Vehicle road noise cancellation reference signal collection method and system based on ai recognition

Abstract

The present application belongs to the field of vehicle noise reduction technology, and specifically relates to a vehicle road noise cancellation reference signal collection method and system based on AI recognition. The present application obtains the vehicle model category information of the vehicle and the road section identification information of the target road section, obtains a reference signal prediction model trained based on historical noise reduction data of vehicles of the same model on the road section from a cloud server, loads the model when the vehicle drives to the target road section, determines the initial coefficients of the local adaptive filter based on the predicted noise signal output by the model to generate an initial reference signal for noise reduction, collects residual noise signals through an error microphone during driving, inputs the preset deviation prediction neural network to output filter coefficient deviation in real time, fine-tunes the filter coefficients to generate a corrected reference signal, uploads the deviation to the cloud after driving away, and performs incremental updating on the model, thereby improving the accuracy of the reference signal and the noise reduction effect.

Description

Technical Field

[0001] This invention belongs to the field of vehicle noise reduction technology, specifically relating to a method and system for acquiring reference signals for vehicle road noise cancellation based on AI recognition. Background Technology

[0002] Road noise generated by the interaction between the tires and the road surface during vehicle operation is one of the main sources of in-vehicle noise. Active road noise control technology cancels out road noise by emitting anti-phase sound waves with the opposite phase from the road noise through speakers. Its noise reduction effect is highly dependent on the quality of the reference signal.

[0003] Existing technologies include methods that utilize neural networks to predict reference signals. For example, Chinese patent application CN119274526A discloses an active road noise control method for vehicles based on Elman neural networks. This method trains an Elman neural network by acquiring vibration acceleration data and in-vehicle noise data, uses the predicted noise signal output by the neural network as a reference signal input to a filter, and adjusts the filter weight parameters using the error signal. However, this method is based on independent operation of a single vehicle, with each model trained and used independently, making it difficult for vehicles to share road noise reduction experience. Once deployed, the model remains fixed and cannot be continuously optimized using actual driving data. Due to differences in suspension wear, manufacturing tolerances, and tire configurations, the same model cannot adapt to the individual physical characteristics of all vehicles. Furthermore, this method does not adequately consider the differences in noise characteristics across different road sections and lacks customized processing for specific road segments.

[0004] Furthermore, existing technologies also include methods for storing noise reduction parameters in the cloud for sharing among multiple vehicles. For example, Chinese patent application CN110211561A discloses a method and device for continuously optimized road noise cancellation, which stores noise reduction parameters such as filter parameters and microphone gain in a cloud database for vehicles to access based on location and road type. However, this type of method stores and shares noise reduction parameters rather than neural network models; it is essentially "parameter access" rather than "model sharing." Different vehicle models have significant differences in physical characteristics such as chassis structure and suspension stiffness, making it difficult for the same set of parameters to be adapted to different models. Moreover, the optimization of this type of method is limited to adjusting existing parameter values, making it difficult to achieve continuous evolution at the model level, thus limiting the ability to improve noise reduction performance.

[0005] Therefore, how to enable the vehicle's reference signal prediction model to learn from the training results of other vehicles on the same road segment, and to make personalized fine-tuning of the model using the actual noise reduction effect during driving, while feeding the fine-tuning results back to the cloud for subsequent vehicles to use, has become an urgent technical problem to be solved. Summary of the Invention

[0006] The purpose of this invention is to provide a vehicle road noise cancellation reference signal acquisition method and system based on AI recognition. This method acquires the vehicle's model type information and the road segment identification information of the target road segment, and sends a model acquisition request to a cloud server. This allows the vehicle to obtain an AI prediction model trained on noise reduction data accumulated by other vehicles of the same model on that road segment before entering the target segment. This solves the problems of existing technologies where independent operation of a single-vehicle AI model leads to difficulties in sharing road segment experience and large initial output deviations. The method loads the reference signal prediction model from the cloud into the local machine and determines the initial coefficients of the filter based on its output prediction value, while simultaneously integrating real-time sensor data from the vehicle to generate parameters. By examining the signal, the vehicle retains the ability to make personalized corrections based on its own physical characteristics while inheriting the training results of the collective AI model, thus solving the problem that a unified AI model is difficult to adapt to the individual differences of different vehicles. By using a pre-set deviation prediction neural network to perform real-time inference on residual noise signals and output the deviation of filter coefficients during driving, the filter coefficients can be rapidly fine-tuned at the millisecond level, solving the problem that existing models cannot respond in time when road characteristics change abruptly. By feeding back the deviation generated during the actual operation of the vehicle to the cloud and incrementally updating the cloud model, the prediction accuracy of the cloud AI model continues to improve with the accumulation of data, realizing the continuous evolution of the AI ​​model.

[0007] The specific technical solution adopted by this invention is as follows: A method for acquiring vehicle road noise cancellation reference signals based on AI recognition, comprising: Obtain the vehicle type information and road sign information for the target road segment; The cloud server sends a request to obtain a reference signal prediction model based on the vehicle type information and the road segment identification information, and receives the reference signal prediction model returned by the cloud server. The reference signal prediction model is generated by the cloud server based on the vehicle type information and the road segment identification information. When the vehicle travels to the target road section, the vibration acceleration signal collected in real time is input into the reference signal prediction model. The initial coefficients of the local adaptive filter are determined based on the predicted noise signal output by the reference signal prediction model. The initial reference signal is generated by combining the vibration acceleration signal collected in real time. The initial reference signal is then input into the active noise reduction system for noise reduction. During the process of driving through the target road section, the residual noise signal after the active noise cancellation system has worked is collected in real time through the vehicle's in-vehicle error microphone, and the residual noise signal is associated with the current geographical location information and timestamp and stored in the vehicle. The residual noise signal is input into the deviation prediction neural network pre-set in the vehicle. The deviation prediction neural network outputs the deviation between the local filter coefficients and the theoretical optimal coefficients based on the spectral characteristics of the residual noise signal. The local filter coefficients are then fine-tuned in real time based on the deviation to generate a corrected reference signal. The correspondence between the deviation and the current geographical location information is recorded. After leaving the target road segment, the deviation and its corresponding geographical location information are uploaded to the cloud server, so that the cloud server can incrementally update the reference signal prediction model based on the received deviation and store the updated model in the cloud database.

