Environment-friendly noise reduction type AI cluster server and use method thereof

Abstract

The invention relates to the technical field of environment noise reduction, and discloses an environment-friendly and noise-reduction AI cluster server and a use method thereof, and the use method of the server comprises the following steps: obtaining multi-dimensional perception data which comprises human traffic data, environment sound data and temperature and humidity data; constructing a multi-dimensional data fusion model, and fusing the obtained multi-dimensional perception data through the multi-dimensional data fusion model to generate a feature vector; based on multi-dimensional data fusion and a deep learning algorithm, the system can automatically adjust the noise reduction strategy according to the dynamic change of the environment to realize accurate noise reduction, and by integrating the multi-dimensional data such as the human traffic, the sound characteristics, the temperature and the humidity, the system can comprehensively sense the environment state and provide a richer information basis for the decision-making of the noise reduction strategy, and at the same time, the noise reduction effect is improved. Advanced algorithms such as deep reinforcement learning are adopted, and the system can continuously optimize a noise reduction strategy through continuous learning environment feedback to adapt to requirements of different scenes.

Description

Technical Field

[0001] The present invention relates to the technical field of environmental noise reduction, and more specifically, it relates to an environmentally friendly noise reduction type AI cluster server and its usage method. Background Art

[0002] With the acceleration of the urbanization process and the increase in the population density in public places, noise pollution has become an important environmental problem affecting people's quality of life and work efficiency. Traditional noise reduction technologies mainly rely on noise reduction equipment with fixed parameters and cannot be adaptively adjusted according to the dynamic changes of the environment, resulting in poor noise reduction effects or resource waste. The current noise reduction systems on the market mainly have the following technical problems: 1) Single noise reduction strategy: Existing noise reduction systems usually adopt preset noise reduction parameters and cannot be dynamically adjusted according to the acoustic characteristics of different scenarios.

[0003] 2) Limited perception dimension: Traditional noise reduction systems mainly rely on a single sound sensor to collect environmental noise data and lack comprehensive perception of multi-dimensional environmental factors such as temperature, humidity, and the number of people, resulting in a lack of comprehensiveness in formulating noise reduction strategies.

[0004] 3) Poor algorithm adaptability: Most existing noise reduction algorithms are designed for specific frequencies or specific scenarios and have insufficient adaptability in complex and changeable environments. When the environmental conditions change, the algorithm cannot automatically adjust the parameters, and the noise reduction effect drops significantly.

[0005] 4) Low energy efficiency: Traditional noise reduction systems often adopt a noise reduction strategy for the entire time period and the entire frequency band without considering the actual needs, resulting in energy waste and excessive wear of equipment.

[0006] 5) Poor user experience: Due to the lack of understanding of the needs of specific user groups and targeted processing, existing noise reduction systems are difficult to provide a customized acoustic environment for users in different scenarios, affecting the user experience. Especially in places with high requirements for sound sensitivity such as the hospital waiting area, the existing technology cannot effectively identify and preferentially process specific frequency noises sensitive to patients, resulting in patients still being disturbed by noise during the waiting process and increasing anxiety. Summary of the Invention

[0007] The present invention provides an environmentally friendly noise reduction type AI cluster server and its usage method to solve the above-mentioned technical problems.

[0008] The first aspect of the present invention provides a usage method of an environmentally friendly noise reduction type AI cluster server, including the following steps: Obtain multi-dimensional perception data, where the multi-dimensional perception data includes the number of people data, environmental sound data, temperature and humidity data; Construct a multi-dimensional data fusion model, and fuse the obtained multi-dimensional perception data through the multi-dimensional data fusion model to generate a feature vector; Establish a noise reduction strategy decision model, and based on the feature vector, output a noise reduction strategy instruction through the noise reduction strategy decision model; Send the noise reduction strategy instruction to the noise reduction device to perform corresponding noise reduction operations, so as to realize the adaptive adjustment of the noise reduction strategy according to the dynamic changes of the environment.

[0009] As a further optimization scheme of the present invention, the multi-dimensional data fusion model is implemented in one of the following ways: Weighted summation method, through the formula: ; Calculate the feature vector, where, is the weight set according to the data importance, is the normalized pedestrian flow data , is the normalized environmental sound data , is the normalized pedestrian flow data , is the normalized temperature data ; Multi-layer perceptron neural network method, extracting and fusing multi-dimensional data features through multi-layer non-linear transformation.

[0010] As a further optimization scheme of the present invention, the noise reduction strategy decision model is implemented in one of the following ways: Rule-based decision-making method, outputting corresponding noise reduction strategy instructions according to the interval range of the feature vector values; Deep reinforcement learning method, learning the optimal noise reduction strategy through the deep Q network or the deep deterministic policy gradient algorithm; Hybrid decision-making method, combining a rule system and a reinforcement learning algorithm, and maintaining special attention to specific frequencies during the learning process.

[0011] As a further optimization scheme of the present invention, the method further includes obtaining specific scenario-related data according to different application scenarios: In the traffic hub scenario, additionally obtain the train / flight schedule flow data; In the large event venue scenario, additionally obtain the event type data; In the shopping mall scenario, additionally obtain the store sound data; In the hospital waiting area scenario, additionally obtain the patient sensitive sound frequency data.

[0012] As a further optimization solution of the present invention, in the hospital waiting area scenario, an attention mechanism is applied to the patient-sensitive sound frequency data: Through the formula , calculate the attention weights; Through the formula , obtain the weighted sensitive frequency data; Input the weighted sensitive frequency data into the multi-dimensional data fusion model, so that the system can give priority to focusing on the frequencies sensitive to patients.

[0013] As a further optimization solution of the present invention, the noise reduction strategy instructions include: Noise reduction intensity instruction, which controls the working intensity of the noise reduction device; Frequency range instruction, which performs fine control on noises of different frequencies; Special frequency processing instruction, which performs additional enhancement or weakening processing on noises of specific frequencies.

[0014] As a further optimization solution of the present invention, the method further includes the following steps: Receive the effect feedback after the noise reduction device executes the noise reduction operation; Update the parameters of the noise reduction strategy decision model based on the effect feedback; Use the updated model parameters to generate optimized noise reduction strategy instructions in the next round of decision-making.

[0015] The second aspect of the present invention provides an environmentally friendly noise reduction type AI cluster server, which is used to execute the usage method of the above-mentioned environmentally friendly noise reduction type AI cluster server to achieve environment-adaptive noise reduction. The server includes: Data acquisition layer, which includes a multi-dimensional sensor array and a data preprocessing unit, and is used to acquire and preprocess multi-dimensional perception data; Computing and processing layer, which includes a main controller, an AI acceleration chip, a memory unit and a storage unit, and is used to execute multi-dimensional data fusion and noise reduction strategy decision-making; Communication control layer, which includes a network module and a control interface, and is used to communicate with the noise reduction device and send noise reduction strategy instructions.

[0016] As a further optimization solution of the present invention, the computing and processing layer further includes: Data preprocessing module, which is used to clean, standardize and extract features from the original data; Multi-dimensional data fusion module, which is used to fuse data of different dimensions to generate feature vectors; Noise reduction strategy decision module, which is used to decide the optimal noise reduction strategy according to the feature vectors; Strategy execution module, which is used to convert the decided noise reduction strategy into specific control instructions.

