Water heating pipeline pressure self-adaptive regulation and control method and system based on Internet of Things

Abstract

The invention provides a water heating pipeline pressure self-adaptive regulation and control method and system based on the Internet of Things, and belongs to the technical field of intelligent control of pipelines of the Internet of Things. According to the technical scheme, the method comprises the steps of building a pipeline acoustic sensing network, collecting pipeline external voiceprint signals to generate lightweight feature data, building a voiceprint-pressure mapping model based on the feature data, outputting a real-time pressure value, generating a pipe network state diagram based on the real-time pressure value, generating a pressure regulation and control instruction based on the pipe network state diagram, and achieving pressure self-adaptive regulation. Feedback data is collected according to the regulation and control result, and mapping model parameters are updated. The method has the advantages that the pipeline acoustic sensing network is constructed, edge calculation, depth feature extraction and the parallel neural network are combined, the pressure self-adaptive regulation and control capacity which does not need invasive sensing arrangement and has timeliness and robustness is achieved, meanwhile, a feedback closed loop is formed through a voiceprint-pressure mapping model, and the method is suitable for large-scale popularization and application. And the processing capacity of the system on complex dynamic working conditions is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent control of Internet of Things pipelines, and particularly to a method and system for adaptive regulation of water heating pipeline pressure based on the Internet of Things. Background Art

[0002] With the rapid development of Internet of Things (IoT) and edge computing technologies, water heating pipeline systems have gradually achieved informatization and automation management in scenarios such as urban infrastructure, building energy conservation, and smart homes; Traditional water heating systems mostly rely on single-point monitoring devices such as pressure sensors and flow meters, combined with controllers to achieve simple on-off regulation functions. However, in recent years, with the increase in urban pipe network density and the complexity of building systems, it is difficult to accurately reflect the operating state in complex pipeline networks only by point-type pressure monitoring. Especially when facing dynamic working conditions such as water hammer effect, pressure fluctuations, and abnormal flow rates, the existing linear sensing means have obvious shortcomings in response speed and system coordination. In the prior art, common methods are to deploy multiple pressure sensors or flow monitors, collect physical data and upload it to the central server for unified analysis and regulation. However, this method has four technical bottlenecks: First, the cost of laying points is high. Pressure sensors need to be embedded, which is difficult to maintain and has poor stability; Second, data update lags behind. Dependence on the central server for data fusion analysis leads to regulation delay and inability to achieve fast response; Third, the dimension of sensed data is single, and it is unable to capture high-frequency acoustic information such as wall resonance and transient disturbances during pipeline operation, resulting in low accuracy of anomaly recognition; Fourth, it is unable to update the self-learning model based on the system operating state and lacks dynamic adaptation ability; Therefore, researching a wide-range pressure sensing and regulation mechanism with high sensitivity, low cost, and deployable at nodes has become the key direction for the evolution of water heating intelligent systems. Summary of the Invention

[0003] The purpose of the present invention is to provide a method and system for adaptive regulation of water heating pipeline pressure based on the Internet of Things, which realizes the ability of pressure adaptive regulation with timeliness and robustness without invasive sensing layout by constructing a pipeline acoustic sensing network, combining edge computing, deep feature extraction, and parallel neural networks, and at the same time uses a voiceprint-pressure mapping model to form a feedback closed loop, improving the system's processing ability for complex dynamic working conditions.

[0004] The present invention is implemented by the following measures: A method for adaptive regulation of water heating pipeline pressure based on the Internet of Things, characterized by comprising: Build a pipeline acoustic perception network to collect the acoustic fingerprint signals outside the pipeline, and generate lightweight feature data through preprocessing by an edge computing node; Construct an acoustic fingerprint-pressure mapping model based on the feature data and output the real-time pressure value; Generate a pipeline network status map based on the real-time pressure value, generate a pressure regulation instruction based on the pipeline network status map, and control the IoT actuator to achieve pressure adaptive regulation; Collect feedback data according to the regulation result and update the parameters of the acoustic fingerprint-pressure mapping model.

[0005] The specific features of the present invention also include: The construction of the pipeline acoustic perception network to collect the acoustic fingerprint signals outside the pipeline includes: arranging attached acoustic sensors at key positions on the outer wall of the pipeline, and connecting the acoustic sensors to the edge computing node through an IoT communication module for real-time collection of structural sound signals and flow noise signals during operation.

[0006] The generation of lightweight feature data through preprocessing by the edge computing node includes: denoising the collected acoustic fingerprint signals and extracting frequency-domain features.

