A remote monitoring method and system for flammable and explosive gases based on big data

Abstract

This invention discloses a remote monitoring method and system for flammable and explosive gases based on big data. It acquires raw datasets collected in real time from multiple sensors; preprocesses the raw datasets using a denoising algorithm, and obtains a first multi-parameter dataset through normalization; generates a second fusion feature set based on the first multi-parameter dataset; if the parameter values ​​of the second fusion feature set exceed a preset threshold, a classification model is used to identify gas leak events, and a diffusion model is combined to generate leak probability and range data; the leak probability and range data are compressed and encoded, and the transmission frequency and data packet size are dynamically adjusted according to network status parameters to form a stable transmission stream; the stable transmission stream is decoded at the monitoring center, and a dynamic visualization interface is generated using a visualization engine; the system reliability is judged based on the interface parameters to determine whether it meets preset standards. This invention achieves real-time monitoring, intelligent identification, and early warning of gas leaks.

Description

Technical Field

[0001] This invention relates to the field of gas monitoring technology, and in particular discloses a remote monitoring method and system for flammable and explosive gases based on big data. Background Technology

[0002] Safety monitoring of flammable and explosive gases is a critical area in industrial production and environmental protection, directly impacting personnel safety and stable equipment operation; its importance is self-evident. Highly hazardous gases such as silanes, due to their flammable and explosive properties, place extremely high demands on the real-time performance, accuracy, and reliability of monitoring technologies. However, existing monitoring methods still have significant limitations in complex industrial environments. Traditional solutions often rely on single sensors or localized sampling, making it difficult to cope with the dynamic and variable nature of gas diffusion, and data transmission is limited, hindering timely remote centralized monitoring. These shortcomings lead to frequent false alarms and delayed responses, failing to meet the demands of modern industry for efficient safety management.

[0003] The core challenge stems from the inadequacy of multi-parameter collaborative detection. Monitoring a single parameter cannot fully capture the complex characteristics of gas leaks, such as the interaction between concentration, temperature, and pressure, which directly limits the system's accurate judgment of leak events. Consequently, the inefficiency of data fusion and analysis becomes apparent. Existing systems lack effective models and algorithms to support the processing of multi-source heterogeneous data, leading to decreased prediction accuracy. Furthermore, the stability and real-time performance of remote data transmission are limited, especially in network fluctuations or complex environments, where data loss or delay exacerbates the unreliability of monitoring. These factors are interconnected and collectively constitute the technical bottleneck of monitoring systems in dynamic and complex scenarios.

[0004] Therefore, how to integrate multi-parameter collaborative detection, optimize data fusion analysis, and improve the real-time stability of remote transmission has become a key issue in building an efficient silane gas remote monitoring system. Summary of the Invention

[0005] This invention provides a remote monitoring method and system for flammable and explosive gases based on big data, aiming to solve at least one of the defects existing in the prior art.

[0006] One aspect of the present invention relates to a remote monitoring method for flammable and explosive gases based on big data, comprising the following steps:

[0007] Acquire the raw dataset collected in real time by multiple source sensors. The raw dataset includes gas concentration, temperature data, and pressure data.

[0008] A denoising algorithm is used to preprocess the original dataset, and the first multi-parameter dataset is obtained through normalization.

[0009] Based on the first multi-parameter dataset, a feature weighting algorithm is used to fuse heterogeneous data, and a second fused feature set is generated by principal component analysis algorithm.

[0010] The parameter values ​​of the second fusion feature set are compared with a preset threshold. If the parameter values ​​of the second fusion feature set exceed the preset threshold, a classification model is used to identify the gas leak event, and a leakage probability and range data are generated by combining the diffusion model.

[0011] The leakage probability and range data are compressed and encoded, and the transmission frequency and data packet size are dynamically adjusted according to network status parameters to form a stable transmission stream.

[0012] The monitoring center decodes the stable transmission stream and uses a visualization engine to generate a dynamic visualization interface. Based on the interface parameters, it determines whether the system reliability meets the preset standards.

[0013] Furthermore, the steps for obtaining the raw dataset acquired in real time from multiple sensor sources include:

[0014] Acquire gas concentration, temperature, and pressure data collected by multiple source sensors;

[0015] The gas concentration, temperature, and pressure data collected by multiple sensors are aggregated to form the original dataset.

[0016] Furthermore, the steps of preprocessing the original dataset using a denoising algorithm and obtaining the first multi-parameter dataset through normalization include:

[0017] The gas concentration, temperature and pressure data are denoised using a mean filtering algorithm from the raw datasets obtained from multiple sensors to obtain the first processed dataset.

[0018] Based on the first processed dataset, the numerical ranges of gas concentration, temperature data, and pressure data are mapped using the min-max normalization method to obtain the second processed dataset;

[0019] If any parameter value in the second processing dataset exceeds the preset threshold, the parameter value is padded by linear interpolation to obtain the third processing dataset.

[0020] The first multi-parameter dataset is determined by performing a weighted average calculation on the gas concentration, temperature, and pressure data in the third processing dataset.

[0021] Furthermore, based on the first multi-parameter dataset, the steps of fusing heterogeneous data using a feature weighting algorithm and generating a second fused feature set using principal component analysis include:

[0022] From the first multi-parameter dataset, the dimensions of gas concentration, temperature and pressure data are unified by standardization, and the data are normalized by Z-score method to obtain the first standardized dataset.

[0023] Based on the first standardized dataset, the entropy method is used to weight the feature importance of gas concentration, temperature data and pressure data to obtain the first weighted feature set;

[0024] If the weight value of any feature in the first weighted feature set is lower than the preset threshold, the feature value is filled in by the mean imputation method to obtain the second weighted feature set.

[0025] Principal component analysis (PCA) is used to reduce the dimensionality of the features in the second weighted feature set, extract the first few principal components, and generate the second fusion feature set.

[0026] Furthermore, the parameter values ​​of the second fusion feature set are compared with a preset threshold. If the parameter values ​​of the second fusion feature set exceed the preset threshold, a classification model is used to identify the gas leak event. The steps of generating leak probability and range data by combining the diffusion model include:

[0027] Parameter values ​​are obtained from the second fusion feature set, and compared with a preset threshold to determine whether the parameter values ​​exceed the threshold, thus obtaining the comparison result.

[0028] If the comparison results show that the parameter value exceeds the preset threshold, the pre-trained classification model is used to identify the gas leak event and generate a leak event label.

[0029] Based on the leak event label, the probability distribution of gas leaks is calculated using a diffusion model to obtain leak probability distribution data;

[0030] High-probability areas are extracted from the leakage probability distribution data, and the boundary of the impact range is determined by the grid division method to generate impact range data;

[0031] For the data on the scope of impact, a spatial interpolation algorithm is used to estimate the probability of the uncovered areas, resulting in extended probability distribution data;

[0032] Based on the extended probability distribution data and the impact range data, a weighted superposition method is used to generate a comprehensive leakage risk distribution, and the risk distribution data is obtained.

[0033] Key risk areas are extracted from risk distribution data, and high-risk points are grouped using clustering algorithms to generate risk partition data.

[0034] Furthermore, the steps of compressing and encoding the leakage probability and range data, and dynamically adjusting the transmission frequency and data packet size according to network status parameters to form a stable transmission stream include:

[0035] The original dataset is obtained from the leakage probability and range data, and the Huffman coding algorithm is used to compress and encode the original dataset to generate a compressed data stream.

[0036] Based on the network status parameters, determine whether the bandwidth in the network status parameters is lower than the preset threshold. If it is lower than the preset threshold, adjust the data packet size and transmission frequency to obtain the adjusted transmission parameters.

