Automobile seat motor intelligent sliding rail adjusting system based on AI voice interaction

Abstract

The invention discloses an automobile seat motor intelligent sliding rail adjusting system based on AI voice interaction, and the system comprises the following modules: a local voice wake-up module which is used for monitoring the voice input of a user, and activating a voice recognition module after detecting a preset wake-up keyword; the voice recognition module is used for generating a standardized text instruction; the semantic modeling module is used for executing word vector modeling processing on the standardized text instruction; the control parameter mapping module is used for inputting the semantic feature vector into a trained neural network model; the slide rail state detection module is used for collecting slide rail state information in real time; the voice synthesis module is used for converting the content in the prompt information field into a voice output content template; the vehicle-mounted audio output module is used for generating broadcast content; and the voice interaction log module is used for recording information involved in the voice interaction process. Semantic modeling and control mapping are fused, and the intelligent voice-driven seat adjusting system is constructed.

Description

Technical Field

[0001] The present invention relates to the field of natural language processing and intelligent execution technology, and in particular to an intelligent slide rail adjustment system for a car seat motor based on AI voice interaction. Background Art

[0002] Against the backdrop of the rapid development of intelligent and connected vehicles, seat comfort adjustment systems, as an essential component for enhancing the user's driving experience, have gradually evolved from traditional mechanical manual adjustment to electric control, intelligent recognition, and personalized adaptation. Especially in high-end vehicles, electric seat adjustment has become a standard feature, and its control methods have gradually expanded to central control screen operation, physical button control, and even voice interaction. However, although some models have introduced simple voice command control systems for basic operations such as starting and stopping in-vehicle functions and setting navigation settings, systems that deeply integrate natural language semantic understanding capabilities with the seat motor adjustment process are still extremely rare. Existing technologies still have significant limitations in accurately identifying user command intent, parsing adjustment parameter details, and providing feedback prompts after motor adjustment.

[0003] Traditional voice-controlled in-vehicle systems rely primarily on keyword triggering to identify user intent, have weak semantic understanding capabilities, and struggle to cope with diverse natural spoken expressions. For example, when a user issues natural language commands such as "move the seat back a little," "faster," or "lower the backrest to its original position," existing systems can often only process predefined fixed phrases such as "front and back" and "fast and slow." They lack support for the precise extraction and parameterized control of detailed elements such as distance, speed, direction, and mode, resulting in inaccurate or unresponsive voice adjustments. Furthermore, existing systems generally lack a control consistency verification mechanism between voice input and motor execution, making it prone to problems such as ambiguous commands, overshoot in adjustments, and target position exceeding limits, impacting the stability and safety of seat adjustment.

[0004] On the other hand, most current electric slide systems use a fixed-logic parameter configuration method, lacking the ability to understand and adapt to user preferences and context. Slide control typically triggers motor movement through preset positions or static logic, failing to dynamically generate personalized parameter combinations based on the user's current semantic intent. Furthermore, most systems lack a clear feedback mechanism for execution results after adjustments are completed, making it difficult for users to determine whether the slide has been accurately adjusted to the target position. The system interaction process lacks closed-loop confirmation and user-friendly prompts, and lacks the ability to continuously record and analyze user voice operation behaviors.

[0005] In response to the above problems, the present invention proposes a car seat slide adjustment system based on the combination of semantic understanding and control mapping. It uses the improved FastText algorithm to enhance the semantic modeling capability, drives the generation of control parameters through semantic vectors, and combines neural networks to achieve fine mapping of slide displacement distance, speed and adjustment mode. At the same time, it constructs a closed-loop execution feedback path and voice prompt mechanism, thereby solving the problems of fuzzy voice control, rough parameter mapping, and lack of feedback closed loop in existing voice control, and significantly improving the intelligence, accuracy and interactive friendliness of the seat adjustment process.

[0006] Therefore, how to provide an intelligent slide rail adjustment system for car seat motors based on AI voice interaction is an urgent problem that technicians in this field need to solve. Summary of the Invention

[0007] One purpose of the present invention is to propose an intelligent slide rail adjustment system for automobile seat motors based on AI voice interaction. The present invention integrates semantic modeling, neural network mapping and slide rail execution feedback mechanism to construct an intelligent seat adjustment system based on voice interaction. The system can accurately identify user natural language commands, generate precise control parameters and execute slide rail adjustment actions. At the same time, the adjustment results are prompted through a closed-loop voice feedback loop, thereby improving the system intelligence and user experience.

[0008] According to an embodiment of the present invention, an intelligent slide rail adjustment system for a car seat motor based on AI voice interaction includes the following modules:

[0009] The local voice wake-up module is used to continuously monitor user voice input during vehicle operation and activate the voice recognition module after detecting that the user voice input contains a preset wake-up keyword;

[0010] A speech recognition module, used to convert natural speech signals emitted by users into standardized text instructions;

[0011] The semantic modeling module is used to perform word vector modeling processing on standardized text instructions and generate semantic feature vectors using the improved FastText algorithm;

[0012] A control parameter mapping module is used to input the semantic feature vector output by the semantic modeling module into the trained neural network model;

[0013] The slide rail status detection module is used to collect the current physical displacement position of the seat slide rail, motor working status and sensor signal information in real time;

[0014] The speech synthesis module is used to convert the content in the prompt information field into a speech output content template and select the corresponding broadcast language, broadcast volume and speech speed settings based on the current cockpit voice configuration;

[0015] The vehicle audio output module is used to play the announcement content generated by the speech synthesis module through the vehicle speaker system, providing audio prompt feedback on the seat rail adjustment results;

[0016] The voice interaction log module is used to record the voice input content, recognition results, adjustment instructions, execution status and prompt content involved in the voice interaction process, and store them locally.

[0017] According to an embodiment of the present invention, a method for adjusting an automobile seat motor slide rail based on AI voice interaction includes the following steps:

[0018] S1. Collect user voice data through the vehicle microphone, use the local voice wake-up module to detect whether it contains the preset wake-up keyword, and activate the voice recognition module when the preset wake-up keyword is detected;

[0019] S2. In the speech recognition module, the collected user speech data is converted into standardized text instructions and preprocessed;

[0020] S3. Input the processed standardized text instructions into a semantic modeling module, and the semantic modeling module uses an improved FastText algorithm to perform word vector modeling on the standardized text instructions to generate a semantic feature vector of the standardized text instructions;

[0021] S4. Inputting the semantic feature vector into a control parameter mapping module, which maps the semantic feature vector into a control parameter set including a target slide rail displacement distance, a target operating speed, and an adjustment mode control code based on a trained neural network model;

[0022] S5. Obtain the current displacement state of the slide rail and the maximum adjustable range of the slide rail, perform boundary correction based on the target displacement distance and the current position in the control parameter set, and obtain the target position and operating speed of the slide rail;

[0023] S6. Generate a slide rail control instruction, which includes a slide rail target position, operating speed parameters, adjustment mode information, and a communication check bit, and send the instruction to the slide rail motor controller via the vehicle control bus interface;

[0024] S7. After receiving the control command, the slide motor controller drives the seat slide motor to perform an adjustment action according to the specified parameters, and monitors the seat slide motor operating status and position feedback data in real time during the adjustment process;

[0025] S8. After the slide rail adjustment is completed, the actual slide rail position is compared with the target position to confirm the execution result of the adjustment action, and a voice prompt information is generated through the voice synthesis module to be fed back to the user.

