Motor home equipment wireless intelligent control method and system based on AI and voice recognition

Abstract

The invention relates to the technical field of intelligent control, in particular to a limo equipment wireless intelligent control method and system based on AI and voice recognition. The method comprises the following steps: a user sends a voice command, denoising and enhancing voice signals through a beam forming technology and an adaptive filter, enhancing a time-frequency structure of the voice signals by using Fourier transform and dynamic gain, and performing auditing verification by generating an auditing code; extracting multi-modal features, generating high-dimensional hidden representation through a bidirectional long-short term memory network and an attention mechanism after fusion, optimizing text output, and improving text quality by using spelling error correction and semantic consistency verification; then analyzing the instruction and generating a standardized instruction format through an intention recognition and slot filling model, executing rule verification to ensure the reasonability and safety of the instruction, and finally generating a control command in combination with environmental data to ensure the intelligent control of equipment. According to the invention, efficient and intelligent voice control and safety verification of the motor home equipment are realized.

Description

Technical Field

[0001] The present invention relates to the field of intelligent control technology, and in particular to a method and system for wireless intelligent control of RV equipment based on AI and voice recognition. Background Art

[0002] With the rapid development of smart home and automation technology, intelligent control systems based on artificial intelligence (AI) and voice recognition have gradually become an important tool to improve living experience and equipment management efficiency, especially in the RV field. As people's requirements for the quality of mobile life increase, RV intelligence has become an important trend in the development of the industry. RV equipment is diverse and complex. The traditional manual control method is not only cumbersome to operate, but also increases the risk of distraction for the driver during driving.

[0003] A Chinese invention patent application with publication number CN118400214A discloses a smart home control method and system based on AI dialogue, wherein the method comprises the following steps: based on a cloud data center, constructing a control voice recognition model, an AI dialogue model and a voice response generation model; using a smart home control device to connect to the cloud data center; based on the smart home control device, collecting the user's real-time control voice data, and using a reconstructed control voice recognition model to generate corresponding real-time control text data; using a reconstructed AI dialogue model to generate corresponding real-time control instructions and real-time response text data; according to the real-time control instructions, controlling the corresponding smart home device, and according to the real-time response text data, using a reconstructed voice response generation model to generate corresponding real-time response voice data; the application solves the problems of complex use, poor experience and poor control accuracy existing in the prior art.

[0004] However, in real life, how to simplify the operating procedures of RV equipment and improve the accuracy and convenience of control has become an urgent problem to be solved in the industry. The existing control methods and single manual operation cannot meet the RV users' needs for intelligence, convenience and safety. There is an urgent need for an innovative solution that can realize wireless intelligent control of equipment and enhance user experience. Summary of the invention

[0005] The purpose of the present invention is to address the problems existing in the background technology and to propose a wireless intelligent control method and system for RV equipment based on AI and voice recognition.

[0006] The technical solution of the present invention is a wireless intelligent control method for RV equipment based on AI and voice recognition, which includes the following specific implementation steps: S1. The user issues voice commands to the RV system through the built-in multi-microphone array, which captures the voice signal, performs denoising and signal enhancement, and generates a review code; S2. Ensure the legality of the voice signal source through audit verification, extract voice features using Mel-frequency cepstral coefficients and logarithmic Mel-frequency spectrum, and combine convolution fusion to obtain time-frequency features. Build an end-to-end speech recognition model with a two-way attention mechanism, generate text data, optimize the output through a language model, and generate standard instructions using intent recognition and slot filling models after spelling correction and semantic consistency verification, and perform rule verification. S3, intelligent control method based on environmental perception and adaptive optimization, real-time collection of temperature, humidity, and light data, and cleaning and feature extraction, extraction of statistical features after filtering abnormal data, combining environmental status with user preferences, building a multimodal environmental perception model and generating an optimized control strategy, and adjusting operation instructions by optimizing the objective function, reducing energy consumption and time costs, performing adaptive parameter optimization, and generating optimized control instructions; S4, passing the optimized control instruction to the target device, after receiving the control instruction, the target device performs the corresponding operation, and at the same time, the device transmits its latest status data back to the user interface through the status feedback mechanism; S5. The feedback data of the target device is displayed on the visual user interface. The user can view the status of the device in the RV in real time through the interface and issue further control instructions as needed.