[0037] By adjusting the transmission parameters, the compressed data stream is divided into multiple data packets, and the data packets are transmitted using the sliding window protocol to generate a stable transmission stream.

[0038] Extract the receiver feedback information from the stable transmission stream, determine whether the data integrity in the receiver feedback information meets the preset standard, and if not, readjust the transmission frequency to determine the optimized transmission stream.

[0039] Furthermore, the stable transmission stream is decoded at the monitoring center, and a dynamic visualization interface is generated using a visualization engine. The steps to determine whether the system reliability meets preset standards based on the interface parameters include:

[0040] A stable transmission stream is obtained from the monitoring center, and the Huffman decoding algorithm is used to decode the stable transmission stream. If data loss is detected, it is recovered through redundant check codes to obtain the decoded data stream.

[0041] Based on the decoded data stream, a dynamic visualization interface is generated using vector graphics drawing tools. Interface parameters are extracted using preset templates to determine the set of interface parameters.

[0042] The display latency and refresh rate are obtained from the set of interface parameters. If the display latency is higher than the preset threshold or the refresh rate is lower than the preset threshold, the system reliability is determined to be unsatisfactory, and the reliability judgment result is obtained.

[0043] Based on the reliability assessment results, real-time monitoring data is extracted from a pre-established feedback information database, and the interface parameters are adjusted using a logistic regression algorithm to generate an optimized dynamic visualization interface.

[0044] Another aspect of the present invention relates to a remote monitoring system for flammable and explosive gases based on big data, used to implement the aforementioned remote monitoring method for flammable and explosive gases based on big data. The remote monitoring system for flammable and explosive gases based on big data includes:

[0045] The acquisition module is used to acquire the raw dataset collected in real time by multiple source sensors. The raw dataset includes gas concentration, temperature data and pressure data.

[0046] The preprocessing module is used to preprocess the original dataset using a denoising algorithm and obtain the first multi-parameter dataset through normalization.

[0047] The generation module is used to perform data fusion on heterogeneous data based on the first multi-parameter dataset using a feature weighting algorithm, and to generate a second fused feature set using a principal component analysis algorithm.

[0048] The comparison module is used to compare the parameter values ​​of the second fusion feature set with a preset threshold. If the parameter values ​​of the second fusion feature set exceed the preset threshold, a classification model is used to identify the gas leak event, and a leakage probability and range data are generated by combining the diffusion model.

[0049] The module is used to compress and encode the leakage probability and range data, and dynamically adjust the transmission frequency and data packet size according to network status parameters to form a stable transmission stream.

[0050] The judgment module is used to decode stable transmission streams in the monitoring center, and uses a visualization engine to generate a dynamic visualization interface to determine whether the system reliability meets the preset standards based on the interface parameters.

[0051] Furthermore, the acquisition module includes:

[0052] The first acquisition unit is used to acquire gas concentration, temperature and pressure data collected by the multi-source sensors;

[0053] The aggregation unit is used to aggregate gas concentration, temperature, and pressure data collected by multiple sensors to form a raw dataset.

[0054] Furthermore, the preprocessing module includes:

[0055] The second acquisition unit is used to perform noise reduction processing on gas concentration, temperature data and pressure data from the raw dataset acquired by the multi-source sensors using a mean filtering algorithm to obtain the first processed dataset.

[0056] The third acquisition unit is used to map the numerical ranges of gas concentration, temperature data and pressure data using the minimum-maximum normalization method based on the first processing dataset to obtain the second processing dataset.

[0057] The fourth acquisition unit is used to complete the parameter value by linear interpolation if any parameter value in the second processing dataset exceeds a preset threshold, so as to obtain the third processing dataset.

[0058] The determining unit is used to determine the first multi-parameter dataset by performing a weighted average calculation on the gas concentration, temperature, and pressure data in the third processing dataset.

[0059] The beneficial effects achieved by this invention are as follows:

[0060] This invention provides a remote monitoring method and system for flammable and explosive gases based on big data. It collects raw data such as gas concentration, temperature, and pressure from multiple sensors. After denoising and normalization preprocessing, heterogeneous data is fused using a feature weighting algorithm, and a fused feature set is generated using principal component analysis. When feature parameters exceed preset thresholds, a classification model and a diffusion model are combined to identify leakage events and predict their probability range. This invention compresses and encodes leakage data and dynamically adjusts transmission parameters based on network status to achieve stable transmission. After decoding, the monitoring center generates a dynamic interface through a visualization engine to assess system reliability. This invention achieves real-time monitoring, intelligent identification, and early warning of gas leaks, improving the accuracy and timeliness of hazardous gas leak monitoring and providing effective protection for industrial safety. Attached Figure Description

[0061] Figure 1 This is a flowchart illustrating an embodiment of a remote monitoring method for flammable and explosive gases based on big data according to the present invention. Detailed Implementation

[0062] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0063] like Figure 1 As shown, the first embodiment of the present invention proposes a remote monitoring method for flammable and explosive gases based on big data, including the following steps:

[0064] Step S100: Obtain the raw dataset collected in real time by the multi-source sensors. The raw dataset includes gas concentration, temperature data and pressure data.

[0065] Multi-source sensors refer to integrated detection devices that can simultaneously acquire different types of environmental parameters (such as gas concentration, temperature, pressure, images, etc.) through multiple measurement channels or technical means.

[0066] Raw datasets refer to the initial collection of data that has not undergone any processing, encoding, or analysis. They originate directly from sources such as sensors, acquisition devices, or user interactions, and retain the most original physical or behavioral characteristics of the observed objects.

[0067] Step S200: Use a denoising algorithm to preprocess the original dataset and obtain the first multi-parameter dataset through normalization.

[0068] Denoising algorithms are a class of digital methods that use mathematical models or signal processing techniques to separate and eliminate noise components from disturbed observation data, restoring the original signal or data characteristics. Their core objective is to improve the signal-to-noise ratio (SNR) and the reliability of subsequent analysis by suppressing random noise (such as sensor errors and environmental interference) and structured noise (such as periodic interference and artifacts).

[0069] Signal preprocessing is a technical process that involves normalizing, denoising, enhancing, or extracting features from raw signals before data analysis or model training. Its aim is to eliminate interference factors in the data, standardize data scale, and improve the accuracy and efficiency of subsequent processing. Essentially, it uses mathematical or engineering methods to transform the raw signal into a standard form more suitable for algorithm analysis.

[0070] Normalization is a data preprocessing technique that maps data to a specific range (such as [0, 1] or [-1, 1]) or a standardized distribution (such as a mean of 0 and a standard deviation of 1) through mathematical transformations. It aims to eliminate model bias caused by differences in units and numerical ranges, and improve the convergence speed and stability of algorithms. Essentially, it involves linearly or non-linearly scaling the original data to ensure consistent scale and comparability.

[0071] A multi-parameter dataset is a structured collection of data consisting of multiple independent variables (parameters) and their corresponding observations. It is typically organized in tabular form, where each column represents a specific parameter and each row corresponds to an observation instance or sample. Such datasets achieve a comprehensive description of complex systems or phenomena by integrating multidimensional features (such as temperature, pressure, and flow rate), supporting cross-parameter correlation analysis and multivariate modeling.

[0072] Step S300: Based on the first multi-parameter dataset, a feature weighting algorithm is used to fuse the heterogeneous data, and a second fused feature set is generated through principal component analysis.

