A method for regulating the speed of a low-noise motor fan

Abstract

The application relates to the technical field of motor fan control, and discloses a low-noise motor fan rotating speed adjusting method.The method trains a fan rotating speed adjusting model based on a deep learning algorithm of an attention mechanism, and uses historical in-vehicle environment data and air conditioner working state data as training samples.In-vehicle temperature, humidity and air conditioner state data are collected in real time, input into the model, the target rotating speed of the fan is output, and the fan is automatically adjusted to the rotating speed.Meanwhile, the passenger adjustment is recorded through a front seat camera, and individualized fan rotating speed setting is realized.In addition, the application also introduces an environment perception compensation system, acquires vehicle external environment data in real time, adds an environment compensation coefficient in the model, dynamically updates the model by using a federal learning framework, and improves the generalization ability and real-time performance of the model.In extreme environment conditions, a safe operation mode is activated, and the safety operation of the vehicle and the comfort of passengers are ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of motor fan control, and in particular to a speed regulation method for a low-noise motor fan. Background Art

[0002] In automotive interior climate control systems, motor fan speed regulation is crucial for maintaining comfortable temperature and humidity inside the vehicle. Traditional fan speed regulation methods typically rely on simple temperature sensors and fixed speed settings, failing to intelligently adjust to changes in the actual interior environment and the personal preferences of passengers. With the advancement of intelligent technology, intelligent control systems based on deep learning algorithms are gradually being applied to the automotive field. However, existing intelligent control systems still have shortcomings in fan speed regulation, such as insufficient model training, low data utilization, and an inability to personalize adjustments based on passenger preferences. Therefore, a low-noise motor fan control method is needed that can intelligently adjust the speed based on in-vehicle environmental data and passenger preferences. Summary of the Invention

[0003] In response to the above-mentioned technical deficiencies, the purpose of the present invention is to provide a method for adjusting the speed of a low-noise motor fan to solve the problem in the prior art that the fan speed adjustment has a low degree of intelligence and cannot be automatically and accurately adjusted according to the actual environmental changes in the vehicle and the personal preferences of the passengers.

[0004] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0005] In a first aspect, the present invention provides a method for adjusting the speed of a low-noise motor fan, the method comprising:

[0006] Step S100: using historical in-car temperature data, humidity data, and car air conditioning operating status data as training samples, a fan speed adjustment model is obtained through training using a deep learning algorithm based on an attention mechanism;

[0007] Step S200: collecting the current temperature data, humidity data and the working status data of the car air conditioner in real time, inputting the real-time data into the fan speed adjustment model, and outputting the fan target speed;

[0008] Step S300: Automatically adjusting the speed of the low-noise motor fan to the target speed according to the fan target speed;

[0009] Step S400: The camera in front of the seat records the current seat occupant and the adjustment status of the motor fan speed. When the same seat occupant sits again, the adjustment record is read and the motor fan speed is automatically adjusted to the preferred setting.

[0010] Preferably, in a possible implementation manner of the first aspect, the fan speed adjustment model structure includes:

[0011] The input layer is used to receive historical in-vehicle temperature data, humidity data, and car air conditioning working status data;

[0012] The embedding layer encodes the data of the input layer and converts it into a dense vector representation in a high-dimensional space;

[0013] Attention mechanism layer, which assigns different weights based on the importance of input data features;

[0014] Hidden layer, which contains multiple neural network layers and is used to extract deep features of input data and integrate information through nonlinear transformation;

[0015] The output layer maps the data features extracted by the hidden layer to the fan target speed value and outputs the target speed.

[0016] Preferably, in a possible implementation of the first aspect, a dynamic convolution module is deployed in the hidden layer to dynamically adjust the convolution kernel parameters according to the weight output by the attention mechanism layer, specifically including:

[0017] In each neuron node of the hidden layer, a deformable convolution kernel is embedded, and its shape parameters are dynamically calculated by the attention weight;

[0018] The deformation amplitude of the convolution kernel is constrained by the gated recurrent unit to ensure the effectiveness of the convolution operation;

[0019] The dynamic convolution output is fused with the standard convolution output using residual connection to form a hybrid feature expression.

