Big Data-Based In-Vehicle Human-Machine Interaction Methods and Systems

Abstract

This invention provides a vehicle-mounted human-machine interaction method and system based on big data. Through three embodiments, it achieves automatic vehicle settings for the current user in the current scenario. The three embodiments construct candidate sets using different methods, mining the relationships between users, between settings, and between users and settings. User features are represented by user feature vectors, and setting features are represented by setting feature vectors. Simultaneously, the correlation between these features is used to calculate the user's preference for settings, and the settings are ranked according to preference levels for automatic vehicle settings. Compared to existing methods that learn settings independently for each individual user, this human-machine interaction is more intelligent, and the user experience is significantly improved.

Description

Technical Field

[0001] This invention relates to the field of human-computer interaction technology, and specifically to a vehicle-mounted human-computer interaction method and system based on big data. Background Technology

[0002] In in-vehicle human-machine interaction (HMI), the same user may have different preferences for vehicle HMI settings in different scenarios. Taking in-vehicle air conditioning as an example, a male user might prefer to set the temperature to 18℃ on a hot, sunny day, and 26℃ on a chilly autumn day. Different users may also have different preferences for air conditioning settings in the same scenario. For instance, on a hot, sunny day, a female user might prefer to set the temperature to 22℃. To provide personalized HMI services, existing technologies use machine learning models to learn user preferences and habits based on historical HMI settings data. The learned model can then predict the user's preferred HMI settings based on the current scenario.

[0003] However, existing machine learning models learn independently for each user, with each user having their own model. When user interaction habits change or a user makes new settings in a new scenario, the machine learning model updates only for that single user, having no effect on other users. However, when this user makes new settings in a new scenario, other users with similar characteristics are also likely to make the same settings in the same new scenario. Existing machine learning models do not fully explore the relationships between users, between users and settings, and between settings, resulting in less intelligent human-computer interaction and a need for further improvement in user experience. Summary of the Invention

[0004] This invention aims to improve the intelligence level of in-vehicle human-computer interaction and thus enhance the human-computer interaction experience. It provides an in-vehicle human-computer interaction method and system based on big data.

[0005] To achieve this objective, the present invention adopts the following technical solution:

[0006] A method for in-vehicle human-machine interaction based on big data is provided, including the following steps:

[0007] L1, acquire the current user's in-vehicle human-machine interaction habit data, including user information, scene information, setting information, and setting interaction preferences, in order to construct the user feature vector associated with the current user and acquire the setting feature vector;

[0008] L2, construct a candidate set for predicting vehicle settings for the current user;

[0009] L3 extracts the highest preference level among the current user's preferences for each setting in the candidate set, and uses the setting with the highest preference level as the prediction result for the current user's vehicle settings.

[0010] Preferably, the user information includes any one or more of the following: user ID, user age, user gender, clothing index, physiological information, and emotional information;

[0011] The scene information includes any one or more of the following: outside temperature, inside temperature, vehicle speed, remaining battery power, remaining fuel, weather conditions, light intensity, current time, vehicle's geographical location, road navigation information, current road type, and driving gradient.

[0012] The settings information includes any one or more of the following: air conditioning settings, seat settings, fragrance settings, window settings, ambient lighting settings, and audio settings.

[0013] Preferably, the user feature vector is user u i The vector is composed of the preference levels of the n settings that have been interacted with in the past; the obtained setting feature vector is a vector composed of the similarity between setting s1 and each setting s2 among the n settings.

[0014] Preferably, the method for constructing the candidate set in step L2 is as follows:

[0015] Obtain the n settings that the current user has interacted with in the past, and then find k settings that have similarity to each of the n settings, for a total of n×k settings to form the candidate set as the basis for automatically setting the vehicle for the current user.

[0016] Preferably, the similarity between s1 and s2 is set as follows:

[0017]

[0018] S C =S1∩S2,S c This represents the intersection of the user set S1 with setting s1 and the user set S2 with setting s2.

[0019] Preferably, the user feature vector is: user u i The vector is composed of the similarity between the user and the top p users who have similarity in user features and are arranged in descending order of similarity; the obtained setting feature vector is: a vector composed of the preference of p users for the setting s1 that they have interacted with in the current scene.