[0007] Constructing an acoustic fingerprint-pressure mapping model based on the feature data and outputting the real-time pressure value includes: the edge computing node receiving the lightweight acoustic feature data and performing time window division according to the set trigger rule; For each data segment within each valid time window, three types of key features, namely, spectral energy distribution, main frequency band amplitude volatility sequence, and harmonic attenuation gradient sequence, are respectively extracted, and they are organized into a multi-dimensional time series input data in the order of collection time to form a feature data matrix; Subsequently, the feature data matrix is normalized: The acoustic fingerprint-pressure mapping model is loaded and instantiated by the edge computing node calling the model construction module, and the model is composed of a convolutional neural network branch and a long short-term memory network branch connected in parallel; The input end of the convolutional network branch receives the normalized spectral energy matrix and completes the frequency-domain feature combination processing through multiple groups of convolutional layers and pooling layers in sequence; The long short-term memory network branch is used to process the time series vectors composed of the main frequency band volatility and the harmonic attenuation gradient; The model construction module sets the parameters, input formats, and fusion methods of each branch through a configuration file, and the fusion layer splices the output results of the convolutional network branch and the long short-term memory network branch and inputs them into a fully connected structure to complete the feature mapping process and generate the real-time pressure value; The generation of the pipe network status diagram based on the real-time pressure value includes: after the edge computing node outputs the real-time pressure value of each time window, it generates the real-time pressure value corresponding to the current node, encapsulates the pressure value, its acquisition timestamp, and the node number into structured data, and sends it to the data aggregation and processing unit through the Internet of Things communication module; After receiving the structured data of multiple edge computing nodes, the data aggregation and processing unit establishes a node set and a connection relationship set according to the node numbers in the structured data and the preset pipeline topology structure; In the established node set, each node attribute includes the pressure value corresponding to the node, the acquisition time, the node number, and the data validity status, and each connection relationship set defines the adjacent connection, connection direction, and distance attributes between nodes.

[0008] The data aggregation and processing unit receives data at a fixed period and updates the node set attributes. When it is found that a certain node does not upload data in the current period, the data status of this node is marked as a missing status, and based on the valid pressure values of its adjacent nodes in the same period, a linear interpolation method is used to compensate the pressure value of the missing node. After all node information is updated and completed, the data aggregation and processing unit combines and generates the pipe network status diagram corresponding to the current period based on the updated node set and connection relationship set.

[0009] The generation of the pressure regulation instruction based on the pipe network status diagram to control the Internet of Things actuator to achieve pressure adaptive adjustment includes: the data aggregation and processing unit makes an interval judgment according to the real-time pressure values of each node in the pipe network status diagram and the preset upper and lower pressure threshold values, identifies the nodes in the abnormal state, and combines the connection relationship between nodes to identify the abnormal area; When the abnormal area meets the preset judgment conditions, the data aggregation and processing unit generates a pressure regulation instruction according to the abnormal range and the pressure deviation situation. The pressure regulation instruction includes the adjustment type, actuator number, control parameters, and time information, and is sent to the target Internet of Things actuator through the communication module. After receiving the regulation instruction, the Internet of Things actuator completes parameter verification and action execution.

[0010] Collecting feedback data according to the regulation result and updating the parameters of the voiceprint-pressure mapping model includes: collecting the feedback data of the Internet of Things actuator after completing the regulation instruction, and comparing the pressure measurement value of the current node included in the feedback data with the real-time pressure value output by the edge computing node based on the voiceprint-pressure mapping model and the target control parameters set in the regulation instruction; When the comparison result meets the preset deviation conditions, update sample data is constructed based on the lightweight acoustic feature data in the historical period and the pressure measurement value included in the feedback data, and part of the parameters of the voiceprint-pressure mapping model are adjusted by the edge computing node to achieve periodic correction of the model.

[0011] An Internet of Things-based self-adaptive pressure regulation system for water heating pipelines, comprising: A data processing module: constructing a pipeline acoustic perception network, collecting external acoustic fingerprint signals of the pipeline, and preprocessing through an edge computing node to generate lightweight feature data; A mapping model module: constructing an acoustic fingerprint-pressure mapping model based on the feature data and outputting a real-time pressure value; An analysis and processing module: generating a pipeline network status map based on the real-time pressure value, generating a pressure regulation instruction based on the pipeline network status map, and controlling an Internet of Things actuator to achieve self-adaptive pressure regulation; A model update module: collecting feedback data according to the regulation result and updating the parameters of the acoustic fingerprint-pressure mapping model.