[0026] Optionally, the preprocessing operation includes removing invalid stop words, correcting recognition errors, and extracting semantic entities.

[0027] Optionally, the improved FastText algorithm in S3 specifically includes:

[0028] S31. Construct a subword-level training corpus covering automotive voice control scenarios. The subword-level training corpus includes multiple types of text samples consisting of historical commands from users of the in-vehicle voice system, standard semantic templates, and common natural language control expressions, covering various types of colloquial expressions such as action command words, quantifier phrases, directional words, and vague expressions.

[0029] S32, performing subword segmentation processing on the input standardized text instruction, dividing the complete sentence into continuous subword units, and marking their sequential positions and grammatical dependencies in the original sentence to preserve word order information and hyponym semantic relationships;

[0030] S33. Use the extended FastText nested structure to map each subword unit to the word vector space:

[0031]

[0032] Among them, e i is the embedding vector of subword i, which is the basic unit output of word vector modeling. σ is the activation function, which is used to introduce nonlinear expression capabilities. W1 is the weight matrix, which is used to map the concatenated vector to the embedding space. b1 is the bias term, which adjusts the linear offset of the mapping output. concat() is the vector concatenation operation, which is used to merge the three types of information: subword frequency, grammatical position, and context mean. log(1+f i ) is the logarithmic transformation of the frequency of subword i, f i is the frequency of occurrence of the subword in the subword-level training corpus, p i is the grammatical position vector of subword i, including word order and dependency information in the sentence, is the set of context subwords of subword i, and c is the vector representation of each subword in the context. Shared embedding is performed by constructing a word-subword nested index;

[0033] S34. During semantic modeling, a multi-scale context sliding window mechanism is used for subword unit sequences. The short window identifies the core verbs or verb phrases that represent regulatory behaviors in user speech data, and the long window understands the macro-semantic relationships between distance, direction, and speed elements in user speech data.

[0034] S35. Introduce an attention weighting mechanism based on the set of embedding vectors generated within the multi-scale context sliding window:

[0035]

[0036] Among them, v ctx is the weighted fusion vector of multi-scale context, K is the number of scales of different sliding windows, is the subword window set under the kth window scale, γ k is the overall attention fusion weight of the k-th window scale, e t is the embedding representation of the t-th subword in the window, is the transpose of the attention vector at the kth window scale, and softmax() is the normalized attention distribution function, so that the sum of the attention values ​​is 1. Based on the grammatical position, part-of-speech tagging, and historical usage frequency of each subword unit in the sentence, its semantic contribution is dynamically evaluated and weighted fused based on the embedding vector to generate an instruction representation vector with the main sentence characteristics;

[0037] S36: Concatenate the weighted fused command representation vector with the context interaction vector, where the context interaction vector includes similar control commands issued by the user in the past, records of executed adjustment modes, and scene preference labels. By fusing the context information, the multi-round understanding capability of the current sentence is enhanced, and a context enhancement vector is output.

[0038] S37, performing normalization and dimension mapping processing on the context enhancement vector, reconstructing it in a fixed-length vector format, and generating a semantic feature vector with good structural stability and discriminative performance by removing sub-dimensions with low confidence, performing dimension regularization and information compression;

[0039] S38. Output the above-mentioned semantic feature vector as the only input of the control parameter mapping module. The semantic feature vector includes a main sentence representation sub-block, a context preference encoding field, a structural position information flag, and an adjustment scene label field, which is used for the subsequent generation process of driving the target slide rail displacement distance, operating speed, and execution mode.

[0040] Optionally, the S33 specifically includes:

[0041] S331, performing a subword splitting operation on each standard term in the subword-level training corpus, decomposing the standard term into multiple subword units using a fixed-length window sliding method, generating a unique embedding index number for each subword unit, and outputting a set of mapping relationships between the standard term and the subword unit;

[0042] S332. Based on the mapping relationship set, construct a word-subword nested index table, wherein the word-subword nested index table defines all subword indexes contained in each standard term and records word order information of the subwords;

[0043] S333. Call the word-subword nested index table to initialize embedding vector parameters for each subword unit. The embedding vector parameters are weighted and distributed according to the frequency of occurrence of the subword unit in the entire subword-level training corpus to form an initial subword embedding matrix.

[0044] S334. Based on the initial subword embedding matrix, multiple subword unit embedding vectors belonging to the same standard term are combined and calculated to generate a nested word vector representation of the standard term:

[0045]

[0046] Among them, v word is the nested word vector representation of a standard term consisting of multiple nested subwords, n is the total number of subword units that constitute the term, α j is the attention coefficient of the jth subword, indicating its weight in the semantics of the word, e j is the embedding vector of the jth subword, r j is the relative position structure vector of the subword, marking its order and position dependency in the entry. W2 is the mapping matrix, which converts the structural information into the mapping parameters of the semantic space. b2 is the bias term used to adjust the structural mapping result. tanh() is the hyperbolic tangent activation function, which makes the mapping value in the (-1, 1) interval. It saves the relative position structure information of the subword unit in the standard entry.

[0047] S335. For a term whose frequency of appearance in the subword-level training corpus is lower than a preset threshold, extract the set of all subword unit embedding vectors corresponding to the term by querying the term-subword nested index table, and call the FastText nested structure to perform an embedding construction operation on the set of all subword unit embedding vectors to generate a complete word vector for the low-frequency term;

[0048] S336. After the embedding training process is completed, a unified embedding storage structure is established. The embedding storage structure includes three parts: a mapping relationship set, an initial subword embedding matrix, and a word-subword index table.

[0049] Optionally, the S4 specifically includes:

[0050] S41, receiving a semantic feature vector output by a semantic modeling module;

[0051] S42, inputting the semantic feature vector into an embedded neural network mapping model, wherein the embedded neural network mapping model is composed of a multi-layer perceptron neural network and has been supervised trained based on standardized text instructions;

[0052] S43. At the neural network input layer, the semantic feature vectors are distributed to multiple parallel processing channels. Different processing channels focus on semantic substructures related to the slide displacement, adjustment speed, and execution mode, respectively, to improve the parameter decoupling accuracy.

[0053] S44, converting each semantic substructure into a high-dimensional expression through the middle hidden layer of the embedded neural network, and completing the feature nonlinear compression processing under the action of the internal activation mechanism to generate the potential control representation of the action decision;

[0054] S45. In the neural network output layer, the potential control representation is mapped into three control parameter items, namely, the target slide rail displacement distance, the target operating speed, and the adjustment mode control code. Each parameter item corresponds to an independent output channel, and a control parameter set is output.

