambient humidity, and air quality).

[0073] For example, the elderly are more sensitive to temperature. The age and physiological difference regulation model combines historical data analysis with real-time feedback through multimodal learning, and automatically sets the temperature in the venue to a slightly higher range to ensure that the elderly feel comfortable. For children or young people, the age and physiological difference regulation model will meet their needs by increasing ventilation or lowering the temperature. Generative adversarial networks (GANs) are used to simulate changes in user physiological data under different environmental conditions. Especially when data is scarce, GANs can generate simulated data to help the system optimize regulation strategies in different scenarios. In addition, the age and physiological difference regulation model can further refine the adjustment of environmental parameters through real-time physiological data from smart wearable devices to ensure that each user can get a personalized comfort experience.

[0074] At the same time, the somatosensory comfort model adjusts model parameters through reinforcement learning based on feedback information such as user complaints, changes in user expressions captured by AI cameras, and changes in body temperature obtained by WiFi / Bluetooth detectors, so as to gradually improve the accuracy of comfort prediction.

[0075] The motor operation optimization model uses time series prediction network (LSTM) and reinforcement learning (RL) technology, focusing on the efficient operation of the fan.

[0076] The LSTM of the motor operation optimization model is good at processing time series data of fan motor operating parameters (such as current, voltage, speed, etc.), and can predict future load demand based on historical data. For example, before a large-scale event, when the flow of people gradually increases, the LSTM of the motor operation optimization model can predict the future load increase trend, increase the power output of the fan in advance, and prevent energy waste caused by sudden overload operation of the equipment.

[0077] The motor operation optimization model is based on the data collection of human flow by the data acquisition module 1. The camera detects the changes in human flow in each area through video monitoring and the YOLO algorithm. The infrared sensor detects the number of people entering and leaving the stadium, and the WiFi / Bluetooth detector estimates the future human flow and density changes by analyzing the signal strength and density changes of smart devices in the stadium.

[0078] The motor operation optimization model organizes the traffic data collected by each sensor (such as the number of people entering and leaving the venue, the density of people in the area, etc.) into a multidimensional time series according to the timestamp. For example, the traffic data is summarized once every minute to form an hourly and daily change series. For more accurate analysis, multiple time series can be established based on the traffic data of different areas.

[0079] The LSTM of the motor operation optimization model is trained using historical passenger flow data. The LSTM of the motor operation optimization model has a strong ability to capture long-term dependencies and can predict the passenger flow in a certain period of time in the future (such as the next 15 minutes or 30 minutes) by analyzing the past passenger flow change trend.

[0080] The input data of the LSTM of the motor operation optimization model includes historical passenger flow data of multiple time steps, holiday information, large-scale event announcements, weather changes and other additional features. These features will affect the changing trend of passenger flow.

[0081] The output of the LSTM of the motor operation optimization model: the predicted value of the flow of people in each time period in the future (such as every minute or every 5 minutes). These predicted values ​​of the flow of people reflect the increase or decrease trend of the flow of people in the future.

[0082] Based on the LSTM prediction of future passenger flow, the LSTM of the motor operation optimization model will correlate the predicted passenger flow with the load of the air conditioning system. Through training with historical data, the motor operation optimization model will learn how the load of the fans of the air conditioning system and the fresh air system responds to these changes (such as temperature rise, air flow increase, etc.) when the passenger flow increases or decreases.

[0083] The relationship between load change and traffic is usually nonlinear. For example, an increase in traffic may gradually increase the temperature in a stadium and increase the carbon dioxide concentration and humidity in the air. This relationship can be learned by LSTM or other regression models.

[0084] When the LSTM of the motor operation optimization model predicts that the flow of people will increase rapidly in the future, the motor operation optimization model will adjust the speed and power output of the fan in advance to reduce the rapid increase in cooling load. The specific adjustment strategy is as follows:

[0085] Increase wind speed and air volume: When it is predicted that the flow of people will increase rapidly, the system will gradually increase the speed of the fan to ensure that the air can circulate effectively and prevent the stuffy feeling caused by the crowded people.