[0017] As a further optimized solution of the present invention, the AI acceleration chip is a dedicated neural network processing unit that supports 16-bit floating-point operations and is used to accelerate the calculation of deep learning models; The server further includes a power management unit that can intelligently adjust the system power consumption, supports an energy-saving mode, and achieves the energy efficiency goal of environmental protection and noise reduction.

[0018] Advantages of the present invention: Based on multi-dimensional data fusion and deep learning algorithms, the system can automatically adjust the noise reduction strategy according to the dynamic changes of the environment to achieve precise noise reduction. Moreover, by integrating multi-dimensional data such as the number of people, sound characteristics, temperature and humidity, the system can comprehensively perceive the environmental state, providing a richer information basis for noise reduction strategy decision-making. At the same time, by adopting advanced algorithms such as deep reinforcement learning, the system can continuously optimize the noise reduction strategy by continuously learning environmental feedback to meet the requirements of different scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 is a flowchart of a method for using an environmentally friendly noise reduction type AI cluster server of the present invention; Figure 2 is a graph of the number of people flow data in Embodiment 1 of the present invention; Figure 3 is a graph of the train / flight frequency flow data in Embodiment 1 of the present invention; Figure 4 is a graph of the environmental sound data in Embodiment 1 of the present invention; Figure 5 is a graph of the humidity data in Embodiment 1 of the present invention; Figure 6 is a graph of the temperature data in Embodiment 1 of the present invention; Figure 7 is a graph of the data after normalization in Embodiment 1 of the present invention; Figure 8 is a graph of the feature vector value data at each time point in Embodiment 1 of the present invention; Figure 9 is a graph of the noise reduction strategy decision data at each time point in Embodiment 1 of the present invention; Figure 10 is a graph of the comparison of sound intensity (decibel) before and after noise reduction in Embodiment 1 of the present invention; Figure 11 is a graph of the passenger satisfaction survey results data in Embodiment 1 of the present invention; Figure 12 is a graph of the staff work efficiency survey data in Embodiment 1 of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0020] Reference will now be made to exemplary embodiments to discuss the subject matter described herein. It should be understood that the discussion of these embodiments is only to enable those skilled in the art to better understand and thus implement the subject matter described herein, and that changes can be made to the functions and arrangements of the elements discussed without departing from the scope of protection of the content of this specification. Each example may omit, substitute, or add various processes or components as needed. Additionally, features described relative to some examples may be combined in other examples.

[0021] Example 1 This embodiment provides a method for using an environmentally friendly and noise-reducing AI cluster server, which is applied to urban transportation hubs, aiming to solve the technical need of dynamically adjusting noise reduction strategies with the help of multi-dimensional data in a noisy environment, thereby improving the transportation hub environment and enhancing the travel experience of passengers. The specific steps are as follows: Step 1, data acquisition: Use sensors to continuously acquire the number of people , train / flight traffic , environmental sound , humidity and temperature data, where: (1) Pedestrian flow data : The set of the number of people entering the transportation hub within the time period , obtained by the pedestrian flow sensors installed at the entrance of the transportation hub; (2) Train / flight traffic data : The set of trains / flights arriving or departing within the time period , obtained by the train / flight management system of the transportation hub; (3) Environmental sound data : The set of sound intensity at different frequencies , obtained by the sound sensors distributed throughout the transportation hub; (4) Humidity data : The set of humidity , obtained by the humidity sensors; (5) Temperature data : The set of temperature , obtained by the temperature sensors.

[0022] The specific steps of data acquisition are as follows: Step 101, install pedestrian flow sensors at the entrance of the transportation hub, and continuously count the number of people entering the transportation hub to form pedestrian flow data ; Step 102: Connect to the train / flight schedule management system of the transportation hub to obtain the train / flight schedule information arriving at or departing within the time period and form the train / flight schedule flow data .

[0023] Step 103: Deploy a sound sensor array inside the transportation hub to collect the sound intensities in different regions and at different frequencies, and form the ambient sound data .

[0024] Step 104: Install humidity sensors to collect the humidity data inside the transportation hub .

[0025] Step 105: Install temperature sensors to collect the temperature data inside the transportation hub .

[0026] Step 2: Multidimensional data fusion: Build a multidimensional data fusion model , fuse the obtained data, define the fusion function , and through the formula , integrate different types of data into a feature vector .

[0027] In this embodiment, the fusion function integrates various types of data by using the weighted summation method, and the specific expression is: ; where is the weight set according to the importance of the data. In order to enable weighted summation of data with different dimensions, it is necessary to first perform normalization processing on various types of data so that their value ranges are unified to [0, 1].

[0028] The normalization processing adopts the maximum-minimum normalization method. For any data , its normalization formula is: ; where represents the data to be normalized, represents the minimum value of this data , represents the maximum value of the data , and the calculated through this formula is the normalized data, and its value range is unified to [0, 1].

[0029] Specifically, for the passenger flow data , after normalization, we get ; for the train / flight schedule flow data , after normalization, we get ; For the environmental sound data , after normalization, we get ; For the humidity data , after normalization, we get ; For the temperature data , after normalization, we get .

[0030] Then, the calculation formula for the feature vector is as follows: ; In this embodiment, the weights are set as: , , , , , indicating that the pedestrian flow and the vehicle trips / flight volume have a greater impact on the noise reduction strategy.

[0031] Step 3, Noise reduction strategy decision: Establish a noise reduction strategy decision model , using the feature vector as the input, define the decision function , through the formula , output the noise reduction strategy instruction set .

[0032] In this embodiment, the decision function is a rule-based function. When certain eigenvalue in (specifically referring to the comprehensive numerical performance of the data such as pedestrian flow, vehicle trips / flight volume, environmental sound, humidity, temperature, etc. after normalization and weighted summation in the fused feature vector ) meets specific conditions, output the corresponding noise reduction strategy instruction , where represents the th noise reduction strategy, such as increasing the noise reduction intensity, changing the noise reduction frequency range, etc.

[0033] The specific rules are as follows: (1) When , output the instruction : Set the noise reduction intensity to the highest level, covering the full frequency band; (2) When , output the instruction : Set the noise reduction intensity to a high level, focusing on covering the middle and high frequency bands; (3) When , output the instruction : Set the noise reduction intensity to a medium level, evenly covering each frequency band; (4) When occurs, output the instruction : Set the noise reduction intensity to the low level, focusing on covering the low-frequency band; (5) When occurs, output the instruction : Set the noise reduction intensity to the lowest level, only covering the necessary frequency bands.

[0034] In addition, when the sound intensity of certain specific frequencies in the ambient sound data exceeds the preset threshold, a special noise reduction instruction for that frequency will be triggered. For example, when the high-frequency harsh sound generated when the train enters the station exceeds the threshold, the system will output the instruction : Temporarily enhance the noise reduction intensity in the high-frequency band.

[0035] Step 4, Strategy Execution: Send the noise reduction strategy instruction to the noise reduction device to perform the corresponding noise reduction operation. Specifically: Step 401, The AI cluster server transmits the noise reduction strategy instruction to the noise reduction device control system in the transportation hub through the network; Step 402, After receiving the instruction, the noise reduction device control system adjusts the working parameters of the noise reduction device according to the instruction content, such as the noise reduction intensity, frequency range, etc.; Step 403, The noise reduction device performs the noise reduction operation according to the adjusted parameters, changing the acoustic environment in the transportation hub; Step 404, The noise reduction effect is fed back to the AI cluster server as a reference for the next round of decision-making.