[0008] In a preferred embodiment, the specific steps for obtaining the vehicle type information and the road segment identification information of the target road segment are as follows: Obtain the user-defined navigation destination and planned driving route; Based on the driving route, at least one target road segment is determined, and each target road segment corresponds to a unique road segment identification information; Obtain the vehicle type category information, which includes at least the vehicle brand, model, year, and chassis platform identifier; The vehicle type information and the road segment identification information of each target road segment are associated according to a preset format.

[0009] In a preferred embodiment, the steps of sending a reference signal prediction model acquisition request to the cloud server based on the vehicle type information and the road segment identification information, and receiving the reference signal prediction model returned by the cloud server, wherein the reference signal prediction model is generated by the cloud server based on the vehicle type information and the road segment identification information, are as follows: Send a reference signal prediction model acquisition request to the cloud server. The reference signal prediction model acquisition request includes the vehicle type information and the road segment identification information. When a reference signal prediction model corresponding to the vehicle type information and the road segment identification information already exists in the cloud server, the stored reference signal prediction model is directly returned by the cloud server. When the cloud server does not have a reference signal prediction model corresponding to the vehicle type information and the road segment identification information, it receives a null response returned by the cloud server and switches to the sensor direct acquisition mode based on the physical model to generate a reference signal. At the same time, during the process of the vehicle traveling through the target road segment, it collects the vehicle's vibration acceleration signal and the corresponding residual noise signal, and uploads the collected data to the cloud server so that the cloud server can train and generate the reference signal prediction model based on the data.

[0010] In a preferred embodiment, when the vehicle travels to the target road segment, the real-time collected vibration acceleration signal is input into the reference signal prediction model. The initial coefficients of the local adaptive filter are determined based on the predicted noise signal output by the reference signal prediction model. An initial reference signal is generated by combining the real-time collected vibration acceleration signal and input into the active noise reduction system for noise reduction. The specific steps are as follows: The vehicle's current geographical location information is obtained in real time through the onboard positioning module while the vehicle is in motion. The location is matched with the road segment identification information of the target road segment based on the current geographical location information to determine whether to enter the target road segment; When it is determined that the vehicle has entered the target road segment, the reference signal prediction model is loaded into the local system of the vehicle. The real-time acquired vibration acceleration signal is input into the reference signal prediction model, and the reference signal prediction model outputs a prediction noise signal after forward propagation calculation. The initial coefficients of the local adaptive filter are determined based on the predicted noise signal, and the initial coefficients are loaded into the local adaptive filter. The real-time collected chassis vibration acceleration signal is input into the local adaptive filter, and the initial reference signal is generated after filtering. The initial reference signal is output to the feedforward controller of the active noise reduction system.

[0011] In a preferred embodiment, during the process of driving through the target road segment, the residual noise signal after the active noise cancellation system has been activated is collected in real time using the vehicle's in-vehicle error microphone. The residual noise signal is then associated with the current geographical location information and a timestamp and stored in the vehicle. The specific steps are as follows: The in-vehicle error microphone remains continuously activated throughout the entire process of the vehicle traveling through the target road section; The residual noise signal after the active noise cancellation system has been working is collected in real time through the error microphone. Simultaneously collect current vehicle speed, engine speed, and interior temperature information; The residual noise signal is associated with the corresponding current geographical location information and timestamp and then stored in the vehicle's memory. Repeat the above data collection and storage steps until the vehicle leaves the target road section.

[0012] In a preferred embodiment, the residual noise signal is input into a pre-installed deviation prediction neural network at the vehicle end. The deviation prediction neural network outputs the deviation between the local filter coefficients and the theoretical optimal coefficients based on the spectral characteristics of the residual noise signal. The local filter coefficients are then fine-tuned in real time based on the deviation to generate a corrected reference signal. The correspondence between the deviation and the current geographical location information is recorded. The specific steps are as follows: Acquire the residual noise signal collected in real time by the error microphone; The time-domain waveform of the residual noise signal is input into a pre-set deviation prediction neural network at the vehicle end, wherein the deviation prediction neural network is a forward propagation neural network; Obtain the deviation between the local filter coefficients output by the deviation prediction neural network and the theoretical optimal coefficients; The fine-tuning amplitude is determined based on the magnitude of the deviation, the fine-tuning direction is determined based on the direction of the deviation, and the fine-tuning correction value of the local filter coefficient is generated. The fine-tuning correction value is superimposed on the current local filter coefficients to obtain the updated filter coefficients; The real-time acquired vibration acceleration signal is reprocessed using the updated filter coefficients to generate a corrected reference signal; The deviation is recorded in conjunction with the current geographic location information.