[0055] S46. Perform semantic consistency check on the control parameter set and the current state of the vehicle, including comparison with the maximum displacement limit, current speed threshold, current seat position, and system adjustment mode configuration table, and output the final control parameter result that passes the check.

[0056] Optionally, the S5 specifically includes:

[0057] S51, calling the slide rail state detection module to obtain the current slide rail displacement state, the slide rail displacement state is collected by the slide rail position sensor and uploaded to the control processing unit to represent the current physical position of the slide rail;

[0058] S52, accessing the maximum adjustable range parameters of the slide rail stored in the vehicle system, wherein the maximum adjustable range parameters of the slide rail include a starting limit position of the slide rail, a maximum end limit position, and a preset safety tolerance range;

[0059] S53, extracting a target rail displacement distance from the control parameter set, and adding the target rail displacement distance to the current rail displacement state to generate an initial target rail position value as the adjustment target position to be corrected;

[0060] S54: Perform boundary constraint processing on the initial target rail position value. If the initial target rail position value exceeds the maximum adjustable range, correct it to the nearest legal position range according to the safety limit strategy configured by the vehicle system, and trigger the fault tolerance adjustment mode if necessary.

[0061] S55. Generate an operating speed parameter that matches the target slide rail position based on the difference between the corrected target slide rail position and the current actual position and in combination with the execution strategy corresponding to the current adjustment mode;

[0062] S56: Output the target slide rail position after boundary correction and the matching running speed parameters.

[0063] Optionally, the S6 specifically includes:

[0064] S61, receiving the target slide rail position and running speed parameters after boundary correction, and synchronously obtaining the adjustment mode information in the control parameter set to form the core content field of the slide rail control instruction;

[0065] S62. Establishing a data frame format for the slide rail control instruction according to a predefined slide rail control protocol structure of the system, wherein the data frame format includes a fixed frame header, a target position field, a running speed field, an adjustment mode field, and an instruction redundancy check field;

[0066] S63, filling the control parameter contents into the corresponding fields respectively, and packaging each field according to the byte alignment rules, field sequence requirements and data encoding method in the slide control protocol to form a complete slide control instruction frame to be sent;

[0067] S64, performing consistency check and redundancy comparison on the encapsulated complete slide rail control instruction frame to verify the integrity of the instruction structure, the legitimacy of the field values, and the rationality of the tolerance between the target position and the current slide rail state;

[0068] S65, transmitting the verified slide rail control instruction frame to the vehicle control bus, and sending it to the slide rail motor controller through the vehicle control bus interface, wherein the vehicle control bus interface supports CAN, LIN or vehicle Ethernet;

[0069] S66. After receiving the control instruction frame, the slide motor controller parses the field content, completes field decoding and verification logic confirmation based on the parsing result, and generates a control instruction.

[0070] Optionally, the S7 specifically includes:

[0071] S71. Initialize the slide control program according to the field decoding result, call the preset slide motor control channel, configure the drive polarity according to the target displacement direction, configure the PWM modulation parameters according to the target speed, and set the soft limit buffer and safety interrupt flag.

[0072] S72. After receiving the control command, the slide motor controller starts the slide motor, performs position detection and current monitoring, and periodically samples the real-time parameters of the slide motor during displacement.

[0073] S73. During the operation of the slide rail, the slide rail motor controller continuously monitors the progress of the target displacement and collects multiple operation status data including the current position of the slide rail motor, the operating current, the feedback voltage, and the driving load status;

[0074] S74. Perform abnormality detection on the collected multiple operating status data. If abnormal operating characteristics such as overcurrent, stagnation, drift, overtravel, or low-speed no response are found, immediately trigger a stop control command and record the current abnormal status flag code;

[0075] S75. After the slide motor reaches the control target or stops, the final displacement completion status, current position data and operation status record results are output.

[0076] Optionally, the S8 specifically includes:

[0077] S81. After the slide rail motor finishes running, obtain the actual position state of the current slide rail, and compare the actual position state with the target position set in the slide rail control instruction to determine the completion status of the adjustment action;

[0078] S82. Generate an execution status flag based on the result of the adjustment action completion judgment, wherein the execution status flag includes "adjustment successful", "position error exceeds limit", "operation interrupted", and "not in place".

[0079] S83, packaging the execution status flag, the current actual position value of the slide rail, the target position value of the slide rail, and the operation mode data into a prompt information field;

[0080] S84: Calling the speech synthesis module to convert the content in the prompt information field into a speech output content template, and selecting the corresponding broadcast language, broadcast volume, and speech speed settings according to the current cockpit voice configuration;

[0081] S85: Send the synthesized voice data to the vehicle audio output module, and control the vehicle speakers to broadcast prompt messages, including prompts such as "Seat adjustment completed", "Currently moved to the set position", and "Adjustment abnormality aborted, please check".

[0082] S86. Record the trigger source, prompt type, and broadcast content of this round of voice prompt interaction, and store them in the voice interaction log module.

[0083] The beneficial effects of the present invention are:

[0084] The intelligent slide rail adjustment system for automobile seat motors based on AI voice interaction proposed in this invention can effectively overcome the problems of limited voice recognition capabilities, inaccurate adjustment instruction parameters, and lack of control feedback mechanism in the existing technology, bringing significant technological progress and practical value.

[0085] First, by introducing an improved FastText algorithm for word embedding modeling, the system possesses enhanced semantic understanding capabilities, adapting to diverse, colloquial natural language commands. For example, for commands with ambiguous semantics or combined expressions, such as "Move the seat a little to the right," "Lower it a little," or "Hurry up to the front," the system not only accurately identifies the action intent but also extracts key parameters such as distance, direction, and speed. This avoids the misrecognition and unresponsiveness issues often associated with fixed phrase recognition in traditional voice control systems, significantly improving the naturalness and accuracy of voice interaction.

[0086] Secondly, the system built a control parameter mapping module driven by semantic vectors. Through a trained embedded neural network, the semantic feature vectors are efficiently mapped into a set of control parameters, including the target slide displacement distance, operating speed, and adjustment mode. In the subsequent boundary correction module, dynamic corrections are performed based on the current slide state and safety limit rules. This end-to-end semantic control mapping and state fusion mechanism not only enhances the precision of control command generation, but also ensures the safety and rationality of adjustment actions, avoiding operational risks caused by over-positioning or crossing boundaries.

[0087] In addition, the present invention establishes a complete closed-loop control and voice prompt feedback mechanism for the slide rail execution. After completing the slide rail action, the system will automatically compare the actual position with the target position, generate an execution status flag, and call the speech synthesis module to provide voice prompt feedback through the vehicle speaker, so that the user can know the adjustment result without visual confirmation. This mechanism not only improves the human-computer interaction experience of the system, but also enhances the user's perception and trust in the execution of the adjustment action. At the same time, the voice interaction log module also records all voice interaction data and adjustment process information, providing data support for subsequent system learning optimization and user behavior modeling.