[0073] The Feature Weighting Algorithm is a machine learning method that quantifies the contribution of different features in a dataset to a target task by assigning differentiated weights (weight values ​​≥ 0). Its core objective is to improve model performance (such as classification accuracy and regression error) and interpretability by adjusting the importance distribution of the feature space, enhancing the decision-making influence of key features, and suppressing the interference of irrelevant or redundant features.

[0074] Data fusion is an information processing technology that integrates heterogeneous data from multiple sources (such as sensors, databases, and text) to generate more consistent and accurate comprehensive information. Its core is to automatically analyze, correlate, and combine data from different sources to build a unified data view or model to support intelligent decision-making, system optimization, or the analysis of complex phenomena.

[0075] Principal Component Analysis (PCA) is an unsupervised linear dimensionality reduction algorithm. Its core idea is to project the original high-dimensional data into a low-dimensional space (called principal components) through orthogonal transformation, so as to maximize the variance of the data in the new coordinate axis direction, thereby reducing the data complexity while retaining key information.

[0076] The Second-Level Fused Feature Set is a high-level feature set that further optimizes, filters, and combines the feature space after initial feature extraction and fusion during multi-source data fusion. Its core objective is to construct a more discriminative and generalizable low-dimensional representation by eliminating redundant information and strengthening the correlation of key features, so as to adapt to the modeling needs of downstream tasks (such as classification and regression).

[0077] Step S400: Compare the parameter values ​​of the second fusion feature set with a preset threshold. If the parameter values ​​of the second fusion feature set exceed the preset threshold, use a classification model to identify the gas leak event and combine it with a diffusion model to generate leak probability and range data.

[0078] A classification model is a supervised learning algorithm that aims to classify unknown samples into predefined discrete categories by learning the mapping relationship between features and class labels in a dataset.

[0079] A gas leak incident refers to the abnormal release of gas from a closed container, pipeline, or other storage / transportation system into the external environment due to equipment damage, operational errors, corrosion, human sabotage, or other reasons, which may cause safety risks, environmental pollution, or production interruption.

[0080] The diffusion model is a type of generative model based on progressive noise addition and denoising reconstruction. Its core idea is to simulate the diffusion process in physics (such as molecular diffusion) to gradually add noise to the data (such as images and audio) to a random distribution, and then learn the inverse denoising steps to restore the original data distribution, thereby achieving high-quality data generation.

[0081] Leakage probability refers to the quantitative assessment of the likelihood of unplanned leakage of gas or liquid from a closed system under specific conditions (such as equipment operating status, environmental parameters, etc.). Its calculation usually involves comprehensive modeling combining material properties (such as minimum ignition energy, auto-ignition temperature), operating parameters (pressure, temperature), and external environmental factors (ignition source density, control measures).

[0082] Step S500: Compress and encode the leakage probability and range data, and dynamically adjust the transmission frequency and data packet size according to the network status parameters to form a stable transmission stream.

[0083] Compression coding is an information processing technique that reduces data redundancy through specific algorithms. It aims to reduce data storage space requirements or transmission bandwidth consumption while preserving as much of the original data's useful information as possible. Its core objective is to achieve a balance between minimizing data volume and maximizing information integrity without significantly impacting data usability.

[0084] Network status parameters are a set of key indicators that quantify the real-time operational characteristics of a network. They encompass multiple dimensions of attributes, including network structure, performance, traffic, and signal quality, and are used to monitor, diagnose, and optimize network operational efficiency. Their core function is to provide data support for network operations and maintenance, ensuring communication stability and Quality of Service (QoS).

[0085] A stable transport stream refers to a data transmission process in a network that possesses continuous and predictable performance. Its core characteristics are low latency volatility, high bandwidth consistency, and controllable packet loss rate. It can maintain the continuity of critical services (such as real-time audio and video, and industrial control) in dynamic network environments (such as congestion and interference).58 Stable transport streams ensure that data streams meet preset Quality of Service (QoS) thresholds through protocol optimization and resource guarantee mechanisms, and are the technical foundation for achieving reliable communication.

[0086] Step S600: Decode the stable transmission stream in the monitoring center, generate a dynamic visualization interface using a visualization engine, and determine whether the system reliability meets the preset standards based on the interface parameters.

[0087] The monitoring center serves as a centralized and intelligent hub for network and system operation management. Through real-time collection, analysis, and visualization of multi-source data (such as device status, traffic metrics, and alarm events), it enables comprehensive perception and control of infrastructure, business applications, and security posture. Its core functions encompass status monitoring, anomaly alarms, performance optimization, and decision support, providing operations and maintenance personnel with a unified operating interface and automated response capabilities to ensure high system availability and service continuity.

[0088] A visualization engine is a core technology platform integrating data processing, graphics rendering, and interactive logic. It transforms multi-source heterogeneous data (such as device status, business metrics, and spatial information) into graphics, images, or 3D models, enabling dynamic data presentation and in-depth analysis. Its core value lies in lowering the barrier to data understanding, assisting users in quickly identifying patterns, locating anomalies, and driving decision optimization.

[0089] A dynamic visualization interface is a graphical operating platform that updates and interactively presents data changes in real time. By integrating animations, charts, and real-time data streams (such as sensor signals and business metrics), it transforms complex information into dynamic visual elements (such as trend curves and heatmaps), enabling users to intuitively perceive data evolution patterns and quickly respond to anomalies. Its core features are real-time performance, interactivity, and intelligent adaptability, which can overcome the limitations of static charts and achieve data-driven decision optimization.

[0090] Interface parameters are a set of programmable variables used to configure and adjust the functionality, style, and interactive behavior of user interfaces (UI). By preset or dynamically modifying parameter values, personalized display and behavior control of UI components (such as charts, controls, and animations) can be achieved. Its core objective is to reduce development complexity, improve user experience flexibility, and support users in quickly adapting interface performance to different scenarios based on their needs.

[0091] System reliability refers to the ability or probability of a system to continuously and stably perform its preset functions within specified environmental conditions (such as temperature, electromagnetic interference, and operating methods) and a specified time range. Essentially, it reflects the system's comprehensive performance in resisting and recovering from faults and in uninterrupted operation, and is a core indicator for measuring system quality and stability.

[0092] Pre-defined standards refer to a set of normative requirements and guidelines established in advance during the product or system design phase. These standards guide the development process, constrain design boundaries, and ensure that the final product meets specific performance, safety, and reliability objectives. Their core function is to reduce design risks, prevent potential failures, and improve the adaptability and consistency of the product throughout its entire lifecycle through systematic rules.

[0093] Furthermore, this embodiment proposes a remote monitoring method for flammable and explosive gases based on big data, wherein step S100 includes:

[0094] Step S110: Acquire gas concentration, temperature and pressure data collected by multi-source sensors.

[0095] The core of multi-source sensor data acquisition of gas concentration, temperature and pressure data lies in using different types of sensors to obtain environmental parameters and then summarizing the data to form a raw dataset, which provides a foundation for subsequent analysis.

[0096] Gas concentration sensors are typically based on electrochemical or infrared principles to detect the concentration of specific gases such as silane or methane; temperature sensors mostly use thermistors or thermocouples to measure ambient temperature; and pressure sensors obtain gas pressure through piezoresistive or capacitive principles. These sensors work together to ensure that the data comprehensively reflects the environmental conditions. For example, in an industrial gas monitoring scenario, suppose a chemical plant needs to monitor the concentration, temperature, and pressure of silane inside a reactor.

[0097] A gas concentration sensor is deployed inside the reactor to monitor the silane content in real time, outputting a value of 500 ppm; a temperature sensor records the reactor temperature as 45°C; and a pressure sensor measures the pressure as 2.5 bar. This data is converted into digital signals by an analog-to-digital converter and transmitted to the data acquisition system.

