In-vehicle music playback prediction system, method and computer program product

Abstract

The invention provides a vehicle-mounted music playing prediction system, and the system comprises a music type prediction module which is used for inputting the current music interaction data of a vehicle owner into a music type prediction model, and predicting the preference probability of the vehicle owner for various types of music; the sound effect parameter prediction module is used for inputting the current vehicle state data into an adaptive sound effect parameter prediction model and predicting adaptive sound effect parameters of various types of music to the vehicle type in the current vehicle state; and the music playing prediction module is used for inputting the preference probability of the vehicle owner to various types of music and the adaptive sound effect parameters of various types of music to the vehicle model in the current vehicle state into a preference sound effect parameter prediction model, and predicting the preference sound effect parameters of the vehicle owner to various types of music in the current vehicle state. The music interaction data and the vehicle state data are fused at the same time, and the problem that a traditional vehicle-mounted music system only depends on a static song list or a single environment parameter is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of vehicle entertainment system, and particularly relates to a vehicle music playing prediction system, method and computer program product. BACKGROUND

[0002] Since the birth of the vehicle entertainment system, it has the function of music playing. With the progress of technology, people's requirements for music playing function are getting higher and higher.

[0003] The current vehicle entertainment system has the following problems: the traditional system relies on historical playing records to generate fixed playlists, without considering real-time environmental changes; vehicle state parameters (such as noise, air volume) are not included in the sound effect tuning model, and the equalizer needs to be manually adjusted by the vehicle owner, which has safety hazards during driving; new owners or niche music types have insufficient behavior data, resulting in recommendation bias. SUMMARY

[0004] The purpose of the present application is to overcome the defects of the prior art and provide a vehicle music playing prediction system, comprising: A music type prediction module for inputting the current music interaction data of the vehicle owner into a music type prediction model to predict the preference probability of the vehicle owner for various types of music; An audio parameter prediction module for inputting the current vehicle state data into an adaptive audio parameter prediction model to predict the adaptive audio parameters of various types of music for the vehicle model under the current vehicle state; A music playing prediction module for inputting the preference probability of the vehicle owner for various types of music and the adaptive audio parameters of various types of music for the vehicle model under the current vehicle state into a preference audio parameter prediction model to predict the preference audio parameters of the vehicle owner for various types of music under the current vehicle state.

[0005] Further, it further comprises a music playing module for controlling the vehicle entertainment system to play music of the corresponding type according to the predicted preference probability of the vehicle owner for various types of music as the playing probability of each type of music, and simultaneously controlling the equalizer parameter adjustment to be the preference audio parameters of the corresponding type of music.

[0006] Further, in the music type prediction module, the music interaction data includes music playing type, different type music playing time, different type music playing times, music collection behavior, music skipping behavior, and different type music feedback label.

[0007] Further, in the audio parameter prediction module, the vehicle state data includes vehicle speed, vehicle window opening state, vehicle air conditioner air volume, vehicle noise value, and vehicle machine call state.

[0008] Further, in the music type prediction module, the music type prediction model is obtained by the following method: constructing a car owner and music type matrix after preprocessing the historical music interaction behavior data of all car owners of the corresponding vehicle type, wherein the rows and columns of the car owner and music type matrix represent the car owners and music types respectively, and the matrix values in the car owner and music type matrix represent the play count weighted values of each car owner for the corresponding type of music, the play count weighted values being obtained by weighting the corresponding type of music play duration, the corresponding type of music play count, the corresponding music collection behavior, the corresponding music skip behavior, and the corresponding type of music feedback label in the historical music interaction behavior data of the corresponding car owner, training the collaborative filtering algorithm according to the car owner and music type matrix to obtain the music type prediction model. In the car owner and music type matrix, the features of each car owner are represented as , , wherein i represents the car owner index, , wherein j represents the music type index, , wherein represents the play count weighted value of the nth type of music, and is divided into a first training set and a first validation set, the corresponding type of music play duration, the corresponding type of music play count, the corresponding music collection behavior, the corresponding music skip behavior, and the corresponding type of music feedback label corresponding to the first training set are input into the collaborative filtering algorithm to obtain a first music type prediction value , the first music type prediction value and the corresponding in the first training set are input into a preset first loss function to obtain the value of the first loss function, the parameters of the collaborative filtering algorithm are updated through back propagation according to the value of the first loss function, the corresponding type of music play duration, the corresponding type of music play count, the corresponding music collection behavior, the corresponding music skip behavior, and the corresponding type of music feedback label corresponding to the first validation set are input into the collaborative filtering algorithm after the parameter update to obtain a second music type prediction value , the second music type prediction value and the corresponding in the validation set are compared to determine whether the parameter update is completed, if the parameter update is completed, the music type prediction model is obtained, and if the parameter update is not completed, the parameter update is continued through the first training set until the music type prediction model is obtained.

