A noise identification method, medium, device and product of a thermal power plant

Abstract

This application discloses a noise identification method, medium, equipment, and product for thermal power plants. The thermal power plant includes multiple pieces of equipment. The method involves acquiring raw sound signals and historical acoustic feature data; digitizing each raw sound signal to obtain multiple acoustic features; determining a dynamic threshold range based on the historical acoustic feature data; identifying anomalous features among the multiple acoustic features based on the dynamic threshold range; determining a target label for each anomalous feature based on a predefined set of noise labels and the associated attributes corresponding to each anomalous feature; constructing an association graph based on the index information and target labels corresponding to the multiple anomalous features; and performing posterior probability analysis on the nodes in the association graph to obtain the noise source region of the thermal power plant. This application embodiment achieves anomaly detection through multi-dimensional features, dynamic thresholds, and label matching, and constructs an association graph for probability analysis, enabling precise location of noise sources.

Description

Technical Field

[0001] This application belongs to the field of noise detection, specifically relating to a noise identification method, storage medium, electronic equipment, and computer program product for thermal power plants. Background Technology

[0002] With increasing environmental awareness, countries are tightening regulations on industrial noise emissions. Thermal power plants must comply with these regulations to reduce the impact of noise on the environment and surrounding residents. Therefore, implementing effective noise monitoring and management measures has become a crucial task for thermal power plant operations. Modern thermal power plants are equipped with sensors and monitoring equipment that can collect operational data and acoustic signals in real time. Furthermore, with the application of IoT technology, data acquisition systems are becoming increasingly sophisticated, providing a rich data foundation for noise-based equipment condition monitoring. However, traditional methods often use single indicators such as energy, zero-crossing rate, or spectrum analysis to analyze noise, resulting in low accuracy in identifying the noise source area.

[0003] Therefore, it is very important to combine multiple indicators to analyze noise in order to accurately locate the noise source area. Summary of the Invention

[0004] The purpose of this application is to provide a noise identification method, medium, equipment, and product for thermal power plants, which can solve the problem that the accuracy of noise source area location is not high when analyzing noise based on a single index.

[0005] In a first aspect, embodiments of this application provide a noise identification method for a thermal power plant, the thermal power plant including multiple thermal power plant equipment, the method comprising: Acquire raw sound signals of multiple thermal power plant equipment under current operating conditions and historical acoustic characteristic data of multiple thermal power plant equipment under different operating conditions under normal operating conditions; Each original sound signal is digitally processed to obtain multiple acoustic characteristics corresponding to each thermal power plant device; The dynamic threshold range corresponding to each thermal power plant device is determined based on the historical acoustic feature data, and abnormal features are identified among multiple acoustic features corresponding to each thermal power plant device based on the dynamic threshold range; the abnormal features have corresponding association attributes and index information; The target label for each anomaly feature is determined based on a predefined set of noise labels and the associated attributes corresponding to each anomaly feature. An association map is constructed based on the index information corresponding to multiple anomaly features and the target labels corresponding to multiple anomaly features to characterize the noise propagation and fault association structure of the thermal power plant; the nodes in the association map are associated with the equipment of the thermal power plant. By performing posterior probability analysis on the nodes in the correlation graph, the noise source region of the thermal power plant is obtained.

[0006] Optionally, the step of digitizing each original sound signal to obtain multiple acoustic features corresponding to each thermal power plant device includes: Each original audio signal is sequentially subjected to high-pass filtering and noise suppression to obtain a preprocessed signal; Each preprocessed signal is framed and windowed to obtain multiple windowed signal frames; Perform a short-time Fourier transform on each windowed signal frame to obtain the spectrum corresponding to each windowed signal frame, and extract the frequency domain energy distribution features from the spectra corresponding to multiple windowed signal frames; Calculate the Mel frequency cepstral coefficients and the first-order difference coefficients corresponding to the Mel frequency cepstral coefficients for each windowed signal frame based on the spectrum corresponding to each windowed signal frame. Calculate the sum of squares of the amplitudes of the sampling points within each windowed signal frame to obtain the short-time average energy corresponding to each windowed signal frame; Obtain the number of times the signal crosses the zero level within each windowed signal frame, and use the ratio between the number of crossings and the frame length of the windowed signal frame as the zero-crossing rate corresponding to each windowed signal frame; The preprocessed signal is subjected to envelope extraction processing to obtain waveform envelope features; The frequency domain energy distribution characteristics, the Mel frequency cepstral coefficients corresponding to all windowed signal frames, the first-order difference coefficients corresponding to all Mel frequency cepstral coefficients, the short-time average energy corresponding to all windowed signal frames, the zero-crossing rate corresponding to all windowed signal frames, and the waveform envelope characteristics are determined as multiple acoustic characteristics corresponding to each thermal power plant equipment.

[0007] Optionally, the historical acoustic feature data includes multiple historical features and historical operating condition parameters. The step of determining the dynamic threshold range corresponding to each thermal power plant device based on the historical acoustic feature data, and determining abnormal features among the multiple acoustic features corresponding to each thermal power plant device based on the dynamic threshold range, includes: Obtain the equipment type, spatial location, related processes, and real-time operating parameters of the thermal power plant equipment; Based on the aforementioned historical characteristics, calculate the distribution parameters of all thermal power plant equipment under normal operating conditions; Based on the historical operating parameters and the distribution parameters, a dynamic threshold range corresponding to thermal power plant equipment under different operating conditions is generated; The target operating condition of the thermal power plant equipment is determined based on its equipment type, spatial location, related processes, and real-time operating parameters. Acoustic features that are outside the dynamic threshold range corresponding to the target operating condition are identified as the abnormal features.

[0008] Optionally, the noise tag set includes multiple noise tags, each noise tag having corresponding association attributes, including acoustic attributes, spatiotemporal attributes, and operating condition attributes. The step of determining the target tag corresponding to each anomalous feature based on the predefined noise tag set and the association attributes corresponding to each anomalous feature includes: Calculate the similarity between the acoustic attributes corresponding to each abnormal feature and the acoustic attributes corresponding to each noise label, and determine the noise labels with similarity higher than a preset similarity threshold as candidate labels corresponding to the abnormal features; The spatiotemporal and working condition attributes corresponding to each abnormal feature are matched and verified with the spatiotemporal and working condition attributes corresponding to each candidate label to obtain the verification result for each candidate label. Based on the verification results corresponding to each candidate label, at least one target label that meets the preset result is determined from the candidate labels corresponding to each abnormal feature.

[0009] Optionally, the step of constructing an association map characterizing the noise propagation and fault association structure of the thermal power plant based on index information corresponding to multiple abnormal features and target labels corresponding to multiple abnormal features includes: Obtain the physical layout and acoustic environment of the thermal power plant; Based on the index information of all abnormal features, multiple abnormal features are clustered to obtain at least one spatiotemporal triggering event; Based on the physical layout and acoustic environment of the power plant, a graph structure containing one or more nodes is constructed, including equipment nodes, pipeline nodes, and acoustic path nodes. Each spatiotemporal triggering event and the target label corresponding to the abnormal features in each spatiotemporal triggering event are mapped to the nodes in the graph structure, and the connection edges between different nodes in the graph structure are established according to the index information corresponding to the abnormal features, so as to construct the association graph.