[0098] It should be noted that the sensor needs to be calibrated regularly to ensure measurement accuracy. For example, the silane sensor should be calibrated monthly, with the error controlled within ±5ppm.

[0099] Step S120: Summarize the gas concentration, temperature and pressure data collected by the multi-source sensors to form the original dataset.

[0100] In one possible implementation, data aggregation employs a time synchronization mechanism to ensure alignment of multi-source data. For example, data is collected once per second from sensors and transmitted to a central server via an Industrial Internet of Things (IIoT) protocol such as Modbus. The server stores the data by timestamp, forming a raw dataset in CSV file format, containing fields for: time, silane concentration, temperature, and pressure. For instance, the dataset might be recorded as "2025-05-04 08:00:00, 500ppm, 45℃, 2.5bar".

[0101] Preferably, the system will perform preliminary filtering of outliers. For example, if the pressure suddenly changes to 10 bar, an alarm will be triggered and the data will be marked as abnormal.

[0102] Specifically, data aggregation also needs to consider the compatibility between sensors. In one embodiment, if different sensors have different sampling frequencies, such as a gas concentration sensor sampling once per second while a pressure sensor samples once every 2 seconds, the system uses an interpolation algorithm to complete the data and ensure time series consistency. This avoids errors caused by time misalignment during data analysis.

[0103] Understandably, a standardized data format and time base facilitate subsequent modeling and prediction. For example, the original dataset can be used to analyze the gas state inside the reactor and determine whether adjustments to ventilation or pressure valves are necessary.

[0104] The aggregated data can also provide training samples for machine learning models and predict potential security risks. In one embodiment, the dataset is displayed as a real-time graph using visualization tools, allowing operators to intuitively monitor parameter changes.

[0105] Preferably, the system supports data backtracking to facilitate troubleshooting of abnormalities. For example, if the temperature is found to rise abnormally to 60°C during a certain period, it may indicate a cooling system malfunction.

[0106] In another possible implementation, the aggregated data can be preliminarily processed by edge computing devices, reducing the burden on the cloud. For example, the edge device can perform a threshold judgment on the collected 500ppm silane concentration; if it exceeds 600ppm, an alarm is triggered locally. This approach reduces data transmission latency and improves real-time performance.

[0107] It should be noted that edge computing can also denoise sensor data, ensuring the quality of the original dataset.

[0108] The above methods enable the collection and aggregation of multi-source sensor data to form a high-quality raw dataset, which not only provides a reliable basis for real-time monitoring, but also provides data support for subsequent analysis and optimization.

[0109] Measures such as time synchronization, anomaly filtering, and edge computing significantly improve the system's stability and response speed, providing technical support for the safety and efficiency of industrial scenarios.

[0110] Furthermore, this embodiment proposes a remote monitoring method for flammable and explosive gases based on big data, wherein step S200 includes:

[0111] Step S210: From the raw dataset obtained from the multi-source sensors, the gas concentration, temperature data and pressure data are denoised using a mean filtering algorithm to obtain the first processed dataset.

[0112] For example, in industrial gas monitoring scenarios, the gas concentration, temperature and pressure data collected by multi-source sensors often contain noise, which needs to be denoised using a mean filtering algorithm.

[0113] Mean filtering smooths out abrupt noise by calculating the average value of data within a certain time window. For example, suppose the silane concentration data in a chemical plant's reactor is collected at 510ppm, 520ppm, 600ppm, 515ppm, and 518ppm within 5 seconds, where 600ppm might be a noisy point. Mean filtering uses a window size of 3 and calculates the average of the intermediate values. For instance, the average of 520ppm, 515ppm, and 518ppm is 517.67ppm. Replacing 600ppm with this average results in a smoothed first-processed dataset. This method is simple, efficient, preserves data trends, and is suitable for real-time processing.

[0114] Step S220: Based on the first processing dataset, the numerical ranges of gas concentration, temperature data, and pressure data are mapped using the min-max normalization method to obtain the second processing dataset.

[0115] In one possible implementation, the first processed dataset needs further normalization to unify the numerical range, facilitating subsequent analysis. Min-max normalization maps the data to a range of 0 to 1. For example, silane concentration ranges from 400 ppm to 800 ppm, temperature ranges from 20°C to 60°C, and pressure ranges from 1 bar to 3 bar. Taking data at a certain moment as 500 ppm, 45°C, and 2.5 bar, after normalization, the values ​​are 0.25, 0.625, and 0.75 respectively, forming the second processed dataset. This normalization ensures that data with different dimensions are comparable, improving analytical consistency.

[0116] Step S230: If any parameter value in the second processing dataset exceeds a preset threshold, the parameter value is completed by linear interpolation to obtain the third processing dataset.

[0117] It should be noted that if a parameter in the second processing dataset exceeds a threshold, linear interpolation must be used to complete the data to ensure data integrity. For example, if the threshold for the normalized silane concentration is set to 0.9, and the data at a certain moment is 0.95, it is considered an anomaly.

[0118] Assuming the data at consecutive time points are 0.3 and 0.4, linear interpolation is used to calculate the median value of 0.35, which is then used to replace outliers, resulting in the third processed dataset. This method infers reasonable values ​​based on the trend of the preceding and following data, maintaining the continuity of the time series.

[0119] Step S240: Determine the first multi-parameter dataset by performing a weighted average calculation on the gas concentration, temperature, and pressure data in the third processing dataset.

[0120] Specifically, the third processing dataset calculates a multi-parameter dataset using a weighted average to comprehensively reflect the environmental state. For example, assuming the weights are assigned as silane concentration 0.4, temperature 0.3, and pressure 0.3, the normalized data at a certain moment are 0.25, 0.625, and 0.75. The weighted average is 0.25×0.4 + 0.625×0.3 + 0.75×0.3 = 0.5125, forming the first multi-parameter dataset. This value can be used to assess the overall state of the reactor, facilitating quick judgment by the operator.

[0121] Preferably, the window size for mean filtering can be adjusted according to data volatility; for example, the window can be increased when concentration fluctuations are large. For normalization, other methods such as standardization can be selected to suit specific analytical needs.

[0122] The accuracy of linear interpolation depends on the reliability of the preceding and following data, and the completion effect can be improved by combining more contextual data.

[0123] The weights of the weighted average can be dynamically adjusted according to the importance of the parameters; for example, if the concentration has a greater impact on safety, its weight can be increased.

[0124] Understandably, these processing steps are interconnected and work together to ensure data quality. Noise reduction lowers interference, normalization unifies the scale, interpolation ensures integrity, and weighted averaging integrates multi-dimensional information. Each step provides a reliable foundation for subsequent analysis, making it suitable for the efficient data processing needs of industrial scenarios.

[0125] Furthermore, this embodiment proposes a remote monitoring method for flammable and explosive gases based on big data, wherein step S300 includes:

[0126] Step S310: From the first multi-parameter dataset, standardization is used to unify the dimensions of gas concentration, temperature data and pressure data, and the data is normalized using the Z-score method to obtain the first standardized dataset.

[0127] For example, in an industrial gas monitoring scenario, the first multi-parameter dataset contains gas concentration, temperature, and pressure data, which need to be standardized to be unified due to their different dimensions.

[0128] Z-score standardization transforms data into a distribution with a mean of 0 and a standard deviation of 1. For example, in a chemical plant, the concentration of silane in a reactor ranges from 450 ppm to 750 ppm, the temperature ranges from 25°C to 55°C, and the pressure ranges from 1.2 bar to 2.8 bar.

