CN-121980368-A - Low-voltage power supply environment noise tracing edge calculation method and system

CN121980368ACN 121980368 ACN121980368 ACN 121980368ACN-121980368-A

Abstract

The invention discloses a low-voltage power supply environment noise tracing edge calculation method and a system, wherein the method comprises the steps of synchronously collecting data through a multi-mode sensing module, constructing model prediction historical noise space-time distribution, dividing multiple types of noise events by combining statistical anomaly judgment and environment noise limit value, dynamically scheduling a local lightweight model and a cloud high-precision model according to event priority, and completing tracing analysis; the system comprises an edge device layer and a cloud server layer, and resource optimization configuration is achieved through a local and cloud collaborative architecture. According to the invention, through a dual-mode power supply architecture, multi-dimensional dynamic analysis and an intelligent computational power scheduling mechanism, the detection rate of the hidden noise event is remarkably improved, the cloud load is reduced, the accurate classification of five kinds of noise such as nature and traffic and fine granularity subclasses is supported, and a high-efficiency, low-power consumption and high-precision technical scheme is provided for environmental noise treatment.

Inventors

ZHU YUN
HE HAOWEN
YANG WENWEI
LI KUNJIE
CAI JINHUI
ZHAN YIHAO
LI ZEYU
Chen Qianer

Assignees

华南理工大学
华云创信(广东)生态环境科技有限公司

Dates

Publication Date: 20260505
Application Date: 20251231

Claims (10)