[0036] In this embodiment, the noise reduction devices in the transportation hub dynamically adjust the noise reduction strategy based on multi-dimensional perception data, reducing the noise level in the transportation hub, making the passenger information broadcast clearer and more audible, and ultimately improving the travel experience of passengers in the transportation hub. The specific effects are as follows: (1) During the peak hours of train arrivals and departures, the system automatically strengthens the noise reduction to ensure that the passenger information broadcast is clear and audible.

[0037] (2) In areas with a large flow of people, the system specifically enhances the noise reduction effect to reduce the impact of crowd noise on the environment.

[0038] (3) According to the changes in temperature and humidity, the system adjusts the noise reduction strategy to adapt to the sound propagation characteristics under different environmental conditions.

[0039] (4) Overall, reduce the noise level in the transportation hub and create a more comfortable waiting and transfer environment for passengers.

[0040] Real application example of Embodiment 1 To more intuitively demonstrate the application of the self - adjusting noise reduction system of the multi - dimensional perception - driven environmental protection and noise reduction AI cluster server in urban transportation hubs, a real - world application example of a waiting hall at Beijing West Station during the period from 9:00 to 10:00 am on a weekday is given below.

[0041] 1. Data acquisition example: A variety of sensors are installed in the waiting hall of Beijing West Station, and the following data are collected: (1) Passenger flow data (sampled once every 10 minutes), as Figure 2 shown; (2) Train / flight frequency data (sampled once every 10 minutes), as Figure 3 shown; (3) Environmental sound data (unit: decibel, sampled once every 10 minutes), as Figure 4 shown; (4) Humidity data (unit: %, sampled once every 10 minutes), as Figure 5 shown; (5) Temperature data (unit: °C, sampled once every 10 minutes), as Figure 6 shown.

[0042] 2. Multi - dimensional data fusion example: First, normalize all types of data, and set the data range as follows: (1) Passenger flow range: 0 - 2000 people / 10 minutes; (2) Train frequency range: 0 - 10 trains / 10 minutes; (2) Sound intensity range: 40 - 90 decibels; (4) Humidity range: 30 - 70%; (5) Temperature range: 18 - 28 °C; Taking the time point of 9:30 as an example, the normalized data is as Figure 7 shown; According to the weight settings , , , , , calculate the feature vector : ; Similarly, calculate the feature vector values at other time points: The feature vector value data at each time point is as Figure 8 shown.

[0043] 3. Noise reduction strategy decision example: According to the decision rules, for the feature vector values at each time point Determine the noise reduction strategy: The decision-making data for the noise reduction strategy at each time point, such as Figure 9 shown; In addition, at 9:30, since the intermediate frequency sound intensity reached 82 dB, exceeding the preset threshold of 80 dB, the system additionally triggered a special noise reduction instruction for the intermediate frequency : Temporarily increase the noise reduction intensity in the intermediate frequency band by 15%.

[0044] 4. Example of strategy execution: Taking the time point of 9:30 as an example, the AI cluster server sends the following instructions to the noise reduction equipment in the waiting hall: (1) Basic noise reduction strategy: (High-level noise reduction, focusing on covering the intermediate and high frequency bands); (2) Special noise reduction instruction: (Temporarily increase the noise reduction intensity in the intermediate frequency band by 15%); After receiving these instructions, the noise reduction equipment control system adjusted the working parameters of the noise reduction equipment: Low frequency band noise reduction intensity: 65%; Intermediate frequency band noise reduction intensity: 85% (basic 70% + additional 15%); High frequency band noise reduction intensity: 75%; 5. Technical effect verification data: To verify the noise reduction effect, sound monitoring points were set up in different areas of the waiting hall to record the change in sound intensity before and after noise reduction: The data of the sound intensity comparison (dB) before and after noise reduction at 9:30, such as Figure 10 shown; The data of the passenger satisfaction survey results (random sampling of 100 people), such as Figure 11 shown; The data of the staff work efficiency survey (random sampling of 20 people), such as Figure 12 shown; From the above data, it can be seen that the self-adjusting noise reduction system of the environmental protection noise reduction AI cluster server driven by multi-dimensional perception has achieved remarkable results in the waiting hall of Beijing West Station: (1) The average sound intensity has been reduced by about 30%, and the noise reduction effect in the intermediate frequency band is the most obvious, with an average reduction of 33.7%; (2) The passenger satisfaction has been significantly improved, and the average satisfaction rate has increased by 27%; (3) The work efficiency of the staff has been improved, the number of repeated announcements has been reduced by 44%, and the response time to passenger inquiries has been reduced by 29%; (4) The system can dynamically adjust the noise reduction strategy according to multi-dimensional data such as the flow of people and the flow of train trips, and provide stronger noise reduction effects during peak hours (such as 9:30).

[0045] This data fully demonstrates the effectiveness and practical value of this technical solution in the application of urban transportation hubs, not only improving the acoustic environment of the transportation hub, but also enhancing the travel experience of passengers and the work efficiency of staff.

[0046] Example Two This embodiment provides a usage method of an environmentally friendly noise reduction type AI cluster server, which is applied to large urban event venues. The aim is to solve the technical requirement of dynamically adjusting the noise reduction strategy with multi-dimensional data in a crowded and noisy large event venue environment, so as to provide a good acoustic environment, enhance the experience of event participants, and provide an intelligent environmental management means for the venue management party. The specific steps include: Step 5: Data acquisition: Install a pedestrian flow sensor at the entrance of the event venue to obtain pedestrian flow data ; The venue management party enters the event type data before the event ; Arrange sound, humidity, and temperature sensors in the venue to obtain , , data respectively, where: (1) Pedestrian flow data : The set of the number of people entering the event venue within the time period , obtained through the pedestrian flow sensor at the entrance; (2) Event type data : Represents the type of the current event, such as a concert, a sports event, etc., entered into the system by the venue management party before the event; (3) Ambient sound data : The set of sound intensities at different frequencies , obtained by using the sound sensors distributed in the venue; (4) Humidity data : The set of humidity , obtained through the humidity sensor in the venue; (5) Temperature data : Obtained by the temperature sensor in the venue, and the temperature set is ; ; The specific steps of data acquisition are as follows: Step 501: Install pedestrian flow sensors at each entrance of the event venue, and count the number of people entering the venue in real time to form pedestrian flow data ; Step 502: Before the event, the venue management party enters the event type data through the system interface, including information such as the event category (such as concert, sports event, exhibition, etc.), the expected scale, and the duration; Step 503: Deploy a sound sensor array inside the event venue to collect the sound intensities in different areas and at different frequencies, and form environmental sound data. ; Step 504: Install a humidity sensor to collect the humidity data inside the event venue. ; Step 505: Install a temperature sensor to collect the temperature data inside the event venue. .

[0047] Step 6: Multidimensional data fusion: Construct a multidimensional data fusion model. , define a fusion function. Since the event type has a great influence on the sound characteristics, the fusion formula is: ; where is the weight set according to the data importance, can be one-hot encoded and then participate in the operation. Through this step, different types of data are integrated into a feature vector. .

[0048] The specific steps are as follows: Step 601: Normalize the pedestrian flow data , environmental sound data , humidity data and temperature data so that their value ranges are unified to [0, 1]; Step 602: Perform one-hot encoding (One-Hot Encoding) on the event type data . Assume that there are types of event types, then is encoded as a -dimensional vector, where only one element is 1 and the rest are 0. For example, if the event type is "concert" and "concert" corresponds to the second position in the encoding table, then .