[0013] In a preferred embodiment, after leaving the target road segment, the deviation and its corresponding geographical location information are uploaded to the cloud server, so that the cloud server incrementally updates the reference signal prediction model based on the received deviation and stores the updated model in the cloud database. The specific steps are as follows: After detecting that the vehicle has left the target road section, compile all deviations recorded during this journey and their corresponding geographical location information; The deviation and the geographic location information are packaged into a feedback data packet; When a wireless network connection is available, the feedback data packet is uploaded to the cloud server, so that the cloud server uses the deviation as new training data to perform incremental updates on the reference signal prediction model and stores the updated model in the cloud database.

[0014] This invention also provides a vehicle road noise cancellation reference signal acquisition system based on AI recognition, using the above-mentioned vehicle road noise cancellation reference signal acquisition method based on AI recognition, including: The information acquisition module is used to acquire the vehicle type information and the road segment identification information of the target road segment; The model request and receiving module is used to send a reference signal prediction model acquisition request to the cloud server based on the vehicle type information and the road segment identification information, and to receive the reference signal prediction model returned by the cloud server, wherein the reference signal prediction model is generated by the cloud server based on the vehicle type information and the road segment identification information; The model loading and reference signal generation module is used to input the real-time collected vibration acceleration signal into the reference signal prediction model when the vehicle travels to the target road section, determine the initial coefficients of the local adaptive filter based on the predicted noise signal output by the reference signal prediction model, and generate an initial reference signal by combining the real-time collected vibration acceleration signal. The initial reference signal is then input into the active noise reduction system for noise reduction. The noise reduction effect acquisition module is used to collect the residual noise signal after the active noise reduction system has worked in real time through the vehicle's in-vehicle error microphone during the process of driving through the target road section, and to associate the residual noise signal with the current geographical location information and timestamp and store it in the vehicle. The deviation prediction and online calibration module is used to input the residual noise signal into a deviation prediction neural network pre-installed on the vehicle. The deviation prediction neural network outputs the deviation between the local filter coefficients and the theoretical optimal coefficients based on the spectral characteristics of the residual noise signal. The local filter coefficients are then fine-tuned in real time based on the deviation to generate a corrected reference signal, and the correspondence between the deviation and the current geographical location information is recorded. The cloud-based feedback update module is used to upload the deviation amount and its corresponding geographical location information to the cloud server after leaving the target road segment, so that the cloud server can incrementally update the reference signal prediction model based on the received deviation amount and store the updated model in the cloud database.

[0015] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described AI-based vehicle road noise cancellation reference signal acquisition method.

[0016] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the above-described method for acquiring reference signals for vehicle road noise cancellation based on AI recognition.

[0017] The technical effects achieved by this invention are as follows: This invention obtains vehicle type information and target road segment identification information and sends a model acquisition request to a cloud server. This allows the vehicle to acquire an AI prediction model trained on noise reduction data accumulated by other vehicles of the same type on that road segment before entering the target segment. This solves the problems of existing technologies where independent operation of a single-vehicle AI model leads to difficulty in sharing road segment experience and large initial output deviations. By loading the reference signal prediction model from the cloud into the local machine and determining the initial coefficients of the filter based on its output prediction value, and simultaneously integrating real-time sensor data of the vehicle to generate a reference signal, the vehicle can inherit the training results of the group AI model while retaining the ability to make personalized corrections based on its own physical characteristics. This solves the problem that a unified AI model is difficult to adapt to the individual differences of different vehicles. By using a pre-set deviation prediction neural network to perform real-time inference on residual noise signals and output the filter coefficient deviation during driving, the millisecond-level rapid fine-tuning of the filter coefficients is achieved, solving the problem that existing models cannot respond in time when road characteristics change abruptly. By feeding back the deviation generated during the actual operation of the vehicle to the cloud and incrementally updating the cloud model, the prediction accuracy of the cloud AI model continuously improves with data accumulation, realizing the continuous evolution of the AI ​​model. This invention transforms vehicle road noise reference signal acquisition from independent AI prediction by a single vehicle to collaborative AI prediction among vehicle groups, providing higher-quality reference signal input for active noise reduction systems. Compared to methods that store noise reduction parameters in the cloud for multiple vehicles to share, this invention stores and shares neural network models instead of traditional noise reduction parameters, and supports continuous model evolution as data accumulates, achieving precise matching between vehicle type and road segment in two dimensions. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the method flow of the present invention; Figure 2 This is a schematic diagram of the system modules of the present invention; Figure 3 This is a schematic diagram of the computer device structure of the present invention. Detailed Implementation

[0019] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0020] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0021] Secondly, the term "an embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in a preferred embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single embodiment or an embodiment selectively excluded from other embodiments.

[0022] Please see Figure 1 As shown, this invention provides a method for acquiring vehicle road noise cancellation reference signals based on AI recognition, including: S1. Obtain the vehicle type information and road segment sign information for the target road segment; In step S1, by obtaining the vehicle type information and the road segment identification information of the target road segment, a data foundation is provided for subsequently requesting a dedicated reference signal prediction model for this vehicle type and road segment from the cloud server. This enables the model request to accurately match the vehicle type characteristics and road segment characteristics. The specific steps are as follows: Obtain the user-defined navigation destination and planned driving route; Based on the driving route, at least one target road segment is determined, and each target road segment corresponds to a unique road segment identification information; Obtain the vehicle type category information, which includes at least the vehicle brand, model, year, and chassis platform identifier; The vehicle type information and the road segment identification information of each target road segment are associated according to a preset format.