[0020] Preferably, in a possible implementation of the first aspect, step S200 specifically includes:

[0021] Through temperature sensors, humidity sensors and air conditioning status monitoring devices, the current temperature data, humidity data and working status data of the car air conditioner are captured in real time;

[0022] Perform data preprocessing on the collected real-time data, including data cleaning and outlier detection;

[0023] The pre-processed real-time data is input into the trained fan speed control model. The model automatically calculates and outputs the fan target speed required in the current environment based on the input data.

[0024] Verify the fan target speed output by the model to ensure it is within the preset reasonable range.

[0025] Preferably, in a possible implementation of the first aspect, step S300 specifically includes:

[0026] receiving a fan target speed value output by a fan speed regulation model, and using the fan target speed value as a desired fan operating speed;

[0027] Compare the actual speed of the current low-noise motor fan with the target speed to determine the direction and amplitude of speed adjustment;

[0028] According to the comparison result, a speed adjustment instruction is sent to the low-noise motor fan, and the instruction includes a speed adjustment amount;

[0029] The motor adjusts the actual speed of the fan to the target speed according to the received command;

[0030] During the adjustment process, the actual speed of the fan is monitored in real time to ensure that it is within the preset error range with the target speed.

[0031] Preferably, in a possible implementation of the first aspect, step S400 specifically includes:

[0032] Perform facial recognition through the camera in front of the seat to detect and record the information of the person currently sitting in the seat;

[0033] matching the detected person information with previously stored passenger adjustment records, the adjustment records including each passenger's preference setting for the motor fan speed;

[0034] When it is recognized that the same seat occupant is sitting again, the corresponding preference settings are retrieved from the database;

[0035] Use the fan target speed as the initial setting and adjust the low-noise motor fan speed according to your preferences.

[0036] Preferably, in a possible implementation of the first aspect, when the face recognition algorithm recognizes a passenger for the first time, the facial features of the passenger are collected and recorded in a database, and the passenger's preference settings are recorded;

[0037] When the facial features of a certain passenger are identified to match the facial feature records in the database, the corresponding preference settings are retrieved from the database.

[0038] Preferably, in a possible implementation manner of the first aspect, the preference setting includes:

[0039] Read the current fan target speed and set it as the initial reference;

[0040] When a passenger gets on the vehicle for the first time, the fan speed adjusted by the passenger is read, the deviation between the adjusted fan speed and the target fan speed is calculated, and the deviation is recorded in the database as the passenger's preference setting;

[0041] When a passenger takes a ride multiple times, the preference is set to the average of the multiple deviation values.

[0042] Preferably, in a possible implementation manner of the first aspect, the method further includes an environment perception compensation system:

[0043] Obtain real-time external environmental data through the Internet of Vehicles, including weather conditions, road congestion index, and air pollution index;

[0044] Add an environmental compensation coefficient calculation module to the fan speed adjustment model to perform multimodal fusion of external environmental data and in-vehicle sensor data;

[0045] A federated learning framework is used to dynamically update the fan speed control model, and distributed training of the vehicle group's control data is performed through edge computing devices.

[0046] When extreme environmental conditions are detected, the safe operating mode is activated, limiting the fan speed adjustment range and synchronously adjusting the air conditioning compressor operating frequency.

[0047] The beneficial effects of the present invention are as follows: by training a fan speed control model using a deep learning algorithm based on an attention mechanism, it can accurately capture changes in complex factors such as in-vehicle temperature, humidity, and air conditioning operating status, enabling intelligent fan speed adjustment. Real-time in-vehicle environmental data is collected and input into the model, which outputs a target fan speed that matches the current environment, improving the accuracy and timeliness of fan speed adjustment. Seat-mounted cameras record passenger adjustments, allowing for personalized fan speed settings for each passenger, enhancing the passenger experience.

[0048] In addition, the present invention introduces an environmental perception compensation system that can acquire real-time data about the vehicle's external environment. It also adds an environmental compensation coefficient calculation module to the model, making fan speed adjustment more adaptable to changes in the external environment. The model is dynamically updated using a federated learning framework and distributed training is performed using edge computing devices, improving the model's generalization and real-time performance. In extreme environmental conditions, the safe operating mode is activated, limiting the fan speed adjustment range and adjusting the air conditioning compressor operating frequency to ensure safe vehicle operation and passenger comfort. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to make the technical solutions in the embodiments of the present application or the prior art clearer, the accompanying drawings needed in the embodiments or prior art description will be briefly introduced below. Obviously, the accompanying drawings in the following description only represent some of the embodiments of the present application, and all other embodiments obtained by those of ordinary skill in the art without creative efforts based on these drawings also belong to the protection scope of the present application.