[0049] Step 603: Calculate the feature vector : ; In this embodiment, the weights are set as: , , , , , indicating that the event type and pedestrian flow, environmental sound have a greater impact on the noise reduction strategy.

[0050] Step 7, Noise Reduction Strategy Decision: Establish a noise reduction strategy decision model , using the feature vector as the input, define the decision function , and through the formula , output the noise reduction strategy instruction set , where the decision function can establish more complex rules in combination with the activity type. For example, when the activity is a concert and there are more high-frequency components in the ambient sound, output an instruction to increase the high-frequency noise reduction intensity .

[0051] The specific steps are as follows: Step 701, Basic Rules Based on Activity Type: (1) Concert: Focus on reducing mid-high frequency noise and retaining the music frequency band; (2) Sports event: Evenly reduce noise in each frequency band and retain the commentary frequency band; (3) Exhibition: Focus on reducing low-frequency noise and creating a quiet environment; Step 702, Adjustment Rules Based on the Number of People: (1) When : Increase the overall noise reduction intensity by 20%; (1) When : Increase the overall noise reduction intensity by 10%; (3) When : Maintain the basic noise reduction intensity; Step 703, Fine Adjustment Rules Based on Ambient Sound: (1) When the sound intensity of a certain frequency band in exceeds the threshold of 0.9: Increase the noise reduction for that frequency band by an additional 30%; (2) When the sound intensity of a certain frequency band in is between 0.7 and 0.9: Increase the noise reduction for that frequency band by an additional 20%; Step 704, Environmental Adaptation Rules Based on Temperature and Humidity: (1) When and : Adjust the noise reduction algorithm parameters to adapt to the sound propagation characteristics in high temperature and high humidity environments; (2) When and : Adjust the noise reduction algorithm parameters to adapt to the sound propagation characteristics in low temperature and low humidity environments.

[0052] Through the basic rules of activity types, the adjustment rules of the flow of people, the fine adjustment rules of environmental sounds, and the environmental adaptation rules of temperature and humidity, the decision function outputs the final set of noise reduction strategy instructions 。

[0053] Step 8, Strategy Execution: Send the noise reduction strategy instructions to the noise reduction equipment to perform the corresponding noise reduction operations. Specifically: Step 801, The AI cluster server transmits the noise reduction strategy instructions to the noise reduction equipment control system in the event venue through the network.

[0054] Step 802, After receiving the instructions, the noise reduction equipment control system adjusts the working parameters of the noise reduction equipment according to the instruction content, such as the noise reduction intensity, frequency range, etc.

[0055] Step 803, The noise reduction equipment performs noise reduction operations according to the adjusted parameters, changing the acoustic environment in the event venue.

[0056] Step 804, The noise reduction effect is fed back to the AI cluster server as a reference for the next round of decision-making.

[0057] In large urban event venues, the noise reduction equipment dynamically adjusts the noise reduction strategy based on multi-dimensional perception data, effectively reducing the noise level in the venue and providing a suitable acoustic environment for different types of events. Event participants can hear the event-related sounds more clearly, enhancing the event experience. At the same time, the venue management side can better control the venue environment through intelligent environmental management means, and computer technology is optimized in multi-dimensional data fusion and decision-making models, improving its adaptability in diverse scenarios and promoting the further development of the adaptive algorithm of the noise reduction AI cluster server.

[0058] The specific effects are as follows: (1) In concert events, the system can retain the music frequency band while reducing the noise in other frequency bands, enabling the audience to better enjoy the music; (2) In sports events, the system can retain the commentary frequency band while reducing the noisy sounds in the venue, enabling the audience to clearly hear the game commentary; (3) In areas with a large flow of people, the system automatically enhances the noise reduction effect, reducing the impact of crowd noise on the event experience.

[0059] According to different event types and environmental conditions, the system can intelligently adjust the noise reduction strategy to provide the best acoustic environment.

[0060] Example 3 This embodiment provides an environmentally friendly and noise-reducing AI cluster server and its usage method, which is applied to large shopping malls, aiming to solve the technical requirements of dynamically adjusting the noise reduction strategy with multi-dimensional data in a noisy shopping mall environment, improving the shopping environment in the mall, enhancing the shopping experience of customers, and providing a direction for the intelligent management of the mall.

[0061] Step 9, Data acquisition: Use cameras to obtain the pedestrian flow data , use sound sensors to obtain regional sound data , obtain store sound data by connecting to the audio devices in the stores , use humidity sensors to obtain humidity data , use temperature sensors to obtain temperature data , where: (1) Pedestrian flow data : The set of the number of people entering different areas of the shopping mall within the time period , which is obtained by counting the number of people through the cameras installed at the entrances and main channels of the shopping mall. Among them .

[0062] (2) Regional sound data : The set of the sound intensity and frequency characteristics of different areas in the shopping mall, which is obtained by using the sound sensors distributed in each area of the shopping mall .

[0063] (3) Store sound data : The set of sound data such as promotional sounds and background music in each store, which is obtained by connecting to the audio devices in the store .

[0064] (4) Humidity data : The set of humidity in different areas in the shopping mall, which is obtained by using the humidity sensors distributed throughout the shopping mall .

[0065] (5) Temperature data : The set of temperature in different areas in the shopping mall, which is obtained by using temperature sensors .

[0066] The specific steps of data acquisition are as follows: Step 901, Install cameras at the entrances and main channels of the shopping mall, and use computer vision algorithms to count the number of people entering different areas of the shopping mall in real time to form pedestrian flow data ; Step 902, Deploy sound sensors in each area of the shopping mall, collect the sound intensity and frequency characteristics of different areas, and form regional sound data ; Step 903: Establish a connection with the audio devices in each store, obtain sound data such as promotional sounds and background music in the store, and form store sound data ; Step 904: Install humidity sensors in various areas of the mall to collect humidity data in different areas ; Step 905: Install temperature sensors in various areas of the mall to collect temperature data in different areas ; Step 10: Multidimensional data fusion: Construct a multidimensional data fusion model , define a fusion function , and integrate different types of data into a feature vector through the formula , where the fusion function adopts a more complex neural network structure to extract and fuse features from various types of data, rather than simple weighted summation. The input of the neural network is various types of data, and after passing through the hidden layer, a feature vector is output . .

[0067] The specific steps are as follows: Step 1001: Preprocess all input data, including operations such as normalization and denoising, to make it suitable for neural network processing; Step 1002: Construct a multi-layer perceptron (MLP) neural network as the fusion function , and its structure is as follows: Input layer: Receive the preprocessed , , , , data.

[0068] Hidden layer 1: Contains 128 neurons and uses the ReLU activation function; Hidden layer 2: Contains 64 neurons and uses the ReLU activation function; Hidden layer 3: Contains 32 neurons and uses the ReLU activation function; Output layer: Outputs a feature vector , with a dimension of 16.

[0069] Step 1003: The mathematical expression of the neural network is: ; ; ; ; Among them, , , respectively represent the output results of different hidden layers of the multi-layer perceptron neural network. , , , are weight matrices used to perform linear transformations on the input data. For example, map the data of the input layer to hidden layer 1, and through matrix multiplication, perform weighted combinations of different input features. , , , are bias vectors, adding a constant offset to the result of the linear transformation, which helps the model learn more complex functions. Because only linear transformation cannot learn non-linear relationships, the bias term can translate the model in the feature space, thus better fitting the data; the activation function , which performs non-linear transformation on the result of the linear transformation, enabling the neural network to learn non-linear relationships. For example, when is negative, outputs ; when is positive, outputs . By introducing non-linear activation functions, the neural network can learn more complex patterns and features in the input data; is the concatenation vector of the input data, containing the preprocessed , , , , data; Step 1004: Optimize the neural network parameters with training data so that it can effectively extract and fuse the features of multi-dimensional data.