[0009] Further, in the sound effect parameter prediction module, the adaptive sound effect parameter prediction model is obtained by the following method: the historical sound effect tuning data of all owners of the corresponding vehicle model under different music types and the historical vehicle state data of all owners of the corresponding vehicle model are respectively preprocessed to obtain a sound effect tuning feature set and a vehicle state feature set. The sound effect tuning features in the sound effect tuning feature set include the tuning parameters of the equalizer under different music types, the sound field of the equalizer under different music types, and the volume value of the equalizer under different music types. The vehicle state features in the vehicle state feature set include vehicle speed, vehicle window opening state, vehicle-mounted air conditioner air volume, vehicle interior noise value, and vehicle machine call state. Pearson correlation analysis is performed between all sound effect tuning features in the sound effect tuning feature set and all vehicle state features in the vehicle state feature set to obtain a set of related vehicle state features having correlation with the sound effect tuning features. Principal component analysis is performed on the set of related vehicle state features to reduce dimensions. The XGBoost algorithm is trained according to the sound effect tuning feature set and the set of related vehicle state features after dimension reduction to obtain the adaptive sound effect parameter prediction model. The sound effect tuning feature set and the set of related vehicle state features after dimension reduction are divided into a second training set and a second validation set. The related vehicle state features in the second training set are input into the XGBoost algorithm to obtain first sound effect tuning feature prediction values The first sound effect tuning feature prediction values and the corresponding in the second training set are input into a preset second loss function to obtain a value of the second loss function. The parameters of the XGBoost algorithm are updated through back propagation according to the value of the second loss function. The related vehicle state features in the second validation set are input into the XGBoost algorithm after the parameter update to obtain second sound effect tuning feature prediction values The second sound effect tuning feature prediction values and the corresponding in the second validation set are compared to determine whether the parameter update is completed. If it is determined that the parameter update is completed, the adaptive sound effect parameter prediction model is obtained. If it is determined that the parameter update is not completed, the parameter update is continued through the second training set until the adaptive sound effect parameter prediction model is obtained.

[0010] Further, in the music playing prediction module, the preference sound effect parameter prediction model is represented as: the preference sound effect parameter of the owner for various types of music = the adaptive sound effect parameter of various types of music for the vehicle model under the current vehicle state × (the preference probability of the owner for the type of music × α + ε). Wherein, α is a music preference weight coefficient, the music preference weight coefficient α is used to amplify or reduce the influence of the owner's preference, and ε is an error value.

[0011] Further, the least square method is used to dynamically adjust the music preference weight coefficient a and the error value e according to the music interaction data of the vehicle owner and the vehicle state data.

[0012] A vehicle-mounted music playing prediction method, characterized in that it comprises: The current music interaction data of the vehicle owner is input into a music type prediction model to predict the preference probability of the vehicle owner for various types of music. The current vehicle state data is input into an adaptive sound effect parameter prediction model to predict the adaptive sound effect parameters of various types of music for the vehicle model under the current vehicle state. The preference probability of the vehicle owner for various types of music and the adaptive sound effect parameters of various types of music for the vehicle model under the current vehicle state are input into a preference sound effect parameter prediction model to predict the preference sound effect parameters of the vehicle owner for various types of music under the current vehicle state.

[0013] A computer program product comprising computer programs / instructions which, when executed by a processor, implement the vehicle-mounted music playing prediction method described above.

[0014] The beneficial effects of the present application are: 1. The music interaction data and the vehicle state data are fused at the same time to solve the problem that the traditional vehicle-mounted music system only relies on static playlists or single environmental parameters.