[0020] Preferably, in step L2, the method for constructing the candidate set is as follows:

[0021] Find p users who have similar user characteristics to the current user, and then find n settings among the p users whose historical interaction scene information is consistent with the current scene information of the vehicle to be set by the current user, for a total of n×p settings as the candidate set.

[0022] Preferably, the similarity between users u1 and u2 is:

[0023]

[0024] K C =K1∩K2, K c This represents the intersection of the settings set K1 of the settings that user u1 has historically interacted with and the settings set K2 of the settings that user u2 has historically interacted with.

[0025] Preferably, the setting is an extended setting, which is a fusion information that includes the setting information and the scene information for setting the vehicle in the corresponding scenario.

[0026] Preferably, step L3 specifically includes the following steps:

[0027] L31, Remove duplicate settings and settings inconsistent with the current scene information from the candidate set;

[0028] L32, predict the current user's preference for each setting in the remaining candidate set after step L31 by calculating the dot product;

[0029] L33, extract the highest preference level of the current user among the preferences of each setting in the candidate set remaining after step L31, and use the setting with the highest preference level as the prediction result for the vehicle settings of the current user.

[0030] Preferably, in step L1, after distinguishing discrete information and continuous information from the vehicle-mounted human-machine interaction habit data and preprocessing it, the preprocessed vector is input into the neural network, and the user feature vector and the setting feature vector are output.

[0031] The discrete user information of the user information includes user ID and / or user gender, and the continuous user information includes user age and / or clothing index; the discrete scene information of the scene information includes weather conditions and / or the geographical information of the vehicle, and the continuous scene information includes outside temperature and / or inside temperature; the discrete setting information of the setting information includes air outlet mode and / or circulation mode, and the continuous setting information includes set temperature and / or set wind speed.

[0032] The discrete and / or continuous user information and / or scene information are preprocessed and then input into the user neural network to obtain the user feature vector; the discrete and / or continuous setting information is preprocessed and then input into the setting neural network to obtain the setting feature vector.

[0033] The preprocessing method for user discrete information, scene discrete information and setting discrete information is as follows: convert the one-hot vector corresponding to the discrete information into an embedding matrix, and use the embedding vector as the discrete information preprocessing vector.

[0034] The preprocessing methods for continuous user information, continuous scene information, and continuous setting information are as follows: after the continuous information is binned, it is converted into discrete information, and then the one-hot vector corresponding to the discrete information is multiplied by the embedding matrix to obtain the embedding vector as the preprocessing vector of the discrete information; or the continuous information is normalized.

[0035] Preferably, the bucketing process involves grouping the continuous information, and the normalization process involves mapping the continuous values ​​of the continuous information to a specified numerical range using linear transformation.

[0036] Preferably, the method for the neural network to predict the current user's vehicle settings in the current scenario is as follows:

[0037] The cosine similarity between the user feature vector and the setting feature vector is calculated as the current user's preference for the setting in the current scenario, and the setting with the highest preference is used for vehicle control.

[0038] Preferably, the method for calculating the user's preference for the settings is as follows:

[0039] When a user sets the vehicle settings in a specified scenario, the value of the user's preference level for the vehicle settings in the specified scenario is incremented by "1".

[0040] After the system configures the vehicle settings for the user in the specified scenario, if the system detects that the user re-enters the specified scenario and does not change the system's settings after entering the specified scenario again, the user's preference level for the vehicle settings in the specified scenario is incremented by "1"; if the user changes the system's settings, the user's preference level for the settings in the specified scenario is decremented by "1".

[0041] Or map positive preferences to "1" and negative preferences to "-1".

[0042] Preferably, step L3 specifically includes the following steps:

[0043] L301 predicts the current user's preference for each setting in the candidate set by calculating dot product or cosine similarity.

[0044] L302, extract the highest preference level among the current user's preference levels for each setting in the candidate set, and use the setting with the highest preference level as the prediction result for the current user's vehicle settings.

[0045] The present invention also provides a vehicle-mounted human-machine interaction system based on big data, which can realize the aforementioned vehicle-mounted human-machine interaction method based on big data.