[0070] Step 11: Noise reduction strategy decision-making: Establish a noise reduction strategy decision-making model , with the feature vector as the input, define the decision function , and through the formula , output the noise reduction strategy instruction set ; Among them, the decision function is based on the reinforcement learning algorithm. By continuously trying different noise reduction strategies and optimizing the strategies according to the feedback of the mall environment, when certain eigenvalue in satisfies specific conditions, output the corresponding noise reduction strategy instruction , where represents the A noise reduction strategy suitable for the mall environment, such as adjusting the noise reduction frequency band according to the sound characteristics of different stores, etc.; The specific steps are as follows: Step 1101: Construct a reinforcement learning model based on the Deep Q-Network (DQN) as the decision function .

[0071] Step 1102: Define the state space as the feature vector , and the action space as the set of possible noise reduction strategy instructions .

[0072] Step 1103: Define the reward function and calculate the reward value according to factors such as the noise reduction effect and customer satisfaction. For example, if the environmental sound level decreases and the customer satisfaction increases after noise reduction, a positive reward is given; otherwise, a negative reward is given.

[0073] Step 1104: The DQN network structure is as follows: Input layer: Receive the feature vector , with a dimension of 16; Hidden layer 1: Contains 64 neurons and uses the ReLU activation function; Hidden layer 2: Contains 32 neurons and uses the ReLU activation function; Output layer: Output the value of each possible action, with a dimension equal to the size of the action space; Step 1105: The training process of DQN adopts the Experience Replay and Target Network technologies. By continuously interacting with the environment, collect the quadruple data of state, action, reward, and next state, and update the Q-network parameters to enable it to learn the optimal noise reduction strategy decision.

[0074] It should be noted that the Experience Replay and Target Network technologies: Experience Replay stores the quadruple of state, action, reward, and next state obtained by the agent interacting with the environment in the experience replay pool, and then randomly samples for training the Q-network; the target network has the same structure as the main Q-network, but the parameter update is relatively slow and is used to calculate the target value.

[0075] The Experience Replay and Target Network technologies: Experience Replay means that the agent interacts with the environment to obtain the state , action , reward , next state quadruple , store this data in the experience replay pool. Then, randomly sample a batch of data from the pool for training the Q-network. The purpose of doing this is to break the correlation between the data and make the training more stable and effective.

[0076] For example, if the data is directly used for training in sequence, there may be a strong temporal correlation between the data, resulting in the model overfitting to the recent data. Random sampling can enable the model to learn the values of a wider range of state-action pairs. The target network is used to make the training of the Q-network more stable. It has the same structure as the main Q-network, but the parameter update is relatively slow. The main Q-network is used to generate the value estimate of the action, and the target network is used to calculate the target value. The calculation of the target value usually uses a variant of the Bellman equation, and the formula is as follows: ; where, is the reward obtained by the current action, which is calculated based on factors such as the noise reduction effect and customer satisfaction. For example, if the ambient sound level decreases after noise reduction and the customer satisfaction increases, a positive reward is given; otherwise, a negative reward is given. The reward value is used to guide the agent to learn the optimal strategy, that is, to maximize the reward by adjusting the action; is the discount factor, which is used to measure the importance of future rewards. The value range is usually between , and it determines the degree of importance the agent attaches to future rewards. If is close to , the agent pays more attention to long-term rewards; if is close to , the agent pays more attention to immediate rewards; is the value estimate of all actions by the target network in state , is the parameter of the target network. By using the target network to calculate the target value, the fluctuations in the training process of the main Q-network can be reduced because the target value is relatively stable and will not change violently with the frequent update of the parameters of the main Q-network.

[0077] Step 1106, after the training is completed, the decision function selects the action with the highest value as the noise reduction strategy instruction output according to the current feature vector .

[0078] Step 12: Policy execution: Send the noise reduction strategy instruction to the noise reduction device to perform the corresponding noise reduction operation. The specific steps are as follows: Step 1201: The AI cluster server sends the noise reduction strategy instruction Transmitted to the noise reduction equipment control system in the mall via the network; Step 1202: After receiving the instruction, the noise reduction device control system adjusts the working parameters of the noise reduction device according to the instruction content, such as noise reduction intensity, frequency range, etc.; Step 1203: The noise reduction device performs a noise reduction operation according to the adjusted parameters to change the acoustic environment in the mall; Step 1204: The noise reduction effect is fed back to the AI cluster server as a basis for updating the reinforcement learning model.

[0079] In this embodiment, the noise reduction equipment in the mall can more accurately and dynamically adjust the noise reduction strategy based on the multi-dimensional perception data. It reduces the noise level in the mall, allowing customers to hear store promotion information, background music and other sounds more clearly, improving customer shopping comfort. At the same time, it provides data support and strategy reference for the intelligent management of the mall, such as reasonably adjusting the distribution of stores according to the noise level in different areas, and promoting the further application of computer technology in the intelligent management of commercial scenarios. The specific effects are as follows: (1) In areas with dense traffic, the system automatically enhances the noise reduction effect to reduce the impact of crowd noise on the shopping environment; (2) Based on the sound characteristics of different stores, the system can intelligently adjust the noise reduction frequency band, retain useful promotional information and background music, and filter out irrelevant noise; (3) In different functional areas of the mall (such as dining area, rest area, shopping area), the system provides differentiated noise reduction strategies based on regional characteristics to create a more suitable acoustic environment; (4) Through continuous learning and optimization, the system can adapt to the dynamic changes in the shopping mall environment and provide increasingly accurate noise reduction services.

[0080] Embodiment 4 This embodiment provides an environmentally friendly noise-reducing AI cluster server and a method of using the same, which are applied to hospital waiting areas. The purpose is to address the technical needs of more finely and dynamically adjusting noise reduction strategies with the help of multi-dimensional data in an environment where there is noise generated by the flow of hospital personnel, so as to provide patients and their families with a quiet and comfortable waiting environment.

[0081] Step 13: Data acquisition: Use infrared sensor counter to obtain real-time traffic flow , obtain environmental sound through high-precision sound sensors , humidity sensor obtains humidity , the temperature sensor obtains the temperature , and collect and organize patients' sensitive sound frequency data ,in: (1) Traffic flow data : The set of the number of people entering the hospital waiting area during the time period , which is obtained by the infrared induction counter installed at the entrance of the waiting area.

[0082] Ambient sound data : The set of sound intensities at different frequencies , which is obtained by the high-precision sound sensors distributed in the waiting area.

[0083] (2) Humidity data : The set of humidity , which is obtained by the humidity sensor in the waiting area.

[0084] (3) Temperature data : The set of temperature , which is collected by the temperature sensor in the waiting area.

[0085] (4) Patient-sensitive sound frequency data : The set of frequencies to which patients are sensitive to various types of sounds collected in advance, which can be obtained by conducting questionnaires on patients and collecting and sorting clinical data in relevant hospital departments The specific steps are as follows: Step 1301: Install an infrared induction counter at the entrance of the waiting area to count the number of people entering the waiting area in real time and form the passenger flow data .