[0015] 2. The music type prediction model uses the collaborative filtering algorithm to achieve the purpose of predicting new vehicle owners based on similar vehicle owner behaviors through the vehicle owner and music type matrix. The adaptive sound effect parameter prediction model associates the vehicle state and the sound effect parameters through the XGBoost algorithm, and the preference sound effect parameter prediction model realizes the dynamic balance of the sound effect parameters under the subjective factors of the vehicle owner and the objective factors of the vehicle through the weight coefficient a.

[0016] 3. The real music playing times are estimated by weighting each music interaction behavior data to solve the problem of recommendation deviation caused by insufficient playing times of niche music types due to vehicle networking data recording.

[0017] 4. The long-term adaptability of the system is improved through the closed-loop feedback of the least square method.

[0018] 5. The risk of manual operation during driving is avoided by automatically completing sound effect adjustment. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 The system block diagram of the present application.

[0020] Figure 2 The method flowchart of the present application. DETAILED DESCRIPTION

[0021] In order to make the technical problems, technical solutions and beneficial effects to be solved in the present application more clear and explicit, the present application will be further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0022] Embodiment 1 As shown in the figure, a vehicle-mounted music playing prediction system comprises: Figure 1 A music type prediction module for inputting the current music interaction data of the vehicle owner into a music type prediction model to predict the preference probability of the vehicle owner for various types of music; An adaptive sound effect parameter prediction module for inputting the current vehicle state data into an adaptive sound effect parameter prediction model to predict the adaptive sound effect parameters of various types of music for the vehicle model under the current vehicle state; A preference sound effect parameter module for inputting the preference probability of the vehicle owner for various types of music and the adaptive sound effect parameters of various types of music for the vehicle model under the current vehicle state into a preference sound effect parameter prediction model to predict the preference sound effect parameters of the vehicle owner for various types of music under the current vehicle state. A music playing module for controlling the vehicle-mounted entertainment system to play the corresponding type of music according to the predicted preference probability of the vehicle owner for various types of music as the playing probability of each type of music, and simultaneously controlling the equalizer parameter adjustment to be the preference sound effect parameters of the corresponding type of music. The means of controlling the vehicle-mounted entertainment system is automatic control, which adopts existing automatic control methods and devices, such as the vehicle-mounted bus control system, sends control instructions through the CAN bus or AutoSAR architecture, and the device is, for example, the Infineon AURIX series MCU + MOST150 multimedia bus interface module, generates APPD (application message) to control the sound host based on the probability weight.

[0023] In the music type prediction module, the music interaction data comprises the music playing type, the playing time of different types of music, the playing times of different types of music, the music collection behavior, the music skipping behavior, and the feedback label of different types of music.

[0024] By integrating the playing time, the number of times, the collection / skipping behavior and the feedback label, the playing times weighted value is calculated by weighting, the active / passive preference expression is covered, and the vehicle owner behavior is described in multiple dimensions. At the same time, the skipping behavior reversely corrects the preference probability (such as reducing the weight of skipping a certain type of music), which improves the prediction robustness.

[0025]

[0026] ​(2) In the sound effect parameter prediction module, the vehicle status data includes vehicle speed, window opening status, vehicle air conditioning air volume, vehicle noise level, and vehicle call status.

[0027] By associating real-time data such as vehicle speed, window status, and air conditioning fan speed with the noise environment (e.g., enhancing low-frequency compensation when driving at high speeds), the sound effect parameters are dynamically adjusted according to the physical environment.