[0088] In summary, this invention integrates multiple technologies such as semantic understanding, control mapping, execution feedback and human-computer interaction optimization to construct a car seat slide adjustment system with high intelligence, high precision and good user experience, which significantly improves the practical application capabilities of the on-board intelligent cockpit system in the field of voice control. BRIEF DESCRIPTION OF THE DRAWINGS

[0089] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0090] Figure 1 This is a structural diagram of an intelligent slide rail adjustment system for a car seat motor based on AI voice interaction proposed by the present invention;

[0091] Figure 2This is an overall flow chart of the car seat motor intelligent slide rail adjustment system based on AI voice interaction proposed by the present invention. DETAILED DESCRIPTION

[0092] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0093] refer to Figure 1 , an intelligent slide rail adjustment system for car seat motors based on AI voice interaction, including the following modules:

[0094] The local voice wake-up module is used to continuously monitor user voice input during vehicle operation and activate the voice recognition module after detecting that the user voice input contains a preset wake-up keyword;

[0095] A speech recognition module, used to convert natural speech signals emitted by users into standardized text instructions;

[0096] The semantic modeling module is used to perform word vector modeling processing on standardized text instructions and generate semantic feature vectors using the improved FastText algorithm;

[0097] A control parameter mapping module is used to input the semantic feature vector output by the semantic modeling module into the trained neural network model;

[0098] The slide rail status detection module is used to collect the current physical displacement position of the seat slide rail, motor working status and sensor signal information in real time;

[0099] The speech synthesis module is used to convert the content in the prompt information field into a speech output content template and select the corresponding broadcast language, broadcast volume and speech speed settings based on the current cockpit voice configuration;

[0100] The vehicle audio output module is used to play the announcement content generated by the speech synthesis module through the vehicle speaker system, providing audio prompt feedback on the seat rail adjustment results;

[0101] The voice interaction log module is used to record the voice input content, recognition results, adjustment instructions, execution status and prompt content involved in the voice interaction process, and store them locally.

[0102] This system modularly integrates local voice wake-up, speech recognition, semantic modeling, control mapping, state detection, speech synthesis, and log management functions to create a full-link voice interaction control architecture for automotive cockpit environments. It can stably recognize commands and complete slide motor control even without a network connection, effectively enhancing the intelligent level of seat adjustment. It addresses the problem of existing in-vehicle voice control relying on cloud-based recognition and being unable to perform complex semantic mapping, providing technical support for improving user driving safety and comfort.

[0103] refer to Figure 2 , a car seat motor intelligent slide rail adjustment method based on AI voice interaction, comprising the following steps:

[0104] S1. Collect user voice data through the vehicle microphone, use the local voice wake-up module to detect whether it contains the preset wake-up keyword, and activate the voice recognition module when the preset wake-up keyword is detected;

[0105] S2. In the speech recognition module, the collected user speech data is converted into standardized text instructions and preprocessed;

[0106] S3. Input the processed standardized text instructions into a semantic modeling module, and the semantic modeling module uses an improved FastText algorithm to perform word vector modeling on the standardized text instructions to generate a semantic feature vector of the standardized text instructions;

[0107] S4. Inputting the semantic feature vector into a control parameter mapping module, which maps the semantic feature vector into a control parameter set including a target slide rail displacement distance, a target operating speed, and an adjustment mode control code based on a trained neural network model;

[0108] S5. Obtain the current displacement state of the slide rail and the maximum adjustable range of the slide rail, perform boundary correction based on the target displacement distance and the current position in the control parameter set, and obtain the target position and operating speed of the slide rail;

[0109] S6. Generate a slide rail control instruction, which includes a slide rail target position, operating speed parameters, adjustment mode information, and a communication check bit, and send the instruction to the slide rail motor controller via the vehicle control bus interface;

[0110] S7. After receiving the control command, the slide motor controller drives the seat slide motor to perform an adjustment action according to the specified parameters, and monitors the seat slide motor operating status and position feedback data in real time during the adjustment process;

[0111] S8. After the slide rail adjustment is completed, the actual slide rail position is compared with the target position to confirm the execution result of the adjustment action, and a voice prompt information is generated through the voice synthesis module to be fed back to the user.

[0112] The present invention clarifies the data interaction paths and processing steps between each module in the entire process from voice input to slide rail execution feedback, constructs a closed-loop semantic understanding and action execution linkage mechanism, supports key operations such as voice wake-up, structured instruction generation, semantic encoding, boundary correction, feedback broadcasting, etc., overcomes the problems of disconnection between recognition and control chains and no echo in voice control in existing systems, and ensures the accuracy of instruction response and execution consistency of the entire system in complex contexts.

[0113] In this embodiment, the preprocessing operation includes removing invalid stop words, correcting recognition errors, and extracting semantic entities.

[0114] By performing multi-level preprocessing on the speech recognition results, including stop word removal, incorrect word correction and semantic subject extraction, the text quality processed by the semantic modeling module is effectively improved, the downstream control parameter generation error caused by colloquialism, punctuation errors or recognition bias is reduced, and the system's anti-interference ability and robustness in non-standard sentence input scenarios are significantly enhanced, solving the problem of the existing in-vehicle voice control system's weak ability to parse non-standard instructions.

[0115] In this embodiment, the improved FastText algorithm in S3 specifically includes:

[0116] S31. Construct a subword-level training corpus covering automotive voice control scenarios. The subword-level training corpus includes multiple types of text samples consisting of historical commands from users of the in-vehicle voice system, standard semantic templates, and common natural language control expressions, covering various types of colloquial expressions such as action command words, quantifier phrases, directional words, and vague expressions.

[0117] S32, performing subword segmentation processing on the input standardized text instruction, dividing the complete sentence into continuous subword units, and marking their sequential positions and grammatical dependencies in the original sentence to preserve word order information and hyponym semantic relationships;

[0118] S33. Use the extended FastText nested structure to map each subword unit to the word vector space:

[0119]

[0120] Among them, e i is the embedding vector of subword i, which is the basic unit output of word vector modeling. σ is the activation function, which is used to introduce nonlinear expression capabilities. W1 is the weight matrix, which is used to map the concatenated vector to the embedding space. b1 is the bias term, which adjusts the linear offset of the mapping output. concat() is the vector concatenation operation, which is used to merge the three types of information: subword frequency, grammatical position, and context mean. log(1+fi ) is the logarithmic transformation of the frequency of subword i, f i is the frequency of occurrence of the subword in the subword-level training corpus, p i is the grammatical position vector of subword i, including word order and dependency information in the sentence, is the set of context subwords of subword i, and c is the vector representation of each subword in the context. Shared embedding is performed by constructing a word-subword nested index;

[0121] S34. During semantic modeling, a multi-scale context sliding window mechanism is used for subword unit sequences. The short window identifies the core verbs or verb phrases that represent regulatory behaviors in user speech data, and the long window understands the macro-semantic relationships between distance, direction, and speed elements in user speech data.

