Edge computing gateway platform architecture interaction method, device and medium

Abstract

The invention discloses an edge computing gateway platform architecture interaction method and device and a medium, and relates to the technical field of acoustic man-machine interaction, and the method comprises the steps: receiving an original sound wave signal, carrying out the human voice activation detection and wake-up word confirmation of the original sound wave signal through a hierarchical wake-up detection mechanism, and generating an effective sound wave segment; performing fast Fourier transform on the effective sound wave segments to generate frequency domain energy features; extracting energy distribution of the frequency domain energy features, and generating a dominant frequency set according to a preset energy threshold value; constructing a sound wave-instruction rule base, and inputting the dominant frequency set into the sound wave-instruction rule base to generate a primary acoustic instruction; converting the primary acoustic instruction into a text instruction, and generating an instruction data set through an Agent decision agent; and collecting space coordinates and network packet loss rate parameters of the edge equipment, and generating a transmission path scheme through a space field intensity gradient algorithm in combination with the instruction data set. According to the method, the instruction recognition accuracy in the noise environment is improved through anti-noise wave analysis.

Description

Technical Field

[0001] This invention relates to the field of acoustic human-computer interaction technology, and in particular to an edge computing gateway platform architecture interaction method, device, and storage medium. Background Technology

[0002] In recent years, edge computing gateway platform architecture interaction technologies have gradually integrated multimodal sensing capabilities. Among them, acoustic wave interaction channels have become a technological branch in industrial IoT control scenarios due to their non-contact transmission characteristics. Current mainstream solutions achieve basic command recognition through frequency domain analysis of acoustic signals, employ Fast Fourier Transform to extract spectral energy features, and combine this with fixed thresholds to select dominant frequency points to complete command mapping. At the path planning layer, Euclidean distance optimization algorithms based on device physical location have become the basic solution, while state feedback mechanisms generally use single-frequency acoustic wave encoding of state information.

[0003] Existing methods suffer from environmental adaptability defects in acoustic command recognition. Fixed energy threshold mechanisms cannot effectively filter broadband interference in complex acoustic environments in industrial settings. Discrete single-frequency mapping increases the risk of command mismatch due to equipment frequency offset and multipath effects. In addition, path planning relies solely on spatial location parameters while ignoring dynamic fluctuations in network quality. The use of low-information-density single-frequency pulse coding for state feedback lacks noise immunity design. The fragmented operation of each link leads to inaccurate transmission link selection and a decrease in feedback signal recognition rate. This results in insufficient robustness of acoustic command recognition and a lack of cross-domain collaborative optimization capabilities in high-noise industrial scenarios. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an edge computing gateway platform architecture interaction method to solve the problems of insufficient command anti-interference capability and defects in cross-domain collaboration mechanism.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides an interaction method for an edge computing gateway platform architecture, comprising: receiving a raw acoustic wave signal; performing human voice activation detection and wake-up word confirmation on the raw acoustic wave signal through a hierarchical wake-up detection mechanism to generate a valid acoustic wave segment; performing a fast Fourier transform on the valid acoustic wave segment to generate frequency domain energy features; extracting the energy distribution of the frequency domain energy features and generating a dominant frequency set according to a preset energy threshold; constructing an acoustic wave-command rule base, inputting the dominant frequency set into the acoustic wave-command rule base to generate a primary acoustic command; converting the primary acoustic command into a text command through speech recognition processing, and generating a command dataset through an Agent decision-making intelligent agent; collecting the spatial coordinates and network packet loss rate parameters of the edge device, and generating a transmission path scheme through a spatial field strength gradient algorithm in conjunction with the command dataset; distributing the command dataset to the target edge device through the transmission path scheme, parsing the command dataset and executing the corresponding control action to generate a response dataset; A sound wave synthesis model is constructed. The response dataset is input into the sound wave synthesis model, the output is a pulse sequence, and a pulse sound wave signal is generated through an audio output device.

[0007] As a preferred embodiment of the edge computing gateway platform architecture interaction method described in this invention, the specific steps for generating frequency domain energy features by performing a Fast Fourier Transform on the effective acoustic wave segment are as follows: The system receives the raw acoustic signal, performs pre-emphasis processing and frame-segmentation windowing operations to generate a windowed acoustic signal, and performs a fast Fourier transform on the windowed acoustic signal to generate a frequency domain signal. Calculate the original frequency domain energy value of the frequency domain signal to generate the original frequency domain energy feature; perform frequency domain mask multiplication filtering operation on the original frequency domain energy feature to generate the filtered frequency domain energy feature; Calculate the short-time energy value of the filtered frequency domain energy feature for each frame and perform energy threshold comparison; extract Mel frequency cepstral coefficient features from the filtered frequency domain energy features of each frame that pass the threshold detection. The Mel frequency cepstral coefficient features are input into the Gaussian mixture model to calculate the probability density of human voice, generating probability density values ​​of human voice and noise. The probability density values ​​of human voice and noise are compared to select human voice frames; the human voice frames are input into a separable convolutional neural network to perform wake word matching, and the consecutive human voice frames that are successfully matched are spliced ​​into effective sound wave segments; a fast Fourier transform is performed on the effective sound wave segments to generate frequency domain energy features.

[0008] As a preferred embodiment of the edge computing gateway platform architecture interaction method described in this invention, the specific steps for constructing the acoustic wave-command rule base—inputting the dominant frequency set into the acoustic wave-command rule base to generate primary acoustic commands—are as follows: Frequency points with frequency energy values ​​greater than a preset energy threshold in the frequency domain energy features are selected, sorted in descending order of frequency energy values, and a dominant frequency set is generated based on the sorting results. Associate all operation commands supported by the edge gateway with predefined frequency ranges to build a sound wave-command rule base; The primary acoustic commands are generated by matching the frequency points of the dominant frequency set with the predefined frequency ranges in the acoustic wave-command rule base.

[0009] As a preferred embodiment of the edge computing gateway platform architecture interaction method described in this invention, the specific steps for converting primary acoustic commands into text commands and generating a command dataset through an Agent decision-making intelligent agent are as follows: Collect the actual operation manuals, historical instruction records, verbal command records, and device interface documents used in the target scenario, and pre-set the instruction code-text mapping table; The primary acoustic commands are matched with the command encoding-text mapping table to generate text commands; Input text instructions into the constructed Agent decision-making agent to generate an instruction dataset.