[0012] The beneficial effects of the present invention are: By constructing a pipeline acoustic perception network, combining edge computing, deep feature extraction and parallel neural networks, the present invention realizes the self-adaptive pressure regulation ability without invasive sensing arrangement, with timeliness and robustness. At the same time, a feedback closed loop is formed by using the acoustic fingerprint-pressure mapping model, improving the system's processing ability for complex dynamic working conditions. Description of the Drawings

[0013] Figure 1 is a flowchart of an Internet of Things-based self-adaptive pressure regulation method for water heating pipelines provided in Embodiment 1 of the present invention. Detailed Embodiments

[0014] To clearly illustrate the technical features of the present solution, the present solution will be elaborated through specific embodiments.

[0015] Embodiment 1 See Figure 1 , an Internet of Things-based self-adaptive pressure regulation method for water heating pipelines, comprising: Constructing a pipeline acoustic perception network, collecting external acoustic fingerprint signals of the pipeline, and preprocessing through an edge computing node to generate lightweight feature data; Constructing an acoustic fingerprint-pressure mapping model based on the feature data and outputting a real-time pressure value; Generating a pipeline network status map based on the real-time pressure value, generating a pressure regulation instruction based on the pipeline network status map, and controlling an Internet of Things actuator to achieve self-adaptive pressure regulation; Collecting feedback data according to the regulation result and updating the parameters of the acoustic fingerprint-pressure mapping model.

[0016] Construct a pipeline acoustic perception network to collect external acoustic fingerprint signals of the pipeline, including: deploy attached acoustic sensors at key positions on the outer wall of the pipeline. The acoustic sensors are connected to the edge computing nodes through the Internet of Things communication module, and are used to collect structural acoustic signals and flow noise signals during operation in real time.

[0017] The attached acoustic sensor refers to a micro vibration sensing element composed of piezoelectric ceramics, MEMS microphones or waveguide structures, etc. It can convert the weak mechanical vibration on the pipeline wall into an electrical signal without damaging the pipeline body structure. Compared with traditional invasive pressure sensors, this type of sensor does not require hole opening or cutting operations during the deployment process, greatly improving the safety and adaptability for old or high-pressure pipe network systems. In addition, the sensor establishes a low-power wireless communication link with the edge computing node through the Internet of Things communication module (such as LoRa, NB-IoT or WiFi) to meet the actual engineering requirements of real-time data transmission back and low-maintenance operation; Regarding the installation position and action logic of the "attached acoustic sensor", the so-called "key positions" are targeted deployments based on the pressure gradient characteristics, water flow distribution characteristics and fluctuation sensitive areas of the plumbing pipeline system. For example, pipeline elbows, branch confluence points, the middle section of the main line and the area near the end load are all positions with intense flow disturbances or frequent pressure changes, with stronger acoustic responses and higher signal-to-noise ratios. Therefore, they are suitable as acoustic fingerprint collection points. Installing attached sensors at these positions is beneficial to improving the overall acoustic fingerprint collection effectiveness of the system and ensuring the spatial representativeness and dynamic response ability of the data basis; Collecting external acoustic fingerprint signals of the pipeline includes: collecting continuous flow sounds, transient pulse sounds, resonance frequency changes and wall vibration signals caused by medium flow, pressure changes or valve operations, which are used to reflect the pressure state inside the pipeline; Specifically, the structural acoustic signal refers to the acoustic signal formed by the tiny mechanical vibration caused by internal pressure fluctuations, medium impact, valve opening and closing or water hammer effect in the pipeline, which is conducted to the outer wall of the pipeline through the pipeline material and excites the vibration of the structural natural mode. The frequency range of this type of signal is generally low, the distribution is stable, carrying pipeline structure response information, reflecting instantaneous disturbances, resonant changes or stress conduction behaviors in the system; The flow noise signal refers to the broadband acoustic signal radiated outside the pipe wall through coupling caused by continuous turbulence, vortices, cavitation and other flow disturbances generated during the flow of the internal medium (such as hot water, gas) in the pipeline. The spectral characteristics of this signal change with pressure, flow rate, temperature difference and pipe diameter, reflecting the steady-state pressure field and flow characteristics in the pipeline; Regarding the distinction between "structural sound signals" and "flow noise signals", by utilizing the differences in their frequency domain and modal response characteristics, an indirect inversion of the pressure state is achieved. Among them, structural sound signals are mainly distributed in the low-frequency band (usually in the range of 20–500 Hz), and their signal intensity is strongly correlated with the medium impact force and the pipe wall elastic mode. It is particularly suitable for detecting dynamic events such as water hammer effects and sudden pressure pulses. Flow noise signals have a wide-spectrum distribution characteristic in the medium-high frequency band (500–3000 Hz) and can accurately reflect the continuous flow state in the pipeline system, such as flow stability, local blockage, and backflow disturbance.