[0086] Lower the temperature in advance: To mitigate the temperature rise caused by a large influx of people, the motor operation optimization model will gradually lower the temperature by controlling the air conditioning system before predicting an increase in the flow of people, thereby avoiding a rapid rise in temperature caused by the increase in the flow of people.

[0087] Dynamic adjustment strategy: The motor operation optimization model uses the LSTM prediction results as input to evaluate the change trend of passenger flow every minute or every 5 minutes in real time. Based on the real-time evaluation results, the motor operation optimization model can dynamically adjust the fan speed and power output to avoid the pressure on the equipment caused by sudden load changes.

[0088] For example, when the LSTM of the motor operation optimization model predicts that the flow of people will gradually increase to a peak in the next 15 minutes, the system can follow the following steps:

[0089] 1. In the initial stage when the flow of people begins to increase (the first 10-15 minutes), the fan speed and air conditioning system will begin to increase slowly.

[0090] 2. When the flow of people reaches its peak, the fan and air-conditioning system reach maximum power output to ensure ventilation effect.

[0091] 3. When the flow of people gradually decreases, the system will gradually reduce the power output with the same strategy to maintain stable air circulation.

[0092] In addition, the motor operation optimization model uses the fan motor operating parameters collected by the data acquisition module 1, combined with LSTM and RL to monitor the fan health status in real time, predict potential failures, and perform preventive maintenance, thereby extending the service life of the equipment and improving its overall operating efficiency.

[0093] Potential failure prediction: The LSTM of the motor operation optimization model learns the health status and failure trend of the equipment by analyzing the time series data of the motor operation parameters (such as current, voltage, temperature, etc.). Once an abnormal fluctuation or abnormal change pattern is detected, the LSTM will output an alarm signal.

[0094] Long Short-Term Memory (LSTM) is a deep learning model that processes time series data and can capture long-term dependencies and dynamic changes in data. By analyzing historical data, LSTM can learn the behavior patterns and characteristics of motors during normal operation.

[0095] Specifically, the data input of the LSTM of the motor operation optimization model is: the collected fan motor operating parameters, such as current, voltage, vibration signal, temperature, etc. These fan motor operating parameters are input into the LSTM model in time series for training and prediction.

[0096] Normal state training: In the early stage of the system, the LSTM of the motor operation optimization model is trained using the motor's historical operating data (current, voltage, temperature, etc. under normal operating conditions). The LSTM of the motor operation optimization model will learn the normal behavior patterns and characteristics of the motor under different load conditions.

[0097] Abnormal state identification: After training is completed, the LSTM of the motor operation optimization model will continue to predict the real-time collected data. If the real-time data deviates greatly from the normal behavior pattern of the motor operation optimization model, the LSTM of the motor operation optimization model will mark it as abnormal.

[0098] Abnormal fluctuation detection: During the prediction process, the LSTM of the motor operation optimization model predicts the operation data at each moment and compares the predicted value with the actual measured value. If the deviation at a certain moment exceeds the set threshold (determined according to the normal fluctuation range in historical data), the LSTM of the motor operation optimization model will trigger an alarm signal. For example, if the actual current value suddenly increases or fluctuates abnormally during a certain period of current, the LSTM of the motor operation optimization model will detect a large deviation and consider it a potential precursor to failure.

[0099] Output of alarm signal: Once the LSTM of the motor operation optimization model detects abnormal fluctuations or unusual change patterns (such as excessive current, voltage fluctuations beyond the normal range, rapid temperature increase, etc.), the system generates an alarm signal to alert operators of potential failure risks.

[0100] Preventive maintenance: Based on the prediction results and anomaly detection of the LSTM of the motor operation optimization model, the system will remind the operator to arrange maintenance in advance, or adjust the fan load or stop some equipment for maintenance at an appropriate time period based on the suggestions of reinforcement learning to avoid sudden failures.

[0101] Preventive maintenance includes scheduling maintenance plans in advance, using reinforcement learning to optimize maintenance decisions, and proactively adjusting wind turbine load and health management.