[0050] Figure 1 A flowchart of a method for speed regulation of a low-noise motor fan is provided for the present application. DETAILED DESCRIPTION

[0051] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments only represent some of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts also belong to the protection scope of the present application.

[0052] Embodiment one: as shown in the present application, a method for speed regulation of a low-noise motor fan is provided, which comprises: Figure 1

[0053] Step S100: using historical in-vehicle temperature data, humidity data and automobile air conditioner working state data as training samples, a fan speed regulation model is trained by a deep learning algorithm based on an attention mechanism.

[0054] Specifically, the historical in-vehicle temperature data, humidity data and automobile air conditioner working state data are obtained through a car control system. The fan speed regulation model structure includes: an input layer for receiving historical in-vehicle temperature data, humidity data and automobile air conditioner working state data; an embedding layer for encoding the data of the input layer and converting it into a dense vector representation in a high-dimensional space; an attention mechanism layer for assigning different weights according to the importance of the input data features; a hidden layer containing multiple neural network layers for extracting deep features of the input data and integrating information through nonlinear transformation; and an output layer for mapping the data features extracted by the hidden layer to the target speed value of the fan and outputting the target speed.

[0055] A dynamic convolution module is deployed in the hidden layer to dynamically adjust the convolution kernel parameters according to the weights output by the attention mechanism layer, which specifically includes: a deformable convolution kernel is embedded at each neuron node in the hidden layer, and the shape parameters of the convolution kernel are dynamically calculated and generated by the attention weights; the deformation amplitude of the convolution kernel is constrained through a gated recurrent unit to ensure the effectiveness of the convolution operation; and a residual connection method is used to fuse the dynamic convolution output and the standard convolution output to form a hybrid feature representation.

[0056] ​In this embodiment, historical in-vehicle temperature data, humidity data and automobile air conditioning working state data are used as training samples to build a prediction model based on an improved attention mechanism deep learning framework, which includes five parts: data input layer, feature embedding layer, attention weighting layer, dynamic convolution hidden layer and speed output layer. The real-time in-vehicle environmental parameters collected by the automobile control system are transmitted to the model input end through the CAN bus, and the target fan speed instruction is generated after multi-layer feature conversion. The specific implementation process is as follows:

[0057] The training samples are derived from the historical running data collected by the vehicle-mounted sensors, including temperature sensor readings, humidity sensor signals and air conditioner compressor working state codes in time series form. In the data preprocessing stage, the sliding window method is used to construct the sample sequence, the window length is set to 20 time steps, the step size is set to 5 seconds, and each sample contains the environmental parameter variation characteristics of 100 seconds. The input layer is designed as a multi-channel data structure, corresponding to temperature, humidity and air conditioning state three independent channels respectively, and zero mean standardization is used to eliminate the dimension difference.

[0058] The embedding layer uses a learnable encoding method to realize data space mapping. A 32-dimensional embedding vector is designed for continuous variables such as temperature and humidity, and a 16-dimensional one-hot encoding is used for discrete variables such as air conditioning working state. The embedding process introduces a feature interaction mechanism, which sets cross-channel correlation parameters in the embedding matrix and establishes temperature-humidity correlation features, air conditioning mode-temperature coupling features and other combined expressions through matrix dot product operations. The encoded feature vector is elevated to a unified dimensional space through a fully connected network to form a fused feature tensor with a shape of [batch_size, seq_len, 64].

[0059] The attention mechanism layer uses a multi-head scaled dot-product attention architecture with 4 independent attention heads working in parallel. The query matrix Q, key matrix K and value matrix V of each attention head are extracted from the input features through linear transformation, where the dimension of Q matrix is set to 256 and the dimensions of K and V matrices are set to 128. The attention weight calculation introduces a temperature coefficient adjustment mechanism to dynamically adjust the exponential scaling amplitude of the softmax function, enhancing the response sensitivity to sudden environmental changes. The weighted feature vector is normalized by layer after residual connection with the original input to form a context vector with feature selection enhancement.