[0122] S35. Introduce an attention weighting mechanism based on the set of embedding vectors generated within the multi-scale context sliding window:

[0123]

[0124] Among them, v ctx is the weighted fusion vector of multi-scale context, K is the number of scales of different sliding windows, is the subword window set under the kth window scale, γ k is the overall attention fusion weight of the k-th window scale, e t is the embedding representation of the t-th subword in the window, is the transpose of the attention vector at the kth window scale, and softmax() is the normalized attention distribution function, so that the sum of the attention values ​​is 1. Based on the grammatical position, part-of-speech tagging, and historical usage frequency of each subword unit in the sentence, its semantic contribution is dynamically evaluated and weighted fused based on the embedding vector to generate an instruction representation vector with the main sentence characteristics;

[0125] S36: Concatenate the weighted fused command representation vector with the context interaction vector, where the context interaction vector includes similar control commands issued by the user in the past, records of executed adjustment modes, and scene preference labels. By fusing the context information, the multi-round understanding capability of the current sentence is enhanced, and a context enhancement vector is output.

[0126] S37, performing normalization and dimension mapping processing on the context enhancement vector, reconstructing it in a fixed-length vector format, and generating a semantic feature vector with good structural stability and discriminative performance by removing sub-dimensions with low confidence, performing dimension regularization and information compression;

[0127] S38. Output the above-mentioned semantic feature vector as the only input of the control parameter mapping module. The semantic feature vector includes a main sentence representation sub-block, a context preference encoding field, a structural position information flag, and an adjustment scene label field, which is used for the subsequent generation process of driving the target slide rail displacement distance, operating speed, and execution mode.

[0128] Based on the traditional FastText algorithm, subword-level corpus training, word-subword nested modeling, multi-scale sliding window and attention fusion mechanism are introduced, enabling the system to have the ability to model low-frequency words, fuzzy expressions and dialect instructions, and to construct semantic representations with user preference awareness based on contextual historical interactions, overcoming the defects of shallow semantic understanding and context-unawareness of existing word vector models, and significantly improving the accuracy and generalization ability of instruction parsing.

[0129] In this embodiment, the S33 specifically includes:

[0130] S331, performing a subword splitting operation on each standard term in the subword-level training corpus, decomposing the standard term into multiple subword units using a fixed-length window sliding method, generating a unique embedding index number for each subword unit, and outputting a set of mapping relationships between the standard term and the subword unit;

[0131] S332. Based on the mapping relationship set, construct a word-subword nested index table, wherein the word-subword nested index table defines all subword indexes contained in each standard term and records word order information of the subwords;

[0132] S333. Call the word-subword nested index table to initialize embedding vector parameters for each subword unit. The embedding vector parameters are weighted and distributed according to the frequency of occurrence of the subword unit in the entire subword-level training corpus to form an initial subword embedding matrix.

[0133] S334. Based on the initial subword embedding matrix, multiple subword unit embedding vectors belonging to the same standard term are combined and calculated to generate a nested word vector representation of the standard term:

[0134]

[0135] Among them, v word is the nested word vector representation of a standard term consisting of multiple nested subwords, n is the total number of subword units that constitute the term, α j is the attention coefficient of the jth subword, indicating its weight in the semantics of the word, e j is the embedding vector of the jth subword, r jis the relative position structure vector of the subword, marking its order and position dependency in the entry. W2 is the mapping matrix, which converts the structural information into the mapping parameters of the semantic space. b2 is the bias term used to adjust the structural mapping result. tanh() is the hyperbolic tangent activation function, which makes the mapping value in the (-1, 1) interval. It saves the relative position structure information of the subword unit in the standard entry.

[0136] S335. For a term whose frequency of appearance in the subword-level training corpus is lower than a preset threshold, extract the set of all subword unit embedding vectors corresponding to the term by querying the term-subword nested index table, and call the FastText nested structure to perform an embedding construction operation on the set of all subword unit embedding vectors to generate a complete word vector for the low-frequency term;

[0137] S336. After the embedding training process is completed, a unified embedding storage structure is established. The embedding storage structure includes three parts: a mapping relationship set, an initial subword embedding matrix, and a word-subword index table.

[0138] By constructing a sub-word nesting structure, frequency-weighted distribution, and low-frequency word completion mechanism, the system's ability to generate word vectors when faced with cold-start corpus or long-tail control instructions is enhanced, effectively supplementing sparse semantic information and enhancing the integrity of the semantic space. This solves the problem of insufficient support for user personalized terms and industry terminology in traditional voice systems, and improves the adaptability and scalability of the voice system.

[0139] In this embodiment, the S4 specifically includes:

[0140] S41, receiving a semantic feature vector output by a semantic modeling module;

[0141] S42, inputting the semantic feature vector into an embedded neural network mapping model, wherein the embedded neural network mapping model is composed of a multi-layer perceptron neural network and has been supervised trained based on standardized text instructions;

[0142] S43. At the neural network input layer, the semantic feature vectors are distributed to multiple parallel processing channels. Different processing channels focus on semantic substructures related to the slide displacement, adjustment speed, and execution mode, respectively, to improve the parameter decoupling accuracy.

[0143] S44, converting each semantic substructure into a high-dimensional expression through the middle hidden layer of the embedded neural network, and completing the feature nonlinear compression processing under the action of the internal activation mechanism to generate the potential control representation of the action decision;

[0144] S45. In the neural network output layer, the potential control representation is mapped into three control parameter items, namely, the target slide rail displacement distance, the target operating speed, and the adjustment mode control code. Each parameter item corresponds to an independent output channel, and a control parameter set is output.

[0145] S46. Perform semantic consistency check on the control parameter set and the current state of the vehicle, including comparison with the maximum displacement limit, current speed threshold, current seat position, and system adjustment mode configuration table, and output the final control parameter result that passes the check.

[0146] This control parameter mapping module uses a multi-channel neural network to decouple and process semantic substructures, effectively mapping user semantic vectors into action execution parameters and performing consistency verification based on vehicle status. This solves the problems of unclear fusion of different semantic intentions and high parameter error generation rate in traditional methods. At the same time, through neural network training, the nonlinear mapping ability between semantics and control domains is enhanced, thereby improving the decision-making accuracy and execution safety of the entire system.

[0147] In this embodiment, the S5 specifically includes:

[0148] S51, calling the slide rail state detection module to obtain the current slide rail displacement state, the slide rail displacement state is collected by the slide rail position sensor and uploaded to the control processing unit to represent the current physical position of the slide rail;

[0149] S52, accessing the maximum adjustable range parameters of the slide rail stored in the vehicle system, wherein the maximum adjustable range parameters of the slide rail include a starting limit position of the slide rail, a maximum end limit position, and a preset safety tolerance range;

[0150] S53, extracting a target rail displacement distance from the control parameter set, and adding the target rail displacement distance to the current rail displacement state to generate an initial target rail position value as the adjustment target position to be corrected;

[0151] S54: Perform boundary constraint processing on the initial target rail position value. If the initial target rail position value exceeds the maximum adjustable range, correct it to the nearest legal position range according to the safety limit strategy configured by the vehicle system, and trigger the fault tolerance adjustment mode if necessary.