[0028] (3) In the music type prediction module, the music type prediction model is obtained through the following method: After preprocessing (data cleaning, normalization) the historical music interaction behavior data of all car owners of the corresponding car model, a car owner and music type matrix is ​​constructed. In the car owner and music type matrix, the rows and columns represent car owners and music types, respectively. The matrix value in the car owner and music type matrix represents the weighted value of the number of times each car owner plays the corresponding type of music. The weighted value of the number of times the playback is performed is obtained by weighting the playback duration of the corresponding type of music, the number of times the corresponding type of music is played, the collection behavior of the corresponding music, the skip behavior of the corresponding music, and the feedback label of the corresponding type of music in the historical music interaction behavior data of the corresponding car owner. This is because the number of times the corresponding type of music is played often cannot truly reflect the actual number of times it is played. For example, the skip behavior of the corresponding music is recorded as one playback, but this does not represent the actual music playback behavior. By weighting, the actual number of times the music is played is estimated to be closer to the actual number of times the music is played. The collaborative filtering algorithm is trained based on the car owner and music type matrix to obtain the music type prediction model. In the car owner and music type matrix, the features of each car owner are... Represented as , Indicates the vehicle owner index. Indicates the music genre index. This represents the weighted value of the number of times the nth type of music is played. The dataset is divided into a first training set and a first validation set. The playback duration, playback count, music collection behavior, skipping behavior, and feedback labels of the corresponding music type in the first training set are input into a collaborative filtering algorithm to obtain the predicted value of the first music type. The predicted value of the first music genre Corresponding to the first training set The first loss function (e.g., mean squared error) is input to obtain its value. Based on this value, the parameters of the collaborative filtering algorithm are updated via backpropagation (e.g., gradient descent). The collaborative filtering algorithm, updated with the input parameters of the corresponding music type playback duration, playback count, collection behavior, skip behavior, and feedback label from the first validation set, yields the second music type prediction value. The predicted value of the second music genre and the corresponding in the validation set The comparison is made to determine whether the parameter updating is completed, and if the parameter updating is completed, the music type prediction model is obtained, and if the parameter updating is not completed, the parameter updating is continued by using the first training set until the music type prediction model is obtained. The forward propagation of the validation set is the final verification step of model optimization, and is used to confirm whether the model has truly learned the generalization rule. The method for determining whether the parameter updating is completed is as follows: the second music type prediction value and the corresponding in the validation set are close, and the training / validation loss curve is reasonable. The case where the parameter updating is not completed is handled as follows: if overfitting occurs, the parameters of the model are regularized, the owner and music type matrix data are enhanced, and the model training is stopped early; if underfitting occurs, the complexity of the model is increased, and the parameters are adjusted. The updatable parameters of the collaborative filtering algorithm include: (1) similarity calculation parameters: similarity metric selection (cosine / Pearson / Jaccard), neighborhood size (k value in k-NN), similarity weight decay factor (time decay coefficient); (2) matrix decomposition parameters (if MF is used): latent factor dimension (usually 50-300), regularization coefficient (lambda is usually 0.01-0.1), learning rate; (3) real-time updating parameters: time window size, cold start processing weight.

[0029] The collaborative filtering algorithm here is specifically a user-based collaborative filtering algorithm, and the core idea is that if a user A and a user B show similar preferences in historical behavior, when the user A is interested in a certain item, the items that the user B likes and that the user A has not yet contacted can be recommended to the user A.

[0030] Through the aggregation of the group behavior of the owner and the music type matrix, the new owner can predict the preferences based on similar owners, and the data sparsity problem can be alleviated.

[0031] (4) In the sound effect parameter prediction module, the adaptive sound effect parameter prediction model is obtained by the following method: the historical sound effect tuning data of all owners of the corresponding vehicle model under different music types and the historical vehicle state data of all owners of the corresponding vehicle model are respectively preprocessed (data cleaning, normalization) to obtain a sound effect tuning feature set and a vehicle state feature set. The sound effect tuning features in the sound effect tuning feature set include the tuning parameters of the equalizer under different music types, the sound field of the equalizer under different music types, and the volume value of the equalizer under different music types. The vehicle state features in the vehicle state feature set include vehicle speed, vehicle window opening state, vehicle-mounted air conditioner air volume, vehicle interior noise value, and vehicle audio call state. Pearson correlation analysis is performed between all sound effect tuning features in the sound effect tuning feature set and all vehicle state features in the vehicle state feature set to obtain a set of relevant vehicle state features having correlation with the sound effect tuning features, principal component analysis (PCA, an unsupervised statistical dimension reduction method, the core goal of which is to convert high-dimensional data into low-dimensional representation while retaining the main variation characteristics in the data set) is used to reduce the dimension of the set of relevant vehicle state features, and the XGBoost algorithm is trained according to the sound effect tuning feature set and the set of relevant vehicle state features after dimension reduction to obtain an adaptive sound effect parameter prediction model. The sound effect tuning feature set and the set of relevant vehicle state features after dimension reduction are divided into a second training set and a second validation set, the relevant vehicle state features in the second training set are input into the XGBoost algorithm to obtain first sound effect tuning feature prediction values , the first sound effect tuning feature prediction values and the corresponding in the second training set are input into a preset second loss function (such as mean square error) to obtain a value of the second loss function, and the relevant vehicle state features in the second validation set are input into the XGBoost algorithm after parameter update to obtain second sound effect tuning feature prediction values , the second sound effect tuning feature prediction values and the corresponding The comparison is performed to determine whether the parameter update is completed. If it is determined that the parameter update is completed, the adaptive sound effect parameter prediction model is obtained. If it is determined that the parameter update is not completed, the parameter update is continued by using the second training set until the adaptive sound effect parameter prediction model is obtained. The process of determining whether the parameter update is completed is referred to the training method of the music type prediction model, which is not described herein again. The core updateable parameters of XGBoost include (1) tree structure parameters: max_depth (3-10, control tree complexity), min_child_weight (1-10, prevent overfitting), gamma (0-0.5, minimum loss drop in splitting); (2) learning process parameters: learning_rate (0.01-0.3, step control), n_estimators (50-1000, number of trees), subsample (0.5-1, sample sampling ratio), colsample_bytree (0.5-1, feature sampling ratio); (3) regularization parameters: reg_alpha (L1 regularization, 0-1), reg_lambda (L2 regularization, 1-2); (3) task-specific parameters: objective (loss function selection), eval_metric (evaluation index).