[0010] As a preferred embodiment of the edge computing gateway platform architecture interaction method described in this invention, the instruction dataset includes device address, operation instructions, and execution parameters.

[0011] As a preferred embodiment of the edge computing gateway platform architecture interaction method described in this invention, the specific steps for generating the transmission path scheme are as follows: Traverse all edge devices, obtain their three-dimensional spatial coordinates and network packet loss rate parameters within the most recent time window through their registration information, and form an edge device parameter set. Parse the instruction dataset and extract the device address of the target edge device; Calculate the field strength value of each edge device based on the parameter set of the edge devices, and generate the field strength set of the edge devices. A gradient ascent path search is performed based on the field strength set of edge devices and device addresses to generate a transmission path scheme.

[0012] As a preferred embodiment of the edge computing gateway platform architecture interaction method described in this invention, the specific steps for generating the response dataset are as follows: Read the edge device path node sequence from the transmission path scheme, and transmit the instruction dataset to the target edge device node via a wireless communication protocol; The target edge device node parses the instruction dataset to obtain operation instructions and execution parameters, and verifies the integrity of the operation instructions through cyclic redundancy check and field completeness check; Based on the execution parameters, the system executes the control actions corresponding to the operation instructions, collects the operation result status codes and the current status parameters of the device, and generates a response dataset.

[0013] As a preferred embodiment of the edge computing gateway platform architecture interaction method described in this invention, the specific steps for generating the pulse acoustic signal are as follows: A sound wave synthesis model is constructed based on predefined status code-frequency mapping rules and waveform generation functions; The response dataset is input into the acoustic synthesis model, and a pulse sequence is generated through status code frequency mapping, waveform generation, pulse cutting and silence interval insertion operations. The pulse sequence is converted into an analog electrical signal by a digital-to-analog converter, and then the audio output device is driven to convert the analog electrical signal into a pulse sound wave signal.

[0014] In a second aspect, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements any step of the edge computing gateway platform architecture interaction method as described in the first aspect of the present invention.

[0015] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the edge computing gateway platform architecture interaction method as described in the first aspect of the present invention.

[0016] The beneficial effects of this invention are as follows: it improves the accuracy of command recognition in industrial noise environments through a dynamic anti-noise wave analysis mechanism, breaking through the bottleneck of environmental adaptability; it expands the supported command capacity by leveraging wide-band command mapping to meet the needs of complex control scenarios; it achieves dual-dimensional collaborative optimization of physical location and network quality based on the spatial field strength gradient algorithm, dynamically avoiding high packet loss nodes, improving command transmission success rate and reducing end-to-end latency; and it combines an acoustic-network closed-loop collaborative architecture to ensure the recognizability of state feedback pulses in high-noise environments, solving the problem of feedback failure. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart illustrating the interaction methods of the edge computing gateway platform architecture.

[0019] Figure 2 A flowchart for generating primary acoustic instructions.

[0020] Figure 3 A flowchart for generating a transmission path scheme.

[0021] Figure 4 This is a flowchart for generating pulsed acoustic wave signals. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0025] Reference Figures 1-4 As one embodiment of the present invention, this embodiment provides an interaction method for an edge computing gateway platform architecture, including the following steps: S1. Receive the original sound wave signal, and perform human voice activation detection and wake-up word confirmation on the original sound wave signal through a hierarchical wake-up detection mechanism to generate effective sound wave segments; perform fast Fourier transform on the effective sound wave segments to generate frequency domain energy features; S1.1 Receive the original acoustic wave signal, perform pre-emphasis processing and frame windowing operation to generate a windowed acoustic wave signal; perform fast Fourier transform on the windowed acoustic wave signal to generate a frequency domain signal; It should be noted that the gateway collects user voice input in real time through a microphone and speaker device connected via USB or Bluetooth. The microphone and speaker device converts the user's voice input into an analog electrical signal through the principle of sound wave electrical signal conversion, and transmits the analog electrical signal to the analog-to-digital converter built into the gateway. The analog-to-digital converter discretizes the analog electrical signal according to a fixed sampling rate (based on the Nyquist sampling theorem and the characteristics of the human voice spectrum, such as 16kHz) to generate the original sound wave signal in the form of a digital sequence. The original acoustic signal is pre-emphasized using a first-order high-pass filter to enhance its high-frequency components, resulting in an enhanced acoustic signal. This enhanced signal is then segmented into frames of fixed time length (defined based on spectral resolution requirements, such as 25 milliseconds). Each frame is then multiplied point-by-point by a Hanning window function value to generate a windowed acoustic signal, expressed as follows: ; ; in, Indicates the first Windowed acoustic signal at each sampling point; Indicates the first Enhanced acoustic signal at each sampling point; Indicates the first Hanning window function values ​​at each sampling point; This represents the total number of sampling points involved in the calculation; Indicates the sampling point index; The windowed acoustic signal is converted into a frequency domain signal using the Fast Fourier Transform formula, expressed as follows: ; in, Indicates the first Frequency domain signal at a frequency point; Indicates frequency index; It represents the negative of the imaginary unit, i.e., the downward direction.

[0026] S1.2 Calculate the original frequency domain energy value of the frequency domain signal to generate the original frequency domain energy feature; perform frequency domain mask multiplication filtering operation on the original frequency domain energy feature to generate the filtered frequency domain energy feature; It should be noted that the frequency domain signal at each frequency point is a complex number formed by the sum of its imaginary and real parts; the original frequency domain energy value of each signal is calculated based on its real and imaginary parts, expressed as follows: ; in, Indicates the first The original frequency domain energy value of each frequency point; Represents the real part of the frequency domain signal; Represents the imaginary part of a frequency domain signal; By performing the original frequency domain energy value calculation on all frequency points, the original frequency domain energy values ​​of all frequency points are obtained. Then, by combining them in the corresponding frequency point order, the original frequency domain energy value sequence is obtained, which is the original frequency domain energy feature.