[0152] S55. Generate an operating speed parameter that matches the target slide rail position based on the difference between the corrected target slide rail position and the current actual position and in combination with the execution strategy corresponding to the current adjustment mode;

[0153] S56: Output the target slide rail position after boundary correction and the matching running speed parameters.

[0154] By introducing the rail boundary correction mechanism and soft limit logic processing, the system can automatically adjust the control strategy and match the reasonable speed when the target displacement exceeds the boundary or approaches the boundary position, avoiding structural interference and rail impact caused by incorrect adjustment. It overcomes the problem that the existing vehicle seat control system has no judgment mechanism for abnormal adjustment position and poses a safety hazard, and ensures that the system remains controllable and recoverable under various extreme operations.

[0155] In this embodiment, S6 specifically includes:

[0156] S61, receiving the target slide rail position and running speed parameters after boundary correction, and synchronously obtaining the adjustment mode information in the control parameter set to form the core content field of the slide rail control instruction;

[0157] S62. Establishing a data frame format for the slide rail control instruction according to a predefined slide rail control protocol structure of the system, wherein the data frame format includes a fixed frame header, a target position field, a running speed field, an adjustment mode field, and an instruction redundancy check field;

[0158] S63, filling the control parameter contents into the corresponding fields respectively, and packaging each field according to the byte alignment rules, field sequence requirements and data encoding method in the slide control protocol to form a complete slide control instruction frame to be sent;

[0159] S64, performing consistency check and redundancy comparison on the encapsulated complete slide rail control instruction frame to verify the integrity of the instruction structure, the legitimacy of the field values, and the rationality of the tolerance between the target position and the current slide rail state;

[0160] S65, transmitting the verified slide rail control instruction frame to the vehicle control bus, and sending it to the slide rail motor controller through the vehicle control bus interface, wherein the vehicle control bus interface supports CAN, LIN or vehicle Ethernet;

[0161] S66. After receiving the control instruction frame, the slide motor controller parses the field content, completes field decoding and verification logic confirmation based on the parsing result, and generates a control instruction.

[0162] This control command generation mechanism has designed a rigorous data frame structure and communication verification process, supports mainstream vehicle communication protocols (such as CAN, LIN, and automotive Ethernet), and ensures that control commands still have strong anti-interference capabilities and real-time response performance in complex vehicle environments through field packaging and consistency verification. It solves the technical problems of inconsistent control command structures and poor communication fault tolerance in traditional control commands, and improves system deployment compatibility and industrial-grade stability.

[0163] In this embodiment, the S7 specifically includes:

[0164] S71. Initialize the slide control program according to the field decoding result, call the preset slide motor control channel, configure the drive polarity according to the target displacement direction, configure the PWM modulation parameters according to the target speed, and set the soft limit buffer and safety interrupt flag.

[0165] S72. After receiving the control command, the slide motor controller starts the slide motor, performs position detection and current monitoring, and periodically samples the real-time parameters of the slide motor during displacement.

[0166] S73. During the operation of the slide rail, the slide rail motor controller continuously monitors the progress of the target displacement and collects multiple operation status data including the current position of the slide rail motor, the operating current, the feedback voltage, and the driving load status;

[0167] S74. Perform abnormality detection on the collected multiple operating status data. If abnormal operating characteristics such as overcurrent, stagnation, drift, overtravel, or low-speed no response are found, immediately trigger a stop control command and record the current abnormal status flag code;

[0168] S75. After the slide motor reaches the control target or stops, the final displacement completion status, current position data and operation status record results are output.

[0169] By collecting the operating status parameters of the slide motor in real time and setting abnormal detection thresholds, the system can automatically identify fault phenomena such as overcurrent, stalling, sliding drift, and low-speed stagnation during operation, and trigger emergency shutdown and status reporting, effectively preventing motor damage and operational failure. It solves the problem of the existing system adjustment process lacking real-time monitoring and safety interruption mechanism, and improves overall reliability and driving safety assurance capabilities.

[0170] In this embodiment, S8 specifically includes:

[0171] S81. After the slide rail motor finishes running, obtain the actual position state of the current slide rail, and compare the actual position state with the target position set in the slide rail control instruction to determine the completion status of the adjustment action;

[0172] S82. Generate an execution status flag based on the result of the adjustment action completion judgment, wherein the execution status flag includes "adjustment successful", "position error exceeds limit", "operation interrupted", and "not in place".

[0173] S83, packaging the execution status flag, the current actual position value of the slide rail, the target position value of the slide rail, and the operation mode data into a prompt information field;

[0174] S84: Calling the speech synthesis module to convert the content in the prompt information field into a speech output content template, and selecting the corresponding broadcast language, broadcast volume, and speech speed settings according to the current cockpit voice configuration;

[0175] S85: Send the synthesized voice data to the vehicle audio output module, and control the vehicle speakers to broadcast prompt messages, including prompts such as "Seat adjustment completed", "Currently moved to the set position", and "Adjustment abnormality aborted, please check".

[0176] S86. Record the trigger source, prompt type, and broadcast content of this round of voice prompt interaction, and store them in the voice interaction log module.

[0177] The system generates broadcast information by comparing feedback from the slide rail execution results, and combines the speech synthesis module to dynamically generate multi-language, adjustable-speed broadcast content, and records the details of this round of interaction in the log module. This not only provides users with a clear and intuitive voice feedback experience, but also provides data support for the construction of semantic memory for the in-vehicle intelligent assistant. It overcomes the shortcomings of the existing voice adjustment system with single feedback and no data archiving, and improves the transparency of interaction and the closed-loop service capability.

[0178] Example 1:

[0179] In order to verify the feasibility of the present invention in practice, the present invention was applied to the development process of a new electric SUV model that was launched on the market. The R&D team hopes to solve the problems of cumbersome use, slow response, and poor personalization of traditional physical buttons for adjusting seat positions by introducing a more intelligent in-car interaction method. Traditional button or knob-type seat slide adjustment relies on mechanical triggering. The user needs to manually complete a series of key operations, and the adjustment process does not accurately judge the slide boundary, which poses a risk of misadjustment. Voice control has bottlenecks in multi-round semantic understanding, low-frequency expression recognition, and control parameter mapping due to the weak generalization ability of existing models. In this context, the project team applied the car seat motor intelligent slide adjustment system based on AI voice interaction proposed in this invention to the vehicle seat control subsystem.