[0010] Optionally, the step of performing posterior probability analysis on the nodes in the correlation graph to obtain the noise source region of the thermal power plant includes: Obtain the number of device nodes, the prior probability of each device node, and the observation probability of each device node generating observation data; The posterior probability of each device node is generated based on the number of nodes, the prior probability of each device node, and the observation probability of each device node. The target node is determined from multiple device nodes based on the posterior probability corresponding to each device node. Identify at least one target device associated with the target node from among multiple thermal power plant devices; The equipment area of ​​the target device in the thermal power plant is obtained, and the equipment area is used as the noise source area of ​​the thermal power plant.

[0011] Optionally, the posterior probability corresponding to each node is calculated using the following formula:

[0012] in, For the i-th device node, Let D be the j-th device node, and D be the observed data. For device nodes under observation data D conditions The corresponding posterior probability, For device nodes The corresponding prior probability, For device nodes The corresponding prior probability, For device nodes The observation probability that generates the observation data, where N is the number of nodes.

[0013] Secondly, embodiments of this application provide a storage medium that stores computer instructions, which, when executed by a computer, are used to perform the steps of the noise identification method for thermal power plants as described in the first aspect.

[0014] Thirdly, embodiments of this application provide an electronic device, including at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the noise identification method for thermal power plants as described in the first aspect.

[0015] Fourthly, embodiments of this application provide a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the noise identification method for thermal power plants as described in the first aspect.

[0016] In this embodiment, by extracting multiple acoustic feature indicators from the original sound signal in real time and combining them with a dynamic threshold range adaptive to the equipment for comprehensive anomaly judgment, the limitations of single feature analysis are overcome, significantly improving the accuracy and timeliness of noise detection and providing a reliable basis for early equipment fault warning. Furthermore, by automatically matching abnormal features with a predefined set of noise labels, a leap from single signal analysis to multi-source information collaborative judgment is achieved, thereby completing rapid classification of abnormal noise and intelligent identification of source equipment, greatly improving analysis efficiency and the system's adaptability to different complex operating conditions. Simultaneously, by constructing a correlation graph that integrates multi-source index information and target labels, and performing posterior probability analysis based on the graph's topology and node information, the likelihood of different equipment within the power plant acting as noise sources can be quantitatively assessed collaboratively. This represents an advancement from isolated point judgment to systemic reasoning across the entire plant, accurately locating noise source areas at the plant level. This provides intuitive and reliable data support for targeted maintenance and noise reduction optimization, effectively ensuring the safety and stability of power plant equipment operation. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the structure of a noise identification system provided in an embodiment of this application; Figure 2 This is a flowchart illustrating the steps of a noise identification method for a thermal power plant provided in an embodiment of this application. Figure 3 This is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0018] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0020] The noise identification method for thermal power plants provided in this application will be described in detail below with reference to the accompanying drawings, through specific embodiments and application scenarios.

[0021] Reference Figure 1 This is a schematic diagram of the structure of a noise identification system provided in an embodiment of this application. Specifically, it includes a central control module 1, a sound transmission module 2, a data acquisition and analysis module 3, a threshold judgment module 4, a tag definition module 5, a tag identification module 6, an event segmentation module 7, a source matching module 8, and a communication early warning module 9. The central control module is the master node that interfaces with each substation terminal, controlling and issuing commands to the various functional modules within the substation terminals. It should be noted that the substation terminals are typically deployed in a specific equipment area within a thermal power plant to monitor one or more thermal power plant devices within that area.

[0022] In the embodiments of this application, by means of... Figure 1 The noise identification system shown implements a noise identification method for thermal power plants. Specifically, a thermal power plant comprises multiple pieces of equipment. Since the operating noise of a thermal power plant is essentially the superposition and propagation of acoustic signals generated by the operation of its various internal equipment in space, any abnormal noise must originate from the abnormal state of one or more specific pieces of equipment. Therefore, the core of effective noise identification and source localization in thermal power plants lies in extracting and correlating the acoustic features of each piece of equipment constituting the power plant independently, thereby achieving decoupling from the mixed noise field and accurately locating the specific noise source area at the system level.

[0023] Reference Figure 2 This is a flowchart illustrating the steps of a noise identification method for a thermal power plant provided in an embodiment of this application, specifically including the following steps: Step 201: Acquire the original sound signals of multiple thermal power plant equipment under the current operating state and the historical acoustic characteristic data of multiple thermal power plant equipment under different operating conditions under normal operation. In this embodiment, the original sound signals are collected in real time in the equipment area of ​​each substation terminal of the thermal power plant mainly through the sound transmission module, and the historical acoustic feature data of the thermal power plant equipment is collected through the threshold judgment module. The original sound signal refers to the continuous time-domain sound pressure waveform radiated by each thermal power plant equipment during operation and directly recorded by the sound transmission module. The historical acoustic feature data is a long-term accumulated acoustic feature database that covers the equipment of the entire plant under various typical normal operating conditions. It can be understood as a set of acoustic parameters that can characterize the health status of each thermal power plant equipment extracted and saved from its original sound signals under different operating conditions (such as different loads and temperatures) in the past when it was confirmed to be operating normally.

[0024] In one embodiment of this application, the sound transmission module includes a sound transmission component, which can collect raw sound signals in real time at multiple points through several substation terminals. It should be noted that the sound transmission component has a sensitivity greater than 30mV / Pa at 250Hz, and its directional response supports 0° and 90° incident angles.

[0025] Step 202: Digitize each original sound signal to obtain multiple acoustic features corresponding to each thermal power plant device; In this embodiment, the main control module distributes the raw sound signals collected by the transmission module to the acquisition and analysis module, enabling the acquisition and analysis module to digitally process each raw sound signal to obtain multiple acoustic features. It can be understood that after the acquisition and analysis module processes the raw sound signals collected by the transmission module, multiple corresponding acoustic features can be obtained for each raw sound signal. These features together constitute a multi-dimensional digital description of the current acoustic state of the device.

[0026] Step 203: Determine the dynamic threshold range corresponding to each thermal power plant device based on the historical acoustic feature data, and determine the abnormal features among multiple acoustic features corresponding to each thermal power plant device based on the dynamic threshold range; the abnormal features have corresponding association attributes and index information; In this embodiment, a threshold determination module compares several acoustic features obtained from the acquisition and analysis module with a dynamic threshold range to identify abnormal features among multiple acoustic features. If the device is functioning normally, no abnormal features will be identified during this process.

[0027] Specifically, the dynamic threshold range is mainly based on the device's own historical normal data and is dynamically calculated or matched in combination with its operating conditions. Therefore, determining abnormal characteristics based on the dynamic threshold range can achieve accurate early warning that adapts to individual differences in equipment and changes in operating conditions, avoiding false alarms or missed alarms caused by fixed thresholds.

[0028] It should be noted that the dynamic threshold range is generated and applied independently for each piece of equipment in a thermal power plant, reflecting the individualization and refinement of monitoring. When the acoustic characteristics of a piece of equipment exceed its specific dynamic threshold range, that characteristic is marked as an anomalous feature. Each anomalous feature carries index information of its generation time, as well as correlation attributes inherited or analyzed from the original data (original sound signal). This information provides the basis for subsequent correlation analysis.