[0018] The lightweight feature data generated by preprocessing through the edge computing node includes: denoising the collected voiceprint signal and extracting frequency domain features.

[0019] The denoising process includes filtering the voiceprint signal to suppress environmental interference signals. The frequency domain feature extraction includes extracting feature parameters such as resonance frequency, spectral energy, and transient amplitude, and dimensionally compressing the feature parameters based on the principal component analysis method or autoencoding technology to generate lightweight acoustic feature data for modeling. This processing flow mainly includes steps such as signal acquisition and caching, denoising filtering, frequency domain feature extraction, feature compression, and data standardization, which are all completed on the edge computing node; First, in the voiceprint signal acquisition stage, the edge computing node receives the original signal from the attached acoustic sensor through the Internet of Things communication module and continuously caches the signal at a fixed sampling rate (for example, 48 kHz). The sliding time window mechanism (or the trigger mechanism based on signal mutation) is used to segment and cache the signal, and then it enters the denoising processing stage. The system adaptively selects filtering strategies according to the current signal characteristics, including band-pass filters, wavelet denoising, or adaptive background noise suppression methods, to filter out the low-frequency structural resonance and high-frequency electromagnetic interference in the building environment, so as to retain the true voiceprint response characteristics induced by pressure to the greatest extent; Taking the band-pass filter as an example; Its transfer function is defined as:

[0020] The denoised signal enters the frequency domain feature extraction process, and the time-domain signal is converted into spectral data through the fast Fourier transform (FFT); converted into spectral data:

[0021] On this basis, three types of key feature parameters are extracted: One is the resonance frequency, that is, the main energy peak and its harmonic positions in the FFT spectrogram, which are used to capture the resonance frequency of the pipeline modal response (extract the frequency corresponding to the main peak from extract the corresponding frequency of the main peak ) Second, it is the spectral energy. The sub - intervals are divided according to the preset frequency bands, and the energy vector is obtained by integrating each frequency band, reflecting the local flow energy distribution. The spectral energy (the energy value is obtained by summing the squares of the spectral values in the frequency band ); ) Third, it is the transient amplitude, The local time - domain peak value and volatility are extracted by the short - time window method, characterizing the sudden pressure disturbance. The transient amplitude (calculating the short - time energy with the short - time window ); )

[0022] The above features are uniformly normalized, using the min - max normalization formula:

[0023] Thus, the numerical consistency between different dimensions is ensured.

[0024] In the feature compression stage, the linear method of principal component analysis (PCA) or the non - linear method of auto - encoder (AutoEncoder) can be selected for dimensionality reduction.

[0025] In the PCA method, the original feature matrix is used to construct the covariance matrix . After eigenvalue decomposition, the principal component matrix is selected to generate the dimensionality - reduced vector , where is the lightweight acoustic feature data. If the auto - encoder method is used, the high - dimensional features are compressed by the encoding function into:

[0026] where is clearly the lightweight acoustic feature data, and this feature is used as the input of the subsequent voiceprint - pressure mapping model.

[0027] Finally, the edge computing node encapsulates the lightweight acoustic feature data into a structured format and attaches meta - information such as node number, acquisition timestamp, and signal integrity score to ensure that the data can be recognized and traced by the model. where is the transfer function of the band - pass filter, is the input frequency, is the cut - off frequency of the filter; is the voiceprint signal at the th sampling point, is the spectral value at the th frequency point, is the total number of sampling points, is the sampling frequency; is the main peak frequency, is the start and end frequencies of the frequency band, is the frequency index, is the spectral energy; is the th point within the window, is the window length is the short-time energy; is the original eigenvalue; is the minimum eigenvalue, is the normalization result; is the original feature matrix, is the covariance matrix, is the number of samples, is the principal component matrix, is the PCA compression result; is the AutoEncoder input, is the weight matrix, is the bias, is the activation function, is the output lightweight acoustic feature data.