[0007] Preferably, the process of capturing the speech signal and performing denoising and signal enhancement processing is as follows: S21, adopts a multi-microphone array arrangement, collects voice signals through array, captures voice information from different directions, uses beamforming technology to improve the collection effect of target voice, and suppresses environmental noise; The beamforming model formula is: ; Where y(t) represents the output speech signal after beamforming processing; x i (t) represents the speech signal received by the i-th microphone; w i represents the weighting coefficient of the i-th microphone, which is dynamically adjusted to enhance the target direction speech; τ i represents the propagation delay between the i-th microphone and the speech source; N represents the number of microphones; S22, based on the minimum mean square error adaptive filter method, remove noise and retain the speech signal, the adaptive filter model is: ; ; Where e(t) represents the clean speech signal after filtering; d(t) represents the signal mixed with speech and noise, that is, the main microphone signal; n j (t) represents the pure noise signal collected by the auxiliary microphone; hj represents the jth filter weight coefficient, which is dynamically updated through learning; M represents the filter order; μ represents the learning rate, which is used to control the step size of weight update; S23, segment the time domain signal and add a window, perform Fourier transform (SIFT) on each segment of the signal to obtain a frequency domain representation. The SIFT formula is: ; Where, X(t,f) represents the frequency domain representation of the signal; x(n) represents the nth time domain sampling point; w(n) represents the windowing function; f represents the frequency component; N represents the number of sampling points of each signal segment; j' represents the imaginary unit; Dynamically adjust the gain to enhance the time-frequency structure of the speech signal. The improved gain formula is: ; ; Where G(t,f) represents the dynamic gain function; |X(t,f)| 2 represents speech power density; |N(t,f)| 2 represents the noise power spectrum density; α represents the adaptive balance factor, which dynamically adjusts the gain weights of speech and noise signals; β represents the noise suppression factor, which is used to control the degree of noise reduction; F c (f) represents the frequency tracking smoothing function, which enhances the details of the target frequency range; min and f max represents the target speech frequency range; γ represents the attenuation factor, which controls the smoothness of the frequency boundary; S24, reconstruct the enhanced frequency domain signal back to the time domain to avoid speech distortion caused by high-frequency noise or excessive filtering. The inversion formula is: ; Where S'(t,f) represents the enhanced frequency domain signal; ϕ(t,f) represents the phase information of the speech signal, which is corrected by frequency domain phase tracking to avoid phase distortion; S25, convert the enhanced frequency domain signal S'(t,f) back to the time domain through inverse short-time Fourier transform, and the inverse short-time Fourier transform formula is: .

[0008] Preferably, the review code generation process is as follows: S31. In the conic section C n (a,b):y 2 =ax 2 -Select a point on bx (mod n) and calculate the review generated code C g =Pp×point(mod n); In the formula, n=p×q, p and q represent two different large prime numbers; a, b∈Z n And (a,n)=(b,n)=1, (a / p)=(a / q)=-1; #C n =(p+1)×(q+1), p+1=2r, q+1=2r, r and s represent prime numbers; G is C n (a,b) base point; Pp represents the authority parameter, (Pp,N n )=1,N n =2×r×s; Z n represents the remainder class ring modulo n; S32, calculate the first-order review parameter A1=H(data||C g ); Where data represents the binary string of the speech signal; H is the hash function, H: {0,1} * ×C n → ; ={1,2,…,N n -1};|| indicates cascade operation; S33. Calculate the second-order review parameter A2 = A1 × S A +point (mod n); Among them, S A represents the second-order parameter factor, S A =-D A ×G(mod n),D A represents the first-order parameter factor and Pp×D A =1(mod N n ); S34. Generate review code A=(A1, A2).

[0009] Preferably, the audit verification process to ensure the legality of the source of the voice signal is as follows: S41. Calculate the first-order audit factor fc1=Pp×A2+A1×G (mod n); Where A1 represents the first-order review parameter; A2 represents the second-order review parameter; S42, calculate the second-order audit factor fc2=H(data||fc1); S43. If A1=fc2, the audit verification is passed, that is, the source of the received voice signal data is legal.

[0010] Preferably, the extraction process of extracting speech features using Mel-frequency cepstral coefficients and logarithmic Mel-frequency spectrum is as follows: S51, Mel frequency cepstrum coefficients: ; Wherein, MFCC(k) represents the kth Mel frequency cepstrum coefficient; Y(m) represents the spectrum value of the Mel filter bank; M' represents the number of Mel filter banks; m represents the mth filter in the Mel filter bank; S52, Logarithmic Mel Spectrum: LMS(k)=log(|Y(k)| 2 ); Where LMS(k) represents the kth logarithmic Mel spectrum feature; S53, perform multi-layer convolution fusion on the Mel frequency cepstral coefficients and the logarithmic Mel spectrum features to obtain the multimodal time-frequency features of the speech: ; In the formula, F fusion represents the fused feature matrix; W1 and W2 represent the convolution kernel weights; b represents the bias; σ represents the activation function.

[0011] Preferably, the end-to-end speech recognition model of the bidirectional attention mechanism includes: Encoder: Transform the multimodal features F fusion Encoded into a high-dimensional hidden representation using a stacked bidirectional LSTM network: ; In the formula, h t represents the hidden state at the tth moment; Attention mechanism: Capture the global context information in the speech signal and generate a context vector through weighted sum: ; ; ; ; ; In the formula, c i represents the context vector, which is the information extracted from the encoder hidden layer through the attention mechanism to provide richer context information; W a and v a represents the weight parameter of the attention mechanism; β t represents the time step adaptive weight; represents the improved attention weight; represents the attention weight; W t represents the time step adaptive weight matrix; [h i ;h j ] represents the encoder hidden state h i and the decoder hidden state h j Connection; Ffusion (t) represents the output after multimodal feature fusion; T represents the total time steps of the encoder; Decoder: Combines context vector and language modeling information to generate text data: y i =Decoder(y i-1 ,c i ); In the formula, y i represents the i-th output word or character generated by the decoder, generated by the decoder based on the previous output and the context vector; y i-1 Represents the output of the previous time step, which is passed as input to the next time step during the decoding process.