[0046] This invention provides three different solutions through three embodiments to automatically configure the vehicle for the current user in the current scenario. The three embodiments construct candidate sets using different methods, mining the relationships between users, between settings, and between users and settings. User features are represented using user feature vectors, and setting features are represented using setting feature vectors. Simultaneously, the correlation between these features is used to calculate the user's preference for settings, and the settings are ranked according to preference levels for automatic vehicle configuration. Compared to existing methods that learn settings independently for each individual user, this approach results in more intelligent human-computer interaction and a significantly improved user experience. Attached Figure Description

[0047] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments of the present invention will be briefly described below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0048] Figure 1 This is a diagram illustrating the implementation steps of the vehicle-mounted human-machine interaction method based on big data provided in this embodiment of the invention. Detailed Implementation

[0049] The technical solution of the present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0050] The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual images. They should not be construed as limiting the scope of this patent. To better illustrate the embodiments of the present invention, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual dimensions of the product. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0051] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," "inner," and "outer" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present patent. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0052] In the description of this invention, unless otherwise explicitly specified and limited, the term "connection" or similar designation indicating a connection between components should be interpreted broadly. For example, it can refer to a fixed connection, a detachable connection, or an integral part; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium; it can refer to the internal communication between two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0053] First, the technical principle of the in-vehicle human-machine interaction method based on big data provided in this application is explained. Then, the technical implementation process of this application is specifically illustrated through three embodiments.

[0054] The in-vehicle human-machine interaction method based on big data provided in this application includes the following technical components:

[0055] 1. Collect in-vehicle human-machine interaction user habits and obtain big data on in-vehicle human-machine interaction user habits, including: user information, scenario information, setting information, and setting interaction preferences;

[0056] User information includes any one or more of the following: user ID, user age, user gender, clothing index, physiological information, and emotional information; the clothing index is a weather index for clothing, determined by the thickness of clothing, as shown in Table 1 below.

[0057] Table 1 Clothing Weather Index and Corresponding Service Phrases

[0058] Clothing Weather Index Clothing thickness H Service language 1 H＞25.0 Suitable for wearing thick down jackets, gloves, and other winter clothing. 2 15.0＜H≤25.0 It is appropriate to wear cotton-padded coats, leather jackets, and thick sweaters underneath as winter clothing. 3 8.0＜H≤15.0 Suitable for wearing early winter clothes such as jackets and coats 4 4.0＜H≤8.0 Suitable for wearing suits, jackets, and other early spring and late autumn attire. 5 1.5＜H≤4.0 Suitable clothing for spring and autumn includes cotton-padded shirts, long-sleeved T-shirts, and denim clothing. 6 -1.2＜H≤1.5 Lightweight shirts and long skirts are suitable for summer wear. 7 H≤-1.2 Suitable summer outfits include short-sleeved T-shirts, skirts, and shorts.

[0059] Physiological information includes respiratory rate, heart rate, blood oxygen saturation, etc.

[0060] The scenario information includes any one or more of the following: outside temperature, inside temperature, vehicle speed, remaining battery power, remaining fuel, weather conditions (sunny, cloudy, rainy, etc.), light intensity, current time, vehicle location information (latitude and longitude, city, etc.), road navigation information (road congestion, etc.), current road type (elevated road, tunnel, residential area, etc.), and driving gradient.

[0061] The settings information includes any one or more of the following: air conditioning settings (air conditioning temperature, air conditioning air outlet mode, air conditioning fan speed, etc.), seat settings (seat massage, seat heating, seat ventilation, etc.), fragrance settings (fragrance mixing ratio, fragrance concentration, etc.), window settings (side window opening, sunroof opening, etc.), ambient lighting settings (ambient lighting dynamic effects, brightness, color, etc.), and audio settings (volume level, reverberation effect, etc.).

[0062] A single data entry contains corresponding user information, scene information, setting information, and setting interaction preferences. For example, user A, aged 35 and female, in a scenario with an outside temperature of 36°C and a light intensity of 500 lx, has a preference of setting the temperature to 22.5°C, the fan speed to 3, and the air conditioning interaction mode to "blow air onto face" with internal circulation. This constitutes a data entry. In this data entry, user A, age 35, and female are user information; the outside temperature of 36°C and light intensity of 500 lx are scene information; the set temperature of 22.5°C, the fan speed of 3, and the setting to "blow air onto face" with internal circulation are setting information; and the setting air conditioning interaction preference is "preference 3," where "3" represents the setting interaction preference. Another example is user B, who, when passing near a chemical plant, has a preference of setting all car windows to 0% opening. This preference (preference 2) also constitutes a data entry. Big data is a collection of numerous data entries.