[0023] Specifically, the vehicle obtains the user-set navigation destination and planned driving route through the in-vehicle navigation system. The driving route consists of multiple continuous road segments, each with unique road segment identification information. The vehicle parses the road segment identification information of at least one target road segment based on the driving route. This road segment identification information is determined by the latitude and longitude range, road code, and driving direction of the road segment. The vehicle also obtains its own vehicle type information, which includes the vehicle brand, model, year, and chassis platform identifier. The chassis platform identifier is used to characterize the vehicle's chassis architecture and suspension system type and is a key parameter for determining road noise transmission characteristics. The vehicle associates the vehicle type information and the road segment identification information of each target road segment according to a preset format for subsequent sending of model acquisition requests to the cloud server.

[0024] S2. Send a reference signal prediction model acquisition request to the cloud server based on the vehicle type information and the road segment identification information, and receive the reference signal prediction model returned by the cloud server. The reference signal prediction model is generated by the cloud server based on the vehicle type information and the road segment identification information. In step S2, by sending a model acquisition request to the cloud server based on vehicle type information and road segment identification information and receiving the returned reference signal prediction model, the vehicle obtains an AI prediction model trained on noise reduction data accumulated by other vehicles of the same model on that road segment before entering the target road segment. This avoids the inability to acquire cloud data due to poor network signal during driving, and also realizes the sharing of noise reduction experience among vehicles of the same model and the aggregation of collective wisdom. The specific steps are as follows: Send a reference signal prediction model acquisition request to the cloud server. The reference signal prediction model acquisition request includes the vehicle type information and the road segment identification information. When a reference signal prediction model corresponding to the vehicle type information and the road segment identification information already exists in the cloud server, the stored reference signal prediction model is directly returned by the cloud server. When the cloud server does not have a reference signal prediction model corresponding to the vehicle type information and the road segment identification information, it receives a null response returned by the cloud server and switches to the sensor direct acquisition mode based on the physical model to generate a reference signal. At the same time, during the process of the vehicle traveling through the target road segment, it collects the vehicle's vibration acceleration signal and the corresponding residual noise signal, and uploads the collected data to the cloud server so that the cloud server can train and generate the reference signal prediction model based on the data.

[0025] Specifically, the vehicle sends a request to the cloud server via its onboard communication module to obtain a reference signal prediction model. This request includes the vehicle's model category information and the road segment identification information of the target road segment. Upon receiving the request, the cloud server uses the model category information and road segment identification information as a combined index to search the model database for the existence of a corresponding reference signal prediction model. When the cloud server finds a corresponding model, it directly returns the model to the vehicle. When the cloud server does not find a corresponding model, it returns a null response. After receiving a null response, the vehicle switches to a physical model-based sensor direct acquisition mode, which directly uses the vibration acceleration signal collected by the chassis acceleration sensor and filters it with a local adaptive filter to generate a reference signal, without relying on an AI prediction model. Simultaneously, as the vehicle travels through the target road segment, it continuously collects vibration acceleration signals and corresponding residual noise signals, uploading the collected data to the cloud server as initial historical noise reduction data for the vehicle model on that road segment. This data is then used by the cloud server to train and generate a reference signal prediction model for the vehicle model on that road segment for subsequent vehicle use.

[0026] The reference signal prediction model is generated by the cloud server based on vehicle type information and road segment identification information. For example, the cloud server's historical noise reduction database stores a large number of raw data records uploaded by vehicles of the same model traveling on various road segments. Each record includes the road segment identification information, vehicle type information, vibration acceleration signal time-domain waveform, and corresponding residual noise signal spectral characteristics. The cloud server retrieves data from this database using a combined index of vehicle type information and road segment identification information. When at least one set of historical noise reduction data is retrieved, the cloud server uses this data as a training set and inputs it into a pre-set reference signal prediction neural network for training. This reference signal prediction neural network is a long short-term memory network, and its input layer connects to... The model receives a time-domain waveform sequence of vibration acceleration signals, with a sequence length of 512 sampling points and a sampling frequency of 2000 Hz. The hidden layer contains two layers of long short-term memory units, each with 128 neurons. The output layer outputs a time-domain waveform sequence of the predicted noise signal. During training, the vibration acceleration signals from historical denoised data are used as input, and the corresponding residual noise signals are used as labels. The backpropagation algorithm is used for training, with the mean squared error loss function, the Adam optimizer, and a learning rate of 0.001. Training stops when the loss function value of the model on the validation set no longer decreases for 10 consecutive training epochs. After training, the cloud server stores the trained model as a reference signal prediction model for this vehicle type on this road segment.

[0027] S3. When the vehicle travels to the target road section, the vibration acceleration signal collected in real time is input into the reference signal prediction model. The initial coefficients of the local adaptive filter are determined based on the predicted noise signal output by the reference signal prediction model. The initial reference signal is generated by combining the vibration acceleration signal collected in real time. The initial reference signal is then input into the active noise reduction system for noise reduction. In step S3, the reference signal prediction model sent from the cloud is loaded locally, and the initial coefficients of the filter are determined based on the predicted value output by the model. This enables the noise reduction system of the vehicle to have the prediction capability based on the training results of the group AI model at the initial moment of entering the road segment, avoiding the problem of large initial prediction deviation caused by the single vehicle model being calculated from scratch in the existing technology. The specific steps are as follows: The vehicle's current geographical location information is obtained in real time through the onboard positioning module while the vehicle is in motion. The location is matched with the road segment identification information of the target road segment based on the current geographical location information to determine whether to enter the target road segment; When it is determined that the vehicle has entered the target road segment, the reference signal prediction model is loaded into the local system of the vehicle. The real-time acquired vibration acceleration signal is input into the reference signal prediction model, and the reference signal prediction model outputs a prediction noise signal after forward propagation calculation. The initial coefficients of the local adaptive filter are determined based on the predicted noise signal, and the initial coefficients are loaded into the local adaptive filter. The real-time collected chassis vibration acceleration signal is input into the local adaptive filter, and the initial reference signal is generated after filtering. The initial reference signal is output to the feedforward controller of the active noise reduction system.