[0012] Preferably, the process of generating standard instructions using the intent recognition and slot filling model is as follows: S71. Add a context-dependent dynamic weight adjustment module to the standard Transformer model: ; ; In the formula, represents the semantic vector at time t; h t Represents the encoder output of Transformer at time t; Represents the global semantic features of the text, which is calculated as the weighted average of the features of the text at all times; α t Represents the dynamic weight of the feature at the current moment; W α represents the weight matrix; The classification probability formula is: P(C|T')=softmax(W); S72. Construct a slot filling model, extract the device name and operation type key information from the text as the slot, and the slot filling formula is: ; In the formula, P(S j |T') represents S j The probability distribution of W s and b s They represent the weight and bias of the slot prediction model respectively; S73. Generate an executable instruction format based on the intent classification and slot filling results, and perform verification. The verification process is as follows: S7301. Define the standardized instruction format as: I=(D,O,P); In the formula, I represents the generated instruction; D represents the target device; O represents the operation type; P represents additional parameters; S7302, instruction generation formula: I=f(C,S)=Map(C,S); Where f(C,S) represents the mapping from intent C and slot S to instruction I; Map represents the mapping function from intent and slot to instruction; S7303. To ensure the rationality and security of the instruction, perform rule verification. The rule verification formula is: ; In the formula, V represents the instruction verification result, 1 is valid and 0 is invalid; R i (I) represents the verification result of the ith rule on instruction I; δ(x) represents the logical AND operation.

[0013] Preferably, the generation process of the optimized control instruction is as follows: S81, collecting environmental data E related to the control instruction in real time through the environmental sensor network; S82, cleaning and feature extraction of the collected environmental data E to eliminate noise and obtain key influencing factors; Abnormal data filtering: Filter abnormal data by setting dynamic thresholds: ; In the formula, Represents the filtered environmental data; μ i and σ i They represent the mean and standard deviation of the data respectively; k represents the sensitivity parameter; e i Represents a single environmental data indicator; Feature extraction: Extract statistical features from time series data, including: Environmental data average: ; Data change trend: ; In the formula, T'' represents the length of the time interval or the total duration of data sampling, that is, the time span of the data; i (t) represents the value of the i-th environmental data at time t; S83. Construct a multimodal environment perception model based on the correlation between environmental data and instructions: Environmental state classification: The environmental state E is mapped to a discrete state space S' through a classification model: ; In the formula, s represents the environmental state category; h E Represents the feature vector of environmental data based on Transformer; W s and b s Represent the classifier weight and bias respectively; User preference modeling: Combine historical operation records to build a user preference function U(D,s): ; Where U(D,s) represents the user's preference value for device D in environment state s; Freq(D,s) represents the historical usage frequency of device D in state s; Rec(D,s) represents the recommendation weight of device D in state s; α' and β' represent weight parameters; S84, generating an optimized control strategy according to the current environment state s, the instruction I and the user preference U; Optimization objective function: ; In the formula, O * represents the optimized operation instruction; C(O,E) represents the resource consumption of executing instruction O; Adaptively optimize the parameter P in the control instruction, and optimize the parameter P * for: ; Where P * represents the optimized instruction parameter; γ' represents the adjustment step; ΔP(E) represents the parameter increment of the environmental data E; S85, generate optimized control instruction I * =(D,O * ,P * ).

[0014] The technical solution of the present invention is a wireless intelligent control system for RV equipment based on AI and voice recognition, which is used to execute the above-mentioned wireless intelligent control method for RV equipment based on AI and voice recognition, including: The speech recognition module is used to collect user speech signals through a microphone and perform noise removal and signal enhancement processing; The speech analysis module is used to convert speech into text data, and to perform semantic understanding and command analysis on the text data to identify specific control instructions; Artificial intelligence decision-making module, which uses AI algorithms to intelligently analyze the analysis results and generate control commands based on the current status of the RV equipment; The equipment control module is used to control various equipment in the RV through wireless communication technology according to the control commands output by the artificial intelligence decision module; The user interface module is used to provide a user interface for users to interact with the system.

[0015] Compared with the prior art, the above technical solution of the present invention has the following beneficial technical effects: The present invention designs a wireless intelligent control method and system for RV equipment based on AI and voice recognition. By comprehensively applying multi-microphone array technology, AI voice recognition, dynamic gain optimization, voice data security protection, and intelligent control decision-making, the intelligence, reliability and user experience of RV equipment are significantly improved, and it has important technical innovation value: (1) Improving speech recognition accuracy and robustness: By adopting multi-microphone array and beamforming technology, it can effectively capture speech signals from different directions and suppress environmental noise, significantly improving the quality and recognition accuracy of speech signals. Adaptive filters and dynamic gain adjustment further optimize signal processing and enhance the clarity and reliability of the target speech. (2) Speech signal enhancement and noise reduction: The present invention uses time-frequency processing and Fourier transform combined with noise suppression technology to effectively reduce the interference of background noise, retain the core information of the speech signal, and avoid the common speech distortion problem in traditional single-microphone devices. This technology effectively improves the speech recognition capability in complex environments; (3) Ensure the security and legitimacy of voice data: Through the generation and verification of first-order and second-order review parameters, combined with the generation of hash algorithms and review codes, the integrity and source legitimacy of voice data are effectively guaranteed, the risks of malicious attacks and data tampering are prevented, and the security of the system is ensured; (4) Multimodal feature fusion improves speech recognition accuracy: Combining Mel-frequency cepstral coefficient (MFCC) and log-Mel spectrum (Log-Mel) features, the time-frequency features of speech signals are extracted through multi-layer convolution fusion, which greatly improves the accuracy of the speech recognition model. In addition, the context information is weighted and optimized through the bidirectional attention mechanism, so that the model can capture more comprehensive semantic information, further improving the accuracy of speech recognition. (5) Optimizing language modeling and command generation: The present invention significantly improves the adaptability and accuracy of speech recognition in RV equipment control by constructing a domain-specific language model and dynamically adjusting the modeling weights of general and special corpora. It also ensures the accuracy and consistency of recognition results through semantic consistency checking and spelling correction, thus avoiding possible misrecognition and misunderstanding errors. (6) Intelligent decision-making and safety control: The artificial intelligence decision-making module combines environmental perception and adaptive optimization technology to generate precise control commands based on the real-time status of the RV equipment and external environmental data, ensuring the accuracy and rationality of equipment control. The verification mechanism ensures the safety of instructions and the effectiveness of execution, avoiding potential risks of misoperation and equipment damage. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a system architecture diagram of a wireless intelligent control system for RV equipment based on AI and voice recognition proposed by the present invention; Figure 2 This is a method flow chart of a wireless intelligent control method for RV equipment based on AI and voice recognition proposed by the present invention. DETAILED DESCRIPTION