[0063] There are three ways to determine interaction preferences:

[0064] The first method is: when a user actively makes a certain setting to the vehicle in a certain scenario, the value of the user's setting interaction preference (setting preference level) for that setting to the vehicle in that scenario is increased by "1".

[0065] The second approach is as follows: When a user makes a certain setting to the vehicle in a certain scenario, if the user does not change the system settings, the system increments the user's setting interaction preference value for that setting in that scenario by "1"; if the user changes the system settings, the system decrements the user's setting interaction preference value for that setting in that scenario by "1".

[0066] The third method is to map positive values ​​of the setting interaction preference to "1" and negative values ​​to "-1". For example, if a user closes the car window once when entering a highway, and then the system automatically sets the window to be closed twice more when entering the highway, the setting interaction preference value is 3, which is positive. Therefore, this setting interaction preference with a positive value is mapped to "1". Similarly, if user D changes the system's recommended foot vent mode twice when the temperature is below -10°C, the setting interaction preference value is -2, which is negative. Therefore, this setting interaction preference with a negative value is mapped to "-1".

[0067] 2. Construct user feature vectors and set feature vectors based on user information, scenario information, setting information, and interaction preferences;

[0068] In different embodiments, the expression form of the user feature vector is different, and the expression form of the feature vector is also different. These will be explained in detail in the following specific embodiments.

[0069] 3. Based on user feature vectors and set feature vectors, predict the user's preference for the settings through correlation calculation;

[0070] The method for calculating the correlation is as follows: calculate the dot product of the user feature vector and the set feature vector, or calculate the cosine similarity between the user feature vector and the set feature vector. The specific calculation process will be described in the following embodiments.

[0071] 4. Configure vehicles based on predicted preference levels.

[0072] The following three embodiments illustrate the specific implementation process of the big data-based in-vehicle human-machine interaction method provided in this application.

[0073] Example 1

[0074] The technical implementation of the big data-based in-vehicle human-machine interaction method provided in Example 1 includes the following steps:

[0075] 1. Construct user feature vectors and setting feature vectors based on user information, scenario information, setting information, and setting interaction preferences. Specifically, the user feature vector is user u. i (i represents the i-th user) The vector is composed of the preference levels of the n settings that the user has interacted with in the past (or the most recent in the past). The obtained setting feature vector is the vector composed of the similarity between setting s1 and each of the aforementioned n settings s2.

[0076] 2. The similarity between settings is determined by the audiences of the two settings. The higher the overlap of the audiences, the more similar the two settings are. The specific calculation process is as follows: Let the user set of setting s1 be S1, the user set of setting s2 be S2, and the intersection of their users be S1. C =S1∩S2, then the similarity between setting s1 and setting s2 is:

[0077]

[0078] 3. Obtain the n settings that the current user has interacted with in the past to form a setting feature vector. Then, find the k settings in the setting feature vector associated with the current user, arranged from high to low similarity, to form a candidate set as the basis for automatic vehicle settings for the user.

[0079] For example, user u i If the feature vector contains n settings, then for each setting in the feature vector associated with a user, k settings are found that are arranged in descending order of similarity. A total of n×k settings constitute the set of settings associated with user u. i Candidate set for vehicle configuration.

[0080] 4. Configure the vehicle based on the predicted preference levels, including ranking the preference levels of multiple settings and selecting the setting with the highest preference level for vehicle control. The specific method for obtaining the preference levels of multiple settings is as follows:

[0081] Get a user u i The n most recently interacted settings;

[0082] For each of the n settings, find the corresponding k settings. The similarity calculation method for the settings is as follows:

[0083]

[0084] Duplicate settings in the candidate set are removed, as well as extended settings in the candidate set that are inconsistent with the current scenario information. Then, the user's preference for each remaining setting in the candidate set is predicted by multiplying the settings. The setting with the highest associated preference is used as the prediction result of how the current user will set up the vehicle in the current scenario.