[0028] Specifically, during operation, the vehicle acquires its current geographical location information in real time through an onboard positioning module, which uses either the Global Positioning System (GPS) or the BeiDou Navigation Satellite System. The vehicle then performs location matching between its current geographical location and the road segment identification information of each target road segment. This matching involves comparing the current location coordinates with the latitude and longitude range in the road segment identification information. If the location coordinates fall within the latitude and longitude range of a road segment and the driving direction matches that road segment's driving direction, the vehicle is determined to have entered that target road segment. Upon determining that the vehicle has entered a target road segment, the aforementioned reference signal prediction model is loaded into the vehicle's local computing environment. The vehicle inputs the real-time collected chassis vibration acceleration signal into this reference signal prediction model. After forward propagation calculation, the model outputs a predicted noise signal waveform. This predicted value integrates the training results of historical noise reduction data from multiple vehicles of the same model on that road segment, exhibiting high initial accuracy. The vehicle determines the initial coefficients of the local adaptive filter based on the predicted noise signal and loads these initial coefficients into the filter, enabling it to have near-optimal filtering characteristics upon startup. Simultaneously, the vehicle collects chassis vibration acceleration signals in real time via acceleration sensors mounted on the front subframe, rear axle, or shock absorber towers. The vehicle inputs the real-time collected chassis vibration acceleration signals into the local adaptive filter, which uses the loaded initial coefficients to filter the input signal and generate an initial reference signal. This initial reference signal reflects the estimated waveform of the noise from the road surface excitation transmitted through the vehicle body to the listener's ear on the target road segment. The vehicle outputs the initial reference signal to the feedforward controller of the active noise cancellation system. The feedforward controller generates an anti-phase sound wave with the opposite phase to the road noise based on the reference signal, which is played through the vehicle's speakers to achieve active road noise cancellation.

[0029] S4. During the process of driving through the target road section, the residual noise signal after the active noise cancellation system has worked is collected in real time through the vehicle's in-vehicle error microphone, and the residual noise signal is associated with the current geographical location information and timestamp and stored in the vehicle. In step S4, actual noise reduction effect data is collected synchronously as the vehicle travels through the target road segment, providing input data for the local bias prediction neural network. This enables the vehicle to use a pre-set AI model to evaluate the current noise reduction effect in real time and make rapid adjustments, while simultaneously accumulating training data for cloud model updates. The specific steps are as follows: The in-vehicle error microphone remains continuously activated throughout the entire process of the vehicle traveling through the target road section; The residual noise signal after the active noise cancellation system has been working is collected in real time through the error microphone. Simultaneously collect current vehicle speed, engine speed, and interior temperature information; The residual noise signal is associated with the corresponding current geographical location information and timestamp and then stored in the vehicle's memory. Repeat the above data collection and storage steps until the vehicle leaves the target road section.

[0030] Specifically, after entering the target road segment, the vehicle continuously activates the in-vehicle error microphone, which is positioned near the heads of the occupants to collect residual noise signals after the active noise cancellation system has been activated. This residual noise signal reflects the difference between the generated anti-phase sound wave and the actual road noise, serving as a direct basis for evaluating the current noise cancellation effect. Throughout the entire process of driving through the target road segment, the error microphone continuously collects residual noise signals at a preset sampling frequency. Simultaneously, the vehicle synchronously collects vehicle speed, engine speed, and in-vehicle temperature information via the vehicle controller local area network bus as auxiliary information for noise cancellation effect data. This auxiliary information is used to distinguish noise cancellation performance under different driving conditions during subsequent deviation prediction neural network inference. The vehicle associates the residual noise signal with the current geographical location information and timestamp and stores it in the vehicle's on-board memory, forming a noise cancellation effect data record with location and time tags. The vehicle repeats the above collection and storage steps, forming a continuous data record sequence throughout the entire journey through the target road segment until the vehicle leaves the target road segment.

[0031] S5. Input the residual noise signal into the deviation prediction neural network preset in the vehicle. The deviation prediction neural network outputs the deviation between the local filter coefficients and the theoretical optimal coefficients based on the spectral characteristics of the residual noise signal. The local filter coefficients are fine-tuned in real time based on the deviation to generate a corrected reference signal, and the correspondence between the deviation and the current geographical location information is recorded. In step S5, a preset deviation prediction neural network is used to perform real-time inference on the residual noise signal and output the filter coefficient deviation, enabling millisecond-level rapid fine-tuning of the filter coefficients. This allows the local filter to perform personalized online optimization based on the current actual state of the vehicle, building upon the results of the collective AI model. The specific steps are as follows: Acquire the residual noise signal collected in real time by the error microphone; The time-domain waveform of the residual noise signal is input into a pre-set deviation prediction neural network at the vehicle end, wherein the deviation prediction neural network is a forward propagation neural network; Obtain the deviation between the local filter coefficients output by the deviation prediction neural network and the theoretical optimal coefficients; The fine-tuning amplitude is determined based on the magnitude of the deviation, the fine-tuning direction is determined based on the direction of the deviation, and the fine-tuning correction value of the local filter coefficient is generated. The fine-tuning correction value is superimposed on the current local filter coefficients to obtain the updated filter coefficients; The real-time acquired vibration acceleration signal is reprocessed using the updated filter coefficients to generate a corrected reference signal; The deviation is recorded in conjunction with the current geographic location information.