[0029] Step 204: Determine the target label corresponding to each anomaly feature based on the predefined set of noise labels and the associated attributes corresponding to each anomaly feature; In this embodiment, the label definition module pre-constructs a structured noise label knowledge base (noise label set) based on equipment type, process location, and typical fault modes. The label recognition module receives the abnormal features marked by the threshold judgment module and the predefined noise label set in the label definition module through the central control module. Then, by comparing the matching degree between the abnormal features and each noise label, it identifies one or more most likely target labels corresponding to each abnormal feature, thereby providing a preliminary qualitative assessment of the abnormal noise.

[0030] Step 205: Construct an association map to characterize the noise propagation and fault association structure of the thermal power plant based on the index information corresponding to multiple abnormal features and the target labels corresponding to multiple abnormal features; the nodes in the association map are associated with the equipment of the thermal power plant; In this embodiment, for each piece of equipment in a thermal power plant, the event decomposition module analyzes the index information corresponding to the abnormal features, aggregates the abnormal features, obtains aggregated information, and transmits it to the source matching module. The source matching module also combines the target tags identified by the tag recognition module to construct an association graph. The association graph is a graph model where nodes represent events formed by at least one piece of thermal power plant equipment, and edges represent physical connections, spatial adjacency relationships, or temporal and causal relationships between events. The association graph binds abstract acoustic anomalies to specific physical equipment locations and plant structures, thereby visually representing the noise propagation and potential fault association structure of the entire thermal power plant.

[0031] It should be noted that although the detection of abnormal features is carried out independently for each piece of equipment in a thermal power plant, the propagation of noise in a complex industrial environment is correlated. An anomaly in a single piece of equipment may originate from itself or may be conducted or triggered by adjacent equipment. The construction of the correlation map is precisely to break through the limitations of the perspective of a single piece of equipment and to place all the marked abnormal features and their labels in a unified network of physical and logical relationships of the thermal power plant for overall judgment.

[0032] Step 206: Perform posterior probability analysis on the nodes in the correlation graph to obtain the noise source region of the thermal power plant.

[0033] In this embodiment of the application, the source matching module will also perform posterior probability analysis on each node in the constructed association graph.

[0034] Specifically, posterior probability refers to the likelihood that each node representing a physical device in the graph is the true noise source, given that all current anomalous events (dynamic nodes in the graph) have been observed. By calculating the posterior probability of each node, the degree of anomalousness in different device areas can be quantitatively assessed. The system determines a physical spatial range based on the posterior probability and identifies it as the most likely noise source area, thereby achieving precise location of the noise source from a probabilistic statistical perspective.

[0035] In this embodiment, by extracting multiple acoustic feature indicators from the original sound signal in real time and combining them with a dynamic threshold range adaptive to the equipment for comprehensive anomaly judgment, the limitations of single feature analysis are overcome, significantly improving the accuracy and timeliness of noise detection and providing a reliable basis for early equipment fault warning. Furthermore, by automatically matching abnormal features with a predefined set of noise labels, a leap from single signal analysis to multi-source information collaborative judgment is achieved, thereby completing rapid classification of abnormal noise and intelligent identification of source equipment, greatly improving analysis efficiency and the system's adaptability to different complex operating conditions. Simultaneously, by constructing a correlation graph that integrates multi-source index information and target labels, and performing posterior probability analysis based on the graph's topology and node information, the likelihood of different equipment within the power plant acting as noise sources can be quantitatively assessed collaboratively. This represents an advancement from isolated point judgment to systemic reasoning across the entire plant, accurately locating noise source areas at the plant level. This provides intuitive and reliable data support for targeted maintenance and noise reduction optimization, effectively ensuring the safety and stability of power plant equipment operation.

[0036] In one embodiment of this application, the step of digitizing each original sound signal to obtain multiple acoustic features corresponding to each thermal power plant device includes: Each original audio signal is sequentially subjected to high-pass filtering and noise suppression to obtain a preprocessed signal; Each preprocessed signal is framed and windowed to obtain multiple windowed signal frames; Perform a short-time Fourier transform on each windowed signal frame to obtain the spectrum corresponding to each windowed signal frame, and extract the frequency domain energy distribution features from the spectra corresponding to multiple windowed signal frames; Calculate the Mel frequency cepstral coefficients and the first-order difference coefficients corresponding to the Mel frequency cepstral coefficients for each windowed signal frame based on the spectrum corresponding to each windowed signal frame. Calculate the sum of squares of the amplitudes of the sampling points within each windowed signal frame to obtain the short-time average energy corresponding to each windowed signal frame; Obtain the number of times the signal crosses the zero level within each windowed signal frame, and use the ratio between the number of crossings and the frame length of the windowed signal frame as the zero-crossing rate corresponding to each windowed signal frame; The preprocessed signal is subjected to envelope extraction processing to obtain waveform envelope features; The frequency domain energy distribution characteristics, the Mel frequency cepstral coefficients corresponding to all windowed signal frames, the first-order difference coefficients corresponding to all Mel frequency cepstral coefficients, the short-time average energy corresponding to all windowed signal frames, the zero-crossing rate corresponding to all windowed signal frames, and the waveform envelope characteristics are determined as multiple acoustic characteristics corresponding to each thermal power plant equipment.

[0037] In this embodiment, the acquisition and analysis module performs parallel digitization processing on the raw audio signals uploaded by multiple substation terminals to obtain multiple acoustic features corresponding to each raw audio signal. To enable those skilled in the art to better understand the feature extraction process of the acquisition and analysis module, a typical signal processing chain is used here to illustrate the digitization process for each raw audio signal.

[0038] First, each raw audio signal is sequentially subjected to high-pass filtering and noise suppression to obtain a preprocessed signal. High-pass filtering refers to using a digital filter to remove low-frequency components below a certain cutoff frequency (e.g., 50Hz) from the signal to eliminate the influence of low-frequency environmental noise (such as power grid interference) and highlight the mid-to-high frequency vibration noise of the equipment itself. Noise suppression refers to using methods such as spectral subtraction or adaptive filtering to suppress periodic or steady-state background noise in the signal and improve the signal-to-noise ratio.

[0039] Subsequently, each preprocessed signal needs to be framed and windowed. Each preprocessed signal is divided into continuous or partially overlapping short time intervals with a fixed duration (e.g., 20-40 milliseconds), and the signal of each short time interval is multiplied by a window function (e.g., a Hamming window) to obtain multiple windowed signal frames. Framed processing is used to approximate non-stationary signals as short-time stationary signals, while windowing is used to reduce spectral leakage caused by signal truncation.

[0040] The specific process for extracting each acoustic feature based on the preprocessed signal and multiple windowed signal frames is as follows: Frequency domain energy distribution characteristics: The spectrum of each windowed signal frame is obtained by performing a short-time Fourier transform. Then, the spectrum is statistically analyzed. For example, the energy proportion, spectral centroid, and spectral roll-off point of a specific frequency band (such as an octave band) are calculated. These statistics together constitute the frequency domain energy distribution characteristics, which are used to describe the distribution pattern of noise energy on the frequency axis.