[0027] Set a set of fixed weight values ​​that correspond one-to-one with each frequency point; the length of the fixed weight value sequence is the same as the original frequency domain energy characteristics. The weight value of each position is set according to whether the corresponding frequency point belongs to the target frequency band to be retained (defined based on the physiological vocal characteristics of human speech, such as the 300-3400Hz human voice frequency band): the weight value of the frequency point belonging to the target frequency band is set to 1.0 (for complete retention), and the weight value of the frequency point not belonging to the target frequency band is set to 0.0 or a very small value close to 0.0 (for complete suppression or significant attenuation); the fixed weight value sequence is the frequency domain mask sequence; Calculate the product of the frequency domain mask sequence and all corresponding frequency points in the original frequency domain energy features to generate the filtered frequency domain energy value. Arrange all the filtered frequency domain energy values ​​into a new sequence according to the frequency point order, which is the filtered frequency domain energy feature.

[0028] S1.3 Calculate the short-time energy value of the filtered frequency domain energy feature for each frame and perform energy threshold comparison; extract Mel frequency cepstral coefficient features from the filtered frequency domain energy features of each frame that pass the threshold detection. It should be noted that the filtered frequency domain energy values ​​of all frequency points of the filtered frequency domain energy characteristics of each frame are summed to generate the short-time energy value of the filtered frequency domain energy characteristics of each frame. During periods without user commands, 500 consecutive audio frames are continuously collected as background noise samples. The arithmetic mean of the energy values ​​of all frequency points in the consecutive audio frames is calculated. The product of the arithmetic mean and the environmental safety factor (defined based on the statistical characteristics of environmental noise and the reliability requirements of the application scenario, such as 3.0 in industrial scenarios) is weighted to generate a noise energy threshold. The short-time energy value of the filtered frequency domain energy feature of each frame is compared with the noise energy threshold: when the short-time energy value exceeds the noise energy threshold, the corresponding frame of the filtered frequency domain energy feature is marked as a valid audio frame; otherwise, it is discarded directly. Mel-frequency cepstral coefficient feature extraction is performed on valid audio frames: The valid audio frames are input into a Mel-scale triangular filter bank, which consists of 40 triangular window functions covering the core frequency band of human voice (defined based on the physiological vocal characteristics of human speech, 300-3400Hz). Each filter performs a nonlinear weighted summation on the core frequency band of human voice, the low-frequency noise band, and the high-frequency noise band, and outputs the Mel-band energy value of each frequency band. Natural logarithm operation is performed on all Mel-band energy values ​​to enhance the contrast of energy differences between different frequency bands. Discrete cosine transform is used to eliminate the correlation between feature dimensions and reduce dimensionality to generate Mel-frequency cepstral coefficient features.

[0029] S1.4 Input the Mel frequency cepstral coefficient features into the Gaussian mixture model to calculate the probability density of human voice, and generate the probability density values ​​of human voice and noise; It should be noted that the Gaussian mixture model consists of multiple Gaussian components, each containing three Gaussian parameters: mixture weight coefficients, mean vector, and covariance matrix. The number of Gaussian components in the Gaussian mixture model is determined based on the complexity of the target data distribution and prior knowledge. An independent Gaussian parameter structure is created for each Gaussian component, containing a mixture weight coefficient, a mean vector, and a covariance matrix. The constraint type of the covariance matrix is ​​defined, and different forms such as complete covariance, diagonal covariance, or spherical covariance can be selected to control the complexity and expressive power of the Gaussian mixture model. The Gaussian parameter structures of all Gaussian components are combined to form a complete probability density function expression, which is the weighted sum of the probability density functions of each Gaussian component and the corresponding mixture weight coefficients, thus completing the theoretical framework of the Gaussian mixture model. A professional recording environment was selected, with background noise controlled below 30 decibels (based on acoustic environment quality standards and speech acquisition technical specifications). Speakers of different ages, genders, and dialect regions were recruited. Each speaker recorded basic speech units including monophthongs, diphthongs, and consonant-vowel combinations, as well as commonly used vocabulary and short phrases in the context. The recording equipment used a high-fidelity condenser microphone, with the microphone 20 cm away from the speaker's mouth and equipped with a windscreen. The original recordings were acoustically analyzed to select qualified samples with a signal-to-noise ratio (SNR) greater than the SNR threshold (based on human auditory perception characteristics, such as 25 decibels). Segments containing interference such as coughing and breathing sounds were manually removed. The qualified samples were then segmented into phonemes and annotated with text to form a clean speech library. Human voice training samples are obtained by calling a clean speech library; environmental noise samples are collected from real-world scenarios; and the human voice training samples and environmental noise samples are used as training samples. The Gaussian parameters are iteratively optimized using the expectation-maximization algorithm. Specifically, in the initialization phase, the Gaussian parameters of each Gaussian component are randomly set (based on the principle of probabilistic randomness). In the expectation step, the posterior probability of each training sample belonging to each Gaussian component is calculated by calling the posterior probability formula. In the maximization step, the Gaussian parameters are recalculated based on the posterior probability using the formulas for calculating each Gaussian parameter. When the change in Gaussian parameters is less than the change threshold (defined based on the balance between convergence accuracy and computational efficiency of the Gaussian mixture model) or the maximum number of iterations is reached, the training of the Gaussian mixture model is complete. The expression for the Gaussian mixture model is: ; ; in, The probability density value representing the output of the Gaussian mixture model refers to the probability that the input Mel frequency cepstral coefficient features belong to human voice or noise in the Gaussian mixture model. This represents the characteristics of the input Mel-frequency cepstral coefficients; Indicates the number of Gaussian components; The index represents the Gaussian component, with a value of ; Indicates the first The mixing weight coefficient of each Gaussian component refers to the proportion of importance of the current Gaussian component in the entire Gaussian mixture model; Indicates the first The mean vector of each Gaussian component describes the center position of the Gaussian component in the feature space. The covariance matrix represents the correlation and distribution pattern among the feature dimensions; It represents the Gaussian density function, describing the distribution pattern of human voice or noise; This represents the characteristic dimension of the Mel frequency cepstral coefficients; This represents the matrix transpose operator; The Mel frequency cepstral coefficient features are input into the Gaussian mixture model. By substituting the mean vector and covariance matrix of the Mel frequency cepstral coefficient features into the Gaussian density function, the probability density values ​​of each human voice Gaussian component and noise Gaussian component are calculated. The product of each probability density value and the corresponding mixing weight coefficient is calculated to obtain the weighted probability value. All weighted probability values ​​are summed to obtain the total probability of belonging to human voice and the total probability of belonging to noise, respectively, thus generating the probability density values ​​of human voice and noise.