[0028] Based on the feature data, a voiceprint-pressure mapping model is constructed and the real-time pressure value is output, including: the edge computing node receives the lightweight acoustic feature data , and performs time window division according to the set trigger rules, including continuously collecting data for a fixed duration or detecting a signal mutation exceeding the threshold; For each data segment within the valid time window, the spectral energy distribution , the main frequency band amplitude volatility sequence and the harmonic attenuation gradient sequence are respectively extracted, and the three types of key features are organized into a multi-dimensional time series input data in the order of acquisition time to form a feature data matrix X;

[0029] Subsequently, the input data matrix X is normalized. Through the min-max normalization method, all data dimensions are uniformly distributed in the [0,1] interval:

[0030] The voiceprint-pressure mapping model is called and instantiated by the model construction module of the edge computing node. The model consists of a convolutional neural network branch and a long short-term memory network branch connected in parallel; The input end of the convolutional network branch receives the normalized spectral energy matrix and completes the frequency domain feature combination processing through multiple sets of convolutional layers and pooling layers in sequence; The long short-term memory network branch is used to process the time series vector composed of the main frequency band volatility and the harmonic attenuation gradient; The normalized spectral energy matrix As the input of the convolutional neural network branch, the main frequency band amplitude volatility And the harmonic attenuation gradient Together serve as the input of the long short-term memory network branch; Among them, the convolutional network branch inputs the spectral energy matrix , and realizes the frequency domain feature combination processing through multiple groups of convolutional and pooling layers. Its feature output The mathematical expression is:

[0031] The long short-term memory network branch aims at the input time series feature vector , and models the dynamic evolution process of the features through a multi-layer LSTM structure. Its feature output Is expressed as:

[0032] Among them, Is the LSTM hidden state at the t-th moment, Is the output of the hidden state at the last time step. The model construction module sets the parameters, input formats and fusion methods of each branch through a configuration file. The fusion layer splices the output results of the convolutional network branch and the long short-term memory network branch and inputs them into the fully connected structure to complete the feature mapping process and generate the real-time pressure value; Specifically: The model construction module sets parameters such as the network structure, input dimension, convolutional kernel size, number of LSTM hidden units, learning rate, and fusion strategy of the convolutional network and the long short-term memory network branch through an external configuration file; Subsequently, in the fusion layer, the output features of the convolutional network branch And the output features of the long short-term memory network branch Are spliced into a unified feature vector , and the mathematical expression is:

[0033] Finally, the fused feature vector is input into the fully connected layer mapping structure, and after linear mapping through the weight matrix And the bias , and then through the Sigmoid activation function , finally outputs the real-time pressure value corresponding to the current time window , and the specific expression is:

[0034] Among them, is lightweight acoustic feature data; are the original feature data of spectral energy distribution, main frequency band amplitude volatility, and harmonic attenuation gradient respectively; , and are the normalized spectral energy matrix, main frequency band amplitude volatility, and harmonic attenuation gradient respectively; X is a multi-dimensional input feature data matrix; is the normalized feature data matrix; are the minimum and maximum values of the feature data matrix respectively; are the weight matrix and bias of the convolutional network respectively; MaxPool is the max pooling layer; is the output feature of the convolutional network; is the LSTM hidden state feature, are the LSTM input weight, hidden state weight, and bias term respectively; is the final LSTM time feature output; is the fused feature vector; are the weight matrix and bias vector of the fully connected layer respectively; is the Sigmoid function; p is the real-time pressure value of the final mapping output, taking values in the interval (0, 1), and corresponding to the actual physical pressure range after anti-normalization.

[0035] Generating a pipeline network status map based on the real-time pressure value includes: after the edge computing node completes the output of the real-time pressure value for each time window, generating the real-time pressure value corresponding to the current node, and encapsulating the pressure value, its acquisition timestamp, and the node number into structured data, and sending it to the data aggregation and processing unit through the Internet of Things communication module; After receiving the structured data of multiple edge computing nodes, the data aggregation and processing unit establishes a node set and a connection relationship set according to the node numbers in the structured data and the preset pipeline topology structure, and the pipeline topology structure is used to record the physical connection relationship, connection direction, and physical distance between nodes; In the established node set, each node attribute includes the pressure value corresponding to the node, acquisition time, node number, and data validity status, and each connection relationship set defines the adjacent connection, connection direction, and distance attributes between nodes.