1. The low-voltage power supply environment noise tracing edge calculation method is characterized by comprising the following steps of: S1, collecting multi-mode noise data; S2, preprocessing the acquired multi-mode noise data to acquire noise time sequence characteristics; s3, inputting the noise time sequence characteristics into a trained historical noise distribution prediction model to predict historical noise distribution; s4, classifying the dynamic noise event based on the predicted historical noise distribution; And S5, performing intelligent power calculation scheduling traceability on the divided noise events.
2. The low-voltage power supply environment noise traceability edge computing method is characterized by comprising the steps of carrying out data cleaning on original data, deleting samples with the deletion rate exceeding a set threshold, carrying out interpolation complementation on residual data, filling the deletion value by adopting a mean value interpolation method, carrying out normalization processing on continuous data after interpolation, carrying out supervised learning format conversion on the normalized data, constructing normalized time sequence data into input and output pairs usable for supervised learning by using a time sequence sliding window method, forming an input feature matrix containing historical time sequence information by creating a hysteresis term of a specific time step for each feature, and taking an actual observed value at a corresponding moment as an output target, thereby converting the continuous time sequence data into supervised learning data with a clear feature-tag corresponding relation.
3. The method for calculating the tracing edge of low voltage power supply environmental noise according to claim 1, wherein the step of training the historical noise distribution prediction model comprises the steps of: preprocessing the acquired multi-modal noise data; constructing a multi-layer long-short-term memory network LSTM architecture, comprising an input layer, a multi-level stacked LSTM hidden layer and a fully connected output layer, and realizing the dynamic capture of time sequence characteristics through a gating mechanism; establishing a search space containing learning rate, batch size and LSTM stacking layer number parameters, constructing a probability model by adopting a Bayesian optimization algorithm, and carrying out parameter optimization by taking a decision coefficient as an optimization objective function; model training and optimization, namely performing model iterative training in a super-parameter search space, and outputting a historical noise distribution prediction model with strong generalization capability through a cross verification strategy; Historical noise distribution interval calculation based on model prediction result and 3 In principle, a dynamic confidence interval updated in real time is constructed for characterizing the reasonable fluctuation range of the noise value, Is the residual standard deviation.
4. The method for calculating the tracing edge of low voltage power supply environmental noise according to claim 3, wherein the historical noise distribution interval is calculated as: residual sequence extraction, namely calculating the difference value between a real value sequence and a predicted value sequence of a historical noise distribution prediction model on a test set, and constructing a residual sequence; residual statistical analysis, calculating residual standard deviation based on residual sequence ; Real-time prediction, namely inputting the noise time sequence characteristic data of the current period to a historical noise distribution prediction model, and outputting a noise reference value of the current period ; Historical noise distribution interval generation based on current noise reference value And 3 In principle, a dynamic confidence interval is constructed and used for representing the reasonable fluctuation range of the noise value in the current period and providing a data reference for the statistic anomaly judgment of the subsequent dynamic noise event classification.
5. The method for computing the tracing edge of low voltage power supply environmental noise according to claim 4, wherein classifying the dynamic noise event comprises the steps of: step 41, historical noise distribution prediction: according to the historical noise distribution interval, a dynamic confidence interval reflecting the historical noise distribution of the current period is constructed; step 42, statistical anomaly determination: by comparing current noise monitoring values And the historical noise distribution interval, the statistical characteristics of noise events are judged, and three-level statistical abnormal weights are defined to reflect the current noise monitoring value The degree of deviation from the historical noise distribution interval is defined as: abnormally high weights: The weight is 4, which indicates that the current noise value significantly exceeds the historical noise distribution; Abnormally low weights: the weight is 3, which indicates that the current noise value is obviously lower than the historical noise distribution; Conventional weights: The weight is 1, which indicates that the current noise value accords with the historical noise distribution; Step 43, environmental noise limit value judgment: ambient noise limit defined in accordance with acoustic ambient quality standards Determining compliance of noise events and defining limit decision weights to reflect current noise values Whether the corresponding environmental noise standard is met is defined as: exceeding the standard weight: The weight is 2, which indicates that the current noise value exceeds the standard and needs to be treated in time; Weight up to standard: the weight is 1, which indicates that the current noise value reaches the standard and accords with the standard; Step 43, event joint classification: By combining the dual condition of statistical anomaly determination and ambient noise limit determination, combining the product of the two weights to dynamically determine the event processing priority and classifying the noise event as: an abnormally high superscalar event T1 with a weight product of 4 x 2 = 8, The method is characterized in that the current noise value is obviously higher than the historical noise distribution and the sound pressure level exceeds the standard, and belongs to a high-priority event; an abnormally low superscalar event T2 with a weight product of 3 x 2 = 6, The method is characterized in that the current noise value is obviously lower than the historical noise distribution but the sound pressure level is still out of standard, and belongs to a secondary priority event; abnormally high up-to-standard events T3 with a weight product of 4 x 1 = 4, The noise value is characterized as being obviously higher than the historical noise distribution but not exceeding the standard, and belongs to potential risk events needing further verification; abnormally low up-to-standard events T4 with a weight product of 3 x 1 = 3, The noise value is obviously lower than the historical noise distribution and reaches the standard, and the noise value is characterized as an abnormal event needing further verification; conventional superscalar event T5-weight product 1 x 2 = 2, And is also provided with The noise value accords with the historical noise distribution but continuously exceeds the standard, and belongs to a conventional exceeding event; conventional qualifying event T6 weight product 1 x1 = 1, And is also provided with The noise value accords with the historical noise distribution and meets the standard, and the noise value is characterized as belonging to a normal event.
6. The method for calculating the noise tracing edge of the low-voltage power supply environment according to claim 5, wherein step S5 is based on classification of noise events, and a double-end cooperative strategy driven by the classification result of the noise events is constructed, and according to the event characteristics, calculation resources are dynamically mobilized to execute differentiation processing, and the specific strategy is as follows: the ultra-high standard exceeding event T1 has the dual characteristics that the noise value is obviously higher than the historical noise distribution and the sound pressure level exceeds the standard, and is preferentially uploaded to a cloud server for ensuring the tracing accuracy, and a cloud high-precision model is used for noise source classification and tracing analysis; the noise value is obviously lower than the historical noise distribution, but the sound pressure level is still out of standard, and the abnormal low out-of-standard event T2 belongs to a secondary priority event, wherein the local light-weight model is firstly used for primary classification, and if the confidence coefficient of the local model is lower than a threshold value, the local model is uploaded to the cloud end in an idle period to be used for secondary verification by using a high-precision model; The noise value is obviously higher than the historical noise distribution but is not out of standard, and belongs to a potential risk event needing further verification, trend analysis is carried out through a local model, and if a continuous rising trend is detected or sporadic noise source characteristics exist, a cloud model is triggered to carry out traceability analysis; The abnormal low up-to-standard event T4 is characterized in that the noise value is obviously lower than the historical noise distribution and reaches the standard, the abnormal event is to be further verified, the context data is combined for comprehensive judgment, if the abnormal event is judged, the abnormal event is locally archived, and if the abnormal event is judged to be potential abnormality, the local model is used for tracing analysis; the noise value accords with the historical noise distribution but continuously exceeds the standard, and is usually interpretable conventional noise, and the noise source classification and recording are directly completed by a local lightweight model; and the conventional up-to-standard event T6 is that the noise value accords with the historical noise distribution and reaches the standard, belongs to a normal event, does not need real-time analysis and processing, and is only uploaded to a cloud archive backup in a system idle period for long-term trend analysis.
7. The low-voltage power supply environment noise tracing edge calculation method according to claim 6 is characterized in that tracing analysis results are stored in a database in real time, a visual interface is constructed through Vue3+ Fastapi, and key information of noise source duty ratio and sound pressure level distribution is dynamically displayed.
8. An apparatus for implementing the low-voltage power supply environmental noise tracing edge computing method according to any one of claims 1 to 7, comprising: The multi-mode sensing module integrates a high-precision digital omnidirectional microphone, a sound pressure level sensor and a multi-parameter weather sensor, realizes millisecond-level space-time synchronous acquisition of sound pressure level data, weather parameters and environmental audio through a time synchronous circuit, and supports parallel processing of multi-channel data streams; the network communication module integrates the 4G/5G dual-mode communication module and adopts a dynamic bandwidth allocation strategy, namely, monitoring data and a local identification result use a low bandwidth transmission channel, and an abnormally high exceeding event T1 or an abnormally low exceeding event T2 are automatically upgraded to a high-speed channel; the real-time display module comprises a capacitive touch screen and is used for integrating a visual system and displaying multidimensional noise characteristics in real time, wherein the multidimensional noise characteristics comprise a second-level sound pressure level time sequence curve, a noise source duty ratio pie chart and noise event time period information; the edge computing unit is preloaded with a bimodal computing framework, supports concurrent deployment of a local lightweight model and a historical noise distribution prediction model, and meets real-time analysis requirements; The power management module comprises a hot plug lithium battery pack and an intelligent power management unit, and is provided with a mains supply priority power supply strategy and a battery endurance optimization algorithm.
9. A system for implementing the low-voltage power supply environmental noise tracing edge computing method according to any one of claims 1 to 7, comprising: the edge equipment layer is used as a data sensing and decision unit deployed on a monitoring site and used for collecting data in real time, quickly deciding and responding, and responding and processing conventional events in real time according to noise event classification results; The cloud server layer is built on a cloud platform and comprises the steps of running a high-precision analysis model to process complex tracing tasks, continuously carrying out algorithm optimization and version iteration on the model, and actively pushing the optimized model to the edge side.
10. A computer-readable storage medium, characterized in that a computer program is stored, which, when being executed by a processor, carries out the steps of the method according to any one of claims 1 to 7.