[0129] We take data at a certain moment as 600 ppm, 40℃, and 2 bar, assuming the mean concentration is 600 ppm with a standard deviation of 100 ppm, the mean temperature is 40℃ with a standard deviation of 10℃, and the mean pressure is 2 bar with a standard deviation of 0.4 bar. The Z-scores are calculated to be 0, 0, and 0 respectively, forming the first standardized dataset. This method eliminates dimensional differences, facilitating subsequent feature analysis.

[0130] Step S320: Based on the first standardized dataset, the entropy method is used to calculate the weighted importance of the features of gas concentration, temperature data and pressure data to obtain the first weighted feature set.

[0131] In one possible implementation, the entropy method is used to assess the importance of gas concentration, temperature, and pressure as features. The entropy method determines the contribution of each feature to the system state by calculating its information entropy. For example, concentration data fluctuates significantly, resulting in a lower information entropy and a weight of 0.5; temperature and pressure fluctuate less, resulting in higher entropy values ​​and weights of 0.3 and 0.2, respectively. If the standardized data at a certain moment are 0.5, 0.2, and 0.1, the weighted calculation is 0.5 × 0.5 + 0.2 × 0.3 + 0.1 × 0.2 = 0.33, forming the first weighted feature set. This method highlights key features and improves the relevance of the analysis.

[0132] Step S330: If the weight value of any feature in the first weighted feature set is lower than the preset threshold, the feature value is filled in by the mean imputation method to obtain the second weighted feature set.

[0133] It should be noted that if the weight of a feature in the first weighted feature set is lower than a threshold (e.g., if the threshold is set to 0.25 and the pressure weight of 0.2 is lower than the threshold), it needs to be filled by using the mean. Assuming the standardized values ​​of the pressure data in previous time periods are 0.3, 0.4, and 0.2, and the mean is 0.3, the lower weight values ​​are replaced with 0.3 to form the second weighted feature set. This filling method utilizes historical data trends to maintain the integrity of the feature set.

[0134] Step S340: Dimensionality reduction of the features in the second weighted feature set is performed using principal component analysis algorithm to extract the first few principal components and generate the second fusion feature set.

[0135] The covariance matrix of the second weighted feature set is:

[0136] (1)

[0137] In formula (1), Let the covariance matrix of the second weighted feature set be represented. Indicates the number of samples. Indicates the first The feature vector of each sample This represents the mean vector of all sample features. Formula (1) calculates the covariance relationship between features in the feature set, providing a basis for subsequent principal component extraction.

[0138] Specifically, principal component analysis is used to reduce the dimensionality of the second weighted feature set and extract the main features. For example, if the second weighted feature set contains three dimensions: concentration, temperature, and pressure, correlation analysis shows that concentration and temperature are highly correlated.

[0139] Principal component analysis extracts the first two principal components. It is assumed that the first principal component reflects the combined changes in concentration and temperature, and the second principal component reflects the changes in pressure, generating a second fused feature set. For example, data at a certain moment, after projection, might have values ​​of 0.4 and 0.1, retaining 90% of the information. This dimensionality reduction method reduces redundancy and highlights the core trends in change.

[0140] Preferably, the weight calculation of the entropy method can be dynamically adjusted in conjunction with real-time data, such as increasing its weight when the concentration is abnormal. Mean imputation can introduce more historical data to improve accuracy. Principal component analysis can adjust the number of principal components retained according to the scenario; for example, in a safety monitoring scenario, more components can be retained to ensure information integrity.

[0141] Understandably, the above processing steps are progressive, standardizing and unifying units, using entropy to highlight feature importance, mean imputation to ensure data integrity, and principal component analysis to reduce complexity. These steps collectively provide a high-quality data foundation for subsequent analysis, suitable for the real-time needs of industrial gas monitoring.

[0142] Furthermore, this embodiment proposes a remote monitoring method for flammable and explosive gases based on big data, wherein step S400 includes:

[0143] Step S410: Obtain parameter values ​​from the second fusion feature set, compare them with a preset threshold, determine whether the parameter values ​​exceed the threshold, and obtain the comparison result.

[0144] The following formula is used to determine whether a parameter value exceeds a threshold:

[0145] (2)

[0146] In formula (2), Indicates the first The comparison results of the parameters Indicates the first obtained from the second fusion feature set Parameter values, This represents the corresponding preset threshold. When the parameter value exceeds the threshold, the comparison result is 1; otherwise, it is 0.

[0147] For example, in industrial gas monitoring scenarios, obtaining parameter values ​​from the second fusion feature set typically involves extracting feature vectors after principal component analysis. For instance, in a chemical plant monitoring silane concentration, temperature, and pressure, the second fusion feature set contains two principal components: 0.4 and 0.1, reflecting the combined changes in concentration and temperature, and pressure changes, respectively. A preset threshold might be 0.5. Comparison reveals that 0.4 does not exceed the threshold, and 0.1 is significantly lower, indicating a normal system state. This comparison method quickly filters out abnormal states by setting reasonable thresholds.

[0148] Step S420: If the comparison result shows that the parameter value exceeds the preset threshold, the pre-trained classification model is used to identify the gas leak event and generate a leak event label.

[0149] In one possible implementation, if the parameter value exceeds a threshold, such as a principal component value reaching 0.6, it indicates a potential risk of gas leakage. In this case, a pre-trained classification model, such as a support vector machine, is trained based on historical leakage data, inputting a feature vector of 0.6 and outputting a leakage event label, such as "high-risk leakage." The classification model accurately distinguishes between normal and abnormal states by learning historical data patterns.

[0150] Step S430: Based on the leak event label, calculate the probability distribution of gas leak using a diffusion model to obtain leak probability distribution data.

[0151] The probability distribution of tag-based gas leaks is calculated as follows:

[0152] (3)

[0153] In formula (3), Indicates a given leak event label Gas leakage probability under certain conditions The probability density, Indicates the first Gas concentration observations at each location, Indicates label The corresponding expected concentration value, This represents the standard deviation parameter of the diffusion model.

[0154] It should be noted that, based on the leak event label, the diffusion model is used to calculate the leak probability distribution. For example, using the label "high-risk leak" as a condition, combined with reactor space data, the diffusion model generates a probability distribution showing that the leak probability in a certain area is 80%. By simulating the gas diffusion process, the diffusion model predicts the likelihood of a leak, providing a data foundation for subsequent analysis.

[0155] Step S440: Extract high-probability areas from the leakage probability distribution data, use a grid division method to determine the boundaries of the impact range, and generate impact range data.

[0156] Specifically, high-probability areas, such as those with a probability greater than 70%, are extracted from the leakage probability distribution, and the impact range is determined using a grid division method. For example, the reactor space is divided into a 10x10 grid, and the grids with a probability greater than 70% are marked to form impact range data, such as covering five grids in the southeast corner of the reactor. This method clearly defines the risk area through spatial division.

[0157] Step S450: For the impact range data, the probability of the uncovered area is estimated by spatial interpolation algorithm to obtain extended probability distribution data.

[0158] The weights of the influence of each known point on the unknown point in spatial interpolation are calculated as follows:

[0159] (4)

[0160] In formula (4), Indicates the first Spatial interpolation weights for known data points Indicates the point to be estimated up to the th Euclidean distance between known data points Indicates the point to be estimated up to the th Euclidean distance between known data points This represents the distance decay exponent parameter. This represents the total number of known data points involved in the interpolation calculation.