[0032] The principal component analysis method is to summarize all the features in the related vehicle state feature set into a smaller number of new features. The new features can cover most of the information of the original features, so that the dimensionality reduction and redundancy removal purposes can be achieved, and the XGBoost training efficiency and generalization can be improved. At the same time, the vehicle state features after dimensionality reduction can inherit the correlation between the vehicle state features and the sound effect tuning features obtained by the Pearson correlation analysis, so the vehicle state feature set after dimensionality reduction has a one-to-one correspondence with the feature values in the sound effect tuning feature set .

[0033] The historical music interaction behavior data of all car owners of the corresponding vehicle model and the historical sound effect tuning data of all car owners of the corresponding vehicle model under different music types are processed as follows: 1. Remove the music with a playing time <10s; 2. The car owner may trigger a sudden increase or decrease in volume due to a false trigger, so the data with a sudden change in volume within 10s and then a recovery are removed.

[0034] (5) The music playing prediction module, the preference sound effect parameter prediction model is expressed as: the preference sound effect parameter of the car owner to various types of music = the adaptive sound effect parameter of various types of music to the vehicle model under the current vehicle state x (the preference probability of the car owner to the type of music x a + e); Wherein, a is the music preference weight coefficient, the music preference weight coefficient a is used to amplify or reduce the influence of the car owner's preference, e is the error value.

[0035] The preference probability (subjective) and the adaptive sound effect (objective) are dynamically coupled through the weight coefficient a, and the individualized adaptation of the car owner is realized by adjusting the music preference weight coefficient a, the error value e absorbs the model prediction error, and the extreme fluctuation of the parameter is avoided. The initial value of the weight coefficient a is set by the car owner, and the initial value of the error value e is preset.

[0036] According to the music interaction data of the car owner and the vehicle state data, the least square method is used to dynamically adjust the music preference weight coefficient a and the error value e. When the car owner actively tunes, the value of a is increased to respond to the subjective intention of the car owner, when the environment suddenly changes (such as suddenly closing the window), the value of a is reduced and the value of e is increased, and the sound effect stability is preferentially ensured.

[0037] Embodiment 2 As shown in Figure 2 A vehicle-mounted music playing prediction method, characterized in that it comprises: Input the current music interaction data of the car owner into the music type prediction model to predict the preference probability of the car owner to various types of music; Input the current vehicle state data into the adaptive sound effect parameter prediction model to predict the adaptive sound effect parameter of various types of music to the vehicle model under the current vehicle state; Input the preference probability of the car owner to various types of music and the adaptive sound effect parameter of various types of music to the vehicle model under the current vehicle state into the preference sound effect parameter prediction model to predict the preference sound effect parameter of the car owner to various types of music under the current vehicle state.

[0038] Embodiment 3 A computer program product comprising computer programs / instructions which, when executed by a processor, implement the vehicle-mounted music playing prediction method of embodiment 2.