[0086] Step 1302: Deploy a high-precision sound sensor array inside the waiting area to collect the sound intensities in different regions and at different frequencies and form the ambient sound data .

[0087] Step 1303: Install a humidity sensor to collect the humidity data in the waiting area .

[0088] Step 1304: Install a temperature sensor to collect the temperature data in the waiting area .

[0089] Step 1305: Through questionnaire surveys and clinical data collection, sort out the frequency data of patients' sensitivity to various types of sounds , and these data include the sensitivity of different types of patients (such as heart disease patients, neurological disease patients, etc.) to specific frequency sounds.

[0090] Step 14: Multidimensional data fusion: Construct a multidimensional data fusion model , define the fusion function , and through the formula , integrate different types of data into a feature vector 。

[0091] In this embodiment, the fusion function adopts a multi-layer perceptron (MLP) neural network structure to better capture the complex relationships between data. The specific implementation is as follows: Step 1401: Normalize the pedestrian flow data , environmental sound data , humidity data and temperature data so that their value ranges are unified to [0, 1]; Step 1402: Process the patient's sensitive sound frequency data and convert it into a vector with the same dimension as the environmental sound data to represent the sensitivity of each frequency; Step 1403: Construct a multi-layer perceptron (MLP) neural network as the fusion function , and its structure is as follows: Input layer: Receive the preprocessed , , , , data; Hidden layer 1: Contains 64 neurons and uses the ReLU activation function; Hidden layer 2: Contains 32 neurons and uses the ReLU activation function Output layer: Output the feature vector , with a dimension of 16; Step 1404: The mathematical expression of the neural network is: ; ; ; where is the concatenated vector of the input data, containing the normalized , , , and the processed data, represents the data after being processed by the hidden layer 1, represents the data after being processed by the hidden layer 2. , , are the weight matrices used for linear transformation of the input data. For example maps the data in the input layer to the hidden layer 1, , , is a bias vector that adds a constant offset to the result of a linear transformation, which helps the model learn more complex functions. is an activation function that performs a non - linear transformation on the result of a linear transformation, enabling the neural network to learn non - linear relationships. For example, when is negative, outputs ; when is positive, outputs , is the finally output feature vector.

[0092] Step 1405: Optimize the neural network parameters with training data so that it can effectively extract and fuse the features of multi - dimensional data, where the training data includes various types of sensor data collected historically and the corresponding noise reduction effect evaluation results.

[0093] Step 1406: To maintain special attention to the patient - sensitive sound frequency data , apply an attention mechanism to at the input layer of the neural network, enabling the network to pay more attention to the sensitive frequency data: ; ; where is the attention weight matrix, which is used to weight the patient - sensitive sound frequency data . By multiplying with , it adjusts the importance of different frequency data in the model. The function is used to convert the result of ; where is the -th element in , is the dimension of . After being processed by the function, each element is between and , and the sum of all elements is . This allows the model to allocate attention according to the importance of different frequencies. represents element - wise multiplication, multiplying and element - wise to obtain the weighted sensitive frequency data In this way, the attention to the sensitive frequency data of the patient is highlighted, so that in the subsequent model processing, the frequencies sensitive to the patient can be given more attention, which helps to better implement the noise reduction processing for the patient's sensitive sounds.

[0094] Step 15, Noise reduction strategy decision: Establish a noise reduction strategy decision model , with the feature vector as the input. Define the decision function , through the formula , output the noise reduction strategy instruction set .

[0095] In this embodiment, the decision function adopts the deep reinforcement learning method and combines the rule system to better adapt to the dynamic environment of the hospital waiting area: Step 1501, Construct a reinforcement learning model based on the Deep Q-Network (DQN) as the core of the decision function .

[0096] Step 1502, Define the state space as the feature vector , and the action space as the set of possible noise reduction strategy instructions .

[0097] Step 1503, Define the reward function, and calculate the reward value according to factors such as the noise reduction effect and the patient's comfort. For example, if the environmental sound level decreases and the patient's comfort improves after noise reduction, a positive reward is given; otherwise, a negative reward is given.

[0098] Step 1504, The DQN network structure is as follows: Input layer: Receive the feature vector , with a dimension of 16; Hidden layer 1: Contains 32 neurons and uses the ReLU activation function; Hidden layer 2: Contains 16 neurons and uses the ReLU activation function; Output layer: Output the value of each possible action, with a dimension equal to the size of the action space; Step 1505, The training process of DQN adopts the Experience Replay and Target Network technologies. By continuously interacting with the environment, collect the quadruple data of state, action, reward, and next state, and update the Q-network parameters to enable it to learn the optimal noise reduction strategy decision; Step 1506, In order to ensure that the system can still maintain special attention to the patient's sensitive frequencies during the learning process, integrate the rule-based priority noise reduction rules in Steps 1501 to 1505 into the decision-making process of DQN: When the noise reduction strategy recommended by DQN conflicts with the rule-based priority noise reduction rule, the system will give priority to the rule-based strategy; As DQN learning progresses, the system gradually increases the weight of DQN decisions and reduces the weight of rule decisions, achieving a smooth transition from rule-driven to data-driven; Step 1507: After training is completed, the decision function According to the current feature vector , combined with the DQN output The final set of noise reduction strategy instructions is determined by combining the recommendations of the value and rule system. .

[0099] The specific rules remain the same: 1. Priority noise reduction rules based on patient sensitive frequencies: when The frequency and When the medium and high sensitivity frequencies overlap and the sound intensity exceeds the threshold of 0.6, the noise reduction priority of this frequency is set to the highest; when The frequency and When the medium-sensitivity frequency overlaps and the sound intensity exceeds the threshold of 0.7, the noise reduction priority of this frequency is set to the second highest; when The frequency and When the medium and low sensitivity frequencies overlap and the sound intensity exceeds the threshold of 0.8, the noise reduction priority of the frequency is set to medium; 2. Regulation rules based on human traffic: when : Enhance the overall noise reduction strength by 25%; when : Enhance the overall noise reduction strength by 15%; when : Enhance the overall noise reduction strength by 5%; when : Maintain basic noise reduction strength; 3. Fine-tuning rules based on ambient sound: For non-sensitive frequencies, when If the sound intensity of a certain frequency band exceeds the threshold of 0.9, the noise reduction will be enhanced by 20% for this frequency band; For non-sensitive frequencies, when The sound intensity of a certain frequency band is between 0.7 and 0.9: the noise reduction is enhanced by 10% for this frequency band; For non-sensitive frequencies, when The sound intensity of a certain frequency band is between 0.5-0.7: the noise reduction is enhanced by 5% for this frequency band; 4. Environmental adaptation rules based on temperature and humidity: When and : Adjust the noise reduction algorithm parameters to adapt to the sound propagation characteristics in high temperature and high humidity environments; When and : Adjust the noise reduction algorithm parameters to adapt to the sound propagation characteristics in low temperature and low humidity environments; Based on the priority noise reduction rule for the patient's sensitive frequency, the adjustment rule for the number of people, the fine adjustment rule for the ambient sound, and the environmental adaptation rule for temperature and humidity, the decision function Outputs the final set of noise reduction strategy instructions ; Step 16. Strategy execution: Send the noise reduction strategy instructions to the noise reduction device to perform the corresponding noise reduction operations. The specific steps are as follows: Step 1601. The AI cluster server transmits the noise reduction strategy instructions to the noise reduction device control system in the waiting area through the network; Step 1602. After receiving the instructions, the noise reduction device control system adjusts the working parameters of the noise reduction device according to the instruction content, such as the noise reduction intensity, frequency range, etc.; Step 1603. The noise reduction device performs noise reduction operations according to the adjusted parameters, changing the acoustic environment in the waiting area; Step 1604. The noise reduction effect is fed back to the AI cluster server as a reference for the next round of decision-making.