[0161] In one embodiment, a spatial interpolation algorithm estimates the probability of uncovered areas based on the impact range data. For example, inverse distance-weighted interpolation is used to calculate the probability of unmonitored grids based on known grid probabilities, such as estimating the probability of an uncovered grid as 60%. This method utilizes neighboring data to expand the probability distribution, generating expanded probability distribution data and improving coverage integrity.

[0162] Step S460: Based on the extended probability distribution data and the impact range data, a weighted superposition method is used to generate a comprehensive leakage risk distribution, and the risk distribution data is obtained.

[0163] The probability distribution considering diffusion effects is calculated as follows:

[0164] (5)

[0165] In formula (5), Indicates position The extended probability distribution value at that location, This represents the initial leakage probability distribution. This indicates the radial distance from the leak source. The parameter representing the standard deviation of the extended distribution, Indicates the concentration effect coefficient. This indicates the concentration of pollutants at that location. This indicates the maximum concentration value.

[0166] Preferably, the comprehensive leakage risk distribution is generated using a weighted overlay method. For example, the extended probability distribution data has a weight of 0.6, the impact range data has a weight of 0.4, and the weighted risk value of a certain grid is calculated to be 0.68, forming the risk distribution data. This method combines multi-source data to generate a comprehensive risk view.

[0167] Step S470: Extract key risk areas from the risk distribution data, group high-risk points using a clustering algorithm, and generate risk partition data.

[0168] The comprehensive risk distribution value for each point in the computational space is:

[0169] (6)

[0170] In formula (6), Indicates the first Risk value at each location point This represents the total number of risk factors. Indicates the first The weighting coefficients of each risk factor, Indicates the first Each risk factor on the coordinate Risk contribution function at the location.

[0171] Points that meet the high-risk criteria will be assigned to the nearest cluster center:

[0172] (7)

[0173] In formula (7), Indicates the first A high-risk cluster, Indicates the first A high-risk point, Point To the cluster center distance, This represents the total number of clusters. Point The risk value, This indicates a high-risk threshold.

[0174] For example, key risk areas can be extracted from risk distribution data, such as grids with risk values ​​greater than 0.65. Using the K-means clustering algorithm, high-risk points can be divided into three groups, generating risk partitioning data, such as the southeast corner being a high-risk group. This method clearly defines risk levels through clustering, facilitating precise responses.

[0175] Understandably, the above methods progress step by step from feature extraction to risk partitioning, using techniques such as comparison, classification, diffusion, interpolation, and clustering to form a complete risk assessment chain, which is suitable for industrial gas monitoring needs.

[0176] Furthermore, this embodiment proposes a remote monitoring method for flammable and explosive gases based on big data, wherein step S500 includes:

[0177] Step S510: Obtain the original dataset from the leakage probability and range data, and use the Huffman coding algorithm to compress and encode the original dataset to generate a compressed data stream.

[0178] For example, in industrial gas monitoring scenarios, the raw dataset obtained from leakage probability and range data typically includes gridded probability values ​​and spatial coordinates. For instance, the reactor area of ​​a chemical plant might be divided into a 10x10 grid, with each grid recording the leakage probability. For example, a grid in the southeast corner might have a probability of 75% and coordinates (8, 2). This data constitutes the raw dataset, which is quite large, and direct transmission might consume excessive bandwidth.

[0179] Using Huffman coding to compress the original dataset effectively reduces data size. Huffman coding analyzes the frequency of probability values ​​in the data, assigning short codewords to high-frequency values ​​and long codewords to low-frequency values. For example, a 75% probability value occurs frequently and is encoded as 01, while a 10% probability value occurs less frequently and is encoded as 1101. The resulting compressed data stream significantly reduces the data size, facilitating subsequent transmission.

[0180] Step S520: Based on the network status parameters, determine whether the bandwidth in the network status parameters is lower than the preset threshold. If it is lower than the preset threshold, adjust the data packet size and transmission frequency to obtain the adjusted transmission parameters.

[0181] The following formula is used to determine whether the current network bandwidth is lower than a preset threshold:

[0182] (8)

[0183] In formula (8), This indicates the current network bandwidth measurement. The preset bandwidth threshold is represented by formula (8), which is used to determine whether the current network bandwidth is lower than the preset threshold. When the inequality is true, the transmission parameter adjustment mechanism is triggered.

[0184] In one possible implementation, the bandwidth is determined to be below a preset threshold based on network status parameters.

[0185] Network status parameters include real-time bandwidth and latency. For example, the current bandwidth is 2Mbps, and the preset threshold is 5Mbps. Bandwidth below the threshold indicates network congestion, and directly transmitting compressed data streams may result in packet loss.

[0186] It's important to note that adjusting packet size and transmission frequency can optimize transmission efficiency. For example, reducing the packet size from 1024 bytes to 512 bytes and decreasing the transmission frequency from 10 times per second to 5 times per second yields the adjusted transmission parameters. This adjustment reduces the amount of data transmitted per transmission, thus reducing network load.

[0187] Step S530: Using the adjusted transmission parameters, the compressed data stream is divided into multiple data packets, and the data packets are transmitted using the sliding window protocol to generate a stable transmission stream.

[0188] Specifically, by adjusting the transmission parameters, the compressed data stream is divided into multiple data packets and transmitted using a sliding window protocol. The sliding window protocol allows for the simultaneous transmission of four data packets, such as a window size of 4, with subsequent packets sent only after the receiver confirms the transmission. For example, if the compressed data stream is divided into 100 data packets, four packets are sent at a time, and the sliding window moves forward after the receiver confirms, generating a stable transmission stream. This method ensures orderly data transmission and reduces the need for retransmissions.

[0189] Step S540: Extract the receiver feedback information from the stable transmission stream, determine whether the data integrity in the receiver feedback information meets the preset standard, and if not, readjust the transmission frequency and determine the optimized transmission stream.

[0190] The overall integrity level of the receiver's feedback information is evaluated by calculating the weighted average of the packet integrity rate within each time window.

[0191] (9)

[0192] In formula (9), Indicators representing data integrity This indicates the total number of time windows for sampling feedback information. Indicates the first The number of data packets lost within a time window. Indicates the first Total number of data packets sent within a time window Indicates the first The weighting coefficients for each time window.

[0193] In one embodiment, receiver feedback information is extracted from a stable transport stream to determine whether data integrity meets a preset standard. The feedback information includes the data packet reception status and checksum. For example, the receiver feedback may indicate that 98 data packets were received, with 2 packets lost, indicating that integrity did not meet the standard.

[0194] Preferably, the transmission frequency is readjusted, such as reducing it from 5 times per second to 3 times per second, to increase the stability of each transmission and determine the optimized transmission flow. This adjustment reduces packet loss caused by network congestion by lowering the transmission rate. It can be understood that the above method forms a complete data processing and transmission chain, from data compression to transmission optimization.

[0195] Huffman coding reduces data volume, bandwidth determination and parameter adjustment adapt to network conditions, sliding window protocol ensures transmission stability, and feedback information and frequency optimization ensure data integrity. For example, the optimized transport stream successfully transmits all data packets at a bandwidth of 2Mbps, and the receiving end completely reconstructs the leakage probability data. This method, through multi-stage collaboration, ensures efficient data transmission and provides reliable data support for gas leak monitoring.

[0196] Furthermore, this embodiment proposes a remote monitoring method for flammable and explosive gases based on big data, wherein step S600 includes:

[0197] Step S610: Obtain a stable transmission stream from the monitoring center, and use the Huffman decoding algorithm to decode the stable transmission stream. If data loss is detected, it is recovered through the redundancy check code to obtain the decoded data stream.