[0060] The dynamic convolution hidden layer uses a gated deformable convolution architecture for feature extraction, integrating an attention-guided mechanism with a standard convolutional neural network. Each hidden layer contains 32 deformable convolution kernels, whose spatial deformation parameters are dynamically generated by the attention weight matrix. The kernel offset generator is constructed using a two-layer fully connected network. The first layer maps the 64-dimensional attention features to a 128-dimensional intermediate space, and the second layer outputs a coordinate offset that is equal to the square of the kernel size. A gating mechanism generates a spatial mask using a sigmoid function to constrain the offset, ensuring that the effective receptive field of the convolution operation does not exceed 7x7 pixels. The convolution output features are compressed by depthwise separable convolution and then fused with the standard convolution path output in a channel-wise weighted manner. The fusion coefficients are dynamically calculated using a gated recurrent unit.

[0061] The residual connection module uses a cross-layer feature reuse strategy, introducing skip connections after each set of dynamic convolution operations. An adaptive weighting mechanism is designed for the feature fusion stage, which nonlinearly combines the original features and the convolution output using a learnable parameter matrix. The combination weights are generated by applying a sigmoid function to the L2 norm of the current feature map.

[0062] The output layer is designed as a multi-scale prediction structure, consisting of an immediate speed prediction branch and a trend prediction branch. The immediate branch uses a fully connected layer to directly map the final hidden features to speed values. The trend branch uses a temporal convolutional network to extract the speed gradient for the next five time steps. The outputs of these two branches are weighted and summed along the channel dimension, with the weight parameters dynamically adjusted based on the current air conditioner operating mode: in rapid cooling mode, the immediate branch weight is increased to 0.7, while in energy-saving mode, the trend branch weight is increased to 0.6. The final output layer uses a hyperbolic tangent function to constrain the predicted value to the interval [0, 1] and, after linear transformation, matches the actual fan speed range.

[0063] Model training utilizes a phased optimization strategy. Initially, the deformation parameters of the dynamic convolution module are frozen, and pre-training with a standard convolutional network is performed for 100 rounds. In the second phase, the gated unit parameters are unfrozen and fine-tuned for 50 rounds using a cosine annealing learning rate scheduling algorithm, with a base learning rate set to 3e-4. In the final phase, all-parameter joint training is enabled, with dynamic gradient clipping introduced and the gradient norm capped at 0.5 to prevent parameter oscillation. The loss function is designed as a composite function, consisting of the mean squared error of speed prediction, a smoothness constraint for trend prediction, and a regularization term for the convolution kernel deformation, with a weight ratio of 5:2:1. A data augmentation strategy incorporates random masking, randomly masking local segments of the input sequence with a 20% probability, to improve the model's tolerance to sensor failures.

[0064] Step S200: collecting the current temperature data, humidity data and working status data of the car air conditioner in real time, inputting the real-time data into the fan speed adjustment model, and outputting the fan target speed.

[0065] Specifically, temperature sensors, humidity sensors, and air conditioning status monitoring devices are installed inside the vehicle. These sensors and monitoring devices capture real-time data on the vehicle's temperature and humidity, as well as the operating status of the air conditioning. This raw data often contains noise and outliers, so data preprocessing is required before inputting it into the model.

[0066] During the data preprocessing phase, data cleaning is performed to remove invalid or missing values ​​due to sensor failure or data transmission errors. Next, outlier detection is performed to identify and address values ​​that significantly deviate from the normal range, ensuring that the data input into the model is accurate and reliable.

[0067] In this embodiment, data cleaning is performed first. First, the df.dropna method is used to remove rows containing NaN values ​​to ensure the accuracy of subsequent analysis. Next, outlier detection is performed. For temperature data, the normal range is set to 0 to 40 degrees, and values ​​outside this range are considered outliers. For humidity data, we set the normal range to 5% to 90%. Similarly, any value outside this range is considered an outlier. For detected outliers, we choose to replace them with the upper and lower limits of the corresponding range to ensure the rationality and consistency of the data.

[0068] The preprocessed real-time data's format and dimensions already match the fan speed control model's input requirements, allowing it to be directly fed into the trained model. Within the model, the embedding layer converts the data into a dense vector representation in a high-dimensional space. Subsequently, the attention mechanism layer assigns weights based on the importance of input data features, emphasizing those that are most critical for fan speed control. The hidden layer further extracts deeper features and integrates them through nonlinear transformations. Finally, the output layer maps these features into the fan target speed value and outputs it.