[0030] S1.5. Compare the probability density values ​​of human voice and noise to select human voice frames; input the human voice frames into a separable convolutional neural network to perform wake word matching, and splice the successfully matched consecutive human voice frames into effective sound wave segments; perform fast Fourier transform on the effective sound wave segments to generate frequency domain energy features.

[0031] It should be noted that the dimension of the input Mel-frequency cepstral coefficient features is determined, a deep convolutional layer is defined to use multiple independent filters to process each input channel to extract spatial features, a point convolutional layer is added to fuse channel information through 1x1 convolution and reduce the number of convolution parameters, a batch normalization layer and a ReLU activation function are introduced to accelerate training stability and enhance nonlinear expressive power, a global average pooling layer is connected to reduce feature dimension to prevent overfitting, a final output feature map is generated, a fully connected layer is configured to map the final output feature map to class scores (such as wake words or non-wake words), and a Softmax layer is used to convert the class scores into wake word classification probabilities, thus completing the construction of the separable convolutional neural network; Acoustic datasets for training separable convolutional neural networks were collected through multi-environment sampling and structured annotation. The specific process involved deploying high-fidelity microphone arrays in target scenarios (such as machine tool workshops and pump rooms) to collect background environmental noise and operator spoken commands. Clear speech samples were recorded from speakers of different ages and genders at varying noise levels for each wake word. A sound level meter was used to measure the ambient noise in decibels at the sampling points. The raw recordings were pre-screened, retaining qualified samples with a signal-to-noise ratio (SNR) higher than a threshold (based on human auditory perception characteristics, such as 25 dB). Segments containing sudden interference such as coughing or equipment collisions were manually removed. Phoneme-level text annotations were performed on the filtered raw data, marking the start and end times, wake word text label, and corresponding environmental noise type for each qualified sample. This resulted in an acoustic dataset containing positive samples (valid wake words) and negative samples (background noise / invalid commands), with positive samples labeled as 1 and negative samples as 0, forming a binary label set corresponding to the audio data. The acoustic dataset was then divided into training, validation, and test sets. The Mel-frequency cepstral coefficient features of the training set samples are input into the separable convolutional neural network for forward propagation to obtain the predicted probability distribution. The cross-entropy loss between the predicted probability distribution and the true labels (derived from the binary label set) is calculated, and the gradient of the cross-entropy loss with respect to all network weight parameters is calculated using the backpropagation algorithm. The Adam optimizer is used to dynamically adjust the learning rate according to the gradient direction and iteratively update the weights of the deep convolutional kernel, point convolutional kernel, and fully connected layer. After each training cycle, the accuracy and cross-entropy loss are calculated using the validation set to monitor the performance of the separable convolutional neural network and prevent overfitting. When the validation set accuracy no longer improves for several consecutive cycles, the early stopping mechanism is activated to terminate the training. The final accuracy and recall of the trained separable convolutional neural network are evaluated using a completely independent test set to ensure that the separable convolutional neural network has high precision and generalization ability, thus completing the training of the separable convolutional neural network. The expression for converting to wake word classification probability is: ; in, Indicates category The probability of classifying the wake word; Represents the number of categories in the total number of categories. The category refers to the category to be calculated, for example... When the value is 1, it indicates the first category (such as a wake word). When the value is 2, it indicates the second category (such as a non-wake word); This represents a temporary variable for summing the index, used for looping through all categories from 1 to... ; Indicates the first The score for the number of categories; Indicates the first The score for the number of categories; When the probability density value of human voice is higher than that of noise, the current frame is determined to be a human voice frame. All human voice frames are combined into a human voice frame sequence and input into a separable convolutional neural network to generate wake word matching probabilities. When the wake word matching probability exceeds the probability threshold, the current human voice frame is marked as a wake word frame. According to the principle of timestamp continuity, adjacent wake word frames are spliced ​​together in chronological order to form a continuous sound wave segment. A fast Fourier transform is performed on the continuous sound wave segment to generate a frequency domain energy feature vector.

[0032] S2. Extract the energy distribution of frequency domain energy features and generate a dominant frequency set based on a preset energy threshold; construct a sound wave-command rule base, input the dominant frequency set into the sound wave-command rule base, and generate primary acoustic commands; convert the primary acoustic commands into text commands, and generate a command dataset through an Agent decision-making intelligent agent; S2.1 Filter the frequency points in the frequency domain energy features whose frequency energy value is greater than the preset energy threshold, sort them in descending order of frequency energy value, and generate a dominant frequency set based on the sorting results; It should be noted that the energy threshold is defined based on the overall statistical characteristics of the frequency domain energy features. Specifically, the arithmetic mean of all frequency domain energy values ​​in the entire frequency domain energy features is calculated, and the product of the arithmetic mean and the amplification factor (defined based on the signal-to-noise ratio requirements and background noise statistical characteristics of the application scenario, usually taking a value of 2 to 5) is used as the energy threshold. Traverse the frequency domain energy values ​​of all frequency points in the frequency domain energy feature, compare each frequency domain energy with the energy threshold, and select all frequency points greater than the energy threshold as candidate points; sort the selected candidate points in descending order of frequency domain energy value to form the dominant frequency set.