[0198] For example, in an industrial gas monitoring scenario, after receiving a stable transmission stream, the monitoring center needs to decode it to reconstruct the data. The Huffman decoding algorithm reads the codewords from the compressed data stream and, combined with a pre-constructed Huffman tree, maps the codewords back to their original probability values ​​and coordinates. For instance, codeword 01 corresponds to a 75% probability value, and codeword 1101 corresponds to a 10% probability value. The decoding process starts from the header of the data stream, parsing bit by bit to generate a decoded data stream containing gridded probability values ​​and coordinates.

[0199] It should be noted that if data loss occurs during transmission, such as if a certain grid probability value is not decoded correctly, it can be recovered using redundant check codes.

[0200] Redundancy check codes record data characteristics by adding extra check bits during encoding. For example, if a check code detects missing data at coordinates (8,2), it can estimate the missing value to be 70% based on the regularity of the probability values ​​of adjacent grids, thereby recovering the complete data stream.

[0201] Step S620: Based on the decoded data stream, generate a dynamic visualization interface using a vector graphics drawing tool, extract interface parameters through a preset template, and determine the interface parameter set.

[0202] In one possible implementation, the decoded data stream is used to generate a dynamic visualization interface.

[0203] Vector graphics drawing tools create gridded heatmaps by parsing probability values ​​and coordinates from a data stream. For example, in a 10x10 grid, the grid with a 75% probability is displayed in red, and the 10% in green, forming an intuitive distribution map of gas leak probability. Preset templates extract interface parameters, such as color mapping rules, a resolution of 1920x1080, and a scaling ratio of 1:1, forming a set of interface parameters.

[0204] Step S630: Obtain the display delay and refresh rate from the interface parameter set. If the display delay is higher than the preset threshold or the refresh rate is lower than the preset threshold, it is determined that the system reliability does not meet the preset standard, and the reliability judgment result is obtained.

[0205] The reliability assessment result is calculated using the following formula:

[0206] (10)

[0207] In formula (10), This indicates the result of the display delay reliability assessment. This indicates the display delay value of the current measurement. This represents a preset threshold for display latency. The result is 0 (unreliable) when the display latency exceeds the preset threshold, and 1 (reliable) otherwise.

[0208] Specifically, display latency and refresh rate in the interface parameter set directly affect user experience. Assume the display latency is 200ms with a preset threshold of 100ms; the refresh rate is 20Hz with a preset threshold of 30Hz. If both fail to meet the standards, it indicates insufficient system reliability, generating a reliability assessment result.

[0209] Step S640: Based on the reliability judgment result, extract real-time monitoring data from the pre-established feedback information database, adjust the interface parameters using a logistic regression algorithm, and generate an optimized dynamic visualization interface.

[0210] Preferably, based on the reliability assessment results, real-time monitoring data is extracted from the feedback information database, including historical latency, refresh rate, and user interaction records. For example, the database shows that the latency has repeatedly exceeded 150ms recently, and the refresh rate is often below 25Hz. A logistic regression algorithm analyzes this data to predict parameter adjustment directions, such as reducing the resolution to 1280x720 and optimizing color mapping complexity. After adjustment, the display latency is reduced to 80ms, the refresh rate is increased to 35Hz, and an optimized dynamic visualization interface is generated.

[0211] Understandably, this optimized interface can run smoothly on low-performance devices, and the grid heatmap clearly presents the probability of leakage, assisting monitoring personnel in making quick decisions.

[0212] The entire process, from decoding to interface optimization, forms a complete data processing chain, with each step closely linked to ensure the efficient presentation of monitoring data.

[0213] This invention relates to a remote monitoring system for flammable and explosive gases based on big data, used to implement the aforementioned remote monitoring method for flammable and explosive gases based on big data. The remote monitoring system includes an acquisition module, a preprocessing module, a generation module, a comparison module, a formation module, and a judgment module. The acquisition module acquires raw datasets collected in real time from multiple sensors, including gas concentration, temperature, and pressure data. The preprocessing module performs signal preprocessing on the raw dataset using a denoising algorithm, and obtains a first multi-parameter dataset through normalization. The generation module, based on the first multi-parameter dataset, uses a feature weighting algorithm to analyze the data. The system performs data fusion, generating a second fused feature set using principal component analysis. A comparison module compares the parameter values ​​of the second fused feature set with preset thresholds. If the parameter values ​​exceed the preset thresholds, a classification model is used to identify the gas leak event, and a diffusion model is combined to generate leak probability and range data. A formation module compresses and encodes the leak probability and range data, dynamically adjusting the transmission frequency and data packet size based on network status parameters to form a stable transmission stream. A judgment module decodes the stable transmission stream at the monitoring center, generating a dynamic visualization interface using a visualization engine, and judging whether the system reliability meets preset standards based on the interface parameters.

[0214] Furthermore, the remote monitoring system for flammable and explosive gases based on big data provided in this embodiment includes a first acquisition unit and a summarization unit in its acquisition module. The first acquisition unit is used to acquire gas concentration, temperature data and pressure data collected by multiple source sensors. The summarization unit is used to summarize the gas concentration, temperature data and pressure data collected by multiple source sensors to form an original dataset.

[0215] Preferably, the remote monitoring system for flammable and explosive gases based on big data provided in this embodiment includes a preprocessing module comprising a second acquisition unit, a third acquisition unit, a fourth acquisition unit, and a determination unit. The second acquisition unit is used to perform noise reduction processing on gas concentration, temperature, and pressure data from the raw dataset acquired by multiple source sensors using a mean filtering algorithm to obtain a first processed dataset. The third acquisition unit is used to map the numerical ranges of gas concentration, temperature, and pressure data using a minimum-maximum normalization method based on the first processed dataset to obtain a second processed dataset. The fourth acquisition unit is used to complete the parameter value by linear interpolation if any parameter value in the second processed dataset exceeds a preset threshold, to obtain a third processed dataset. The determination unit is used to determine a first multi-parameter dataset by performing a weighted average calculation on the gas concentration, temperature, and pressure data in the third processed dataset.

[0216] This embodiment provides a remote monitoring method and system for flammable and explosive gases based on big data. Compared with existing technologies, it collects raw data such as gas concentration, temperature, and pressure from multiple sources of sensors. After denoising and normalization preprocessing, it uses a feature weighting algorithm to fuse heterogeneous data and generates a fused feature set using principal component analysis. When the feature parameters exceed a preset threshold, it combines classification and diffusion models to identify leakage events and predict probability ranges. This embodiment compresses and encodes leakage data and dynamically adjusts transmission parameters according to network status to achieve stable transmission. After decoding, the monitoring center generates a dynamic interface through a visualization engine to evaluate system reliability. This embodiment achieves real-time monitoring, intelligent identification, and early warning of gas leaks, improving the accuracy and timeliness of hazardous gas leak monitoring and providing effective protection for industrial safety production.

[0217] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its spirit and scope. Thus, if these modifications and modifications of the invention fall within the scope of the claims and their equivalents, the invention is also intended to include these modifications and modifications.