[0069] To ensure that the fan target speed output by the model meets actual requirements and is within a reasonable range, it is verified. This includes comparing it with the preset speed range and backtesting using historical data to evaluate the accuracy and stability of the model output.

[0070] In this example, the speed range is first set to 0 to 2000 rpm. The model then outputs a target fan speed based on real-time in-vehicle environmental data (temperature, humidity, and air conditioning status). This target speed is then compared with the preset range to ensure that the output is reasonable and meets actual requirements. Next, backtesting is performed using historical data, including past in-vehicle environmental data and corresponding fan speed records. After preprocessing, this historical data is input into the model to obtain the target speed. This target speed is then compared with the actual speed, and the model's accuracy is evaluated using metrics such as error rate and accuracy.

[0071] Step S300: Automatically adjust the speed of the low-noise motor fan to the target speed according to the fan target speed.

[0072] Specifically, the fan target speed value output by the fan speed regulation model is received and used as the expected fan operating speed; the actual speed of the current low-noise motor fan is compared with the target speed to determine the direction and amplitude of the speed regulation; based on the comparison result, a speed regulation instruction is sent to the low-noise motor fan, and the instruction includes a speed adjustment amount; the motor adjusts the actual speed of the fan to the target speed according to the received instruction; during the adjustment process, the actual speed of the fan is monitored in real time to ensure that it is within a preset error range with the target speed.

[0073] In this embodiment, after receiving the fan target speed value (set to N rpm) output by the fan speed control model, the system obtains the actual speed of the low-noise motor fan (set to M rpm). M is compared with N to determine the direction and magnitude of the speed adjustment. If M is less than N, the system generates an acceleration instruction containing the adjustment required to increase the speed to N. If M is greater than N, a deceleration instruction is generated containing the adjustment required to reduce the speed to N.

[0074] The motor gradually adjusts the fan's actual speed based on the received command through its internal speed regulation mechanism. During the adjustment process, the system monitors the fan's actual speed in real time to ensure that the error between the actual speed and the target speed N remains within a preset range of ±10 rpm.

[0075] Step S400: The camera in front of the seat records the current seat occupant and the adjustment status of the motor fan speed. When the same seat occupant sits again, the adjustment record is read and the motor fan speed of the person is automatically adjusted to his or her preferred setting.

[0076] Specifically, facial recognition is performed through a camera in front of the seat to detect and record the information of the person currently occupying the seat; the detected person information is matched with the previously stored passenger adjustment records, which include each passenger's preference settings for the motor fan speed; when it is recognized that the same person is sitting in the seat again, the corresponding preference settings are retrieved from the database; the fan target speed is used as the initial setting, and the speed of the low-noise motor fan is adjusted according to the preference settings.

[0077] When the facial recognition algorithm recognizes a passenger for the first time, it collects the passenger's facial features and records them in the database, and also records the passenger's preference settings; when the facial features of a passenger are identified and match the facial feature records in the database, the corresponding preference settings are retrieved from the database.

[0078] The preference setting step includes reading the current fan target speed and setting the current fan target speed as the initial benchmark; when the passenger rides the vehicle for the first time, reading the fan speed adjusted by the passenger, calculating the deviation between the adjusted fan speed and the fan target speed, and recording this deviation as the passenger's preference setting in the database; when the passenger rides the vehicle multiple times, the preference setting is the average of the multiple deviation values.

[0079] In this embodiment, the camera in front of the seat uses an IR monocular camera and employs facial recognition technology to capture and identify the person currently occupying the seat in real time. Through the facial recognition algorithm, the system can detect and record the passenger's information, including facial features and identity information.

[0080] This embodiment utilizes the ArcFace algorithm. The ArcFace algorithm is a deep learning-based facial recognition method that extracts facial features by building a deep neural network and uses these features for facial recognition and comparison. When the system detects a passenger, it matches their facial features with passenger adjustment records stored in a database. These adjustment records include each passenger's previous fan speed preferences. These preferences were generated by manually adjusting the fan speed during previous rides and saving the adjustment results.