[0039] Those aspects of the specification that are not otherwise explicitly described will be understood as comprising part of the known art to which the application pertains. Those skilled in the art will appreciate that the embodiments of the application can be practiced with embodiments other than those described, which will vary with respect to the specific implementation details. That is, it will be appreciated that the described embodiments are intended to be illustrative only and in no way limiting of the scope of the application. It is therefore contemplated to cover by the appended claims all such alternatives, modifications, permutations, and equivalents. Thus, if there is a conflict between what is described in the disclosure and anything that has

[0040] The present application is described in reference to the drawings using a flowchart and / or a block diagram of the method, apparatus (system) and computer program product according to the embodiments of the application. It will be understood that each block of the flowchart and / or block diagram, and combinations of blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks. Figure 1 means for performing the function specified by the flowchart and / or block diagram block or blocks.

[0041] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions which implement the function specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks. Figure 1 means for performing the function specified by the flowchart and / or block diagram block or blocks.

[0042] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart and / or block diagram block or blocks. Figure 1 one or more functions specified in the flowchart and / or block diagram block or blocks. Figure 1 means for performing the function specified by the flowchart and / or block diagram block or blocks.

[0043] It should be pointed out finally that the above examples are only used for illustrating the technical solutions of the present application but not for limiting the protection scope thereof, and although the present application has been described in detail with reference to the above examples, it should be understood by those skilled in the art that the specific embodiments of the present application can be changed, modified or replaced equivalently by those skilled in the art after reading the present application, but these changes, modifications or equivalent replacements are all within the protection scope of the claims to be filed of the present application.

Claims

1. A vehicle-mounted music playback prediction system, characterized in that, include: The music genre prediction module is used to input the car owner's current music interaction data into the music genre prediction model to predict the probability of the car owner's preference for various types of music. The sound effect parameter prediction module is used to input the current vehicle status data into the appropriate sound effect parameter prediction model to predict the appropriate sound effect parameters for various types of music for the vehicle model under the current vehicle status. The music playback prediction module is used to input the car owner's preference probability for various types of music and the appropriate sound effect parameters for various types of music for the vehicle model under the current vehicle status into the preference sound effect parameter prediction model, and predict the car owner's preference sound effect parameters for various types of music under the current vehicle status.

2. The in-vehicle music playback prediction system according to claim 1, characterized in that, Also includes: The music playback module is used to determine the playback probability of each type of music based on the predicted probability of the car owner's preference for various types of music. Based on the playback probability of each type of music, the module controls the in-vehicle entertainment system to play the corresponding type of music, while simultaneously adjusting the equalizer parameters to the preferred sound effect parameters for the corresponding type of music.

3. The in-vehicle music playback prediction system according to claim 1, characterized in that: In the music type prediction module, music interaction data includes music playback type, playback duration of different types of music, number of times different types of music are played, music collection behavior, music skipping behavior, and feedback tags for different types of music. Vehicle status data includes vehicle speed, window opening status, vehicle air conditioning fan speed, in-vehicle noise level, and vehicle-to-vehicle call status.

4. The in-vehicle music playback prediction system according to claim 1, characterized in that, In the music genre prediction module, the music genre prediction model is obtained through the following method: After preprocessing the historical music interaction behavior data of all car owners for the corresponding vehicle model, a car owner and music type matrix is ​​constructed. In the car owner and music type matrix, the rows and columns represent car owners and music types, respectively. The matrix values ​​in the car owner and music type matrix represent the weighted values ​​of the number of times each car owner plays music of the corresponding type. The weighted values ​​of the number of times the number of times the number of times the number of times the number of times the number of times the number of times the number of times the number of times the number of times the owner of the corresponding type of music plays music, the corresponding music collection behavior, the corresponding music skipping behavior, and the corresponding music feedback tags in the historical music interaction behavior data of the corresponding car owner are weighted. The collaborative filtering algorithm is trained based on the car owner and music type matrix to obtain a music type prediction model.