Claims

1. A remote monitoring method for flammable and explosive gases based on big data, characterized in that, Includes the following steps: Acquire raw datasets collected in real time by multiple source sensors, the raw datasets including gas concentration, temperature data and pressure data; The original dataset is preprocessed using a denoising algorithm, and the first multi-parameter dataset is obtained through normalization. Based on the first multi-parameter dataset, a feature weighting algorithm is used to fuse the heterogeneous data, and a second fused feature set is generated through principal component analysis, specifically including: From the first multi-parameter dataset, the dimensions of gas concentration, temperature data and pressure data are unified by standardization processing, and the data is normalized by the Z-score method to obtain the first standardized dataset. Based on the first standardized dataset, the entropy method is used to weight the feature importance of gas concentration, temperature data and pressure data to obtain the first weighted feature set; If the weight value of any feature in the first weighted feature set is lower than a preset threshold, the feature is filled in by the mean imputation method to obtain the second weighted feature set. Principal component analysis algorithm is used to reduce the dimensionality of the features in the second weighted feature set, extract the first few principal components, and generate the second fusion feature set. The parameter values ​​of the second fused feature set are compared with a preset threshold. If the parameter values ​​of the second fused feature set exceed the preset threshold, a classification model is used to identify the gas leak event, and a leakage probability and range data are generated by combining the diffusion model. Specifically, this includes: Parameter values ​​are obtained from the second fusion feature set, and compared with a preset threshold to determine whether the parameter values ​​exceed the threshold, thus obtaining the comparison result. If the comparison result indicates that the parameter value exceeds the preset threshold, a pre-trained classification model is used to identify the gas leak event and generate a leak event label. Based on the leak event label, the probability distribution of gas leaks is calculated using a diffusion model to obtain leak probability distribution data; High-probability areas are extracted from the leakage probability distribution data, and the boundary of the impact range is determined by a grid division method to generate impact range data. For the aforementioned impact range data, a spatial interpolation algorithm is used to estimate the probability of the uncovered areas, resulting in extended probability distribution data. Based on the extended probability distribution data and the impact range data, a weighted superposition method is used to generate a comprehensive leakage risk distribution, resulting in risk distribution data. Key risk areas are extracted from the risk distribution data, and high-risk points are grouped using a clustering algorithm to generate risk partition data. The leakage probability and range data are compressed and encoded, and the transmission frequency and data packet size are dynamically adjusted according to network status parameters to form a stable transmission stream; The stable transmission stream is decoded at the monitoring center, and a dynamic visualization interface is generated using a visualization engine. The system reliability is then judged based on the interface parameters to determine whether it meets the preset standards.

2. The remote monitoring method for flammable and explosive gases based on big data as described in claim 1, characterized in that, The steps for obtaining the raw dataset acquired in real time from multiple sensors include: Acquire gas concentration, temperature, and pressure data collected by multiple source sensors; The gas concentration, temperature, and pressure data collected by the multi-source sensors are summarized to form the original dataset.

3. The remote monitoring method for flammable and explosive gases based on big data as described in claim 1, characterized in that, The step of preprocessing the original dataset using a denoising algorithm and obtaining the first multi-parameter dataset through normalization includes: The gas concentration, temperature, and pressure data are denoised using a mean filtering algorithm from the raw dataset acquired from multiple sensors to obtain the first processed dataset. Based on the first processed dataset, the numerical ranges of the gas concentration, temperature data, and pressure data are mapped using the min-max normalization method to obtain the second processed dataset. If any parameter value in the second processing dataset exceeds a preset threshold, the parameter value is completed by linear interpolation to obtain the third processing dataset. The first multi-parameter dataset is determined by performing a weighted average calculation on the gas concentration, temperature, and pressure data in the third processed dataset.

4. The remote monitoring method for flammable and explosive gases based on big data as described in claim 1, characterized in that, The steps of compressing and encoding the leakage probability and range data, and dynamically adjusting the transmission frequency and data packet size according to network status parameters to form a stable transmission stream include: The original dataset is obtained from the leakage probability and range data, and the original dataset is compressed and encoded using the Huffman coding algorithm to generate a compressed data stream. Based on the network status parameters, determine whether the bandwidth in the network status parameters is lower than a preset threshold. If it is lower than the preset threshold, adjust the data packet size and transmission frequency to obtain the adjusted transmission parameters. By adjusting the transmission parameters, the compressed data stream is divided into multiple data packets, and the data packets are transmitted using a sliding window protocol to generate a stable transmission stream. Extract the receiver feedback information from the stable transmission stream, determine whether the data integrity in the receiver feedback information meets the preset standard, and if not, readjust the transmission frequency to determine the optimized transmission stream.

5. The remote monitoring method for flammable and explosive gases based on big data as described in claim 1, characterized in that, The steps of decoding the stable transmission stream at the monitoring center, generating a dynamic visualization interface using a visualization engine, and determining whether the system reliability meets preset standards based on the interface parameters include: The stable transmission stream is obtained from the monitoring center, and the stable transmission stream is decoded using the Huffman decoding algorithm. If data loss is detected, it is recovered through the redundancy check code to obtain the decoded data stream. Based on the decoded data stream, a vector graphics drawing tool is used to generate the dynamic visualization interface, and interface parameters are extracted through a preset template to determine the interface parameter set; The display latency and refresh rate are obtained from the set of interface parameters. If the display latency is higher than a preset threshold or the refresh rate is lower than a preset threshold, the system reliability is determined to be unsatisfactory, and a reliability judgment result is obtained. Based on the reliability assessment results, real-time monitoring data is extracted from a pre-established feedback information database, and the interface parameters are adjusted using a logistic regression algorithm to generate an optimized dynamic visualization interface.

6. A remote monitoring system for flammable and explosive gases based on big data, used to implement the remote monitoring method for flammable and explosive gases based on big data as described in any one of claims 1 to 5, characterized in that, The aforementioned remote monitoring system for flammable and explosive gases based on big data includes: The acquisition module is used to acquire the raw dataset collected in real time by the multi-source sensors. The raw dataset includes gas concentration, temperature data and pressure data. The preprocessing module is used to preprocess the original dataset using a denoising algorithm and obtain the first multi-parameter dataset through normalization. The generation module is used to perform data fusion on heterogeneous data using a feature weighting algorithm based on the first multi-parameter dataset, and to generate a second fused feature set using a principal component analysis algorithm. The comparison module is used to compare the parameter values ​​of the second fusion feature set with a preset threshold. If the parameter values ​​of the second fusion feature set exceed the preset threshold, a classification model is used to identify the gas leak event, and a diffusion model is combined to generate leak probability and range data. A forming module is used to compress and encode the leakage probability and range data, and dynamically adjust the transmission frequency and data packet size according to network status parameters to form a stable transmission stream. The judgment module is used to decode the stable transmission stream in the monitoring center, generate a dynamic visualization interface using a visualization engine, and judge whether the system reliability meets the preset standards based on the interface parameters.

7. The remote monitoring system for flammable and explosive gases based on big data as described in claim 6, characterized in that, The acquisition module includes: The first acquisition unit is used to acquire gas concentration, temperature and pressure data collected by the multi-source sensors; The aggregation unit is used to aggregate the gas concentration, temperature and pressure data collected by the multi-source sensors to form a raw dataset.

8. The remote monitoring system for flammable and explosive gases based on big data as described in claim 6, characterized in that, The preprocessing module includes: The second acquisition unit is used to perform noise reduction processing on the gas concentration, temperature data and pressure data from the raw dataset acquired by the multi-source sensors using a mean filtering algorithm to obtain a first processed dataset. The third acquisition unit is used to map the numerical ranges of the gas concentration, temperature data and pressure data using the minimum-maximum normalization method based on the first processing dataset to obtain the second processing dataset; The fourth acquisition unit is used to complete the parameter value by linear interpolation if any parameter value in the second processing dataset exceeds a preset threshold, thereby obtaining a third processing dataset. The determining unit is used to determine the first multi-parameter dataset by performing a weighted average calculation on the gas concentration, temperature data and pressure data in the third processed dataset.

Images