[0032] Specifically, the vehicle acquires the residual noise signal collected in real time by the error microphone. This residual noise signal is the residual sound pressure signal resulting from the superposition of the anti-phase sound wave output by the active noise cancellation system and the actual road noise reaching the ear. The vehicle inputs the time-domain waveform of this residual noise signal into a pre-set deviation prediction neural network. For example, this deviation prediction neural network is a three-layer forward propagation neural network. Its input layer receives the power spectrum feature vector of the residual noise signal, with a dimension of 256. The hidden layer contains 64 neurons, with the activation function being a modified linear unit. The output layer outputs a deviation vector, with the dimension being the same as the number of local filter coefficients, and the value range being -1 to 1. This deviation prediction neural network has been pre-trained offline. Its training data comes from a large number of paired records of residual noise signals and corresponding theoretical optimal coefficient deviations under different road surfaces, vehicle speeds, and vehicle models. The theoretical optimal coefficient refers to the coefficient that minimizes the power of the residual noise signal collected by the in-vehicle error microphone under the current driving conditions. At that time, the ideal coefficient vector corresponding to the local adaptive filter, the theoretically optimal coefficient, changes dynamically with the changes in road excitation and vehicle driving state. In the actual system, the local adaptive filter updates the filter coefficients step by step based on the real-time acquired residual noise signal using an iterative learning algorithm to approximate the theoretical optimal value, where the iterative learning algorithm is such as the least mean square algorithm. After the vehicle obtains the deviation amount output by the deviation prediction neural network, it determines the fine-tuning amplitude according to the magnitude of the deviation amount and the fine-tuning direction according to the direction of the deviation amount, generating a fine-tuning correction value for the local filter coefficients. The vehicle superimposes this fine-tuning correction value onto the current local filter coefficients to obtain the updated filter coefficients. The real-time acquired vibration acceleration signal is reprocessed using the updated filter coefficients to generate a corrected reference signal. The corrected reference signal is more suitable for the current physical characteristics of the vehicle and the road condition than the aforementioned initial reference signal. The vehicle also binds and records the deviation amount with the current geographical location information to form a deviation amount record with a location tag.

[0033] S6. After leaving the target road segment, the deviation amount and its corresponding geographical location information are uploaded to the cloud server, so that the cloud server can incrementally update the reference signal prediction model according to the received deviation amount, and store the updated model in the cloud database. In step S6, by feeding back the deviation generated during the actual operation of the vehicle to the cloud and incrementally updating the cloud AI model, the prediction accuracy of the cloud reference signal prediction model continuously improves with data accumulation, realizing the continuous evolution of the AI ​​model. The reference signal prediction model obtained by the vehicle in the future will become more and more accurate. The specific steps are as follows: After detecting that the vehicle has left the target road section, compile all deviations recorded during this journey and their corresponding geographical location information; The deviation and the geographic location information are packaged into a feedback data packet; When a wireless network connection is available, the feedback data packet is uploaded to the cloud server, so that the cloud server uses the deviation as new training data to perform incremental updates on the reference signal prediction model and stores the updated model in the cloud database.

[0034] Specifically, after the vehicle determines it has left the target road segment through the aforementioned location matching, it compiles all the aforementioned bound deviations and their corresponding geographical location information records during this journey. The vehicle packages the deviations and geographical location information into a feedback data packet according to a preset data format. The vehicle checks the current wireless network connection status; when the wireless network connection is available, it uploads the feedback data packet to the cloud server via the wireless network. After receiving feedback data packets from multiple vehicles, the cloud server extracts the deviation records, corresponding vehicle type information, and road segment identification information from each data packet. The cloud server merges the newly received deviation data with historical data and performs incremental updates to the reference signal prediction model for that vehicle type on that road segment. Incremental updates refer to the cloud server merging the newly received deviation data with existing historical data and performing further updates based on existing model parameters. The model undergoes a limited number of iterations of training to gradually approximate the optimal values ​​for the vehicle type on the road segment, rather than retraining the entire model each time. For example, incremental updates use the same loss function and optimizer as the initial training, with 3 to 5 iteration rounds. The cloud server stores the updated reference signal prediction model in the cloud database, in the model entry corresponding to the combination of vehicle type and road segment identifier. When subsequent vehicles send a model retrieval request to the cloud server on the same road segment, they will obtain the updated reference signal prediction model from the cloud server, which has higher prediction accuracy than before the update. As the number of vehicles traveling on the same road segment and uploading feedback data continues to increase, the accuracy of the cloud reference signal prediction model continues to improve, and the initial noise reduction effect of subsequent vehicles also continues to improve, forming a positive feedback loop of continuous evolution of the AI ​​model.