[0180] The user doesn't need to touch any buttons during the ride; simply voice commands such as "Adjust the seat back 20 cm," "Please reduce the seat speed," or "Restore comfort mode" are issued. The system activates the speech recognition module through the local voice wake-up module, then extracts semantic vectors using an improved FastText model and generates precise, executable adjustment parameters through the neural network control mapping module. The system automatically determines the current position and adjustment range of the slide rail, completes boundary corrections, issues control commands, and obtains real-time feedback on the slide rail's status and position data. Once the adjustment is complete, the user is notified through speech synthesis and the onboard speakers that the "seat has been adjusted to the set position," creating a complete intelligent voice closed-loop interactive experience.

[0181] The system underwent three rounds of voice adjustment testing on 50 prototype vehicles at a vehicle manufacturer's proving ground in South China. Each round compared the performance of traditional physical adjustment methods with AI voice adjustment across multiple dimensions, including recognition accuracy, adjustment precision, adjustment time, human-computer interaction satisfaction, system response latency, and the number of test samples. Table 1 shows a performance comparison and evaluation table for the intelligent rail adjustment system.

[0182] Table 1: Comparative evaluation of intelligent slide rail adjustment system performance

[0183]

[0184] As shown in Table 1, this system significantly outperforms traditional methods in terms of recognition accuracy and semantic understanding. In particular, in terms of adjustment precision, the average error has been reduced from ±1.2 cm to ±0.4 cm, improving adjustment efficiency by over 50%. System response latency has also been reduced to approximately one-quarter that of traditional methods. In passenger satisfaction evaluations, this system achieved an average score of 9.1, nearly a 3-point improvement over traditional methods, demonstrating that AI voice interaction enhances overall comfort and user experience.

[0185] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. An intelligent slide rail adjustment system for car seat motors based on AI voice interaction, characterized in that: Includes the following modules: The local voice wake-up module is used to continuously monitor user voice input during vehicle operation and activate the voice recognition module after detecting that the user voice input contains a preset wake-up keyword; A speech recognition module, used to convert natural speech signals emitted by users into standardized text instructions; The semantic modeling module is used to perform word vector modeling processing on standardized text instructions and generate semantic feature vectors using the improved FastText algorithm; A control parameter mapping module is used to input the semantic feature vector output by the semantic modeling module into the trained neural network model; The slide rail status detection module is used to collect the current physical displacement position of the seat slide rail, motor working status and sensor signal information in real time; The speech synthesis module is used to convert the content in the prompt information field into a speech output content template and select the corresponding broadcast language, broadcast volume and speech speed settings based on the current cockpit voice configuration; The vehicle audio output module is used to play the announcement content generated by the speech synthesis module through the vehicle speaker system, providing audio prompt feedback on the seat rail adjustment results; The voice interaction log module is used to record the voice input content, recognition results, adjustment instructions, execution status and prompt content involved in the voice interaction process, and store them locally.

2. The car seat motor intelligent slide rail adjustment system based on AI voice interaction according to claim 1 is characterized in that: The modules are implemented as follows: S1. Collect user voice data through the vehicle microphone, use the local voice wake-up module to detect whether it contains the preset wake-up keyword, and activate the voice recognition module when the preset wake-up keyword is detected; S2. In the speech recognition module, the collected user speech data is converted into standardized text instructions and preprocessed; S3. Input the processed standardized text instructions into a semantic modeling module, and the semantic modeling module uses an improved FastText algorithm to perform word vector modeling on the standardized text instructions to generate a semantic feature vector of the standardized text instructions; S4. Inputting the semantic feature vector into a control parameter mapping module, which maps the semantic feature vector into a control parameter set including a target slide rail displacement distance, a target operating speed, and an adjustment mode control code based on a trained embedded neural network model; S5. Obtain the current slide rail displacement state and the maximum adjustable range of the slide rail, perform boundary correction based on the target slide rail displacement distance and the current position in the control parameter set, and obtain the target slide rail position and operating speed; S6. Generate a slide rail control instruction, which includes a slide rail target position, operating speed parameters, adjustment mode information, and a communication check bit, and send the instruction to the slide rail motor controller via the vehicle control bus interface; S7. After receiving the control command, the slide motor controller drives the seat slide motor to perform an adjustment action according to the specified parameters, and monitors the seat slide motor operating status and position feedback data in real time during the adjustment process; S8. After the slide rail adjustment is completed, the actual slide rail position is compared with the target position to confirm the execution result of the adjustment action, and a voice prompt information is generated through the voice synthesis module to be fed back to the user.

3. The car seat motor intelligent slide rail adjustment system based on AI voice interaction according to claim 1 is characterized in that: The preprocessing operation includes removing invalid stop words, correcting recognition errors, and extracting semantic entities.

4. The car seat motor intelligent slide rail adjustment system based on AI voice interaction according to claim 1 is characterized in that: The improved FastText algorithm in S3 specifically includes: S31. Construct a subword-level training corpus covering vehicle voice control scenarios, wherein the subword-level training corpus includes multiple types of text samples consisting of historical commands of vehicle voice system users, standard semantic templates, and common natural language control expressions; S32, performing subword segmentation processing on the input standardized text instruction, dividing the complete sentence into continuous subword units, and marking their sequential positions and grammatical dependencies in the original sentence; S33, using the extended FastText nested structure, maps each subword unit to the word vector space, and performs shared embedding by constructing a word-subword nested index; S34. During semantic modeling, a multi-scale context sliding window mechanism is used for subword unit sequences. The short window identifies the core verbs or verb phrases that represent regulatory behaviors in user speech data, and the long window understands the macro-semantic relationships between distance, direction, and speed elements in user speech data. S35. An attention weighting mechanism is introduced based on the set of embedding vectors generated within the multi-scale context sliding window. The semantic contribution of each subword unit is dynamically evaluated based on its grammatical position, part-of-speech tagging, and historical usage frequency in the sentence. The weighted fusion is then performed based on the embedding vectors to generate an instruction representation vector with main sentence characteristics. S36: Concatenate the weighted fused command representation vector with the context interaction vector, where the context interaction vector includes similar control commands issued by the user in the past, records of executed adjustment modes, and scene preference labels. By fusing the context information, the multi-round understanding capability of the current sentence is enhanced, and a context enhancement vector is output. S37, performing normalization and dimension mapping processing on the context enhancement vector, reconstructing it in a fixed-length vector format, and generating a semantic feature vector with good structural stability and discriminative performance by removing sub-dimensions with low confidence, performing dimension regularization and information compression; S38. Output the above semantic feature vector as the only input of the control parameter mapping module, wherein the semantic feature vector includes a main sentence representation sub-block, a context preference encoding field, a structural position information flag, and an adjustment scene label field.