[0085] The method for predicting a user's preference for a setting using dot product calculation is as follows: Using the known user preferences for n settings (i.e., user feature vectors) and the similarity between a given setting and these n settings (i.e., setting feature vectors), we obtain the user's preference for that specific setting. For example, the user feature vectors... Set feature vectors The dot product of the two is

[0086]

[0087] It should also be noted that in Embodiment 1, the settings are preferably extended settings, which are fused information that includes setting information and scene information for configuring the vehicle in the corresponding scenario. That is, the settings do not include scene information, while the extended settings do.

[0088] Example 2

[0089] One difference between Example 2 and Example 1 lies in the user feature vector and the setting of the feature vector. In Example 2, the user feature vector is: User u i This is a vector formed by comparing the similarity scores of user u with the scores of the top p users who share similarity in user features, ranked from highest to lowest. i The obtained feature vector is a vector consisting of the preference of p users for each setting s1 that they have interacted with in the past for the same scenario.

[0090] Another difference between Example 2 and Example 1 is that in Example 2, the similarity between two users is determined by their preferences. The higher the overlap in preferences, the more similar the two users are. The specific calculation process is as follows:

[0091] Let K1 be the set of settings that user u1 has interacted with in the past, and K2 be the set of settings that user u2 has interacted with in the past. The intersection of their settings is K. C =K1∩K2, the similarity between user u1 and user u2 is:

[0092]

[0093] Then, find p users who have similar user characteristics to the current user, and find n settings among the p users whose historical interaction scene information is consistent with the current scene information of the vehicle to be set by the current user, for a total of n×p settings as candidate set.

[0094] For example, with user u i If there are p users with similar user characteristics, then we find n settings for each of the p users whose historical interaction scene information matches the current scene information of the vehicle to be set for the current user. A total of n×p settings constitute the settings for the current user u. i Candidate set for automatic vehicle configuration.

[0095] Then, as in Example 1, the vehicle is configured according to the predicted preference level, including sorting the preference levels of n×p settings and selecting the setting with the highest preference level as the current setting for controlling the vehicle as predicted by the current user. The method for obtaining the preference levels of multiple settings is the same as in Example 1, and will not be repeated here.

[0096] It should also be noted that in Embodiment 2, the settings are preferably extended settings, which are fused information that includes setting information and scene information for configuring the vehicle in the corresponding scenario. That is, the settings do not include scene information, while the extended settings do.

[0097] Example 3

[0098] The in-vehicle human-machine interaction method based on big data provided in Example 3 specifically includes the following steps:

[0099] 1. Based on user information, scene information, settings information, and interaction preferences, construct the current user's feature vector and setting feature vector. Specifically, distinguish between discrete and continuous information in the user information, scene information, and settings information, and perform preprocessing on each to obtain preprocessed vectors. Then, feed the preprocessed vectors into the neural network to obtain the feature vectors.

[0100] Discrete user information includes user ID and / or user gender; continuous user information includes user age and / or clothing index. Discrete scene information includes weather conditions and / or vehicle location; continuous scene information includes outside temperature and / or inside temperature. Discrete setting information includes airflow mode and / or recirculation mode; continuous setting information includes set temperature and / or set fan speed.

[0101] Discrete and / or continuous user information and / or scene information are preprocessed and then input into the user neural network to obtain user feature vectors; discrete and / or continuous setting information and / or scene information are preprocessed and then input into the setting neural network to obtain setting feature vectors.

[0102] The preprocessing method for user discrete information, scene discrete information and setting discrete information is as follows: multiply the one-hot vector corresponding to the discrete information by the Embedding matrix to obtain the Embedding vector as the discrete information preprocessing vector.

[0103] Examples of one-hot vectors corresponding to discrete information are as follows:

[0104] Suppose a simplified dataset contains 10 users; in this case, the one-hot vector would be a 10-dimensional vector. For example:

[0105] The one-hot vector corresponding to user ID A is (1,0,0,0,0,0,0,0,0,0,0). T ,

[0106] The one-hot vector corresponding to user ID B is (0,1,0,0,0,0,0,0,0,0,0). T ,

[0107] The one-hot vector corresponding to user ID C is (0,0,1,0,0,0,0,0,0,0). T ,

[0108] …

[0109] The one-hot vector corresponding to user ID J is (0,0,0,0,0,0,0,0,0,1). T ,