[0102] Among them, arranging maintenance plans in advance includes maintenance reminder mechanism and maintenance plan optimization.

[0103] Maintenance reminder mechanism: When the LSTM of the motor operation optimization model predicts that the motor's health status is abnormal and lasts for a period of time, the system will record the time and characteristics of the fault trend. Based on the predicted alarm information, the system can automatically generate maintenance tasks and remind operators in advance.

[0104] Maintenance reminders can be set up based on the severity of the predicted failure. For example, a minor deviation may only require monitoring, while a significant deviation will immediately trigger a maintenance plan.

[0105] Maintenance plan optimization: The LSTM of the motor operation optimization model can predict the deterioration trend of the equipment based on historical data and existing maintenance records. The motor health status can be quantitatively analyzed using regression models or Bayesian optimization methods to determine the specific time point when the motor and other equipment need to be repaired, so that the maintenance plan can be included in the schedule.

[0106] Using reinforcement learning to optimize maintenance decisions includes modeling of states and actions, optimizing maintenance strategies, and dynamically adjusting maintenance strategies.

[0107] Modeling of states and actions: In the reinforcement learning model, the operating state of the motor (such as current, voltage, temperature, vibration, etc.) is used as the "state" of the system, and the power output, speed adjustment, and air volume adjustment of the fan are used as the "action". The goal of the system is to select the appropriate action based on the current state to optimize the health of the equipment.

[0108] Maintenance strategy optimization: Reinforcement learning models (such as DQN or policy gradient algorithms) learn how to select appropriate maintenance actions in different states based on the health status of the equipment. For example, when the LSTM of the motor operation optimization model predicts that the health status of equipment such as fans has deteriorated, the system may choose to reduce the load of the fan during off-peak hours or arrange maintenance on weekends to reduce the impact on users.

[0109] Dynamically adjust maintenance strategies: Through reinforcement learning, the system will continuously optimize maintenance decisions. After each device operation, the system will update the reinforcement learning model based on the feedback after maintenance (such as whether the health status has improved, whether the energy efficiency has increased, etc.), so that the maintenance strategy can be gradually optimized to achieve efficient equipment management.

[0110] Active adjustment of fan load and health management includes: dynamic adjustment of fan load and predictive health management.

[0111] Dynamic adjustment of fan load: If the LSTM of the motor operation optimization model detects abnormal motor operation, the system can automatically reduce the speed or power output of the abnormal fan based on the fault prediction results to prevent further damage to the equipment.

[0112] During peak hours, the system can adjust the load between different fans through load distribution strategies to avoid a single fan being in a high-load state for a long time.

[0113] Predictive health management: The system continuously analyzes the motor degradation based on the trend data predicted by the LSTM of the motor operation optimization model. For equipment with long-term high load or frequent fluctuations, preventive inspections and maintenance can be automatically arranged to avoid sudden failures.

[0114] Reinforcement learning optimizes the power output control strategy of the fan motor through a policy gradient algorithm or a deep Q network (DQN). RL can adjust the motor speed and power output based on real-time feedback to achieve energy efficiency optimization under different load conditions. When the flow of people increases or an activity starts, the system automatically increases the power output of the motor. Conversely, when the flow of people decreases or the activity ends, the system automatically reduces the power output of the motor to save energy.

[0115] By defining a reward function, the fan is rewarded for minimizing power output and energy consumption while meeting user comfort requirements. The policy gradient algorithm optimizes the fan control strategy and finds the optimal action (such as adjusting wind speed, turning on or off some fans) by calculating the policy gradient for the current environment (people flow, air quality, temperature, etc.).

[0116] The policy gradient algorithm directly learns a policy function, which is the probability distribution of selecting an action in each state. The goal of the policy gradient method is to find an optimal policy that maximizes the cumulative reward obtained during the entire system operation. The core idea is to use the gradient ascent method to optimize the policy function so that the policy can take the optimal action according to different states.