[0017] Embodiment 1, as Figure 1 As shown, the present invention proposes a wireless intelligent control system for RV equipment based on AI and voice recognition, including: a voice recognition module, a voice analysis module, an artificial intelligence decision-making module, an equipment control module, and a user interface module.

[0018] The speech recognition module collects user speech signals through a microphone and performs noise removal and signal enhancement processing; The speech analysis module converts speech into text data, performs semantic understanding and command analysis on the text data, and identifies specific control instructions; The artificial intelligence decision-making module uses AI algorithms to intelligently analyze the analysis results and generates control commands based on the current status of the RV equipment (including but not limited to temperature, humidity, and power); The device control module controls various devices in the RV (including but not limited to air conditioning, lighting, and audio) through wireless communication technology (including but not limited to Wi-Fi, Bluetooth, and ZigBee) according to the control commands output by the artificial intelligence decision-making module; The user interface module provides an interface for users to interact with the system, including but not limited to mobile apps and vehicle-mounted touch screens, and supports voice input and manual input.

[0019] Embodiment 2, as Figure 2 As shown, the present invention proposes a wireless intelligent control method for RV equipment based on AI and voice recognition, which is applied to a wireless intelligent control system for RV equipment based on AI and voice recognition proposed in Example 1, and its specific implementation steps are as follows: S1. The user issues a voice command to the RV system through the built-in multi-microphone array (including but not limited to "turn on the lights in the car"), and the voice recognition module is activated to capture the voice signal and perform denoising and signal enhancement processing. The specific implementation process is as follows: S11, adopts a multi-microphone array arrangement, collects voice signals through array, captures voice information from different directions, uses beamforming technology to improve the collection effect of target voice, and suppresses environmental noise; The beamforming model formula is: ; Where y(t) represents the output speech signal after beamforming processing; x i (t) represents the speech signal received by the i-th microphone; w irepresents the weighting coefficient of the i-th microphone, which is dynamically adjusted to enhance the target direction speech; τ i represents the propagation delay between the i-th microphone and the speech source; N represents the number of microphones; S12, based on the minimum mean square error adaptive filter method, remove noise and retain the speech signal, the adaptive filter model is: ; ; Where e(t) represents the clean speech signal after filtering; d(t) represents the signal mixed with speech and noise, that is, the main microphone signal; n j (t) represents the pure noise signal collected by the auxiliary microphone; h j represents the filter weight coefficient, which is dynamically updated through learning; M represents the filter order; μ represents the learning rate, which is used to control the step size of weight update. In this embodiment, the value is between 0.001 and 0.01; S13, segment the time domain signal and add a window, perform Fourier transform (SIFT) on each segment of the signal to obtain a frequency domain representation. The SIFT formula is: ; Where X(t,f) represents the frequency domain representation of the signal; x(n) represents the nth time domain sampling point; w(n) represents the windowing function; f represents the frequency component; N represents the number of sampling points of each signal segment; j' represents the imaginary unit; Dynamically adjust the gain to enhance the time-frequency structure of the speech signal. The improved gain formula is: ; ; Where G(t,f) represents the dynamic gain function; |X(t,f)| 2 represents speech power density; |N(t,f)| 2 represents the noise power spectrum density; α represents the adaptive balance factor, which dynamically adjusts the gain weights of speech and noise signals; β represents the noise suppression factor, which is used to control the degree of noise reduction; F c (f) represents the frequency tracking smoothing function, which enhances the details of the target frequency range; min and f max represents the target speech frequency range; γ represents the attenuation factor, which controls the smoothness of the frequency boundary; S14, reconstruct the enhanced frequency domain signal back to the time domain to avoid speech distortion caused by high-frequency noise or excessive filtering. The inversion formula is: ; Where S'(t,f) represents the enhanced frequency domain signal; ϕ(t,f) represents the phase information of the speech signal, which is corrected by frequency domain phase tracking to avoid phase distortion; S15. The enhanced frequency domain signal S'(t,f) is converted back to the time domain through the inverse short-time Fourier transform (ISTFT). The ISIFT formula is: ; S16: Generate a review code for the processed voice signal. The generation process is as follows: S1601, in the conic section C n (a,b):y 2 =ax 2 -Select a point on bx (mod n) and calculate the review generated code C g =Pp×point(mod n); In the formula, n=p×q, p and q represent two different large prime numbers; a, b∈Z n And (a,n)=(b,n)=1, (a / p)=(a / q)=-1; #C n =(p+1)×(q+1), p+1=2r, q+1=2r, r and s represent prime numbers; G is C n (a,b) base point; Pp represents the authority parameter, (Pp,N n )=1,N n =2×r×s; Z n represents the remainder class ring modulo n; S1602, calculate the first-order review parameter A1=H(data||C g ); Where data represents the binary string of the speech signal; H is the hash function, H: {0,1} * ×C n → ; ={1,2,…,N n -1};|| indicates cascade operation; S1603, calculate the second-order review parameter A2 = A1 × S A +point (mod n); Among them, S A represents the second-order parameter factor, S A =-D A ×G(mod n),D A represents the first-order parameter factor and Pp×D A =1(mod N n ); S1604, generate review code A=(A1, A2); S17. Transmit {A, data} to the speech analysis module.