[0110] The process of multiplying a one-hot vector by the embedding matrix to obtain the embedding vector is illustrated below:

[0111] The embedding matrix is ​​an m×n matrix, where n is the same dimension as the one-hot vector. m is a quantity less than n. Similarly, in a large dataset with 10 users, if m is set to 3, then the embedding matrix E is a 3×10 matrix. The embedding vector for user C is calculated as follows:

[0112]

[0113] The preprocessing method for continuous user information, continuous scene information, and continuous setting information is as follows: the continuous information is converted into discrete information after being binned, and then the one-hot vectors corresponding to the discrete information are multiplied by the embedding matrix to obtain the embedding matrix. g Vectors can be used as preprocessing vectors for discrete information or for normalizing continuous information.

[0114] Taking user age as an example, binning involves dividing users into groups based on age: children, youth, middle-aged, and elderly. After binning, continuous information is converted into discrete information and processed according to the discrete information processing method described above. Normalization uses linear transformations to map continuous values ​​to a specified interval. For example, a set temperature t∈[15, 30] is mapped to the interval [0, 1] using f(t)=(t-15) / 15.

[0115] The neural network processes the preprocessed information to obtain the user feature vector, and the method for setting the feature vector is as follows:

[0116] The parameters in the neural network and the parameters in the embedding matrix are obtained by an optimization algorithm. A neural network is composed of various basic operational modules, including linear layers, non-linear activation functions, normalization layers, and attention layers. The parameters of the neural network are the parameters of these operational modules. The parameters of the embedding matrix are the elements of this matrix. For example, obtaining the parameters W in the neural network and E in the embedding matrix transforms into the following optimization problem:

[0117]

[0118] Where p represents the setting interaction preference information, u is the user feature vector output by the user neural network, s is the setting feature vector output by the setting neural network, and cosine(u, s) is used to calculate the cosine similarity between the two vectors, with the specific expression as follows:

[0119]

[0120] L is the loss function, which can be mean squared loss, cross-entropy loss, or triplet loss. The optimization problem described above can be solved using various optimization algorithms to determine W and E. These algorithms include stochastic gradient descent, momentum method, AdamW method, and RMSprop method. Simply put, the optimization objective is to make the algorithm's predicted preference cosine(u, s) as close as possible to the actual interaction preference p in the big data dataset. For example, consider this data: User A, age 25, male, in a scenario with an outside temperature of 36℃, has a preference of "1" for a set temperature of 22.5℃, a fan speed of 3, and a setting of blowing air onto the face with internal circulation. Based on the user information, scenario information, and setting information, the prediction result obtained after data preprocessing and neural network processing should be as close to "1" as possible. By using optimization algorithms to update the parameters in the neural network and the embedding matrix, the predicted result can be made increasingly closer to the true value of the interaction preference.

[0121] The user neural network and the setting neural network predict the vehicle settings for the current user in the current scenario as follows:

[0122] First, based on the input preprocessed vector, the user feature vector and the set feature vector are output after data processing. The specific method is as follows:

[0123] The user preprocessing vector and the scene preprocessing vector are combined and used as the input vector u of the user neural network. i The output of the user neural network is the user feature vector u. o An example of a user neural network is as follows:

[0124] l1=W1u i +b1

[0125] l n =a(W n l n-1 +b n ), n = 2, ..., N-1

[0126] u o =a(W N l N-1 +b N )

[0127] In this context, the basic operational module of the form Wl+b is the linear layer, where W and b are the parameters of the linear layer. a(·) is the non-linear activation function, which here is tanh, but can also be sigmoid, ReLU, or PReLU.

[0128] Set the preprocessed vector as the input vector s of the neural network. i Setting the output of the neural network is equivalent to setting the feature vector s. o Here is an example of setting up a neural network:

[0129] l1=W1s i +b1

[0130] l n =a(W n l n-1 +b n ), n = 2, ..., N-1

[0131] s o =a(W N l N-1 +b N )

[0132] In this context, the basic operational module of the form Wl+b is the linear layer, where W and b are the parameters of the linear layer. a(·) is the non-linear activation function, which here is tanh, but can also be sigmoid, ReLU, or PReLU.