[0035] Please see Figure 2 The present invention also provides a vehicle road noise cancellation reference signal acquisition system based on AI recognition, which uses the above-mentioned vehicle road noise cancellation reference signal acquisition method based on AI recognition, including: The information acquisition module is used to acquire the vehicle type information and the road segment identification information of the target road segment; The model request and receiving module is used to send a reference signal prediction model acquisition request to the cloud server based on the vehicle type information and the road segment identification information, and to receive the reference signal prediction model returned by the cloud server, wherein the reference signal prediction model is generated by the cloud server based on the vehicle type information and the road segment identification information; The model loading and reference signal generation module is used to input the real-time collected vibration acceleration signal into the reference signal prediction model when the vehicle travels to the target road section, determine the initial coefficients of the local adaptive filter based on the predicted noise signal output by the reference signal prediction model, and generate an initial reference signal by combining the real-time collected vibration acceleration signal. The initial reference signal is then input into the active noise reduction system for noise reduction. The noise reduction effect acquisition module is used to collect the residual noise signal after the active noise reduction system has worked in real time through the vehicle's in-vehicle error microphone during the process of driving through the target road section, and to associate the residual noise signal with the current geographical location information and timestamp and store it in the vehicle. The deviation prediction and online calibration module is used to input the residual noise signal into a deviation prediction neural network pre-installed on the vehicle. The deviation prediction neural network outputs the deviation between the local filter coefficients and the theoretical optimal coefficients based on the spectral characteristics of the residual noise signal. The local filter coefficients are then fine-tuned in real time based on the deviation to generate a corrected reference signal, and the correspondence between the deviation and the current geographical location information is recorded. The cloud-based feedback update module is used to upload the deviation amount and its corresponding geographical location information to the cloud server after leaving the target road segment, so that the cloud server can incrementally update the reference signal prediction model based on the received deviation amount and store the updated model in the cloud database.

[0036] The execution process of the AI-based vehicle road noise cancellation reference signal acquisition system corresponds to the implementation steps of the AI-based vehicle road noise cancellation reference signal acquisition method described above, and will not be described again here.

[0037] Please see Figure 3 The present invention also provides a computer device, which may be a server, and its structure may be as follows: Figure 3 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores all data required for the process of acquiring reference signals for vehicle road noise cancellation based on AI recognition. The network interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements the AI-based vehicle road noise cancellation reference signal acquisition method.

[0038] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer equipment on which the present application is applied.

[0039] An embodiment of the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements any of the above-described AI-based vehicle road noise cancellation reference signal acquisition methods.

[0040] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in this application and in the embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus DRAM (RDRAM), direct Rambus DRAM (DRDRAM), etc.

[0041] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0042] The above description is merely a preferred embodiment of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention. Structures, devices, and operating methods not specifically described or explained in this invention are implemented according to conventional methods in the art unless otherwise specified or limited.

Claims

1. A method for acquiring vehicle road noise cancellation reference signals based on AI recognition, characterized in that: include: Obtain the vehicle type information and road sign information for the target road segment; The cloud server sends a request to obtain a reference signal prediction model based on the vehicle type information and the road segment identification information, and receives the reference signal prediction model returned by the cloud server. The reference signal prediction model is generated by the cloud server based on the vehicle type information and the road segment identification information. When the vehicle travels to the target road section, the vibration acceleration signal collected in real time is input into the reference signal prediction model. The initial coefficients of the local adaptive filter are determined based on the predicted noise signal output by the reference signal prediction model. The initial reference signal is generated by combining the vibration acceleration signal collected in real time. The initial reference signal is then input into the active noise reduction system for noise reduction. During the process of driving through the target road section, the residual noise signal after the active noise cancellation system has worked is collected in real time through the vehicle's in-vehicle error microphone, and the residual noise signal is associated with the current geographical location information and timestamp and stored in the vehicle. The residual noise signal is input into the deviation prediction neural network preset in the vehicle. The deviation prediction neural network outputs the deviation between the local filter coefficients and the theoretical optimal coefficients based on the spectral characteristics of the residual noise signal. The local filter coefficients are then fine-tuned in real time based on the deviation to generate a corrected reference signal. The correspondence between the deviation and the current geographical location information is recorded. After leaving the target road segment, the deviation and its corresponding geographical location information are uploaded to the cloud server, so that the cloud server can incrementally update the reference signal prediction model based on the received deviation and store the updated model in the cloud database.

2. The method for acquiring vehicle road noise cancellation reference signals based on AI recognition according to claim 1, characterized in that: The specific steps for obtaining the vehicle type information and the road segment identification information of the target road segment are as follows: Obtain the user-defined navigation destination and planned driving route; Based on the driving route, at least one target road segment is determined, and each target road segment corresponds to a unique road segment identification information; Obtain the vehicle type category information, which includes at least the vehicle brand, model, year, and chassis platform identifier; The vehicle type information and the road segment identification information of each target road segment are associated according to a preset format.

3. The method for acquiring vehicle road noise cancellation reference signals based on AI recognition according to claim 1, characterized in that: The steps are as follows: First, a request to obtain a reference signal prediction model is sent to the cloud server based on the vehicle type information and the road segment identification information. Second, a request to obtain a reference signal prediction model is received from the cloud server. The reference signal prediction model is generated by the cloud server based on the vehicle type information and the road segment identification information. Send a reference signal prediction model acquisition request to the cloud server. The reference signal prediction model acquisition request includes the vehicle type information and the road segment identification information. When a reference signal prediction model corresponding to the vehicle type information and the road segment identification information already exists in the cloud server, the stored reference signal prediction model is directly returned by the cloud server. When the cloud server does not have a reference signal prediction model corresponding to the vehicle type information and the road segment identification information, it receives a null response returned by the cloud server and switches to the sensor direct acquisition mode based on the physical model to generate a reference signal. At the same time, during the process of the vehicle traveling through the target road segment, it collects the vehicle's vibration acceleration signal and the corresponding residual noise signal, and uploads the collected data to the cloud server so that the cloud server can train and generate the reference signal prediction model based on the data.