[0117] In the control of wind turbines, the reward function is very important. In order to optimize the power output of the wind turbine and the user comfort, the reward function can be defined as follows:

[0118] R(s,a)=α×Comfort-β×Power

[0119] Among them, s represents the environmental status (such as human flow, air quality, temperature, etc.), a represents the current action (such as adjusting the wind speed or turning on / off some fans), Comfort represents the user's comfort score (calculated based on parameters such as temperature, humidity, and air quality), Power represents the power consumption of the fan, and α and β are two adjustment weights used to balance the relationship between user comfort and energy consumption.

[0120] The policy function can be represented as a neural network, which calculates the probability of each action aaa by inputting the current state sss. For example, the input environment state (such as temperature, humidity and flow of people) outputs the probability of different actions (such as increasing wind speed, reducing wind speed, and keeping it unchanged).

[0121] Use the policy gradient algorithm (such as the REINFORCE algorithm) to calculate the gradient of the policy. The formula is as follows:

[0122]

[0123] Among them, πθ(a|s) represents the probability of selecting action a in state s, which is controlled by parameter θ.

[0124] The policy gradient algorithm adjusts the parameter θ to maximize the cumulative expectation under the reward function R(s,a).

[0125] In each round of training, the gradient is calculated and the parameters of the policy network are updated based on the sampled state-action pairs. In this way, the policy will gradually approach the optimal state, minimizing energy consumption without reducing user comfort.

[0126] Therefore, under the control of the policy gradient algorithm, when an increase in the flow of people and a high temperature are detected, the policy function may output a higher action probability, instructing the fan to increase the wind speed to enhance air circulation. At the same time, as the reinforcement learning process progresses, the policy gradient algorithm gradually finds an optimal wind speed setting by adjusting the fan output strategy, so as to achieve a balance between user comfort and energy consumption.

[0127] The deep Q network DQN uses the Q value table to learn the optimal power output under different states. Through the feedback of each step (such as energy consumption, user satisfaction), the Q network continuously adjusts and updates the output strategy of the wind turbine, gradually approaching the optimal control solution.

[0128] DQN is based on the Q-learning algorithm and estimates the Q value through a deep neural network. The Q value represents the expected cumulative reward obtained by taking a certain action in a certain state. The goal of DQN is to find a strategy that selects the action with the largest Q value in a given state. The advantage of DQN is that it can handle continuous state and action space, and improve the stability and convergence speed of learning through experience replay and target network.

[0129] The specific steps are as follows:

[0130] 1. Define states, actions, and rewards:

[0131] Among them, state (s): includes environmental data (such as air quality, temperature and humidity, and human flow) and fan data (such as current wind speed, power, etc.). Action (a): represents the control strategy that the fan can take in the current state, such as increasing wind speed, decreasing wind speed, and keeping it unchanged. Reward (R): Similar to the policy gradient algorithm, the reward function can be defined based on user comfort (such as temperature and humidity changes) and power consumption.

[0132] 2. Build a Q network: Use a deep neural network (DNN) to represent the Q function Q(s,a). Input the current state s and output the Q value of each possible action.

[0133] 3. Experience replay and target network:

[0134] Experience replay: DQN stores the state-action-reward-next state (s, a, R, s′) of each training into an experience pool. When updating the Q network at each step, a batch of data is randomly selected from the experience pool for training to break the correlation between the data.

[0135] Target Network: DQN uses two Q networks, one is the main network (Main Network) and the other is the target network (Target Network). The main network is used to select actions, and the target network is used to estimate the Q value of the next step. The parameters of the target network are updated once every period of time to improve the stability of training.

[0136] 4. Q value update: The formula for Q value update is as follows:

[0137]

[0138] Among them: α is the learning rate, which controls the step size of the update; γ is the discount factor, which indicates the importance of future rewards; R is the current reward.

[0139] 5. Select the optimal action: In each state, the Q value of all actions is calculated through the main network, and the action with the largest Q value is selected as the optimal action in the current state.