[0133] The settings information elements are combined to obtain all possible settings as a candidate set. The current user's preference for each setting in the candidate set is predicted using dot product or cosine similarity calculations. Finally, the setting with the highest preference among the current user's preferences in the candidate set is extracted, and associated with this highest preference setting as the predicted vehicle settings for the current user.

[0134] In summary, the in-vehicle human-machine interaction method based on big data provided in the above three embodiments of this application, such as... Figure 1 As shown, the steps include:

[0135] L1 acquires the current user's in-vehicle human-machine interaction habit data, including user information, scene information, setting information, and setting interaction preferences, in order to construct the user feature vector associated with the current user and acquire the setting feature vector;

[0136] L2, construct a candidate set for predicting vehicle settings for the current user;

[0137] L3 extracts the highest preference level among the current user's preferences for each setting in the candidate set, and uses the setting with the highest preference level as the prediction result for the current user's vehicle settings.

[0138] In summary, this invention provides three different solutions through its three embodiments to automatically configure the vehicle for the current user in the current scenario. The three embodiments construct candidate sets using different methods, mining the relationships between users, between settings, and between users and settings. User features are represented by user feature vectors, and setting features are represented by setting feature vectors. Simultaneously, the correlation between these features is used to calculate the user's preference for settings, and the settings are ranked according to preference levels for automatic vehicle configuration. Compared to existing methods that learn settings independently for each individual user, this approach results in more intelligent human-computer interaction and a significantly improved user experience.

[0139] It should be stated that the above-described specific embodiments are merely preferred embodiments of the present invention and the technical principles employed. Those skilled in the art should understand that various modifications, equivalent substitutions, and variations can be made to the present invention. However, such variations, as long as they do not depart from the spirit of the present invention, should be within the scope of protection of the present invention. Furthermore, some terminology used in this specification and claims is not limiting, but merely for ease of description.

Claims

1. A vehicle-mounted human-machine interaction method based on big data, characterized in that, Including the following steps: L1, acquire the current user's in-vehicle human-machine interaction habit data, including user information, scene information, setting information, and setting interaction preferences, in order to construct the user feature vector associated with the current user and acquire the setting feature vector; L2, construct a candidate set for predicting vehicle settings for the current user; L3 extracts the highest preference level among the current user's preferences for each setting in the candidate set, and uses the setting with the highest preference level as the prediction result for the current user's vehicle settings.

2. The in-vehicle human-machine interaction method based on big data according to claim 1, characterized in that, The user information includes any one or more of the following: user ID, user age, user gender, clothing index, physiological information, and emotional information; The scene information includes any one or more of the following: outside temperature, inside temperature, vehicle speed, remaining battery power, remaining fuel, weather conditions, light intensity, current time, vehicle's geographical location, road navigation information, current road type, and driving gradient. The settings information includes any one or more of the following: air conditioning settings, seat settings, fragrance settings, window settings, ambient lighting settings, and audio settings.

3. The in-vehicle human-machine interaction method based on big data according to claim 1, characterized in that, The user feature vector is the user Historical interactions The vector formed by the preference levels of each setting; the obtained setting feature vector is the setting and Each setting in each setting The vector is formed by the similarity of the two.

4. The in-vehicle human-machine interaction method based on big data according to claim 1 or 3, characterized in that, The method for constructing the candidate set in step L2 is as follows: Get the current user's historical interactions Each setting, then find and Each of the settings has a similarity in setting characteristics. There are 1 setting, total Each setting constitutes the candidate set that serves as the basis for automatically configuring the vehicle for the current user.

5. The in-vehicle human-machine interaction method based on big data according to claim 3, characterized in that, set up and settings The similarity is: , Indicates the settings user set With Settings user set The user intersection.

6. The in-vehicle human-machine interaction method based on big data according to claim 1, characterized in that, The user feature vector is: User Compared with the top users who have similarity in user characteristics and are ranked from highest to lowest similarity. The vector is composed of the similarity of each user; the obtained feature vector is: Settings for individual users interacting with the current scenario The vector formed by the degree of preference.

7. The in-vehicle human-machine interaction method based on big data according to claim 1 or 6, characterized in that, In step L2, the method for constructing the candidate set is as follows: Find users with similar user characteristics to the current user. One user, then search The historical interaction scenario information of each user is consistent with the current scenario information of the vehicle to be set for the current user. There are 1 setting, total Each setting serves as the candidate set.