[0020] S2, speech analysis module is based on the speech recognition method of multimodal feature fusion and adaptive language modeling. It extracts multi-layer speech features through deep learning model and optimizes text output in combination with contextual semantics. The specific implementation steps are as follows: S21, the voice analysis module verifies the received voice signal data, and the verification process is as follows: S2101. Calculate the first-order audit factor fc1=Pp×A2+A1×G (mod n); Where A1 represents the first-order review parameter; A2 represents the second-order review parameter; S2102. Calculate the second-order audit factor fc2=H(data||fc1); S2103, if A1=fc2, the audit verification is passed, that is, the source of the received voice signal data is legal; S22, extracting features by combining Mel frequency cepstral coefficient (MFCC) and logarithmic Mel spectrum (Log-Mel); extracting features by combining Mel frequency cepstral coefficient (MFCC) and logarithmic Mel spectrum (Log-Mel); Mel-frequency cepstral coefficients: ; Wherein, MFCC(k) represents the kth Mel frequency cepstrum coefficient; Y(m) represents the spectrum value of the Mel filter bank; M' represents the number of Mel filter banks; Logarithmic Mel spectrum: LMS(k)=log(|Y(k)| 2 ); Where LMS(k) represents the kth logarithmic Mel spectrum feature; S23, the Mel frequency cepstral coefficients and the logarithmic Mel spectrum features are used to obtain the multimodal time-frequency features of the speech: ; In the formula, F fusion represents the fusion feature matrix; W1 and W2 represent the convolution kernel weights; b represents the bias; σ represents the activation function; S24. Build an end-to-end speech recognition model with a bidirectional attention mechanism, including the following modules: Encoder: Encodes the multimodal features F fusion Encoded into a high-dimensional hidden representation using a stacked Bidirectional Long Short-Term Memory (BiLSTM) network: ; In the formula, h trepresents the hidden state at the tth moment; Attention mechanism: captures the global context information in the speech signal and generates a context vector through weighted sum: ; ; ; ; ; In the formula, c i represents the context vector, which is the information extracted from the encoder hidden layer through the attention mechanism to provide richer context information; W a and v a represents the weight parameter of the attention mechanism; β t represents the time step adaptive weight; represents the improved attention weight; represents the attention weight; W t represents the time step adaptive weight matrix; [h i ;h j ] represents the encoder hidden state h i and the decoder hidden state h j Connection; F fusion (t) represents the output after multimodal feature fusion; Decoder: Combines context vector and language modeling information to generate text data: y i =Decoder(y i-1 ,c i ); In the formula, y i represents the i-th output word (or character) generated by the decoder, which is generated by the decoder based on the previous output and the context vector; i-1 Represents the output of the previous time step, which is passed as input to the next time step during the decoding process; S25. Combined with the language model optimization, the output text data text, the language model probability calculation formula is: ; Where P(Y) represents the probability of outputting text data; Represents the conditional probability of the i-th word given all the previous words; Based on this, combined with the semantic context of RV control, a domain-specific language model is constructed, and the language modeling weights of general corpus and special corpus are adjusted by weighting: ; Where P domain (y i ) represents the probability distribution of the domain corpus; P general (y i ) represents the probability distribution of general corpus; λ represents the weight of domain and general model, which is dynamically adjusted to balance recognition accuracy and versatility; S26, after the text data text is generated, the output is optimized through spelling correction and semantic consistency verification; Spelling correction formula: ; ; In the formula, represents the corrected output word, that is, the result corrected by the spelling correction module; ν represents the vocabulary, that is, all possible words used in the language model; EditD(w,y i ) represents the candidate word w and the input word y i The edit distance between them is the similarity between them; w[:-1] means the string w has the last character removed; y[:-1] means the string y i Remove the last character; c represents the cost of replacement, if the two characters are the same, c=0, if they are different, c=1; It should be noted that EditD(w,y i ) is a method to measure the difference between two strings. It means to convert string w into string y i The minimum number of operations required, including: insert (insert a character into a string), delete (delete a character from a string), and replace (replace a character in a string with another character); Semantic consistency check formula: ; ; ; ; In the formula, S represents the semantic matching score; Represents the context vector c i With words The semantic similarity of v(c i ) represents the context vector c i Embedded representation in semantic space; Represents the generated text word The semantic embedding vector of and represents v(c i )and The Euclidean norm (L2 norm) of ; D represents the dimension of the embedding vector; T represents the total number of time steps of the encoder; S27, based on the intent recognition and slot extraction model, the text data text is classified and the slots are filled. The specific implementation process is as follows: S2701. Add a context-dependent dynamic weight adjustment module to the standard Transformer model: ; ; In the formula, represents the semantic vector at time t; h t Represents the encoder output of Transformer at time t; Represents the global semantic features of the text, which is calculated as the weighted average of the features of the text at all times; α t Represents the dynamic weight of the feature at the current moment; W α represents the weight matrix; The classification probability formula is: P(C|T')=softmax(W); S2702. Construct a slot filling model, extract the device name and operation type key information from the text as the slot, and the slot filling formula is: ; In the formula, P(S j |T') represents S j The probability distribution of W s and b s They represent the weight and bias of the slot prediction model respectively; For example, for the text instruction "turn on the air conditioner to 22 degrees", the intent classification is "control the air conditioner", and the slot filling extraction is: Equipment name: air conditioner; Operation type: Open; Temperature setting: 22 degrees; S28. Generate an executable instruction format based on the intent classification and slot filling results, and perform verification. The verification process is as follows: S2801. Define the standardized instruction format as: I=(D,O,P); In the formula, I represents the generated instruction; D represents the target device; O represents the operation type; P represents additional parameters; S2802, instruction generation formula: I=f(C,S)=Map(C,S); Where f(C,S) represents the mapping from intent C and slot S to instruction I; Map represents the mapping function from intent and slot to instruction; For example: the text command "close the curtains" generates the mapping: I=(curtains, close, no parameters); S2803. To ensure the rationality and security of the instruction, rule verification is performed. The rule verification formula is: ; In the formula, V represents the instruction verification result, 1 is valid and 0 is invalid; R i (I) represents the verification result of the ith rule on instruction I; δ(x) represents the logical AND operation; S2804: Send the verified instructions to the artificial intelligence decision-making module for execution. If the verification fails, an error message will be fed back or the user will be prompted to re-enter the instructions.