5. The car seat motor intelligent slide rail adjustment system based on AI voice interaction according to claim 4 is characterized in that: The S33 specifically includes: S331, performing a subword splitting operation on each standard term in the subword-level training corpus, decomposing the standard term into multiple subword units using a fixed-length window sliding method, generating a unique embedding index number for each subword unit, and outputting a set of mapping relationships between the standard term and the subword unit; S332. Based on the mapping relationship set, construct a word-subword nested index table, wherein the word-subword nested index table defines all subword indexes contained in each standard term and records word order information of the subwords; S333. Call the word-subword nested index table to initialize embedding vector parameters for each subword unit. The embedding vector parameters are weighted and distributed according to the frequency of occurrence of the subword unit in the entire subword-level training corpus to form an initial subword embedding matrix. S334. Based on the initial subword embedding matrix, multiple subword unit embedding vectors belonging to the same standard term are combined and calculated to generate a nested word vector representation of the standard term, and the relative position structure information of the subword units in the standard term is saved; S335. For a term whose frequency of appearance in the subword-level training corpus is lower than a preset threshold, extract the set of all subword unit embedding vectors corresponding to the term by querying the term-subword nested index table, and call the FastText nested structure to perform an embedding construction operation on the set of all subword unit embedding vectors to generate a complete word vector for the low-frequency term; S336. After the embedding training process is completed, a unified embedding storage structure is established. The embedding storage structure includes three parts: a mapping relationship set, an initial subword embedding matrix, and a word-subword index table.

6. The car seat motor intelligent slide rail adjustment system based on AI voice interaction according to claim 1 is characterized in that: The S4 specifically includes: S41, receiving a semantic feature vector output by a semantic modeling module; S42, inputting the semantic feature vector into an embedded neural network mapping model, wherein the embedded neural network mapping model is composed of a multi-layer perceptron neural network and has been supervised trained based on standardized text instructions; S43. At the neural network input layer, the semantic feature vector is distributed to multiple parallel processing channels, with different processing channels focusing on semantic substructures related to the slide displacement, adjustment speed, and execution mode respectively; S44, converting each semantic substructure into a high-dimensional expression through the middle hidden layer of the embedded neural network, and completing the feature nonlinear compression processing under the action of the internal activation mechanism to generate the potential control representation of the action decision; S45. In the neural network output layer, the potential control representation is mapped into three control parameter items, namely, the target slide rail displacement distance, the target operating speed, and the adjustment mode control code. Each parameter item corresponds to an independent output channel, and a control parameter set is output. S46. Perform semantic consistency check on the control parameter set and the current state of the vehicle, including comparison with the maximum displacement limit, current speed threshold, current seat position, and system adjustment mode configuration table, and output the final control parameter result that passes the check.

7. The car seat motor intelligent slide rail adjustment system based on AI voice interaction according to claim 1 is characterized in that: The S5 specifically includes: S51, calling the slide rail state detection module to obtain the current slide rail displacement state, the slide rail displacement state is collected by the slide rail position sensor and uploaded to the control processing unit; S52, accessing the maximum adjustable range parameters of the slide rail stored in the vehicle system, wherein the maximum adjustable range parameters of the slide rail include a starting limit position of the slide rail, a maximum end limit position, and a preset safety tolerance range; S53, extracting a target rail displacement distance from the control parameter set, and adding the target rail displacement distance to the current rail displacement state to generate an initial target rail position value; S54: Perform boundary constraint processing on the initial target rail position value. If the initial target rail position value exceeds the maximum adjustable range, correct it to the nearest legal position range according to the safety limit strategy configured by the vehicle system, and trigger the fault tolerance adjustment mode if necessary. S55. Generate an operating speed parameter that matches the target slide rail position based on the difference between the corrected target slide rail position and the current actual position and in combination with the execution strategy corresponding to the current adjustment mode; S56: Output the target slide rail position after boundary correction and the matching running speed parameters.

8. The car seat motor intelligent slide rail adjustment system based on AI voice interaction according to claim 1 is characterized in that: The S6 specifically includes: S61, receiving the target slide rail position and running speed parameters after boundary correction, and synchronously obtaining the adjustment mode information in the control parameter set to form the core content field of the slide rail control instruction; S62. Establishing a data frame format for the slide rail control instruction according to a predefined slide rail control protocol structure of the system, wherein the data frame format includes a fixed frame header, a target position field, a running speed field, an adjustment mode field, and an instruction redundancy check field; S63, filling the control parameter contents into the corresponding fields respectively, and packaging each field according to the byte alignment rules, field sequence requirements and data encoding method in the slide control protocol to form a complete slide control instruction frame to be sent; S64, performing consistency check and redundancy comparison on the encapsulated complete slide rail control instruction frame to verify the integrity of the instruction structure, the legitimacy of the field values, and the rationality of the tolerance between the target position and the current slide rail state; S65, transmitting the verified slide rail control instruction frame to the vehicle control bus, and sending it to the slide rail motor controller through the vehicle control bus interface, wherein the vehicle control bus interface supports CAN, LIN or vehicle Ethernet; S66. After receiving the control instruction frame, the slide motor controller parses the field content, completes field decoding and verification logic confirmation based on the parsing result, and generates a control instruction.

9. The car seat motor intelligent slide rail adjustment system based on AI voice interaction according to claim 1 is characterized in that: The S7 specifically includes: S71. Initialize the slide control program according to the field decoding result, call the preset slide motor control channel, configure the drive polarity according to the target displacement direction, configure the PWM modulation parameters according to the target speed, and set the soft limit buffer and safety interrupt flag. S72. After receiving the control command, the slide motor controller starts the slide motor, performs position detection and current monitoring, and periodically samples the real-time parameters of the slide motor during displacement. S73. During the operation of the slide rail, the slide rail motor controller continuously monitors the progress of the target displacement and collects multiple operation status data including the current position of the slide rail motor, the operating current, the feedback voltage, and the driving load status; S74. Perform abnormality detection on the collected multiple operating status data. If abnormal operating characteristics such as overcurrent, stagnation, drift, overtravel, or low-speed no response are found, immediately trigger a stop control command and record the current abnormal status flag code; S75. After the slide motor reaches the control target or stops, the final displacement completion status, current position data and operation status record results are output.

10. The car seat motor intelligent slide rail adjustment system based on AI voice interaction according to claim 1 is characterized in that: The S8 specifically includes: S81. After the slide rail motor finishes running, obtain the actual position state of the current slide rail, and compare the actual position state with the target position set in the slide rail control instruction to determine the completion status of the adjustment action; S82. Generate an execution status flag based on the result of the adjustment action completion judgment, wherein the execution status flag includes "adjustment successful", "position error exceeds limit", "operation interrupted", and "not in place". S83, packaging the execution status flag, the current actual position value of the slide rail, the target position value of the slide rail, and the operation mode data into a prompt information field; S84: Calling the speech synthesis module to convert the content in the prompt information field into a speech output content template, and selecting the corresponding broadcast language, broadcast volume, and speech speed settings according to the current cockpit voice configuration; S85. Send the synthesized voice data to the vehicle audio output module, and control the vehicle speakers to broadcast prompt messages, including prompts such as "Seat adjustment completed," "Currently moved to the set position," and "Adjustment abnormality aborted, please check." S86. Record the trigger source, prompt type, and broadcast content of this round of voice prompt interaction, and store them in the voice interaction log module.

Images