[0033] S2.2. Associate all operation commands supported by the edge gateway with predefined frequency ranges to build a sound wave-command rule base; It should be noted that a list of all operation commands and corresponding execution parameters that the edge gateway needs to support should be provided. For example, operation commands refer to the categories of control actions that the edge gateway can perform, including device restart, data backup, mode switching, service start / stop, and status query. Execution parameters refer to the specific configuration values ​​required to execute the corresponding operation command, including the target device identifier, latency, mode value, and backup target path. Within the audible sound wave frequency range (defined based on the physiological perception characteristics of human hearing, such as 100Hz to 15000Hz), select non-overlapping continuous frequency bands. Maintain a fixed interval between the lower and upper frequency limits of each continuous frequency band (defined based on the critical bandwidth principle of human hearing, such as 500Hz). Based on acoustic characteristics, bind each continuous frequency band to different operational commands, ensuring no overlap between any two continuous frequency bands. For example, assign operational commands requiring rapid response or important commands (such as emergency stop) to the mid-to-high frequency bands that are sensitive and easily distinguishable to the human ear. Assign operational commands with similar functions to adjacent frequency bands, forming logical groups for easy memorization. Record the lower and upper frequency limits of continuous frequency bands to form mapping rules consisting of the lower and upper frequency limits of the continuous frequency bands, operational commands, and execution parameters. Combine all mapping rules to form a sound wave-command rule library.

[0034] S2.3 Match the frequency points of the dominant frequency set with the predefined frequency ranges in the sound wave-instruction rule base to generate primary acoustic instructions.

[0035] It should be noted that, by traversing each frequency point in the dominant frequency set, the frequency domain energy value corresponding to each frequency point is compared with all continuous frequency segments in the acoustic wave-instruction rule base. When the frequency domain energy value falls within a certain continuous frequency segment, it is determined that the mapping rule corresponding to the current frequency point and the current continuous frequency segment is successfully matched. The operation instructions and execution parameters recorded in the current mapping rule are extracted, and the corresponding device address is read by combining the target device identifier contained in the execution parameters. The command data containing the device address, operation instructions and execution parameters is generated, and all successfully matched command data are integrated into the primary acoustic command.

[0036] S2.4. Convert the primary acoustic commands into text commands through speech recognition processing, and input the text commands into the Agent decision-making agent to generate a command dataset.

[0037] It should be noted that the process involves collecting operation manuals, historical instruction records, verbal command records, and device interface documents actually used in the target scenario, extracting all frequently occurring operation verbs and device object names, and combining them to form a basic instruction set; analyzing all device control interface protocols in the target scenario, listing all executable machine operation commands; filtering out remotely executable commands based on safety standards, excluding high-risk operations, forming a control command list, assigning a unique frequency encoding pattern to each control command, and recording the correspondence between control commands and frequency encoding patterns in tabular form to form an instruction encoding-text mapping table. The primary acoustic command is used as the query key and compared with the command encoding-text mapping table to find the table entry that completely matches the input primary acoustic command, and the corresponding text command is read and output. We collect massive amounts of unlabeled text data as training corpus, dividing it into training and validation sets. The training corpus undergoes word segmentation, denoising, and vectorization, converting each word into a numerical vector representation in a high-dimensional space. A deep Transformer computational structure is defined as the foundation of the large language model. The training set is input into the Transformer computational structure in batches, and the predicted probability distribution of each word is calculated through forward propagation. A masking language model task is used to randomly mask a portion of the words in the input training set, training the large language model to predict the masked words based on context, generating prediction results, and simultaneously combining this with the next sentence prediction task to determine the logical coherence between sentence pairs. The cross-entropy loss function between the predicted results and the true values ​​of the masked words is calculated, and the gradient of the loss function with respect to all weight parameters is calculated using the backpropagation algorithm. The Adam optimizer is used to dynamically adjust the learning rate, iteratively updating the weight parameters along the gradient's reverse direction, enabling the large language model to gradually learn the semantic representation, grammatical structure, and common-sense knowledge of the words. A hybrid precision computation and distributed training strategy is employed to accelerate the reverse iterative optimization process. At the end of each training cycle, a validation set is used to evaluate the perplexity metric: the text sequence of the validation set is input into the large language model with updated weight parameters, allowing the large language model to predict the probability of the next word's occurrence word by word; for each real word in the text sequence, the predicted probability value given by the large language model is recorded; the negative log-likelihood average of all word prediction probabilities is calculated to obtain the cross-entropy loss; the cross-entropy loss is then exponentially calculated, and the result is the perplexity metric value for this training round. The perplexity metric value directly reflects the uncertainty of the large language model's prediction of the validation set text; the lower the perplexity metric value, the stronger the language model's language modeling ability; when the perplexity no longer decreases, an early stopping mechanism is activated, resulting in large language model weight parameters with convergent loss function and language representation capabilities, forming a large language model with language understanding and generation capabilities. The process involves collecting equipment operation manuals, historical instruction texts, maintenance logs, and safety specification documents from the target domain as training data. This training data is then cleaned and labeled, extracting elements such as operation verbs, equipment names, and execution parameters to form a structured labeled dataset. The trained large language model is then subjected to supervised fine-tuning using this structured labeled dataset. Equipment interface specifications (such as equipment address encoding rules and operation instruction codes), security policies (such as operation permissions and execution parameter thresholds), and operation procedures are embedded into the large language model's contextual learning framework in the form of structured prompts. During deployment, the large language model continuously receives text instruction input, using a self-attention mechanism to parse the semantic intent and parameter constraints of the text instructions, generating structured operation commands that conform to the equipment interface specifications. A thought chain technique is employed to enable the large language model to perform multi-step reasoning, dynamically coordinating the operation logic of multiple devices and avoiding conflicts. This results in an Agent decision-making intelligence with the large language model as its core, achieving end-to-end conversion from natural language to industrial control instructions. The text command is input into the Agent decision-making agent, which parses the command words and execution parameters in the text command, identifies the specific operation action and target device name, queries the device interface specification and security policy to find the corresponding device address code, operation command and execution parameters, and generates a command dataset. Agent decision-making intelligence is a decision-making intelligence with a large language model as its core. It parses the semantic intent of text instructions through the large language model and generates an instruction dataset containing device addresses, operation instructions and execution parameters, thus realizing an architecture that separates computation from agent services. The cyclic redundancy check (CRC) value is calculated using a polynomial based on CRC-32, and stored in the check field of the instruction dataset. The expression for calculating the CRC value is as follows: ; in, Indicates the cyclic redundancy check value; Represents the remainder polynomial; The message polynomial refers to the entire instruction dataset, which is treated as a binary sequence. This indicates a shift operation, which transforms the message polynomial... The corresponding binary sequence is shifted left by 32 bits; It represents the generator polynomial, which is a fixed 32nd degree polynomial.