8. The in-vehicle human-machine interaction method based on big data according to claim 6, characterized in that, user and users The similarity is: , Indicates the user The set of settings for various historical interactions With the user A collection of settings from all previously interacted settings. The intersection of settings.

9. The in-vehicle human-machine interaction method based on big data according to claim 3 or 6, characterized in that, The setting is an extended setting, which is a fusion information that includes the setting information and the scene information for setting the vehicle in the corresponding scenario.

10. The in-vehicle human-machine interaction method based on big data according to claim 1, characterized in that, Step L3 specifically includes the following steps: L31, Remove duplicate settings and settings inconsistent with the current scene information from the candidate set; L32, predict the current user's preference for each setting in the remaining candidate set after step L31 by calculating the dot product; L33, extract the highest preference level of the current user among the preferences of each setting in the candidate set remaining after step L31, and use the setting with the highest preference level as the prediction result for the vehicle settings of the current user.

11. The in-vehicle human-machine interaction method based on big data according to claim 1, characterized in that, In step L1, the in-vehicle human-machine interaction habit data is distinguished into discrete information and continuous information and preprocessed. The preprocessed vector is then input into the neural network, and the user feature vector and the set feature vector are output. The discrete user information of the user information includes user ID and / or user gender, and the continuous user information includes user age and / or clothing index; the discrete scene information of the scene information includes weather conditions and / or the geographical information of the vehicle, and the continuous scene information includes outside temperature and / or inside temperature; the discrete setting information of the setting information includes air outlet mode and / or circulation mode, and the continuous setting information includes set temperature and / or set wind speed.

12. The in-vehicle human-machine interaction method based on big data according to claim 11, characterized in that, The discrete and / or continuous user information and / or scene information are preprocessed and then input into the user neural network to obtain the user feature vector; the discrete and / or continuous setting information and / or scene information are preprocessed and then input into the setting neural network to obtain the setting feature vector. The preprocessing method for user discrete information, scene discrete information and setting discrete information is as follows: convert the one-hot vector corresponding to the discrete information into an embedding matrix, and use the embedding vector as the discrete information preprocessing vector. The preprocessing methods for continuous user information, continuous scene information, and continuous setting information are as follows: after the continuous information is binned, it is converted into discrete information, and then the one-hot vector corresponding to the discrete information is multiplied by the embedding matrix to obtain the embedding vector as the preprocessing vector of the discrete information; or the continuous information is normalized.

13. The in-vehicle human-machine interaction method based on big data according to claim 12, characterized in that, Bucketing involves grouping the continuous information, while normalization involves mapping the continuous values ​​of the continuous information to a specified numerical range using linear transformation.

14. The in-vehicle human-machine interaction method based on big data according to claim 12 or 13, characterized in that, The method for predicting the current user's vehicle settings in the current scenario using a neural network is as follows: The cosine similarity between the user feature vector and the setting feature vector is calculated as the current user's preference for the setting in the current scenario, and the setting with the highest preference is used for vehicle control.

15. The in-vehicle human-machine interaction method based on big data according to claim 1, characterized in that, The method for calculating the user's preference for settings is as follows: When a user sets the vehicle settings in a specified scenario, the value of the user's preference level for the vehicle settings in the specified scenario is incremented by "1". After the system configures the vehicle settings for the user in the specified scenario, if the system detects that the user re-enters the specified scenario and does not change the system's settings after entering the specified scenario again, the user's preference level for the vehicle settings in the specified scenario is incremented by "1"; if the user changes the system's settings, the user's preference level for the settings in the specified scenario is decremented by "1". Or map positive preferences to "1" and negative preferences to "-1".

16. The in-vehicle human-machine interaction method based on big data according to any one of claims 11-13, characterized in that, Step L3 specifically includes the following steps: L301 predicts the current user's preference for each setting in the candidate set by calculating dot product or cosine similarity. L302, extract the highest preference level among the current user's preference levels for each setting in the candidate set, and use the setting with the highest preference level as the prediction result for the current user's vehicle settings.

17. A vehicle-mounted human-machine interaction system based on big data, characterized in that, It can realize the vehicle-mounted human-machine interaction method based on big data as described in any one of claims 1-16.

Images