[0041] Mel-frequency cepstral coefficients and their first-order difference coefficients: First, the spectrum of each windowed signal frame is passed through a set of Mel-scale filters to simulate the nonlinear perception of different frequencies by the human ear. Second, after taking the logarithm of the filter bank output, a discrete cosine transform is performed, and the first N coefficients are taken to obtain the MFCC (Mel-frequency cepstral coefficients) of that frame. The first-order difference coefficients are obtained by performing a first-order difference calculation along the time axis on the MFCC of multiple consecutive frames, and are used to describe the dynamic change trend of the MFCC.

[0042] Short-time average energy: The sum of the squares of the amplitudes of all sampling points within each windowed signal frame is directly calculated as the signal energy of that frame, reflecting the overall level of the signal amplitude. It can be used to detect sudden noise or energy changes.

[0043] Zero-crossing rate: The number of times the signal waveform crosses the zero level within each windowed signal frame is counted, and this number is divided by the duration of the signal frame (or the total number of sampling points) to obtain the short-time zero-crossing rate of the windowed signal frame. This feature is roughly related to the frequency components of the signal. Signals with more high-frequency components have a higher zero-crossing rate, which can be used to distinguish between broadband noise and pure tone.

[0044] Waveform envelope features: These are obtained by extracting the envelope of a preprocessed signal (a continuous signal without framing). For example, the amplitude envelope of the signal can be obtained by Hilbert transform or low-pass filtering after full-wave rectification. Statistical features (such as mean, variance, and peak factor) or time-domain features (i.e., waveform envelope features) can then be extracted from this envelope. These features are particularly sensitive to impulsive and modulated noise.

[0045] It should be noted that the acquisition and analysis module uses the above-mentioned digital processing procedure for each original sound signal to extract multiple acoustic features corresponding to each original sound signal.

[0046] As an example, to ensure that acoustic features with different physical dimensions and numerical ranges can be fairly compared and calculated within the same model, the acquisition and analysis module, after extracting multiple acoustic features for each original sound signal, needs to perform standardization and feature concatenation to generate a corresponding acoustic feature vector for each original sound signal. This acoustic feature vector is then transmitted to the threshold judgment module for further processing. Standardization refers to using methods such as Z-score normalization to transform the value of each feature to a distribution with a mean of 0 and a standard deviation of 1, eliminating biases caused by differences in dimensions and magnitudes. Feature concatenation refers to connecting all standardized acoustic features in a predetermined order into a one-dimensional high-dimensional vector, serving as a unified digital representation of the sound signal.

[0047] This application embodiment, through the above-described multi-level, multi-domain (time domain, frequency domain, cepstral domain) feature extraction process, can comprehensively and robustly extract key information (acoustic features) representing the device status from the original sound signal, providing a feature data foundation for subsequent anomaly detection and accurate identification.

[0048] In one embodiment of this application, the raw audio signal can be processed by high-pass filtering, noise suppression, framing, and windowing using the acquisition component in the acquisition and analysis module. As an example, the frequency range of the acquisition component is 20Hz to 8kHz, the lower limit of measurement is not higher than 23dB, and the upper limit of measurement is not lower than 131dB; the measurement parameters include instantaneous sound pressure level LP, equivalent continuous sound level Leq, cumulative percentage sound level LN, maximum sound level Lmax, minimum sound level Lmin, and standard deviation SD.

[0049] In one embodiment of this application, the historical acoustic feature data includes multiple historical features and historical operating condition parameters. The step of determining the dynamic threshold range corresponding to each thermal power plant device based on the historical acoustic feature data, and determining abnormal features among the multiple acoustic features corresponding to each thermal power plant device based on the dynamic threshold range, includes: Obtain the equipment type, spatial location, related processes, and real-time operating parameters of the thermal power plant equipment; Based on the aforementioned historical characteristics, calculate the distribution parameters of all thermal power plant equipment under normal operating conditions; Based on the historical operating parameters and the distribution parameters, a dynamic threshold range corresponding to thermal power plant equipment under different operating conditions is generated; The target operating condition of the thermal power plant equipment is determined based on its equipment type, spatial location, related processes, and real-time operating parameters. Acoustic features that are outside the dynamic threshold range corresponding to the target operating condition are identified as the abnormal features.

[0050] In this embodiment, the threshold determination module performs the above-described dynamic threshold generation and anomaly determination process.

[0051] First, during system initialization, the system operates based on a long-accumulated historical database (historical acoustic feature data). This database not only contains a large number of historical features collected and calculated by each device under various operating conditions (such as 30%, 50%, 100% load, and different temperature ranges), but also records the historical operating parameters of the device (such as load rate, medium temperature, pressure, etc.) when each historical feature was collected.

[0052] When real-time monitoring of a specific thermal power plant equipment is required, the system obtains the inherent attributes of the equipment (equipment type, spatial location, related processes) as well as its real-time operating status (real-time operating parameters, such as real-time operating load rate, temperature and vibration index).

[0053] The system filters historical features from the database that are of the same type as the equipment, located in the same or similar spatial positions, and belong to the same related processes. Based on these filtered historical features, which represent the long-term normal operation of the equipment, the system calculates the distribution parameters for each acoustic feature dimension (such as MFCC, short-term energy, etc.). The distribution parameters include the mean (representing the center of normal level), variance (representing the normal fluctuation range), and dynamic range (the upper and lower bounds of the normal value). It should be noted that all thermal power plant equipment refers to the collection of all similar equipment in the historical database that meets the current equipment classification criteria, and the statistical results represent the common health benchmark for this type of equipment.

[0054] The system further analyzes the correlation between historical operating condition parameters and distributed parameters. For example, through regression analysis or by creating lookup tables, it determines how the normal mean, variance, or dynamic range of each acoustic characteristic parameter should be adjusted accordingly under different operating condition parameters (such as load rate increasing from 50% to 100%). This generates a corresponding, quantifiable dynamic threshold range for each distinguishable combination of operating conditions (i.e., different operating conditions). It should be noted that this range is not a fixed value, but rather a function or a set of conditional values ​​that vary with the operating conditions.

[0055] Based on the real-time operating parameters of the current equipment, the system maps them to a predefined operating condition category to determine its current target operating condition. Then, it calls upon the set of dynamic threshold ranges that match this target operating condition. The acoustic features extracted in real-time from the current equipment are compared one by one with the corresponding dynamic threshold ranges. If the value of a certain acoustic feature falls outside its threshold range, the feature is determined to be an anomalous feature. Simultaneously, the system records the associated attributes (such as the operating condition at the time of triggering, the initial judgment of the anomalous type tendency) and index information (timestamp, device ID, acquisition terminal ID, etc.) for this anomalous feature.

[0056] Specifically, related attributes refer to a set of contextual information directly related to the physical event represented by the anomalous feature, used to describe its inherent characteristics and background, including but not limited to acoustic attributes, spatiotemporal attributes, and operating condition attributes. Index information refers to a set of key identifiers that can uniquely identify, quickly retrieve, and trace the data record of the anomalous feature, including but not limited to the timestamp, spatial location, and triggering conditions of the anomalous feature.

[0057] It should be noted that the dynamic threshold range is generated based on the historical normal data of the device (or similar devices), rather than a factory-wide uniform standard. It will be dynamically selected or fine-tuned according to the real-time operating parameters of the device. The normal noise level of the same device under high load and low load is different, and its judgment threshold range will also be different. Understandably, as the historical acoustic feature data is updated, the corresponding judgment threshold range will also be updated accordingly.