[0036] The data aggregation and processing unit receives data at fixed intervals and updates the attributes of the node set. When it is found that a certain node does not upload data in the current cycle, the data status of this node is marked as missing, and based on the effective pressure values of its adjacent nodes in the same cycle, a linear interpolation method is used to compensate the pressure value of the missing node. After all node information is updated and completed, the data aggregation and processing unit generates a pipeline network status map corresponding to the current cycle based on the updated node set and connection relationship set.

[0037] Generate pressure regulation instructions based on the pipeline network status map to control the IoT actuator to achieve pressure adaptive adjustment, including: the data aggregation and processing unit makes an interval judgment according to the real-time pressure values of each node in the pipeline network status map and the preset upper and lower pressure threshold values, identifies the nodes in the abnormal state, and combines the connection relationship between nodes to identify the abnormal area; When the abnormal area meets the preset judgment conditions, the data aggregation and processing unit generates pressure regulation instructions according to the abnormal range and pressure deviation situation. The pressure regulation instructions include adjustment type, actuator number, control parameters and time information, and are sent to the target IoT actuator through the communication module. After receiving the regulation instructions, the IoT actuator completes parameter verification and action execution.

[0038] Specifically: After the data aggregation and processing unit completes the generation of the pipeline network status map for the current cycle, it judges the pressure value fields of all nodes in the pipeline network status map. The pressure value is the real-time pressure value generated and uploaded by the edge computing node based on the voiceprint-pressure mapping model in claim 4. The data aggregation and processing unit makes an interval judgment based on the upper and lower pressure threshold values stored in its internal memory. The upper and lower threshold values are loaded by the configuration file in the system initialization stage. The configuration file is set by the deployment personnel according to the pipeline design pressure parameters. During the operation process, the pressure mean and standard deviation of the historical N acquisition cycles are statistically updated through the sliding window mechanism. The update rule is: upper threshold = mean + 3σ, lower threshold = mean - 3σ. The update cycle is the same as the status map generation cycle, and a revision log is recorded for each round of update.

[0039] When the real-time pressure value of a certain node is higher than the upper threshold value of the current cycle, this node is marked as an overpressure node; when it is lower than the lower threshold value, it is marked as an underpressure node; the rest are marked as normal nodes. After all node statuses are marked, the data aggregation and processing unit identifies the abnormal area based on the connection relationship recorded in the pipeline network status map. The abnormal area is defined as: in the current cycle, there are at least 3 continuously connected abnormal nodes of the same type, or the node set spanning 2 or more physical pipe segments is in an abnormal state, and this abnormality has not been eliminated in M consecutive status map cycles, where M is the preset abnormal continuous judgment cycle, defaulting to 3.

[0040] The data aggregation processing unit calculates the regulation task parameters according to the spatial coverage range, pressure deviation degree, and abnormal duration of the abnormal area, and selects the target Internet of Things actuator in combination with the actuator distribution table defined in Claim 1 to generate a pressure regulation instruction. The regulation instruction includes: the unique actuator number, the target adjustment type (pressure increase or pressure reduction), the adjustment amplitude, the action duration, the trigger timestamp, and the priority level; among them, the adjustment amplitude is calculated in a proportional control manner, and the amplitude = pressure deviation value × Kp, where Kp is the control gain coefficient set by the system. All regulation instructions are encapsulated in the MQTT protocol format and have a field signature identifier.

[0041] After receiving the regulation instruction, the Internet of Things actuator sequentially executes a multi-level security verification process, including verifying the consistency of the data packet hash value, verifying the consistency of the actuator firmware version number, and verifying whether the instruction parameter value exceeds the mechanical limit range. After all verifications pass, the execution process is started, and the process includes adjusting the opening of the electric valve, the output frequency of the water pump, or the operating state of the control object mounted on the actuator.

[0042] Collect feedback data according to the regulation result and update the parameters of the voiceprint-pressure mapping model, including: collecting the feedback data of the Internet of Things actuator after completing the regulation instruction, and comparing the measured pressure value of the current node included in the feedback data with the real-time pressure value output by the edge computing node based on the voiceprint-pressure mapping model and the target control parameters set in the regulation instruction; When the comparison result meets the preset deviation condition, construct updated sample data based on the lightweight acoustic feature data in the historical period and the measured pressure value included in the feedback data, and adjust some parameters of the voiceprint-pressure mapping model through the edge computing node to achieve periodic correction of the model.