[0038] Existing technologies suffer from high false trigger rates in industrial scenarios due to the lack of environmental noise filtering mechanisms during frequency domain feature extraction. The use of fixed single-frequency points for instruction mapping results in weak resistance to frequency offset and limited instruction capacity. Furthermore, simple threshold matching exhibits a significantly increased error rate under sudden noise. This solution significantly reduces the false trigger rate by dynamically adjusting the energy judgment threshold using a noise adaptive threshold algorithm. It employs wide-band instruction binding based on the critical bandwidth principle to ensure discernibility by the human ear, thereby improving the frequency tolerance range and multiplying the instruction capacity, while maintaining excellent recognition stability in high-intensity industrial noise environments.

[0039] S3. Collect the spatial coordinates and network packet loss rate parameters of the edge device, and generate a transmission path scheme by combining the instruction dataset and the spatial field strength gradient algorithm. S3.1. Traverse all edge devices, obtain three-dimensional spatial coordinates through edge device registration information, obtain network packet loss rate parameters within the most recent time window through real-time network probes, and form an edge device parameter set. It should be noted that the three-dimensional spatial coordinates of the current edge device are read from the registration configuration information of the edge device (fixed parameters entered when the edge device is deployed to the physical location, representing the actual physical installation location of the edge device); the edge gateway generates multiple probe data packets through the Internet Control Message Protocol within a fixed time window and sends them to the edge device; after receiving the probe data packets, the edge device automatically generates response data packets according to the Internet Control Message Protocol and returns them; the number of response data packets received within the fixed time window (defined based on the dynamic changes in network status, such as 1 second) is counted; the difference between the probe data packets and the response data packets is calculated to obtain the number of lost data packets; the percentage of the number of lost data packets to the number of probe data packets sent is used as the network packet loss rate parameter; the three-dimensional spatial coordinates and network packet loss rate parameter of each edge device are combined into a data structure; the data structures of all edge devices are combined to form the edge device parameter set.

[0040] S3.2 Parse the instruction dataset and extract the device address of the target edge device; calculate the field strength value of each edge device based on the edge device parameter set and generate the edge device field strength set; It should be noted that the device address of the target edge device is obtained by reading the device address field value stored in the instruction dataset; the data of each edge device in the edge device parameter set is traversed, and the distance from each edge device to the edge gateway is calculated based on the three-dimensional spatial coordinates of each edge device. The reciprocal of the distance is then taken, and the field strength value of each edge device is generated based on the reciprocal of the distance and the device packet loss rate parameter. The device addresses of all edge devices are combined with their corresponding field strength values ​​to generate the edge device field strength set; the expression for calculating the distance from each edge device to the edge gateway is as follows: ; in, Indicates the first The distance from each edge device to the edge gateway; , , Indicates the first The three-dimensional spatial coordinates of an edge device; , , The three-dimensional spatial coordinates of the edge gateway are obtained based on the registration configuration information preset during deployment; Represents the edge device index; The expression for calculating the field strength is: ; in, Indicates the first The field strength value of each edge device; Indicates the first The packet loss rate parameter for each edge device.

[0041] S3.3. Perform gradient ascent path search based on the edge device field strength set and device address to generate a transmission path scheme.

[0042] It should be noted that the maximum value of the path node sequence is defined based on network size and device density: the product of the theoretical minimum number of hops between the two farthest edge devices in the network (defined based on network topology and shortest path principle) and the safety factor (defined based on the margin design principle in network reliability engineering, such as 2.0); The algorithm reads the field strength value corresponding to the device address of the target edge device from the edge device field strength set as the path endpoint and the field strength value of the edge gateway as the path start point. Starting from the path start point, the algorithm takes the path start point as the current path node, obtains the device addresses and corresponding field strength values ​​of all adjacent edge devices within the communication range of the current path node, selects the edge device with the highest field strength value as the next-hop path node, records the next-hop path node in the path node sequence and sets it as the new current node, and repeats the search-record process until the path node sequence reaches the path endpoint. The generated path node sequence is the transmission path scheme. When the field strength value of the next-hop path node is not higher than that of the current path node or the length of the path node sequence exceeds the maximum value of the path node sequence, the iteration is terminated and the path search is marked as failed.

[0043] Existing technologies use a single distance or network quality metric for path selection, which has the drawback of failing to consider both physical location and link status, making the transmission path susceptible to spatial obstruction or network fluctuations. This solution calculates the field strength value by fusing the device's three-dimensional spatial coordinates and real-time packet loss rate parameters, quantifies communication reliability by multiplying the inverse of the distance by the network quality factor, and performs a hop-by-hop search along the increasing field strength direction based on a gradient ascent strategy, achieving dual optimization of physical proximity and link stability. This effectively avoids signal blind spots and network congestion nodes, improving command transmission success rate and real-time performance.

[0044] S4. Distribute the instruction dataset to the target edge device through the transmission path scheme, parse the instruction dataset and execute the corresponding control actions to generate a response dataset; S4.1 Read the edge device path node sequence from the transmission path scheme, and transmit the instruction dataset to the target edge device node through the wireless communication protocol; It should be noted that the edge device path node sequence recorded in the transmission path scheme is read. Starting from the edge gateway, the instruction dataset is sent to the first edge device node in the edge device path node sequence through the wireless communication protocol. After receiving the instruction dataset, the current edge device node forwards the instruction dataset to the next edge device node in the edge device path node sequence. The receiving and forwarding operation is repeated until the instruction dataset is delivered to the last edge node, i.e., the target edge device node.