[0058] The embodiments of this application adaptively set dynamic threshold range based on historical big data and real-time operating conditions, which can effectively overcome the problems of high false alarm rate or missed alarm caused by traditional fixed threshold methods that ignore individual differences of equipment and changes in operating status. This significantly improves the accuracy and reliability of anomaly detection and lays a solid first-step judgment foundation for subsequent accurate fault location.

[0059] In one embodiment of this application, the noise tag set includes multiple noise tags, each noise tag having corresponding association attributes, including acoustic attributes, spatiotemporal attributes, and operating condition attributes. The step of determining the target tag corresponding to each anomalous feature based on a predefined noise tag set and the association attributes corresponding to each anomalous feature includes: Calculate the similarity between the acoustic attributes corresponding to each abnormal feature and the acoustic attributes corresponding to each noise label, and determine the noise labels with similarity higher than a preset similarity threshold as candidate labels corresponding to the abnormal features; The spatiotemporal and working condition attributes corresponding to each abnormal feature are matched and verified with the spatiotemporal and working condition attributes corresponding to each candidate label to obtain the verification result for each candidate label. Based on the verification results corresponding to each candidate label, at least one target label that meets the preset result is determined from the candidate labels corresponding to each abnormal feature.

[0060] In this embodiment, the label recognition module completes the above-mentioned intelligent matching process from abnormal features to target labels based on the data obtained by the label definition module and the threshold judgment module.

[0061] First, the label definition module pre-establishes a structured noise label knowledge base (noise label set), including multiple noise labels. Each noise label (such as high-frequency wear of fan bearings, steam leakage of valves) defines its corresponding associated attributes, which include, but are not limited to, acoustic attributes, spatiotemporal attributes, and operating condition attributes. Among them, acoustic attributes describe the typical acoustic feature vector of this type of noise, such as the typical mode of its MFCC coefficient, the main spectral peak positions, and the energy concentration frequency band; spatiotemporal attributes describe the typical spatial location (such as near the blower bearing housing, propagating downstream along the main steam pipeline) and temporal characteristics (such as continuous occurrence, intermittent occurrence with load rises and falls); operating condition attributes describe the typical equipment operating conditions that induce or aggravate this type of noise, such as significant occurrence under high load conditions, and easy occurrence when the cooling water temperature is below the threshold.

[0062] When the tag recognition module receives a tagged abnormal feature, it executes the following two-stage matching process (coarse screening + fine screening).

[0063] In the initial screening stage (coarse screening) based on acoustic attributes, the system calculates the similarity between the feature vector of the abnormal feature (i.e., the digital representation of its acoustic attributes) and the acoustic attribute vector of each noise tag in the knowledge base. The similarity calculation can be quantitatively evaluated using methods such as cosine similarity and Euclidean distance to obtain the quantitative evaluation result (similarity). A preset similarity is set in advance, and all noise tags with similarity higher than the threshold are selected as candidate tags for the abnormal feature.

[0064] In the fine screening stage based on multimodal attributes, for each candidate label, the system further verifies its spatiotemporal and operational attributes. Specifically, it needs to match and verify the spatiotemporal attributes corresponding to the abnormal features with those corresponding to each candidate label to complete the spatiotemporal attribute verification in the fine screening stage; simultaneously, it needs to match and verify the operational attributes corresponding to the abnormal features with those corresponding to each candidate label to complete the operational attribute verification in the fine screening stage.

[0065] Specifically, in the spatiotemporal attribute verification, the actual spatial location (from index information) and temporal pattern of the anomaly feature are checked to see if they match the typical spatiotemporal attributes defined by the candidate label. For example, an anomaly occurring in the boiler area will not be matched with a label whose typical spatiotemporal attribute is the low-pressure cylinder of the steam turbine. In the operating condition attribute verification, the real-time operating parameters at the time the anomaly feature is triggered are checked to see if they meet the operating condition triggering conditions defined by the candidate label. For example, an anomaly occurring under low-load steady-state conditions will not be matched with a label whose operating condition attribute is that it only occurs during high-load transients. This fine-tuning verification will produce a comprehensive verification result (e.g., complete match, partial match, no match).

[0066] The system determines one or more target labels from the candidate labels based on preset decision rules (preset results). For example, the rule could be to select the candidate label with the highest acoustic similarity and a perfect match, or to select all candidate labels with a confidence level higher than the verification result.

[0067] In this embodiment, the search range is quickly narrowed down by using computationally efficient acoustic similarity to avoid complex multi-attribute full comparison of all labels. Based on acoustic similarity, two strong constraints, spatiotemporal and working conditions, are introduced for verification, which greatly eliminates the possibility of mismatch and makes the recognition results more reliable.

[0068] It should be noted that the two-stage matching (coarse screening + fine screening) described above is a complete process executed independently for each anomalous feature. The label recognition module will execute the above acoustic initial screening and multimodal attribute verification process one by one for each anomalous feature reported by the threshold judgment module to complete the label recognition and target label determination for all anomalous features.

[0069] This application embodiment, through the two-stage intelligent matching method that integrates acoustic, spatiotemporal, and operational multimodal attributes, not only achieves rapid and automatic classification of abnormal noise, but also ensures that the classification results are highly consistent with the actual operating scenario of the equipment, thereby providing accurate fault type guidance for subsequent precise location of noise sources.

[0070] In one embodiment of this application, before performing the aforementioned matching, each anomalous feature can undergo feature enhancement processing to generate a multidimensional feature vector containing richer time-frequency domain characteristics. This can be achieved by methods such as calculating the difference coefficients of the original features, combining statistics at different time scales, or introducing wavelet transform coefficients, thereby enhancing the feature's ability to characterize noise transient characteristics and modulation patterns. Label matching can then be performed based on this enhanced multidimensional feature vector, further improving the discriminative power and matching accuracy of acoustic attribute comparisons. This makes the initially selected candidate label set closer to reality, laying a more reliable foundation for the subsequent fine-tuning stage.

[0071] In one embodiment of this application, the spatiotemporal features of the matched target tags are integrated, and the identified target tags and their corresponding associated attributes are reported to the central control module.

[0072] This application embodiment significantly improves the accuracy and stability of anomaly detection by fusing multimodal acoustics and operating condition features and dynamically updating thresholds. It further reduces false alarm and missed alarm rates by using tag matching and spatiotemporal operating condition joint screening. The graph model-based localization method can accurately identify noise source areas. The entire process is automated for monitoring, alarming, and recording, which greatly reduces the cost of manual inspection and improves response efficiency.

[0073] In one embodiment of this application, the step of constructing an association map characterizing the noise propagation and fault association structure of the thermal power plant based on index information corresponding to multiple abnormal features and target labels corresponding to multiple abnormal features includes: Obtain the physical layout and acoustic environment of the thermal power plant; Based on the index information of all abnormal features, multiple abnormal features are clustered to obtain at least one spatiotemporal triggering event; Based on the physical layout and acoustic environment of the power plant, a graph structure containing one or more nodes is constructed, wherein the nodes are equipment nodes, pipeline nodes, and acoustic path nodes. The target labels corresponding to the abnormal features in each spatiotemporal trigger event and each spatiotemporal triggering event are mapped to the nodes in the graph structure, and the connection edges between different nodes in the graph structure are established according to the index information corresponding to the abnormal features, so as to construct the association graph.