[0021] S3, the artificial intelligence decision-making module integrates the intelligent control method of environmental perception and adaptive optimization, and generates control commands based on the current status of the RV equipment. The specific implementation process is as follows: S31, collect environmental data E related to the control instructions in real time through the environmental sensor network, including temperature, humidity, light, and air quality; S32, cleaning and feature extraction of the collected environmental data E to eliminate noise and obtain key influencing factors; Abnormal data filtering: Filter abnormal data by setting dynamic thresholds: ; In the formula, Represents the filtered environmental data; μ i and σ i They represent the mean and standard deviation of the data respectively; k represents the sensitivity parameter; e i Represents a single environmental data indicator; Feature extraction: Extract statistical features from time series data, including but not limited to: Environmental data average: ; Data change trend: ; In the formula, T'' represents the length of the time interval or the total duration of data sampling, that is, the time span of the data; i (t) represents the value of the i-th environmental data at time t; S33. Construct a multimodal environment perception model based on the correlation between environmental data and instructions: Environmental state classification: The environmental state E is mapped to a discrete state space S' through a classification model: ; In the formula, s represents the environmental state category; h E Represents the feature vector of environmental data based on Transformer; Ws and b s Represent the classifier weight and bias respectively; User preference modeling: Combine historical operation records to build a user preference function U(D,s): ; Where U(D,s) represents the user's preference value for device D in environment state s; Freq(D,s) represents the historical usage frequency of device D in state s; Rec(D,s) represents the recommendation weight of device D in state s; α' and β' represent weight parameters; S34, generating an optimized control strategy according to the current environment state s, the instruction I and the user preference U; Optimization objective function: ; In the formula, O * represents the optimized operation instruction; C(O,E) represents the resource consumption of executing instruction O, including but not limited to energy consumption and time cost; Adaptively optimize the parameter P in the control instruction, and optimize the parameter P * for: ; Where P * represents the optimized instruction parameter; γ' represents the adjustment step; ΔP(E) represents the parameter increment of the environmental data E; S35, the optimized control instruction I * =(D,O * ,P * ) is transmitted to the device control module.

[0022] S4, the device control module transmits the optimized control instruction I through a wireless communication network (including but not limited to Wi-Fi, Bluetooth or ZigBee). * =(D,O * ,P * ) is passed to the corresponding device; After receiving the control command, the target device executes the corresponding operation. At the same time, the device transmits its latest status data back to the user interface module through the wireless communication network through the status feedback mechanism.

[0023] S5. The user interface module receives the status feedback of the device and displays it in a visual form on the screen. The user can view the status of the device in the RV in real time through the interface and issue further control instructions as needed.

[0024] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited thereto, and various changes can be made within the knowledge scope of technicians in the relevant technical field without departing from the purpose of the present invention.

Claims

1. A wireless intelligent control method for RV equipment based on AI and voice recognition, characterized in that: The specific implementation steps include the following: S1. The user issues voice commands to the RV system through the built-in multi-microphone array, which captures the voice signal, performs denoising and signal enhancement, and generates a review code; S2. Ensure the legality of the voice signal source through audit verification, extract voice features using Mel-frequency cepstral coefficients and logarithmic Mel-frequency spectrum, and combine convolution fusion to obtain time-frequency features. Build an end-to-end speech recognition model with a two-way attention mechanism, generate text data, optimize the output through a language model, and generate standard instructions using intent recognition and slot filling models after spelling correction and semantic consistency verification, and perform rule verification. S3, intelligent control method based on environmental perception and adaptive optimization, real-time collection of temperature, humidity, and light data, and cleaning and feature extraction, extraction of statistical features after filtering abnormal data, combining environmental status with user preferences, building a multimodal environmental perception model and generating an optimized control strategy, and adjusting operation instructions by optimizing the objective function, reducing energy consumption and time costs, performing adaptive parameter optimization, and generating optimized control instructions; S4, passing the optimized control instruction to the target device, after receiving the control instruction, the target device performs the corresponding operation, and at the same time, the device transmits its latest status data back to the user interface through the status feedback mechanism; S5. The feedback data of the target device is displayed on the visual user interface. The user can view the status of the device in the RV in real time through the interface and issue further control instructions as needed.