[0045] S4.2. The target edge device node parses the instruction dataset to obtain operation instructions and execution parameters, and verifies the integrity of the operation instructions through cyclic redundancy check and field completeness check. It should be noted that after receiving the instruction dataset, the target edge device node extracts the values ​​of the operation instruction field and the execution parameter field, and generates a verification cyclic redundancy check value for the instruction dataset using CRC-32. The verification cyclic redundancy check value is compared with the recorded cyclic redundancy check value, and the existence and conformity of the operation instruction field value and the execution parameter field value are checked. When the verification cyclic redundancy check value matches the recorded cyclic redundancy check value and all field values ​​pass the check, the operation instruction is determined to be complete.

[0046] It should also be noted that the numerical value of the operation instruction field must be a specific string in the sound wave-instruction rule base (such as "DEVICE_REBOOT"), and the execution parameter field must contain a specific key name and the corresponding numerical type must meet the key name requirements (such as "delay" key corresponds to an integer value, and "target" key corresponds to a string value).

[0047] S4.3 Based on the execution parameters, execute the control actions corresponding to the operation instructions, collect the operation result status codes and the current status parameters of the device, and generate a response dataset.

[0048] It should be noted that, based on the execution parameters, the local control logic of the edge device is invoked to complete the control action corresponding to the operation instruction, and the status code value returned after the control action is executed is captured. The current status parameters of the edge device are read, including electrical parameters, environmental parameters, operating status, resource status and business data. The status code value and the current status parameters of the device are combined to generate a response dataset.

[0049] S5. Construct a sound wave synthesis model, input the response dataset into the sound wave synthesis model, generate a pulse sequence, and drive the audio output device to play the pulse sound wave signal.

[0050] S5.1 Constructing an acoustic wave synthesis model based on predefined status code-frequency mapping rules and waveform generation functions; It should be noted that a unique frequency value is assigned to each status code value. The frequency value must be within the audible sound wave frequency range and maintain a sufficient distance from the frequency values ​​of other status code values ​​to avoid confusion, thus forming a status code-frequency mapping rule between status code values ​​and frequency values. The status code-frequency mapping rule is then combined with a sine function to complete the construction of the sound wave synthesis model. The acoustic synthesis model has no hierarchical structure and adopts a flat design with direct single-layer mapping. The acoustic synthesis model is composed of predefined status code-frequency mapping rules and selected sine functions connected in parallel. After inputting the status code value, the corresponding frequency value is directly indexed through the status code-frequency mapping rules, and the sine function is immediately called to output the acoustic signal. The sine function was chosen because the pure tones generated by the sine function have the highest clarity and comfort in human auditory perception, and can accurately represent the sound wave signal corresponding to the state code value through a single frequency value; the sound wave synthesis model does not require a traditional training process, and its construction is entirely based on predefined state code-frequency mapping rules and the sine function; the expression of the sound wave synthesis model is, ; in, Indicates time as The status code value is The sound wave signal at that time; The amplitude is a constant and refers to the volume of the sound wave. This represents a mapping function that maps data based on the input status code value. It searches for and returns the corresponding frequency value from the status code-frequency mapping rules; Indicates a specific moment in time; This indicates the status code value.

[0051] S5.2 Input the response dataset into the acoustic synthesis model, and generate a pulse sequence through status code frequency mapping, waveform generation, pulse cutting and silence interval insertion operations; It should be noted that the status code values ​​are extracted from the response dataset, input into the sound wave synthesis model, and the corresponding frequency values ​​are obtained through the status code-frequency mapping rule. The continuous sound wave signal corresponding to the frequency value is generated through a sine function, and the continuous sound wave signal is cut into multiple equal-length sound wave pulses according to a fixed duration (defined based on auditory perception characteristics and edge device processing efficiency, such as 300-500 milliseconds). A silent segment is generated by creating a signal segment with a zero amplitude and a fixed duration (based on the minimum perceptual threshold of the pulse interval in human factors engineering, such as 100 milliseconds). The silent segment is then directly spliced ​​to the end of each equal-length sound wave pulse to form a complete sequence of alternating equal-length sound wave pulses and silent segments, which is the pulse sequence.

[0052] S5.3 Convert the pulse sequence into an analog electrical signal using a digital-to-analog converter, and drive the audio output device to convert the analog electrical signal into a pulse sound wave signal.

[0053] It should be noted that the pulse sequence is input into the digital-to-analog converter (DAC), and the internal circuit of the DAC outputs the analog voltage corresponding to the magnitude of each pulse in the pulse sequence. The internal circuit of the DAC automatically inserts a transition voltage value between two adjacent analog voltage points, so that the analog voltage value changes linearly from the former analog voltage to the latter analog voltage with a constant slope, generating a continuous voltage signal. The high-frequency abrupt components in the continuous voltage signal are filtered out by a low-pass filter circuit, making the voltage waveform smooth and generating a continuous analog electrical signal. The analog electrical signal is transmitted to the voice coil of the audio output device. The voice coil is placed in a permanent magnetic field and generates a magnetic field of corresponding strength according to the strength of the analog electrical signal. The interaction between the magnetic field and the permanent magnetic field drives the voice coil and the connected diaphragm to reciprocate. The movement of the diaphragm compresses the surrounding air molecules, causing changes in density, thereby generating pulse sound wave signals. It should also be noted that the transition voltage value is determined by the internal circuit of the digital-to-analog converter based on the difference between the values ​​of two adjacent voltage points and the time interval. Specifically, the difference between the previous voltage point and the next voltage point is calculated, and the ratio of the difference to the time interval between the two points is used as the transition voltage value.

[0054] Existing technologies use fixed-frequency encoding of status information, which has drawbacks such as limited feedback information and insufficient resistance to environmental noise. This solution constructs a dynamic mapping relationship between status code values ​​and wide-band acoustic wave characteristics, generates a basic acoustic carrier through a sine function, forms an anti-noise pulse sequence by combining pulse cutting and silence interval insertion, and achieves waveform fidelity in digital-to-analog conversion based on transition voltage smoothing technology. Finally, it drives audio equipment to output a highly recognizable pulsed acoustic wave signal, ensuring the integrity and recognizability of status feedback even in complex industrial noise environments.

[0055] This embodiment also provides a computer device applicable to the edge computing gateway platform architecture interaction method, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the edge computing gateway platform architecture interaction method proposed in the above embodiment.