[0043] Specifically: After the Internet of Things actuator completes the action execution of the pressure regulation instruction, it sends the feedback data containing the control result to the data aggregation processing unit through the Internet of Things communication module. The feedback data includes the execution type of the current regulation action, the control amplitude, the control completion time, the actuator number, the internal status identification information, and the measured pressure value of the current node after the control is completed. The measured pressure value is generated by the built-in sensing module of the Internet of Things actuator; After receiving the feedback data, the data aggregation processing unit compares the measured pressure value of the current node (Internet of Things actuator) included in the feedback data with the real-time pressure value output by the edge computing node based on the voiceprint-pressure mapping model and the target control parameters set in the regulation instruction; Compare it with the real-time pressure value generated and uploaded by the edge computing node in the current period, and combine the target control parameters in the original regulation instruction to calculate the actual execution deviation value; When the deviation value exceeds the error threshold set by the system, or when the cumulative control deviation of the same node exceeds the tolerance upper limit within two or more consecutive cycles, the system triggers the model parameter update process; after the update is triggered, the data aggregation processing unit retrieves the acquisition records of the acoustic sensors within the corresponding time period, extracts the lightweight voiceprint feature data corresponding to the feedback cycle, and forms a new sample pair with the feedback pressure value; the sample pair is used to construct the input-output training pair for model update, where the input is the multi-dimensional time series voiceprint feature within the historical time window, and the output is the pressure result value fed back during this time period. The edge computing node receives the sample pair through the model update interface, and based on the loaded voiceprint-pressure mapping model structure, calls the incremental update module to fine-tune some connection weight parameters on the premise of keeping the original network structure unchanged; during the update process, a local parameter adjustment strategy is adopted to prevent overfitting, and after the model completes a parameter update, the version number and update time information are written into the model configuration file for subsequent version comparison and traceability, completing the periodic correction operation of the model.

[0044] Embodiment 2 The water heating pipeline pressure adaptive control system based on the Internet of Things includes: Data processing module: Construct a pipeline acoustic perception network, collect the external voiceprint signals of the pipeline, and generate lightweight feature data through preprocessing by the edge computing node; Mapping model module: Based on the feature data, construct a voiceprint-pressure mapping model and output the real-time pressure value; Analysis and processing module: Generate a pipeline network state diagram based on the real-time pressure value, generate a pressure control instruction based on the pipeline network state diagram, and control the Internet of Things actuator to achieve pressure adaptive adjustment; Model update module: Collect feedback data according to the control result and update the parameters of the voiceprint-pressure mapping model.

[0045] The technical features not described in the present invention can be realized by or adopted from the prior art, and will not be elaborated here. Of course, the above description is not a limitation of the present invention, and the present invention is not limited to the above examples. Changes, modifications, additions, or substitutions made by those of ordinary skill in the art within the scope of the essence of the present invention should also fall within the protection scope of the present invention.

Claims

1. An adaptive pressure regulation method for water heating pipes based on the Internet of Things, characterized in that, Including: Construct a pipeline acoustic perception network to collect the acoustic fingerprint signals outside the pipeline, and preprocess them through edge computing nodes to generate lightweight feature data; Construct an acoustic fingerprint-pressure mapping model based on the feature data and output the real-time pressure value; Generate a pipeline network status map based on the real-time pressure value, generate a pressure regulation instruction based on the pipeline network status map, and control the IoT actuator to achieve adaptive pressure regulation; Collect feedback data according to the regulation result and update the parameters of the acoustic fingerprint-pressure mapping model.

2. The method for adaptively regulating the pressure of a water heating pipeline based on the Internet of Things according to claim 1, wherein The construction of the pipeline acoustic perception network to collect the acoustic fingerprint signals outside the pipeline includes: arranging attached acoustic sensors at key positions on the outer wall of the pipeline, and connecting the acoustic sensors to the edge computing nodes through the IoT communication module for real-time collection of structural acoustic signals and flow noise signals during operation.

3. The method for adaptively regulating the pressure of a water heating pipeline based on the Internet of Things according to claim 2, wherein The preprocessing through the edge computing nodes to generate lightweight feature data includes: denoising the collected acoustic fingerprint signals and extracting frequency domain features.