2. According to claim 1, a wireless intelligent control method for RV equipment based on AI and voice recognition is characterized in that: The process of capturing speech signals and performing denoising and signal enhancement is as follows: S21, adopts a multi-microphone array arrangement, collects voice signals through array, captures voice information from different directions, uses beamforming technology to improve the collection effect of target voice, and suppresses environmental noise; The beamforming model formula is: ； Where y(t) represents the output speech signal after beamforming processing; x i (t) represents the speech signal received by the i-th microphone; w i represents the weighting coefficient of the i-th microphone, which is dynamically adjusted to enhance the target direction speech; τ i represents the propagation delay between the i-th microphone and the speech source; N represents the number of microphones; S22, based on the minimum mean square error adaptive filter method, remove noise and retain the speech signal, the adaptive filter model is: ； ； Where e(t) represents the clean speech signal after filtering; d(t) represents the signal mixed with speech and noise, that is, the main microphone signal; n j (t) represents the pure noise signal collected by the auxiliary microphone; h j represents the jth filter weight coefficient, which is dynamically updated through learning; M represents the filter order; μ represents the learning rate, which is used to control the step size of weight update; S23, segment the time domain signal and add a window, perform SIFT on each segment of the signal, and obtain a frequency domain representation. The SIFT formula is: ； Where X(t,f) represents the frequency domain representation of the signal; x(n) represents the nth time domain sampling point; w(n) represents the windowing function; f represents the frequency component; N represents the number of sampling points of each signal segment; j' represents the imaginary unit; Dynamically adjust the gain to enhance the time-frequency structure of the speech signal. The improved gain formula is: ； ； Where G(t,f) represents the dynamic gain function; |X(t,f)| 2 represents speech power density; |N(t,f)| 2 represents the noise power spectrum density; α represents the adaptive balance factor, which dynamically adjusts the gain weights of speech and noise signals; β represents the noise suppression factor, which is used to control the degree of noise reduction; F c (f) represents the frequency tracking smoothing function, which enhances the details of the target frequency range; min and f max represents the target speech frequency range; γ represents the attenuation factor, which controls the smoothness of the frequency boundary; S24, reconstruct the enhanced frequency domain signal back to the time domain to avoid speech distortion caused by high-frequency noise or excessive filtering. The inversion formula is: ； Where S'(t,f) represents the enhanced frequency domain signal; ϕ(t,f) represents the phase information of the speech signal, which is corrected by frequency domain phase tracking to avoid phase distortion; S25, convert the enhanced frequency domain signal S'(t,f) back to the time domain through inverse short-time Fourier transform, and the inverse short-time Fourier transform formula is: 。 3. According to the method of wireless intelligent control of RV equipment based on AI and voice recognition according to claim 1, it is characterized in that: The process of generating the review code is as follows: S31. In the conic section C n (a,b):y 2 =ax 2 -Select a point on bx (mod n) and calculate the review generated code C g =Pp×point(mod n); In the formula, n=p×q, p and q represent two different large prime numbers; a、b∈Z n And (a,n)=(b,n)=1, (a / p)=(a / q)=-1; #C n =(p+1)×(q+1), p+1=2r, q+1=2r, r and s represent prime numbers; G is C n The base point of (a,b); Pp represents the permission parameter, (Pp, N n )=1,N n =2×r×s; Z n represents the remainder class ring modulo n; S32, calculate the first-order review parameter A1=H(data||C g ); Where data represents the binary string of the speech signal; H is the hash function, H: {0,1} * ×C n → ; ={1,2,…,N n -1};|| indicates cascade operation; S33. Calculate the second-order review parameter A2 = A1 × S A +point (mod n); Among them, S A represents the second-order parameter factor, S A =-D A ×G(mod n),D A represents the first-order parameter factor and Pp×D A =1(modN n ); S34. Generate review code A=(A1, A2).

4. According to the method of wireless intelligent control of RV equipment based on AI and voice recognition according to claim 1, it is characterized in that: The audit verification process to ensure the legality of the voice signal source is as follows: S41. Calculate the first-order audit factor fc1=Pp×A2+A1×G (mod n); Where A1 represents the first-order review parameter; A2 represents the second-order review parameter; S42, calculate the second-order audit factor fc2=H(data||fc1); S43. If A1=fc2, the audit verification is passed, that is, the source of the received voice signal data is legal.

5. According to the method of wireless intelligent control of RV equipment based on AI and voice recognition according to claim 1, it is characterized in that: The process of extracting speech features using Mel frequency cepstral coefficients and logarithmic Mel spectrum is as follows: S51, Mel frequency cepstrum coefficients: ； Wherein, MFCC(k) represents the kth Mel frequency cepstrum coefficient; Y(m) represents the spectrum value of the Mel filter bank; M' represents the number of Mel filter banks; m represents the mth filter in the Mel filter bank; S52, Logarithmic Mel Spectrum: LMS(k)=log(|Y(k)| 2 ); Where LMS(k) represents the kth logarithmic Mel spectrum feature; S53, perform multi-layer convolution fusion on the Mel frequency cepstral coefficients and the logarithmic Mel spectrum features to obtain the multimodal time-frequency features of the speech: ； In the formula, F fusion represents the fused feature matrix; W1 and W2 represent the convolution kernel weights; b represents the bias; σ represents the activation function.