[0081] To automatically read and adjust passenger preferences, a passenger preference database is established. This database stores each passenger's facial features, identification information, and their motor fan speed preferences. When the system identifies a passenger's facial features that match those in the database, it automatically retrieves their corresponding preferences from the database. When the same seat is recognized again, the system automatically reads their preferences and uses the target fan speed as the initial setting. The low-noise motor fan speed is then adjusted based on the preferences.

[0082] The preference adjustment process is as follows: First, the system reads the current fan target speed and sets it as the initial baseline. Then, when a passenger rides for the first time, the system reads the adjusted fan speed based on the passenger's adjustment. Next, the system calculates the deviation between the adjusted fan speed and the target fan speed and records this deviation as the passenger's preference in the database. This deviation reflects the passenger's personalized fan speed requirements.

[0083] When a passenger takes multiple rides, the system compiles statistics and averages their preferences. Specifically, the system calculates the average of these deviations and uses this average as the passenger's final preference. During fan speed adjustment, the system monitors the actual fan speed in real time and makes fine adjustments based on the preferences to ensure it remains consistent with the passenger's settings. The system also records the results of each adjustment and passenger feedback to continuously optimize and improve the fan speed adjustment algorithm.

[0084] Embodiment 2: The present invention provides a method for adjusting the speed of a low-noise motor fan, which also includes an environmental perception compensation system.

[0085] Specifically, the vehicle's external environmental data, including weather conditions, road congestion index, and air pollution index, is obtained in real time through the Internet of Vehicles. An environmental compensation coefficient calculation module is added to the fan speed adjustment model to perform multimodal fusion of external environmental data and in-vehicle sensor data. The fan speed adjustment model is dynamically updated using a federated learning framework, and distributed training of the adjustment data of the vehicle group is performed through edge computing devices. When extreme environmental conditions are detected, the safe operation mode is activated to limit the fan speed adjustment range and synchronously adjust the air-conditioning compressor operating frequency.

[0086] In this embodiment, the connected vehicle system accesses the meteorological service cloud platform and the traffic management center database via the 5G-V2X communication protocol, collecting real-time detailed environmental parameters such as precipitation intensity, wind speed and direction, and PM2.5 concentration at the vehicle's location. The environmental compensation coefficient calculation module uses an attention mechanism to extract features from heterogeneous data, normalizing discrete weather codes with continuous pollution indices to generate a dynamic compensation factor ranging from 0.8 to 1.2. This factor is combined with operating condition data from the vehicle's in-vehicle temperature sensor and battery thermal management system through a weighted fusion algorithm to form a composite cooling demand index, which serves as a reference input for fan speed adjustment.

[0087] A federated learning framework is employed to continuously optimize model parameters. Vehicle terminals upload local training gradients rather than raw data via edge computing nodes, ensuring privacy and improving model generalization. Differential privacy techniques are introduced during training to add Gaussian noise to gradients, ensuring convergence stability in heterogeneous data scenarios. After the edge server aggregates model updates for a group of vehicles, it pushes the optimized adjustment strategy to the fleet over-the-air (OTA) process, creating a closed-loop learning mechanism of "local training and global sharing."

[0088] When the system detects that the environmental parameters exceed the preset threshold, such as encountering a sandstorm or extremely high temperature weather, the safe operation mode immediately activates the dual protection mechanism. On the one hand, the PID controller limits the fan speed fluctuation to The system maintains a stable operating range, preventing mechanical resonance caused by drastic speed adjustments. Furthermore, it establishes a collaborative channel with the air conditioning control system, adjusting the compressor operating frequency proportionally to the actual fan speed to maintain balanced cabin pressure. This mode prioritizes heat dissipation for key components, maximizing heat dissipation efficiency within a limited speed range by dynamically adjusting the angle of the air deflector at the air conditioning outlet.