[0140] In the early stage of learning, the ∈ greedy strategy is used, that is, actions are randomly selected with a certain probability to prevent falling into the local optimum.

[0141] Therefore, when the stadium is in a state of high traffic and reduced air quality, DQN may choose to increase wind speed as the current optimal action.

[0142] As the system runs, DQN will continuously adjust the Q value to achieve the optimal balance between user comfort and power output. As the data in the experience pool becomes richer, the strategy of the Q network will gradually converge to the optimal solution.

[0143] Step 6: Generate an adjustment strategy and pass it to the control execution module 3 to adjust the air flow rate of the fresh air system and the air temperature of the air conditioning system in the stadium according to the environmental conditions of the big data analysis and deep learning module 2.

[0144] Generate an adjustment strategy and pass it to the control execution module 3. According to the data output by the big data analysis and deep learning module 2, dynamically adjust the operating status of the fresh air system and air-conditioning system in the stadium to ensure comfort and energy saving effects.

[0145] In terms of the regulation of the fresh air system, the regulation strategy is generated and transmitted to the control execution module 3 to adjust the fan speed and air volume of the fresh air system in real time according to the flow of people, air quality and physiological status data. When the flow of people is large, the regulation strategy is generated and transmitted to the control execution module 3 to automatically increase the wind speed to enhance air circulation; when the proportion of elderly people in the stadium is high, the regulation strategy is generated and transmitted to the control execution module 3 to reduce the ventilation intensity to avoid the cold wind directly affecting the comfort of the elderly. Through precise damper control and variable frequency fan regulation, the system can supply fresh air on demand, thereby reducing energy waste.

[0146] In terms of the regulation of the air conditioning system, the regulation strategy is generated and transmitted to the control execution module 3 to automatically adjust the set values ​​of temperature and humidity based on the somatosensory comfort model and age requirements. If the proportion of children in the stadium is high, the regulation strategy generated and transmitted to the control execution module 3 will automatically lower the temperature of the air conditioning system and increase the air circulation of the fan; when the proportion of the elderly is high, the regulation strategy generated and transmitted to the control execution module 3 will moderately increase the temperature to ensure their comfort. The regulation strategy generated and transmitted to the control execution module 3 also has a regional control function, which can dynamically adjust the air conditioning output of each area according to the activity volume and crowd distribution in different areas of the stadium to achieve precise regulation.

[0147] In addition, the frequency conversion speed regulation and load distribution of the fresh air system's motor are achieved through the inverter. The system can flexibly adjust the motor's speed and power output according to the real-time needs of the stadium. When the flow of people decreases, the system will reduce the motor speed to save energy; when the flow of people increases or the environmental requirements are high, the system will increase the motor power output in a timely manner. The load of multiple fans is reasonably distributed according to real-time needs to avoid a single device being in a high-load state for a long time, thereby extending the service life of the equipment and improving overall energy efficiency.

[0148] Take a large concert in a stadium as an example. Before the concert starts, it is expected that the flow of people will increase significantly after 15 minutes. The following are the specific processing steps of the system:

[0149] 1. Data collection and prediction: The AI ​​camera and infrared sensor detect the audience gradually entering the stadium entrance, and the LSTM of the motor operation optimization model predicts that the flow of people will reach a peak in the next 15 minutes (for example, 500 people enter per minute).

[0150] 2. Early response strategy:

[0151] The system calculates the rising trend of cooling load based on the predicted flow of people (e.g. if the temperature rises by 2°C, the CO2 concentration will increase by 0.2%).

[0152] Generate an adjustment strategy and transmit it to the control execution module 3 to gradually increase the fan speed from the existing 50% to 80% in advance, and increase the air conditioning power to reduce the temperature by 1°C to prevent the temperature from rising suddenly during the peak period.

[0153] Dynamic adjustment: The system further fine-tunes the fan speed and air conditioning output power every 5 minutes based on real-time feedback data (temperature, humidity, CO2 concentration, etc.) and the correction results of LSTM prediction.