[0074] In this embodiment, the event splitting module and the source matching module work together to complete the above-mentioned process of constructing a system-level correlation graph from discrete anomaly features.

[0075] The system first acquires the physical layout and acoustic environment information of the power plant, including but not limited to equipment layout drawings, process piping diagrams, building structure models, and sound wave propagation attenuation characteristic data in different media and structures obtained through acoustic simulation or on-site measurements. Then, based on the power plant's physical layout and acoustic environment, an initial, static graph structure is constructed. Specifically, equipment nodes represent specific physical equipment (such as pumps, fans, and valves), with node attributes including equipment ID, type, and three-dimensional coordinates; piping nodes represent pipes, ducts, or cable trays connecting the equipment, with node attributes identifying their respective process system and fluid medium; acoustic path nodes represent critical path points, reflecting surfaces, or virtual monitoring sections for sound wave propagation, with attributes including acoustic transfer function or attenuation coefficient.

[0076] The event segmentation module receives abnormal features and their index information (mainly timestamps and precise spatial coordinates) from all monitoring points across the plant. Based on a predefined clustering algorithm (e.g., the DBSCAN algorithm based on time and spatial density), it aggregates multiple abnormal features that are temporally consecutive or close, spatially adjacent, or related through propagation paths into a higher-level logical unit, called a spatiotemporally triggered event. This transforms a massive number of discrete alarm points into a few meaningful event objects. For example, within five minutes, abnormal features triggered successively by multiple sensors near the #1 blower and its inlet / outlet pipes may be aggregated into a single #1 blower area abnormal event.

[0077] The source matching module associates each spatiotemporal triggering event and target label as a dynamic attribute with the static graph skeleton (graph structure) mentioned above, so that it is associated with one or more key physical nodes.

[0078] Specifically, the source matching module associates the core region of the spatiotemporal triggering event with one or more key physical nodes in the static graph skeleton corresponding to that region (usually the closest or most acoustically associated device node or acoustic path node). Simultaneously, the target label corresponding to the abnormal features contained in the event is recorded as the event's temporal attribute in the attributes of the associated physical nodes, thus mapping the spatiotemporal triggering event and its corresponding target label to the corresponding nodes in the graph structure.

[0079] Subsequently, connections with specific attributes are established or enhanced between physical nodes in the static graph skeleton, or between physical nodes and physical nodes already bound to other events. For example, when establishing an edge between two physical nodes, the edge attribute is recorded as time-high synchronization, determined by the fact that the events bound to these two nodes have synchronized timestamps; this edge is a time-related edge. When establishing or updating edges between physical nodes, the edge weight or attribute is enhanced based on the spatiotemporal proximity of the bound events to represent potential noise propagation paths; this edge is a spatial propagation edge. When establishing an edge between a device node and a pipeline or acoustic path node, the edge attribute may be recorded as being caused by a fault represented by a certain label, based on the fact that the events bound to the device node contain this fault label; this edge is a causal representation edge.

[0080] It is understandable that mapping spatiotemporal triggering events to each physical node of a graph structure can establish the association between events and physical locations. Mapping target labels to each physical node of a graph structure can enable the label information to be mounted as node attributes. By establishing edges with attributes, edge attributes can be used to express complex relationships such as time, space, and causality.

[0081] Through the above node mapping and edge establishment, the resulting association graph is no longer a simple equipment layout diagram, but a graph model that integrates static factory topology, dynamic abnormal events, diagnostic knowledge tags, and rich spatiotemporal and logical relationships between them.

[0082] This application embodiment constructs an association graph to systematically integrate isolated abnormal alarms across the entire plant and multiple monitoring points into a unified, semantically rich relationship network. This provides a crucial structured data foundation for subsequent noise source probability analysis based on graph model reasoning, enabling the noise source location analysis to comprehensively consider equipment connection relationships, sound wave propagation paths, and the synergistic effects of multiple abnormal events.

[0083] In one embodiment of this application, the step of performing posterior probability analysis on the nodes in the correlation graph to obtain the noise source region of the thermal power plant includes: Obtain the number of device nodes, the prior probability of each device node, and the observation probability of each device node generating observation data; The posterior probability of each device node is generated based on the number of nodes, the prior probability of each device node, and the observation probability of each device node. The target node is determined from multiple device nodes based on the posterior probability corresponding to each device node. Identify at least one target device associated with the target node from among multiple thermal power plant devices; The equipment area of ​​the target device in the thermal power plant is obtained, and the equipment area is used as the noise source area of ​​the thermal power plant.

[0084] The source matching module performs posterior probability analysis based on the constructed correlation graph to ultimately locate the noise source region. Nodes in the correlation graph represent physical entities such as equipment nodes, pipeline nodes, or acoustic path nodes, with equipment nodes serving as the primary noise source candidates. Analysis of the correlation graph directly yields the total number of all equipment nodes (node ​​count), the prior probability of each equipment node, and the observation probability of each equipment node generating observation data.

[0085] The prior probability for each device node is an initial probability assigned to each device node in the graph before the analysis begins. It is an initial estimate of the likelihood that the device will become a noise source based on historical experience, and can be set based on historical failure statistics, device importance, or operating time. For example, the prior probability of older or critical equipment can be set slightly higher than that of newer or auxiliary equipment, reflecting the difference in their prior failure probabilities. If there is no special prior knowledge, a uniform prior probability can also be set for all device nodes.

[0086] Specifically, the scoring is based on the equipment's past fault records and the frequency of noise events; the scoring is based on the equipment's current operating load, pressure, and other operating parameters; and the scoring is based on static attributes such as equipment type, service life, structural characteristics, and maintenance status. The three types of scores are then weighted and summed to obtain the prior probability corresponding to each equipment node.

[0087] The observation probability of each device node generating observation data is also called the likelihood probability. It represents the likelihood that, assuming the node is a noise source, the observed data (all anomalous features and their target labels) in the current graph will actually appear. The calculation depth of this probability depends on the topology of the association graph. Specifically, for a device node, its observation probability depends on the connection strength (edge ​​weights) between all anomalous features in the graph and that device node, as well as the characteristics of the events. For example, if multiple anomalous features are closely clustered in space and time (determined based on index information) and point to a certain device node, and the target labels corresponding to the anomalous features match the common failure modes of that device, then the observation probability of that device node will be very high.

[0088] Pipeline nodes or acoustic path nodes are not usually directly considered as noise source candidates; their observation probability may be set to a specific value or influenced by the propagation model in the calculation of equipment nodes.

[0089] Specifically, the sound signature of the current noise is compared with the known typical noise characteristics of the device to obtain the sound feature matching coefficient; the time point of the current noise occurrence is checked to see if it matches the typical time period or operating stage of the device that generates noise to obtain the time pattern matching coefficient; the noise is simulated to propagate from the device location to the actual location captured by the microphone along the preset acoustic propagation path in the spectrum to obtain the spatial propagation matching coefficient; the results of the above three types of matching coefficients are multiplied to obtain the observation probability corresponding to each device node.