[0089] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A method for adjusting the speed of a low-noise motor fan, characterized in that: The method comprises: Step S100: using historical in-car temperature data, humidity data, and car air conditioning operating status data as training samples, a fan speed adjustment model is obtained through training using a deep learning algorithm based on an attention mechanism; Step S200: collecting the current temperature data, humidity data and the working status data of the car air conditioner in real time, inputting the real-time data into the fan speed adjustment model, and outputting the fan target speed; Step S300: Automatically adjusting the speed of the low-noise motor fan to the target speed according to the fan target speed; Step S400: Using a camera in front of the seat, the current seat occupant and the motor fan speed adjustment status are recorded. When the same seat occupant occupies the seat again, the adjustment record is read and the motor fan speed is automatically adjusted to the preferred setting. The method further includes an environment perception compensation system: Obtain real-time external environmental data through the Internet of Vehicles, including weather conditions, road congestion index, and air pollution index; Add an environmental compensation coefficient calculation module to the fan speed adjustment model to perform multimodal fusion of external environmental data and in-vehicle sensor data; A federated learning framework is used to dynamically update the fan speed control model, and distributed training of the vehicle group's control data is performed through edge computing devices. When extreme environmental conditions are detected, the safe operating mode is activated, limiting the fan speed adjustment range and synchronously adjusting the air conditioning compressor operating frequency.

2. The speed regulation method according to claim 1, wherein: The fan speed adjustment model structure includes: The input layer is used to receive historical in-vehicle temperature data, humidity data, and car air conditioning working status data; The embedding layer encodes the data of the input layer and converts it into a dense vector representation in a high-dimensional space; Attention mechanism layer, which assigns different weights based on the importance of input data features; Hidden layer, which contains multiple neural network layers and is used to extract deep features of input data and integrate information through nonlinear transformation; The output layer maps the data features extracted by the hidden layer to the fan target speed value and outputs the target speed.

3. The speed adjustment method according to claim 2, wherein: The dynamic convolution module is deployed in the hidden layer to dynamically adjust the convolution kernel parameters according to the weight output by the attention mechanism layer, specifically including: In each neuron node of the hidden layer, a deformable convolution kernel is embedded, and its shape parameters are dynamically calculated by the attention weight; The deformation amplitude of the convolution kernel is constrained by the gated recurrent unit to ensure the effectiveness of the convolution operation; The dynamic convolution output is fused with the standard convolution output using residual connection to form a hybrid feature expression.

4. The speed regulation method according to claim 1, wherein: The step S200 specifically includes: Through temperature sensors, humidity sensors and air conditioning status monitoring devices, the current temperature data, humidity data and working status data of the car air conditioner are captured in real time; Perform data preprocessing on the collected real-time data, including data cleaning and outlier detection; The pre-processed real-time data is input into the trained fan speed control model. The model automatically calculates and outputs the fan target speed required in the current environment based on the input data. Verify the fan target speed output by the model to ensure it is within the preset reasonable range.

5. The speed regulation method according to claim 1, wherein: The step S300 specifically includes: receiving a fan target speed value output by a fan speed regulation model, and using the fan target speed value as a desired fan operating speed; Compare the actual speed of the current low-noise motor fan with the target speed to determine the direction and amplitude of speed adjustment; According to the comparison result, a speed adjustment instruction is sent to the low-noise motor fan, and the instruction includes a speed adjustment amount; The motor adjusts the actual speed of the fan to the target speed according to the received command; During the adjustment process, the actual speed of the fan is monitored in real time to ensure that it is within the preset error range with the target speed.

6. The speed regulation method according to claim 1, wherein: The step S400 specifically includes: Perform facial recognition through the camera in front of the seat to detect and record the information of the person currently sitting in the seat; matching the detected person information with previously stored passenger adjustment records, the adjustment records including each passenger's preference setting for the motor fan speed; When it is recognized that the same seat occupant is sitting again, the corresponding preference settings are retrieved from the database; Use the fan target speed as the initial setting and adjust the low-noise motor fan speed according to your preferences.

7. The speed regulation method according to claim 6, wherein: When the facial recognition algorithm recognizes a passenger for the first time, it collects the passenger's facial features and records them in the database, and also records the passenger's preference settings; When the facial features of a certain passenger are identified to match the facial feature records in the database, the corresponding preference settings are retrieved from the database.

8. The speed regulation method according to claim 7, wherein: The preferences include: Read the current fan target speed and set it as the initial reference; When a passenger gets on the vehicle for the first time, the fan speed adjusted by the passenger is read, the deviation between the adjusted fan speed and the target fan speed is calculated, and the deviation is recorded in the database as the passenger's preference setting; When a passenger takes a ride multiple times, the preference is set to the average of the multiple deviation values.

Images