6. The method for wireless intelligent control of RV equipment based on AI and voice recognition according to claim 1 is characterized in that: The end-to-end speech recognition model with bidirectional attention mechanism includes: Encoder: Transform the multimodal features F fusion Encoded into a high-dimensional hidden representation using a stacked bidirectional LSTM network: ； In the formula, h t represents the hidden state at the tth moment; Attention mechanism: Capture the global context information in the speech signal and generate a context vector through weighted sum: ； ； ； ； ； In the formula, c i represents the context vector, which is the information extracted from the encoder hidden layer through the attention mechanism to provide richer context information; W a and v a Represents the weight parameter of the attention mechanism; β t represents the time step adaptive weight; represents the improved attention weight; represents the attention weight; W t represents the time step adaptive weight matrix; [h i ;h j ] represents the encoder hidden state h i and the decoder hidden state h j Connection; F fusion (t) represents the output after multimodal feature fusion; T represents the total time steps of the encoder; Decoder: Combines context vector and language modeling information to generate text data: and i =Decoder(and i-1 ,c i ); In the formula, y i represents the i-th output word or character generated by the decoder, generated by the decoder based on the previous output and the context vector; y i-1 Represents the output of the previous time step, which is passed as input to the next time step during the decoding process.

7. The method for wireless intelligent control of RV equipment based on AI and voice recognition according to claim 1 is characterized in that: The process of generating standard instructions using the intent recognition and slot filling model is as follows: S71. Add a context-dependent dynamic weight adjustment module to the standard Transformer model: ； ； In the formula, represents the semantic vector at time t; h t Represents the encoder output of Transformer at time t; Represents the global semantic features of the text, which is calculated as the weighted average of the features of the text at all times; α t Indicates the dynamic weight of the feature at the current moment; W α represents the weight matrix; The classification probability formula is: P(C|T')=softmax(W); S72. Construct a slot filling model, extract the device name and operation type key information from the text as the slot, and the slot filling formula is: ； In the formula, P(S j |T') represents S j The probability distribution of W s and b s They represent the weight and bias of the slot prediction model respectively; S73. Generate an executable instruction format based on the intent classification and slot filling results, and perform verification. The verification process is as follows: S7301. Define the standardized instruction format as: I=(D,O,P); In the formula, I represents the generated instruction; D represents the target device; O represents the operation type; P represents additional parameters; S7302, instruction generation formula: I=f(C,S)=Map(C,S); Where f(C,S) represents the mapping from intent C and slot S to instruction I; Map represents the mapping function from intent and slot to instruction; S7303. To ensure the rationality and security of the instruction, perform rule verification. The rule verification formula is: ； In the formula, V represents the instruction verification result, 1 is valid and 0 is invalid; R i (I) represents the verification result of the ith rule on instruction I; δ(x) represents the logical AND operation.

8. The method for wireless intelligent control of RV equipment based on AI and voice recognition according to claim 1 is characterized in that: The optimized control instruction generation process is as follows: S81, collecting environmental data E related to the control instruction in real time through the environmental sensor network; S82, cleaning and feature extraction of the collected environmental data E to eliminate noise and obtain key influencing factors; Abnormal data filtering: Filter abnormal data by setting dynamic thresholds: ； In the formula, Represents the filtered environmental data; μ i and σ i They represent the mean and standard deviation of the data respectively; k represents the sensitivity parameter; e i Represents a single environmental data indicator; Feature extraction: Extract statistical features from time series data, including: Environmental data average: ; Data change trend: ; In the formula, T'' represents the length of the time interval or the total duration of data sampling, that is, the time span of the data; i (t) represents the value of the i-th environmental data at time t; S83. Construct a multimodal environment perception model based on the correlation between environmental data and instructions: Environmental state classification: The environmental state E is mapped to a discrete state space S' through a classification model: ； In the formula, s represents the environmental state category; h E Represents the feature vector of environmental data based on Transformer; W s and b s Represent the classifier weight and bias respectively; User preference modeling: Combine historical operation records to build a user preference function U(D,s): ； Where U(D,s) represents the user's preference value for device D in environment state s; Freq(D,s) represents the historical usage frequency of device D in state s; Rec(D,s) represents the recommendation weight of device D in state s; α' and β' represent weight parameters; S84, generating an optimized control strategy according to the current environment state s, the instruction I and the user preference U; Optimization objective function: ; In the formula, O * represents the optimized operation instruction; C(O,E) represents the resource consumption of executing instruction O; Adaptively optimize the parameter P in the control instruction, and optimize the parameter P * for: ； Where P * represents the optimized instruction parameter; γ' represents the adjustment step; ΔP(E) represents the parameter increment of the environmental data E; S85, generate optimized control instruction I * =(D,O * ,P * ).

9. A wireless intelligent control system for RV equipment based on AI and voice recognition, which is used to execute a wireless intelligent control method for RV equipment based on AI and voice recognition as described in any one of claims 1 to 8, characterized in that: include: The speech recognition module is used to collect user speech signals through a microphone and perform noise removal and signal enhancement processing; The speech analysis module is used to convert speech into text data, and to perform semantic understanding and command analysis on the text data to identify specific control instructions; Artificial intelligence decision-making module, which uses AI algorithms to intelligently analyze the analysis results and generate control commands based on the current status of the RV equipment; The equipment control module is used to control various equipment in the RV through wireless communication technology according to the control commands output by the artificial intelligence decision module; The user interface module is used to provide a user interface for users to interact with the system.

Images