[0090] The system calculates the posterior probability of each device node as a noise source based on the number of nodes, prior probability, and observation probability obtained above. The posterior probability combines prior knowledge and current observation evidence to quantify the degree of anomaly of each node.

[0091] The system sorts all nodes based on the calculated posterior probabilities and identifies the node with the highest posterior probability as the target node, representing the most likely noise source. Since there is a mapping relationship between device nodes and physical devices in the correlation graph, the system uses this mapping relationship to find one or more physical devices corresponding to the target node and identifies them as target devices. Finally, the system obtains the three-dimensional spatial coordinate range (equipment area) of these target devices in the power plant from the plant's digital model and outputs this area as the noise source area identified in this study.

[0092] It should be noted that the entire posterior probability analysis process is carried out within the relationships constructed by the correlation graph. The edges (connections) of the graph are key to calculating the probability of observation, enabling the analysis to comprehensively consider the synergistic evidence from multiple events and locations, rather than viewing each outlier in isolation.

[0093] It should also be noted that anomalous features and target labels themselves do not participate in probability calculations as independent nodes, but rather as observational data that influences the observation probabilities of associated device nodes. The entire analysis process is conducted on the physical relationship network of the correlation graph, realizing reasoning from discrete evidence to source probabilities.

[0094] In this embodiment, the goal is to locate the source equipment generating anomalous noise. Although the correlation map fully models the physical structure of the power plant (including equipment, pipelines, and acoustic paths), the pipelines and acoustic paths are essentially the propagation mediums or channels of noise, not the source itself. Therefore, in the posterior probability analysis, this embodiment sets the equipment nodes that are potential noise sources as candidate hypotheses and calculates the posterior probability that each equipment node is a noise source. The role of pipeline nodes and acoustic path nodes is mainly reflected in the calculation of the observation probability of equipment nodes. As key transmission networks, they determine the path and attenuation relationship of anomalous features that can propagate from the source equipment to the monitoring point, thus affecting the likelihood (i.e., the probability of observing current evidence) when assuming that a certain equipment is a noise source. In short, pipeline and path nodes are not candidate nodes for the noise source region, but they determine the probability calculation in the process of determining the noise source region.

[0095] The embodiments of this application are based on the posterior probability analysis method of the correlation graph, which deeply integrates prior experience, real-time observation and system topology. This enables the quantitative inference of the most likely noise source device and its location from multi-source and heterogeneous alarm information. This location result is no longer a simple list of alarm points, but a precise spatial area indication with probability confidence, which greatly improves the efficiency and pertinence of maintenance personnel in troubleshooting.

[0096] In one embodiment of this application, the posterior probability corresponding to each node is calculated using the following formula:

[0097] in, For the i-th device node, Let D be the j-th device node, and D be the observed data. For device nodes under observation data D conditions The corresponding posterior probability, For device nodes The corresponding prior probability, For device nodes The corresponding prior probability, For device nodes The observation probability that generates the observation data, where N is the number of nodes.

[0098] In this embodiment, the system calculates the posterior probability of each node as a noise source using Bayes' theorem based on the number of nodes, prior probability, and observation probability obtained above. The specific formula is as follows:

[0099] in, For the i-th device node, Let D be the j-th device node, and D be the observed data. For device nodes under observation data D conditions The corresponding posterior probability, For device nodes The corresponding prior probability, For device nodes The corresponding prior probability, For device nodes The probability of generating observation data, where N is the number of nodes.

[0100] It should be noted that the device node This can be understood as the target node, that is, the specific node where the posterior probability needs to be calculated, or the device node. It can be understood as any one of the device nodes (candidate noise source node), that is, a temporary variable used to traverse all device nodes when calculating the posterior probability corresponding to the target node.

[0101] This application embodiment organically integrates the prior state of the device with the physical model of acoustic propagation through a Bayesian framework. This not only reflects the dynamic correlation between the spatial propagation characteristics of noise and the operating conditions of the device, but also eliminates the influence of environmental interference through normalization processing, thereby achieving a balance between the robustness of noise source localization and computational efficiency in high-noise aliasing scenarios.

[0102] In one embodiment of this application, the communication early warning module is used to report the original sound data, acoustic features, threshold judgment results (abnormal features) and identification output information (noise source area) to the central control module. When the threshold judgment module determines that there is an abnormality, it triggers the communication early warning module to synchronously start recording and transmit the recording file.

[0103] Traditional methods often employ single energy, zero-crossing rate, or spectral analysis, lacking in-depth exploration of multi-dimensional acoustic characteristics and their combined distribution. This makes it difficult to distinguish between similar noise sources and fails to unify the modeling of multi-modal information such as acoustics, spatiotemporal factors, and operating conditions, thus limiting identification accuracy. Traditional single-point or multi-point detection often stops at abnormal alarms, without combining topological structures such as pipelines, equipment, and acoustic paths for graphical model analysis, making it impossible to provide accurate noise source regions.

[0104] This application's embodiments, through deep learning analysis of raw sound signals, can extract and process various acoustic features in real time. Combined with a dynamic adaptive threshold judgment mechanism, it significantly improves the detection accuracy and sensitivity of abnormal noise, helping to identify equipment faults in a timely manner and reduce potential safety risks. Furthermore, by introducing an automated noise tag extraction and recognition mechanism, abnormal features are classified and filtered, enabling rapid matching with related noise sources. This intelligent processing not only improves the efficiency of noise identification but also enhances the system's adaptability, making it applicable to noise analysis of different types of equipment and operating conditions. Simultaneously, by constructing a graph structure of equipment, pipelines, and acoustic paths, abnormal features can be divided into independent spatiotemporal triggering events. Based on posterior probability analysis, noise sources can be accurately located, providing reliable data support for subsequent maintenance and optimization, and promoting the safety and stability of equipment operation.

[0105] It should be noted that, for the sake of simplicity, the method embodiments are all described as a series of actions. However, those skilled in the art should understand that the embodiments of this application are not limited to the described order of actions, because according to the embodiments of this application, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also understand that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily required by the embodiments of this application.

[0106] This application also provides a storage medium that stores computer instructions. When the computer executes the computer instructions, it is used to perform the various processes of the above-described noise identification method embodiment for thermal power plants and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0107] The processor is the processor in the electronic device described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0108] This application also provides an electronic device, including a processor 3010, a memory 309, and a program or instructions stored in the memory 309 and executable on the processor 3010. When the program or instructions are executed by the processor 3010, they implement the various processes of the above-described noise identification method embodiment for thermal power plants and achieve the same technical effect. To avoid repetition, they will not be described again here.

[0109] It should be noted that the electronic devices in the embodiments of this application include the mobile electronic devices and non-mobile electronic devices described above.

[0110] Figure 3 A schematic diagram of the hardware structure of an electronic device to implement an embodiment of this application.

[0111] The electronic device 300 includes, but is not limited to, components such as: radio frequency unit 301, network module 302, audio output unit 303, input unit 304, sensor 305, display unit 306, user input unit 307, interface unit 308, memory 309, and processor 3010.