4. The method for adaptively regulating the pressure of a water heating pipeline based on the Internet of Things according to claim 3, characterized in that, Constructing an acoustic fingerprint-pressure mapping model based on the feature data and outputting the real-time pressure value includes: the edge computing node receives the lightweight acoustic feature data and performs time window division according to the set trigger rule; For each data segment within each valid time window, three types of key features, namely the spectral energy distribution, the main frequency band amplitude volatility sequence, and the harmonic attenuation gradient sequence, are respectively extracted, and they are organized into a multi-dimensional time series input data in the order of collection time to form a feature data matrix; Subsequently, perform normalization processing on the feature data matrix: The acoustic fingerprint-pressure mapping model is loaded and instantiated by the edge computing node calling the model construction module. The model consists of a convolutional neural network branch and a long short-term memory network branch connected in parallel; The input end of the convolutional network branch receives the normalized spectral energy matrix and completes the frequency domain feature combination processing through multiple groups of convolutional layers and pooling layers in sequence; The long short-term memory network branch is used to process the time series vector composed of the main frequency band volatility and the harmonic attenuation gradient; The model construction module sets the parameters, input formats, and fusion methods of each branch through the configuration file. The fusion layer splices the output results of the convolutional network branch and the long short-term memory network branch and inputs them into the fully connected structure to complete the feature mapping process and generate the real-time pressure value.

5. The method for adaptively regulating the pressure of a water heating pipeline based on the Internet of Things according to claim 4, wherein The generation of the pipeline network status map based on the real-time pressure value includes: after the edge computing node completes the output of the real-time pressure value for each time window, it generates the real-time pressure value corresponding to the current node, and encapsulates the pressure value, its acquisition timestamp, and the node number into structured data, and sends it to the data aggregation and processing unit through the IoT communication module; After receiving the structured data of multiple edge computing nodes, the data aggregation and processing unit establishes a node set and a connection relationship set according to the node numbers in the structured data and the preset pipeline topology structure; In the established node set, each node attribute includes the pressure value corresponding to the node, the acquisition time, the node number, and the data validity status. Each connection relationship set defines the adjacent connections, connection directions, and distance attributes between nodes; The data aggregation processing unit receives data at a fixed period and updates the attributes of the node set. When it is found that a certain node does not upload data in the current period, the data status of this node is marked as the missing status, and based on the effective pressure values of its adjacent nodes in the same period, a linear interpolation method is used to compensate the pressure value of the missing node. After all node information is updated and completed, the data aggregation processing unit combines the updated node set and the connection relationship set to generate the pipeline state diagram corresponding to the current period.

6. The method for adaptively regulating the pressure of a water heating pipeline based on the Internet of Things according to claim 5, characterized in that, Generating a pressure regulation instruction based on the pipeline state diagram and controlling the IoT actuator to achieve pressure adaptive regulation includes: the data aggregation processing unit makes an interval judgment according to the real-time pressure values of each node in the pipeline state diagram and the preset upper and lower pressure threshold values, identifies the nodes in the abnormal state, and combines the connection relationship between nodes to identify the abnormal area; When the abnormal area meets the preset determination condition, the data aggregation processing unit generates a pressure regulation instruction according to the abnormal range and the pressure deviation situation. The pressure regulation instruction includes the regulation type, actuator number, control parameter and time information, and is sent to the target IoT actuator through the communication module. After receiving the regulation instruction, the IoT actuator completes parameter verification and action execution.

7. The method for adaptively regulating the pressure of a water heating pipeline based on the Internet of Things according to claim 6, characterized in that, Collecting feedback data according to the regulation result and updating the parameters of the voiceprint-pressure mapping model includes: collecting the feedback data of the IoT actuator after completing the regulation instruction, and comparing the pressure measurement value of the current node included in the feedback data with the real-time pressure value output by the edge computing node based on the voiceprint-pressure mapping model and the target control parameter set in the regulation instruction; When the comparison result meets the preset deviation condition, update sample data is constructed based on the lightweight acoustic feature data in the historical period and the pressure measurement value included in the feedback data, and partial parameters of the voiceprint-pressure mapping model are adjusted by the edge computing node to achieve periodic correction of the model.

8. An Internet of Things-based water heating pipeline pressure adaptive regulation system adopting the method described in any one of claims 1-7, characterized in that, Including: Data processing module: constructing a pipeline acoustic perception network, collecting the external voiceprint signal of the pipeline, and preprocessing through the edge computing node to generate lightweight feature data; Mapping model module: constructing a voiceprint-pressure mapping model based on the feature data and outputting the real-time pressure value; Analysis and processing module: generating a pipeline state diagram based on the real-time pressure value, generating a pressure regulation instruction based on the pipeline state diagram, and controlling the IoT actuator to achieve pressure adaptive regulation; Model update module: collecting feedback data according to the regulation result and updating the parameters of the voiceprint-pressure mapping model.

Images