[0100] In this embodiment, the noise reduction devices in the hospital waiting area more finely and dynamically adjust the noise reduction strategy based on multi-dimensional perception data and patient-sensitive sound frequency data, effectively reducing the noisy sounds sensitive to patients in the waiting area, creating a quiet and comfortable waiting environment for patients and their families, alleviating patients' anxiety, improving the hospital service quality, and at the same time promoting the improvement of the noise reduction algorithm fineness and the development of intelligent applications in the medical service scenario; the specific effects are as follows: (1) For the sound frequencies sensitive to patients, the system can preferentially reduce the noise of these frequencies, reducing patients' discomfort.

[0101] (2) During peak hours with a large number of people, the system automatically enhances the noise reduction effect to maintain a quiet environment in the waiting area; (3) According to the characteristics of the waiting areas of different departments, the system can provide differentiated noise reduction strategies. For example, in the pediatric waiting area, it focuses on reducing high-frequency noise, and in the cardiology waiting area, it focuses on reducing specific frequency noises that may cause heart rate changes, etc.

[0102] (4) Overall reduce the noise level in the waiting area, creating a more conducive environment for patients to rest and for doctor-patient communication.

[0103] Example 5: This embodiment provides an environmentally friendly and noise-reducing AI cluster server, whose hardware composition structure includes: (1) Data acquisition layer: Multi-dimensional sensor array: includes sound sensors, temperature sensors, humidity sensors, human flow sensors, etc., for collecting various environmental data required in the foregoing embodiments; Data preprocessing unit: performs preprocessing such as noise reduction and filtering on the raw data collected by the sensors to improve data quality; Data cache module: temporarily stores the preprocessed data to ensure the continuity of data processing; (2) Computing and processing layer: Main controller: a processor with an ARM Cortex-A76 architecture, a main frequency of 3.0 GHz, responsible for system scheduling and basic operations; AI acceleration chip: uses a dedicated neural network processing unit (NPU), supports 16-bit floating-point operations, and is used to accelerate the calculation of deep learning models; Memory unit: 32GB DDR4 memory, used for data caching and model calculation, supporting high-speed data access; Storage unit: 512GB solid-state drive, storing system and model files to ensure data persistence; (3) Communication control layer: Network module: supports Gigabit Ethernet and Wi-Fi 6 communications to ensure real-time communication with noise reduction devices in various scenarios; Control interface: supports industrial control interfaces such as RS485 and CAN bus to adapt to the device control requirements of different scenarios; Power management unit: intelligently adjusts the system power consumption, supports the energy-saving mode, and achieves the energy efficiency goal of environmental protection and noise reduction.

[0104] An environmentally friendly and noise-reducing AI cluster server is provided, and its software function modules include: (1) Data acquisition module: Function: Real-time collection of environmental data, including sound, temperature, humidity, human flow, etc., corresponding to the data acquisition steps in the foregoing embodiments; Implementation method: Adopts multi-threaded parallel acquisition, and processes sensor data through the interrupt method to ensure data real-time performance; Data format: Stores data in a standardized JSON format for subsequent processing and analysis; (2) Data preprocessing module: Function: Cleans, standardizes, and extracts features from the raw data to prepare for multi-dimensional data fusion; Implementation method: Use the sliding window algorithm for data smoothing and adopt the Min-Max normalization method to unify the data scale; Processing flow: Data cleaning: Remove outliers and missing values to improve data quality; Data normalization: Unify data with different dimensions into the interval [0, 1] for easy model processing; Feature extraction: Extract time-domain and frequency-domain features to enhance data expression ability; (3)Multi-dimensional data fusion module: Function: Fuse data with different dimensions to generate feature vectors, corresponding to the multi-dimensional data fusion step in the foregoing embodiment; Implementation method: Adopt an improved DNN (Deep Neural Network) model to adapt to the data characteristics of different scenarios; Network structure: Input layer: Input nodes corresponding to data of each dimension, receiving preprocessed multi-dimensional data; Hidden layer 1: 256 neurons, using the ReLU activation function to extract low-level features; Hidden layer 2: 128 neurons, using the ReLU activation function to extract middle-level features; Hidden layer 3: 64 neurons, using the ReLU activation function to extract high-level features; Output layer: 32 neurons, using the Sigmoid activation function to generate the final feature vector; Loss function: Adopt the mean squared error (MSE) as the loss function to optimize model parameters; Optimization algorithm: Use the Adam optimizer with a learning rate set to 0.001 to balance the convergence speed and stability; (4)Noise reduction strategy decision-making module: Function: According to the fused feature vectors, decide the optimal noise reduction strategy, corresponding to the noise reduction strategy decision-making step in the foregoing embodiment; Implementation method: Adopt an improved DDPG (Deep Deterministic Policy Gradient) algorithm to adapt to the decision-making requirements of continuous action spaces; Network structure: 1) Actor network: Used to generate noise reduction strategies and map states to actions; 2) Critic network: Used to evaluate the value of the strategy and guide the optimization of the Actor network; 3) Reward function: Comprehensively consider the noise reduction effect, energy consumption, and environmental comfort , where , , are weight coefficients that can be adjusted according to different scenario requirements; (5)Policy Execution Module: Function: Convert the noise reduction policy of the decision-making into specific control instructions, corresponding to the policy execution steps in the foregoing embodiments; Implementation method: Adopt a fuzzy control algorithm to realize the mapping from the policy to the control quantity and improve the control accuracy; Control parameters: 1) Noise reduction intensity: A continuous value in the range of [0, 1], which controls the working intensity of the noise reduction device; 2) Frequency range: The adjustable range is 20 Hz - 20 kHz, and fine control is performed on noises of different frequencies; 3) Response time: ≤100 ms, ensuring the rapid response of the system to environmental changes.

[0105] The working process of the environmental protection noise reduction type AI cluster server provided in this embodiment is as follows: 1. System initialization: (1) Load the configuration file and set the system parameters; (2) Initialize each hardware module and check the system status; (3) Start the data acquisition thread and prepare to receive environmental data; 2. Data acquisition and preprocessing: (1) Collect data from each sensor in multiple threads to ensure the real-time nature of the data; (2) Preprocess and standardize the data to improve the data quality; (3) Temporarily store the processed data for subsequent processing; 3. Feature extraction and fusion: (1) Extract the features of the data in each dimension to enhance the data expression ability (2) Use the DNN model for feature fusion to generate comprehensive features; (3) Generate a feature vector as the input for the noise reduction policy decision-making; 4. Policy decision-making and execution: (1) Use the DDPG algorithm to generate a noise reduction policy and optimize the noise reduction effect; (2) Convert the policy into control instructions to adapt to the noise reduction device; (3) Execute the noise reduction operation to improve the environmental acoustic conditions.