[0056] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0057] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the interaction method for implementing the edge computing gateway platform architecture as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0058] In summary, this invention improves the accuracy of command recognition in industrial noise environments through a dynamic anti-noise wave analysis mechanism, overcoming environmental adaptability bottlenecks; expands the supported command capacity through wide-band command mapping to meet the needs of complex control scenarios; achieves dual-dimensional collaborative optimization of physical location and network quality based on a spatial field strength gradient algorithm, dynamically avoiding high packet loss nodes, improving command transmission success rate and reducing end-to-end latency; and combines an acoustic-network closed-loop collaborative architecture to ensure the recognizability of state feedback pulses in high-noise environments, solving the problem of feedback failure.

[0059] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An interaction method for an edge computing gateway platform architecture, characterized in that: include, The system receives raw sound wave signals and uses a graded wake-up detection mechanism to detect human voice activation and confirm wake words, thereby generating effective sound wave segments. Perform a Fast Fourier Transform on the effective acoustic wave segment to generate frequency domain energy characteristics; Extract the energy distribution of frequency domain energy features and generate a dominant frequency set based on a preset energy threshold; construct a sound wave-command rule base, input the dominant frequency set into the sound wave-command rule base to generate primary acoustic commands; convert the primary acoustic commands into text commands and generate a command dataset through an Agent decision-making intelligent agent; Collect the spatial coordinates of edge devices and network packet loss rate parameters, and generate transmission path schemes using the spatial field intensity gradient algorithm in conjunction with the instruction dataset; The instruction dataset is distributed to the target edge device through a transmission path scheme, the instruction dataset is parsed and the corresponding control actions are executed to generate a response dataset; A sound wave synthesis model is constructed. The response dataset is input into the sound wave synthesis model, the output is a pulse sequence, and a pulse sound wave signal is generated through an audio output device.

2. The edge computing gateway platform architecture interaction method as described in claim 1, characterized in that: The specific steps for performing a Fast Fourier Transform on the effective acoustic wave segment to generate frequency domain energy features are as follows. The system receives the raw acoustic signal, performs pre-emphasis processing and frame-segmentation windowing operations to generate a windowed acoustic signal, and performs a fast Fourier transform on the windowed acoustic signal to generate a frequency domain signal. Calculate the original frequency domain energy value of the frequency domain signal and generate the original frequency domain energy characteristics; Perform frequency domain mask multiplication filtering on the original frequency domain energy features to generate filtered frequency domain energy features; Calculate the short-time energy value of the filtered frequency domain energy feature for each frame and perform energy threshold comparison; extract Mel frequency cepstral coefficient features from the filtered frequency domain energy features of each frame that pass the threshold detection. The Mel frequency cepstral coefficient features are input into the Gaussian mixture model to calculate the probability density of human voice, generating probability density values ​​of human voice and noise. The probability density values ​​of human voice and noise are compared to select human voice frames; the human voice frames are input into a separable convolutional neural network to perform wake word matching, and the consecutive human voice frames that are successfully matched are spliced ​​into effective sound wave segments; a fast Fourier transform is performed on the effective sound wave segments to generate frequency domain energy features.

3. The edge computing gateway platform architecture interaction method as described in claim 2, characterized in that: The construction of the sound wave-command rule base involves inputting the dominant frequency set into the sound wave-command rule base to generate primary acoustic commands. The specific steps are as follows: Frequency points with frequency energy values ​​greater than a preset energy threshold in the frequency domain energy features are selected, sorted in descending order of frequency energy values, and a dominant frequency set is generated based on the sorting results. Associate all operation commands supported by the edge gateway with predefined frequency ranges to build a sound wave-command rule base; The primary acoustic commands are generated by matching the frequency points of the dominant frequency set with the predefined frequency ranges in the acoustic wave-command rule base.

4. The edge computing gateway platform architecture interaction method as described in claim 3, characterized in that: The specific steps for converting primary acoustic commands into text commands and generating a command dataset through an Agent decision-making intelligent agent are as follows. Collect the actual operation manuals, historical instruction records, verbal command records, and device interface documents used in the target scenario, and pre-set the instruction code-text mapping table; The primary acoustic commands are matched with the command encoding-text mapping table to generate text commands; Input text instructions into the constructed Agent decision-making agent to generate an instruction dataset.

5. The edge computing gateway platform architecture interaction method as described in claim 4, characterized in that: The instruction dataset consists of device address, operation instructions, and execution parameters.

6. The edge computing gateway platform architecture interaction method as described in claim 5, characterized in that: The specific steps for generating the transmission path scheme are as follows: Traverse all edge devices, obtain their three-dimensional spatial coordinates and network packet loss rate parameters within the most recent time window through their registration information, and form an edge device parameter set. Parse the instruction dataset and extract the device address of the target edge device; Calculate the field strength value of each edge device based on the parameter set of the edge devices, and generate the field strength set of the edge devices. A gradient ascent path search is performed based on the field strength set of edge devices and device addresses to generate a transmission path scheme.

7. The edge computing gateway platform architecture interaction method as described in claim 6, characterized in that: The specific steps for generating the response dataset are as follows. Read the edge device path node sequence from the transmission path scheme, and transmit the instruction dataset to the target edge device node via a wireless communication protocol; The target edge device node parses the instruction dataset to obtain operation instructions and execution parameters, and verifies the integrity of the operation instructions through cyclic redundancy check and field completeness check; Based on the execution parameters, the system executes the control actions corresponding to the operation instructions, collects the operation result status codes and the current status parameters of the device, and generates a response dataset.

8. The edge computing gateway platform architecture interaction method as described in claim 7, characterized in that: The specific steps for generating the pulsed acoustic signal are as follows. A sound wave synthesis model is constructed based on predefined status code-frequency mapping rules and waveform generation functions; The response dataset is input into the acoustic synthesis model, and a pulse sequence is generated through status code frequency mapping, waveform generation, pulse cutting and silence interval insertion operations. The pulse sequence is converted into an analog electrical signal by a digital-to-analog converter, and then the audio output device is driven to convert the analog electrical signal into a pulse sound wave signal.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the edge computing gateway platform architecture interaction method according to any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the edge computing gateway platform architecture interaction method according to any one of claims 1 to 8.

Images