[0112] Those skilled in the art will understand that the electronic device 300 may also include a power supply (such as a battery) for supplying power to various components. The power supply may be logically connected to the processor 3010 through a power management system, thereby enabling functions such as managing charging, discharging, and power consumption through the power management system. Figure 3 The electronic device structure shown does not constitute a limitation on the electronic device. The electronic device may include more or fewer components than shown, or combine certain components, or have different component arrangements, which will not be elaborated here.

[0113] This application also provides a computer program product, including a computer program / instruction. When the computer program / instruction is executed by a processor, it implements the various processes of the above-described noise identification method embodiment for thermal power plants and achieves the same technical effect. To avoid repetition, it will not be described again here.

[0114] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0115] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0116] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A noise identification method for a thermal power plant, characterized in that, The thermal power plant includes multiple thermal power plant equipment, and the method includes: Acquire raw sound signals of multiple thermal power plant equipment under current operating conditions and historical acoustic characteristic data of multiple thermal power plant equipment under different operating conditions under normal operating conditions; Each original sound signal is digitally processed to obtain multiple acoustic characteristics corresponding to each thermal power plant device; The dynamic threshold range corresponding to each thermal power plant device is determined based on the historical acoustic feature data, and abnormal features are identified among multiple acoustic features corresponding to each thermal power plant device based on the dynamic threshold range; the abnormal features have corresponding association attributes and index information; The target label for each anomaly feature is determined based on a predefined set of noise labels and the associated attributes corresponding to each anomaly feature. An association map is constructed based on the index information corresponding to multiple anomaly features and the target labels corresponding to multiple anomaly features to characterize the noise propagation and fault association structure of the thermal power plant; the nodes in the association map are associated with the equipment of the thermal power plant. By performing posterior probability analysis on the nodes in the correlation graph, the noise source region of the thermal power plant is obtained.

2. The method according to claim 1, characterized in that, The process of digitizing each original sound signal yields multiple acoustic features corresponding to each piece of equipment in a thermal power plant, including: Each original audio signal is sequentially subjected to high-pass filtering and noise suppression to obtain a preprocessed signal; Each preprocessed signal is framed and windowed to obtain multiple windowed signal frames; Perform a short-time Fourier transform on each windowed signal frame to obtain the spectrum corresponding to each windowed signal frame, and extract the frequency domain energy distribution features from the spectra corresponding to multiple windowed signal frames; Calculate the Mel frequency cepstral coefficients and the first-order difference coefficients corresponding to the Mel frequency cepstral coefficients for each windowed signal frame based on the spectrum corresponding to each windowed signal frame. Calculate the sum of squares of the amplitudes of the sampling points within each windowed signal frame to obtain the short-time average energy corresponding to each windowed signal frame; Obtain the number of times the signal crosses the zero level within each windowed signal frame, and use the ratio between the number of crossings and the frame length of the windowed signal frame as the zero-crossing rate corresponding to each windowed signal frame; The preprocessed signal is subjected to envelope extraction processing to obtain waveform envelope features; The frequency domain energy distribution characteristics, the Mel frequency cepstral coefficients corresponding to all windowed signal frames, the first-order difference coefficients corresponding to all Mel frequency cepstral coefficients, the short-time average energy corresponding to all windowed signal frames, the zero-crossing rate corresponding to all windowed signal frames, and the waveform envelope characteristics are determined as multiple acoustic characteristics corresponding to each thermal power plant equipment.

3. The method according to claim 1, characterized in that, The historical acoustic feature data includes multiple historical features and historical operating condition parameters. The step of determining the dynamic threshold range corresponding to each thermal power plant device based on the historical acoustic feature data, and determining abnormal features among the multiple acoustic features corresponding to each thermal power plant device based on the dynamic threshold range, includes: Obtain the equipment type, spatial location, related processes, and real-time operating parameters of the thermal power plant equipment; Based on the aforementioned historical characteristics, calculate the distribution parameters of all thermal power plant equipment under normal operating conditions; Based on the historical operating parameters and the distribution parameters, a dynamic threshold range corresponding to thermal power plant equipment under different operating conditions is generated; The target operating condition of the thermal power plant equipment is determined based on its equipment type, spatial location, related processes, and real-time operating parameters. Acoustic features that are outside the dynamic threshold range corresponding to the target operating condition are identified as the abnormal features.

4. The method according to claim 1, characterized in that, The noise tag set includes multiple noise tags, each with corresponding association attributes, including acoustic attributes, spatiotemporal attributes, and operating condition attributes. Determining the target tag corresponding to each anomaly feature based on the predefined noise tag set and the association attributes corresponding to each anomaly feature includes: Calculate the similarity between the acoustic attributes corresponding to each abnormal feature and the acoustic attributes corresponding to each noise label, and determine the noise labels with similarity higher than a preset similarity threshold as candidate labels corresponding to the abnormal features; The spatiotemporal and working condition attributes corresponding to each abnormal feature are matched and verified with the spatiotemporal and working condition attributes corresponding to each candidate label to obtain the verification result for each candidate label. Based on the verification results corresponding to each candidate label, at least one target label that meets the preset result is determined from the candidate labels corresponding to each abnormal feature.

5. The method according to claim 1, characterized in that, The construction of an association map characterizing the noise propagation and fault association structure of the thermal power plant based on index information corresponding to multiple abnormal features and target labels corresponding to multiple abnormal features includes: Obtain the physical layout and acoustic environment of the thermal power plant; Based on the index information of all abnormal features, multiple abnormal features are clustered to obtain at least one spatiotemporal triggering event; Based on the physical layout and acoustic environment of the power plant, construct a graph structure containing one or more nodes, where the nodes are equipment nodes, pipeline nodes, or acoustic path nodes. The target labels corresponding to the abnormal features in each spatiotemporal trigger event and each spatiotemporal triggering event are mapped to the nodes in the graph structure, and the connection edges between different nodes in the graph structure are established according to the index information corresponding to the abnormal features, so as to construct the association graph.

6. The method according to claim 5, characterized in that, The step of performing posterior probability analysis on the nodes in the correlation graph to obtain the noise source region of the thermal power plant includes: Obtain the number of device nodes, the prior probability of each device node, and the observation probability of each device node generating observation data; The posterior probability of each device node is generated based on the number of nodes, the prior probability of each device node, and the observation probability of each device node. The target node is determined from multiple device nodes based on the posterior probability corresponding to each device node. Identify at least one target device associated with the target node from among multiple thermal power plant devices; The equipment area of ​​the target device in the thermal power plant is obtained, and the equipment area is used as the noise source area of ​​the thermal power plant.

7. The method according to claim 6, characterized in that, The posterior probability corresponding to each node is calculated using the following formula: in, For the i-th device node, Let D be the j-th device node, and D be the observed data. For device nodes under observation data D conditions The corresponding posterior probability, For device nodes The corresponding prior probability, For device nodes The corresponding prior probability, For device nodes The observation probability that generates the observation data, where N is the number of nodes.

8. A storage medium, characterized in that, The storage medium stores computer instructions, which, when executed by the computer, are used to perform a noise identification method for a thermal power plant as described in any one of claims 1-7.

9. An electronic device, characterized in that, Includes at least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform a noise identification method for a thermal power plant as described in any one of claims 1-7.

10. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instruction is executed by the processor, it implements a noise identification method for a thermal power plant as described in any one of claims 1-7.

Images