[0106] The performance indicators of the environmental protection noise reduction type AI cluster server provided in this embodiment are as follows 1. Processing capacity: Data sampling rate: ≥1000 Hz, ensuring high-precision data acquisition; Response time: ≤100 ms, ensuring the rapid response of the system to environmental changes; Concurrent processing ability: Supports ≥100 data streams, meeting the requirements for large-scale deployment; 2. Noise reduction effect: (1) Noise reduction range: 20 Hz - 20 kHz, covering the audible range of the human ear; (2) Maximum noise reduction: ≥30 dB, significantly improving the acoustic environment; (3) Noise reduction uniformity: ±3 dB, ensuring the consistency of noise reduction effects across different frequency bands; 3. Energy consumption control: (1) Standby power consumption: ≤10 W, reducing energy consumption; (2) Maximum power consumption: ≤200 W, ensuring system performance; (3) Supports dynamic power consumption adjustment, optimizing energy usage according to actual needs.

[0107] The environmentally friendly noise reduction type AI cluster server provided in this embodiment, as the technical implementation basis of the foregoing embodiments, can meet the requirements of different application scenarios: (1) Application in transportation hubs: Collect environmental sound data in transportation hubs through high-precision sound sensors, combine with passenger flow data, use the multi-dimensional data fusion module to generate feature vectors, and the noise reduction strategy decision module outputs a noise reduction strategy suitable for the transportation hub environment based on the feature vectors, enhancing the travel experience of passengers; (2) Application in large event venues: The system can adjust the weight configuration of the multi-dimensional data fusion module according to event type data, generating differentiated noise reduction strategies for different types of events (such as concerts, sports events, etc.), retaining useful sounds while filtering out interfering noises; (3) Application in large shopping malls: Utilize the distributed multi-dimensional sensor array to collect environmental data in each area of the shopping mall, perform feature fusion through the deep neural network model, and combine with the reinforcement learning algorithm to provide customized noise reduction services for different commercial areas; (4) Application in hospital waiting areas: The system pays special attention to the sensitive sound frequency data of patients, applies the attention mechanism in the multi-dimensional data fusion module to make the network more focused on sensitive frequency data, and the noise reduction strategy decision module preferentially processes specific frequency noises that are sensitive to patients, creating a more comfortable waiting environment.

[0108] Through the collaborative work of the above-mentioned hardware structure, software modules, and work processes, the environmentally friendly noise reduction type AI cluster server of this embodiment can achieve intelligent environmental perception and noise reduction control. While ensuring the noise reduction effect, it can also achieve the goal of energy conservation and environmental protection. The modular design and standardized interfaces of the system endow it with good scalability and adaptability, enabling it to be applied to different scenario requirements.

[0109] The embodiments of the present invention have been described above, but these embodiments are not limited to the above specific implementation manners. The above specific implementation manners are merely illustrative rather than restrictive. Under the inspiration of these embodiments, those of ordinary skill in the art can also make many forms, all of which fall within the protection scope of these embodiments.

Claims

1. A method for using an environmentally friendly and noise-reducing AI cluster server, characterized in that, It includes the following steps: Obtain multi-dimensional perception data, which includes pedestrian flow data, environmental sound data, and temperature and humidity data; Construct a multi-dimensional data fusion model, and fuse the obtained multi-dimensional perception data through the multi-dimensional data fusion model to generate feature vectors; Establish a noise reduction strategy decision model, and based on the feature vectors, output a noise reduction strategy instruction through the noise reduction strategy decision model; Send the noise reduction strategy instruction to the noise reduction device to perform corresponding noise reduction operations, so as to realize the adaptive adjustment of the noise reduction strategy according to the dynamic changes of the environment.

2. The usage method of an environmentally friendly and noise-reducing AI cluster server according to claim 1, characterized in that The multi-dimensional data fusion model is implemented in one of the following ways: The weighted summation method, through the formula: ； Calculate the feature vector, Among them, is the weight set according to the data importance, is the normalized pedestrian flow data , is the normalized environmental sound data , is the normalized pedestrian flow data , is the normalized temperature data ; The multi-layer perceptron neural network method, which extracts and fuses the multi-dimensional data features through multi-layer non-linear transformation.

3. The usage method of an environmentally friendly and noise-reducing AI cluster server according to claim 1, characterized in that, The noise reduction strategy decision model is implemented in one of the following ways: The rule-based decision method, which outputs corresponding noise reduction strategy instructions according to the interval range of the feature vector values; The deep reinforcement learning method, which learns the optimal noise reduction strategy through the deep Q network or the deep deterministic policy gradient algorithm; The hybrid decision method, which combines the rule system and the reinforcement learning algorithm and maintains special attention to specific frequencies during the learning process.

4. The usage method of an environmentally friendly and noise-reducing AI cluster server according to claim 1, characterized in that, It also includes obtaining specific scenario-related data according to different application scenarios: In the traffic hub scenario, additionally obtain train / flight schedule flow data; In the large event venue scenario, additionally obtain event type data; In the shopping mall scenario, additionally obtain store sound data; In the hospital waiting area scenario, additionally obtain patient sensitive sound frequency data.

5. The usage method of an environmentally friendly and noise-reducing AI cluster server according to claim 4, characterized in that, In the hospital waiting area scenario, apply the attention mechanism to the patient sensitive sound frequency data: Through the formula: ； Calculate the attention weight; Through the formula: ； Obtain the weighted sensitive frequency data; Input the weighted sensitive frequency data into the multi-dimensional data fusion model to make the system give priority to the frequencies sensitive to patients.

6. The usage method of an environmentally friendly and noise-reducing AI cluster server according to claim 1, characterized in that The noise reduction strategy instruction includes: The noise reduction intensity instruction, which controls the working intensity of the noise reduction device; The frequency range instruction, which performs fine control on noises of different frequencies; The special frequency processing instruction, which performs additional enhancement or weakening processing on noises of specific frequencies.

7. An environmentally friendly and noise-reducing AI cluster server and its usage method according to claim 1, characterized in that, It also includes the following steps: Receive the effect feedback after the noise reduction device performs the noise reduction operation; Update the parameters of the noise reduction strategy decision model based on the effect feedback; Use the updated model parameters to generate optimized noise reduction strategy instructions in the next round of decision-making.

8. An environmentally friendly and noise-reducing AI cluster server for implementing the usage method of the environmentally friendly and noise-reducing AI cluster server according to any one of claims 1-7, characterized in that, It includes: The data acquisition layer, which includes a multi-dimensional sensor array and a data preprocessing unit, and is used to obtain and preprocess multi-dimensional perception data; The computing and processing layer, which includes a main controller, an AI acceleration chip, a memory unit, and a storage unit, and is used to perform multi-dimensional data fusion and noise reduction strategy decision-making; The communication control layer, which includes a network module and a control interface, and is used to communicate with the noise reduction device and send noise reduction strategy instructions.

9. An environmentally friendly and noise-reducing AI cluster server according to claim 8, characterized in that, The computing and processing layer also includes: The data preprocessing module, which is used to clean, standardize, and extract features from the original data; The multi-dimensional data fusion module, which is used to fuse data of different dimensions to generate feature vectors; The noise reduction strategy decision module, which is used to decide the optimal noise reduction strategy according to the feature vectors; The strategy execution module, which is used to convert the decided noise reduction strategy into specific control instructions.

10. An environmentally friendly and noise-reducing AI cluster server according to claim 8, characterized in that, The AI acceleration chip is a dedicated neural network processing unit that supports 16-bit floating-point operations and is used to accelerate the calculation of deep learning models; The server further includes a power management unit that can intelligently adjust the system power consumption, supports the energy-saving mode, and achieves the energy efficiency goal of environmental protection and noise reduction.

Images