[0154] The control method and method of the present invention can realize intelligent optimization of environmental regulation in sports stadiums, which not only improves user comfort, but also significantly improves the energy efficiency of the system, and has broad application prospects and economic benefits.

[0155] The above are only preferred embodiments of the present invention and are not intended to limit the scope of protection of the invention. Therefore, any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A flexible fresh air control system for intelligent sports stadiums, characterized by: include: A data acquisition module (1) is used to detect the flow of people, detect age and body comfort, and collect fan motor operating parameters and environmental parameters; A big data analysis and deep learning module (2), including a body comfort model, a motor operation optimization model, and an age and physiological difference adjustment model, for dynamically outputting control parameters of a fresh air system and an air conditioning system based on the data collected by the data collection module (1); Generate an adjustment strategy and transmit it to the control execution module (3), which is used to adjust the fresh air system, air conditioning system and motor variable frequency speed regulation and load distribution of the sports venue according to the control parameters output by the big data analysis and deep learning module (2).

2. The intelligent sports stadium flexible fresh air control system according to claim 1 is characterized by: The data acquisition module (1) comprises: infrared sensors installed at the entrance and exit of a sports stadium and used for detecting the flow of people, a plurality of AI cameras and a plurality of WiFi / Bluetooth detectors installed in the sports stadium and used for detecting the flow of people, a receiving module installed in the sports stadium and used for collecting data from smart wearable devices, a plurality of environmental sensors installed in the sports stadium and used for detecting environmental information, and a voltage sensor, a current sensor and a wind speed sensor for collecting operating parameters of a fan motor.

3. A control method using the intelligent sports stadium flexible fresh air control system according to claim 1, characterized in that: The following steps are involved: Step 1: The data collection module (1) detects the flow of people in the stadium through infrared sensors and AI cameras; Step 2: The data acquisition module (1) detects the signals emitted by the smart devices in the stadium through the WiFi / Bluetooth detector, and estimates the crowd density and crowd flow direction in the area where the WiFi / Bluetooth detector is located; Step 3: Data collection module (1) identifies the age group of users in the stadium based on the images collected by the AI ​​camera based on facial recognition technology; Step 4: The data acquisition module (1) collects environmental information of the area where the environmental sensor is located through the environmental sensor; Step 5: Big data analysis and deep learning module (2) determines the user's comfort level based on the crowd density, crowd flow direction, user age group, and environmental information in each area, and dynamically adjusts the environmental conditions; Step 6: Generate an adjustment strategy and pass it to the control execution module (3) to adjust the air flow rate of the fresh air system and the air temperature of the air conditioning system in the stadium according to the environmental conditions of the big data analysis and deep learning module (2).

4. The control method according to claim 3, characterized in that: In the step 1, the flow of people is detected first, and the flow of people in the stadium is monitored by infrared sensors, AI cameras combined with AI analysis and WiFi / Bluetooth detectors; infrared sensors are installed at the entrance and exit of the stadium, and the flow of people is monitored in real time by detecting the infrared radiation released by the human body, providing an accurate number of people entering and leaving, thereby calculating the number of people in the stadium; The AI ​​camera combines AI technology to detect and count dynamic crowds by capturing images inside the stadium and using YOLO and convolutional neural network algorithms. At the same time, the interior of the stadium is divided into zones and AI cameras are installed, with each AI camera capturing images within the corresponding area.

5. The control method according to claim 3, characterized in that: In step 3, the image captured by the AI ​​camera is combined with facial recognition technology to capture the person in the image, and the age group of the user is identified by the AgeNet algorithm after the facial image is extracted, and the age group is divided into children, youth, middle-aged, and elderly; For situations where users of different age groups, or children and the elderly coexist in the same area, the system makes comprehensive adjustments based on ambient temperature, air flow rate and the user's real-time physiological state, so that the comfort of users of different age groups can be taken into account.

6. The control method according to claim 3, characterized in that: In step 4, the environmental information includes ambient temperature, ambient humidity, and air quality data; wherein the control quality data includes PM2.5 concentration and CO2 concentration.