Description

Low-voltage power supply environment noise tracing edge calculation method and system Technical Field The invention relates to the technical field of environmental noise identification, in particular to a low-voltage power supply environmental noise tracing edge computing method and system. Background Currently, noise abatement faces pain points such as insufficient monitoring coverage, response lag and the like. In order to ensure timeliness and accuracy of noise treatment and construct a treatment closed loop for acoustic environment quality real-time evaluation and accurate noise tracing, an intelligent monitoring device with rapid deployment capability and supporting multi-dimensional efficient analysis is needed to provide technical support. However, when the existing noise tracing device is used for coping with actual demands, three core contradictions still exist to be broken through: first is a mismatch between power mode and deployment flexibility. Current devices are largely divided into two types, stationary and mobile. The fixed equipment has the advantages of stable power supply, but is limited by the condition of mains supply access, and is difficult to rapidly deploy in temporary or powerless areas, the mobile equipment can be flexibly deployed, but is limited by a low-voltage power supply environment, and is difficult to support a high-calculation-force computing unit required by high-precision noise tracing, so that the monitoring depth is limited, and a device capable of rapidly responding to complaint events is needed to meet the noise pollution prevention and control requirement (China noise pollution prevention and control report (2025)). And secondly, the disconnection between the static threshold judgment and the noise pollution requirement. The existing noise tracing monitoring systems take static environment noise limit values as judgment references, such as Song Weihua and the like, trigger noise monitoring data analysis based on threshold linkage (a noise monitoring data analysis method based on a threshold linkage triggering strategy and a device thereof, CN 118329188B) only carry out tracing analysis on the audio exceeding the threshold value, and neglect the elements which possibly exist in the standard reaching period and interfere with normal life, work and study of other people and form noise pollution. For example, night construction noise may be misinterpreted as a compliance event because the threshold is not breached, or bursty noise may be ignored by the system because the fixed threshold is not triggered. The lack of a multi-dimensional dynamic analysis judging mechanism is difficult to comprehensively evaluate the noise pollution condition, so that the tracing precision is reduced, the potential pollution source is more likely to be covered, and the precision of noise treatment is weakened. And thirdly, identifying unbalance between the calculation force demand and the resource allocation. The current noise tracing identification mainly depends on cloud high-precision model or local lightweight model processing. The cloud model has high precision, but along with the establishment of an automatic monitoring network for the sound environment quality of a functional area, the real-time processing requirement of mass data and the bottleneck of cloud computing power are obvious, 90 out-of-standard noise records are acquired in the city of Guangzhou in 2024 for 1 month-11 months, the analysis data size is large, a server faces huge pressure (Guangzhou Japanese, 20 monitoring points in the whole city are comprehensively established into an automatic monitoring network [ EB/OL ] (2024-11-04) for the sound environment of the functional area, while the local model can efficiently process conventional noise such as traffic noise in a fixed period, but has difficulty in distinguishing specific noise sources due to insufficient computing power under complex scenes such as multi-sound source mixed living noise, burst noise and the like, the recognition reliability is obviously reduced (Li Xia. The current situation of noise pollution control analysis and improvement measures [ J ]. Chinese resource comprehensive utilization in Guangdong province, 2024,42 (10): 159-161). Disclosure of Invention The invention provides a low-voltage power supply environment noise tracing edge computing method and system, which effectively solve core problems of inflexible deployment, incomplete tracing analysis, unbalanced calculation force distribution and the like in the prior art through a multidimensional data analysis and dynamic calculation force scheduling mechanism. The invention is realized at least by one of the following technical schemes. A low-voltage power supply environment noise tracing edge computing method comprises the following steps: S1, collecting multi-mode noise data; S2, preprocessing the acquired multi-mode noise data to acquire noise time sequence characteristics; s3, inputt