4. The method for acquiring vehicle road noise cancellation reference signals based on AI recognition according to claim 1, characterized in that: When the vehicle reaches the target road section, the real-time collected vibration acceleration signal is input into the reference signal prediction model. The initial coefficients of the local adaptive filter are determined based on the predicted noise signal output by the reference signal prediction model. An initial reference signal is generated by combining the real-time collected vibration acceleration signal and then input into the active noise reduction system for noise reduction. The specific steps are as follows: The vehicle's current geographical location information is obtained in real time through the onboard positioning module while the vehicle is in motion. The location is matched with the road segment identification information of the target road segment based on the current geographical location information to determine whether to enter the target road segment; When it is determined that the vehicle has entered the target road segment, the reference signal prediction model is loaded into the local system of the vehicle. The real-time acquired vibration acceleration signal is input into the reference signal prediction model, and the reference signal prediction model outputs a prediction noise signal after forward propagation calculation. The initial coefficients of the local adaptive filter are determined based on the predicted noise signal, and the initial coefficients are loaded into the local adaptive filter. The real-time collected chassis vibration acceleration signal is input into the local adaptive filter, and the initial reference signal is generated after filtering. The initial reference signal is output to the feedforward controller of the active noise reduction system.

5. The method for acquiring vehicle road noise cancellation reference signals based on AI recognition according to claim 1, characterized in that: During the process of driving through the target road segment, the residual noise signal after the active noise cancellation system has been activated is collected in real time through the vehicle's in-vehicle error microphone. The residual noise signal is then associated with the current geographical location information and timestamp and stored in the vehicle. The specific steps are as follows: The in-vehicle error microphone remains continuously activated throughout the entire process of the vehicle traveling through the target road section; The residual noise signal after the active noise cancellation system has been working is collected in real time through the error microphone. Simultaneously collect current vehicle speed, engine speed, and interior temperature information; The residual noise signal is associated with the corresponding current geographical location information and timestamp and then stored in the vehicle's memory. Repeat the above data collection and storage steps until the vehicle leaves the target road section.

6. The method for acquiring vehicle road noise cancellation reference signals based on AI recognition according to claim 1, characterized in that: The residual noise signal is input into a pre-installed deviation prediction neural network on the vehicle side. The deviation prediction neural network outputs the deviation between the local filter coefficients and the theoretical optimal coefficients based on the spectral characteristics of the residual noise signal. The local filter coefficients are then fine-tuned in real time based on the deviation to generate a corrected reference signal. The correspondence between the deviation and the current geographical location information is recorded. The specific steps are as follows: Acquire the residual noise signal collected in real time by the error microphone; The time-domain waveform of the residual noise signal is input into a pre-set deviation prediction neural network at the vehicle end, wherein the deviation prediction neural network is a forward propagation neural network; Obtain the deviation between the local filter coefficients output by the deviation prediction neural network and the theoretical optimal coefficients; The fine-tuning amplitude is determined based on the magnitude of the deviation, the fine-tuning direction is determined based on the direction of the deviation, and the fine-tuning correction value of the local filter coefficient is generated. The fine-tuning correction value is superimposed on the current local filter coefficients to obtain the updated filter coefficients; The real-time acquired vibration acceleration signal is reprocessed using the updated filter coefficients to generate a corrected reference signal; The deviation is linked to the current geographical location information and recorded.

7. The method for acquiring vehicle road noise cancellation reference signals based on AI recognition according to claim 1, characterized in that: After leaving the target road segment, the deviation and its corresponding geographical location information are uploaded to the cloud server, so that the cloud server can incrementally update the reference signal prediction model based on the received deviation and store the updated model in the cloud database. The specific steps are as follows: After detecting that the vehicle has left the target road section, compile all deviations recorded during this journey and their corresponding geographical location information; The deviation and the geographic location information are packaged into a feedback data packet; When a wireless network connection is available, the feedback data packet is uploaded to the cloud server, so that the cloud server uses the deviation as new training data to perform incremental updates on the reference signal prediction model and stores the updated model in the cloud database.

8. A vehicle road noise cancellation reference signal acquisition system based on AI recognition, characterized in that: The method for acquiring vehicle road noise cancellation reference signals based on AI recognition as described in any one of claims 1 to 7 includes: The information acquisition module is used to acquire the vehicle type information and the road segment identification information of the target road segment; The model request and receiving module is used to send a reference signal prediction model acquisition request to the cloud server based on the vehicle type information and the road segment identification information, and to receive the reference signal prediction model returned by the cloud server, wherein the reference signal prediction model is generated by the cloud server based on the vehicle type information and the road segment identification information; The model loading and reference signal generation module is used to input the real-time collected vibration acceleration signal into the reference signal prediction model when the vehicle travels to the target road section, determine the initial coefficients of the local adaptive filter based on the predicted noise signal output by the reference signal prediction model, and generate an initial reference signal by combining the real-time collected vibration acceleration signal. The initial reference signal is then input into the active noise reduction system for noise reduction. The noise reduction effect acquisition module is used to collect the residual noise signal after the active noise reduction system has worked in real time through the vehicle's in-vehicle error microphone during the process of driving through the target road section, and to associate the residual noise signal with the current geographical location information and timestamp and store it in the vehicle. The deviation prediction and online calibration module is used to input the residual noise signal into a deviation prediction neural network pre-installed on the vehicle. The deviation prediction neural network outputs the deviation between the local filter coefficients and the theoretical optimal coefficients based on the spectral characteristics of the residual noise signal. The local filter coefficients are then fine-tuned in real time based on the deviation to generate a corrected reference signal, and the correspondence between the deviation and the current geographical location information is recorded. The cloud-based feedback update module is used to upload the deviation amount and its corresponding geographical location information to the cloud server after leaving the target road segment, so that the cloud server can incrementally update the reference signal prediction model based on the received deviation amount and store the updated model in the cloud database.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images