7. The control method according to claim 3, characterized in that: In step 5, the body comfort model of the big data analysis and deep learning module (2) is based on the temperature, humidity, and air quality data obtained from the environmental sensor and the user's physiological state data obtained from the data acquisition module (1), and is comprehensively analyzed through a convolutional neural network and a long short-term memory network; The convolutional neural network of the somatosensory comfort model is used to process environmental data from different areas, including the distribution of temperature, humidity, and air quality, and extract local spatial features from them to identify the comfort requirements of different areas. The long short-term memory network of the somatosensory comfort model is used to process the user's time series physiological data, including physiological signals of changes in heart rate and sweating, and capture the time dependency of these physiological signals. The somatosensory comfort model is based on the output of convolutional neural networks and long short-term memory networks, and is fine-tuned according to historical data and real-time data to automatically adjust the temperature and humidity in sports stadiums. The somatosensory comfort model adjusts model parameters through reinforcement learning based on user complaints, changes in user expressions captured by AI cameras, and feedback on body temperature changes obtained by WiFi / Bluetooth detectors to improve the accuracy of comfort prediction.

8. The control method according to claim 3, characterized in that: In step 5, if there are a large number of children and a small number of elderly people in one area, the system adopts a multi-regional control strategy according to the temperature requirements of different users, and physically divides the area into multiple small areas; for the small area where children are more active, the temperature is lowered and the ventilation is enhanced; and for the small area where there are relatively more elderly people, the temperature is increased and the air flow rate is slowed down; The age and physiological difference adjustment model of the big data analysis and deep learning module (2) uses multimodal deep learning and generative adversarial networks to process the physical differences of users of different age groups. The age and physiological difference adjustment model sets weights according to the age distribution of the population, with the weighting coefficient for children being 0.4 and that for the elderly being 0.

6. Multimodal learning automatically adjusts and outputs control parameters by integrating the user's age, physiological state, and environmental data.

9. The control method according to claim 3, characterized in that: In the step 5, the motor operation optimization model of the big data analysis and deep learning module (2) processes the time series data of the fan motor operation parameters through the time series prediction network, and predicts the future load demand based on the historical data; The motor operation optimization model of the big data analysis and deep learning module (2) is based on the data collection module (1) on the flow of people. The camera detects the flow of people in each area through video monitoring and the YOLO algorithm. The infrared sensor detects the number of people entering and leaving the stadium, and the WiFi / Bluetooth detector analyzes the signal strength and density changes of smart devices in the stadium to estimate the future flow of people and density changes, and organizes them into multi-dimensional time series according to timestamps. The time series prediction network of the motor operation optimization model is trained using historical passenger flow data. The time series prediction network of the motor operation optimization model predicts the passenger flow in a certain period of time in the future by analyzing the past passenger flow change trend; Input data for the time series prediction network of the motor operation optimization model: The input includes historical passenger flow data for multiple time steps, holiday information, large-scale event announcements, and additional features of weather changes; The output of the time series prediction network of the motor operation optimization model: the predicted value of passenger flow in each time period in the future; When it is predicted that the flow of people will increase rapidly, the fan speed will be gradually increased to ensure effective air circulation, and the air conditioning system will be controlled to gradually lower the temperature.

10. The control method according to claim 3, characterized in that: In step 6, in terms of the regulation of the fresh air system, a regulation strategy is generated and transmitted to the control execution module (3) to adjust the fan speed and air volume of the fresh air system in real time according to the flow of people, air quality and physiological status data; when the flow of people is large, a regulation strategy is generated and transmitted to the control execution module (3) to automatically increase the wind speed to enhance air circulation; in terms of the regulation of the air conditioning system, a regulation strategy is generated and transmitted to the control execution module (3) to automatically adjust the set values ​​of temperature and humidity based on the body comfort model and age requirements; a regulation strategy is generated and transmitted to the control execution module (3) to dynamically adjust the air conditioning output of each area according to the activity volume and crowd distribution in different areas of the stadium.

Images