5. The in-vehicle music playback prediction system according to claim 4, characterized in that, The specific method for training the collaborative filtering algorithm based on the car owner and music type matrix to obtain the music type prediction model is as follows: In the car owner and music genre matrix, the characteristics of each car owner Represented as , Indicates the vehicle owner index. Indicates the music genre index. This represents the weighted value of the number of times the nth type of music is played. The dataset is divided into a first training set and a first validation set. The playback duration, playback count, music collection behavior, skipping behavior, and feedback labels of the corresponding music type in the first training set are input into a collaborative filtering algorithm to obtain the predicted value of the first music type. The predicted value of the first music genre Corresponding to the first training set The first loss function is input to obtain its value. Based on this value, the parameters of the collaborative filtering algorithm are updated through backpropagation. The collaborative filtering algorithm, with the updated input parameters including the playback duration, playback count, favorites, skips, and feedback labels of the corresponding music type from the first validation set, yields the second music type prediction value. The predicted value of the second music genre and the corresponding verification set The comparison is used to determine whether the parameter update is complete. If the parameter update is complete, the music type prediction model is obtained. If the parameter update is incomplete, the parameter update is continued to be performed using the first training set until the music type prediction model is obtained.

6. The in-vehicle music playback prediction system according to claim 1, characterized in that, In the sound effect parameter prediction module, the appropriate sound effect parameter prediction model is obtained through the following method: The audio tuning feature set is obtained by preprocessing the historical audio tuning data of all owners of the corresponding model under different music types. The vehicle status feature set is obtained by preprocessing the historical vehicle status data of all owners of the corresponding model. The audio tuning features in the audio tuning feature set include the equalizer tuning parameters under different music types, the sound field of the equalizer under different music types, and the volume value of the equalizer under different music types. The vehicle status features in the vehicle status feature set include vehicle speed, window opening status, vehicle air conditioning fan speed, in-vehicle noise level, and vehicle call status. Pearson correlation analysis is performed between all sound effect tuning features in the sound effect tuning feature set and all vehicle state features in the vehicle state feature set to obtain a set of related vehicle state features that are correlated with the sound effect tuning features. Principal component analysis is used to reduce the dimensionality of the related vehicle state feature set. The XGBoost algorithm is trained based on the sound effect tuning feature set and the dimensionality-reduced related vehicle state feature set to obtain a suitable sound effect parameter prediction model.

7. The in-vehicle music playback prediction system according to claim 6, characterized in that, The specific method for training the XGBoost algorithm based on the sound effect tuning feature set and the dimensionality-reduced relevant vehicle state feature set to obtain the appropriate sound effect parameter prediction model is as follows: Sound effect tuning feature set and the related vehicle state feature set after dimensionality reduction The dataset is divided into a second training set and a second validation set. The relevant vehicle state features in the second training set are then... The first sound effect tuning feature prediction value is obtained by inputting the XGBoost algorithm. The predicted value of the first sound effect adjustment feature. The corresponding one in the second training set The preset second loss function is input to obtain its value. Based on this value, the parameters of the XGBoost algorithm are updated through backpropagation, incorporating relevant vehicle state features from the second validation set. The XGBoost algorithm, after updating the input parameters, obtains the predicted value of the second sound effect tuning feature. The predicted value of the second sound effect adjustment feature. The corresponding second verification set The comparison is used to determine whether the parameter update is complete. If the parameter update is complete, the appropriate sound effect parameter prediction model is obtained. If the parameter update is incomplete, the parameter update is continued through the second training set until the appropriate sound effect parameter prediction model is obtained.

8. The in-vehicle music playback prediction system according to claim 1, characterized in that, In the music playback prediction module, the preferred sound effect parameter prediction model is represented as follows: The car owner's preferred sound effect parameters for various types of music = the appropriate sound effect parameters for various types of music for this car model under the current vehicle status × (the car owner's probability of preferring this type of music × α + ε). Where α is the music preference weighting coefficient, which is used to amplify or reduce the influence of car owner preferences, and ε is the error value; Based on the owner's music interaction data and vehicle status data, the least squares method is used to dynamically adjust the music preference weight coefficient α and the error value ε.

9. A method for predicting in-vehicle music playback, characterized in that, include: Input the car owner's current music interaction data into the music type prediction model to predict the probability of the car owner's preference for various types of music; Input the current vehicle status data into the appropriate sound effect parameter prediction model to predict the appropriate sound effect parameters for various types of music for the vehicle model under the current vehicle status. The car owner's preference probability for various types of music and the appropriate sound effect parameters for various types of music for the current vehicle model are input into the preference sound effect parameter prediction model to predict the car owner's preference sound effect parameters for various types of music under the current vehicle condition.

10. A computer program product, characterized in that, The system includes a computer program / instruction, characterized in that the computer program / instruction, when executed by a processor, implements the in-vehicle music playback prediction method of claim 9.

Images