Laboratory safety state evaluation method and system based on multi-source data fusion

Abstract

The invention discloses a laboratory safety state evaluation method and system based on multi-source data fusion, and relates to the field of laboratory safety management.The method comprises the steps that multi-source sensing data of a laboratory are collected and aligned to a unified time axis, and a laboratory sensing data matrix is constructed; grey correlation analysis is introduced, the risk correlation degree between data of each sensor and a preset risk mode is calculated, and identification of a potential risk mode is achieved; the reliability weight is dynamically determined according to the historical stability, baseline deviation and calibration period of the sensor so as to suppress data distortion caused by aging and drifting; and comprehensively calculating the risk correlation degree, the reliability weight and the risk severity coefficient to form weighted risk energy, and generating laboratory risk field energy through space-time coupling superposition so as to calculate a corresponding laboratory risk state level. Therefore, the dynamic quantification and grading judgment of the overall safety state of the laboratory are realized, and the risk identification capability of the complex laboratory environment is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of laboratory safety management, and particularly relates to a laboratory safety state evaluation method and system based on multi-source data fusion. BACKGROUND

[0002] Modern laboratories, especially chemical, biological and material science laboratories, usually have multiple risk sources such as toxic and harmful chemicals, flammable and explosive substances, precision instruments and personnel activities. In recent years, with the development of Internet of Things (IoT) and industrial internet technology, more and more hardware and software sensors are applied to the monitoring of industrial processes and laboratory environments, so as to master key variables in real time to improve product quality and ensure process safety.

[0003] In chemical production processes, multiple sensors monitor variables such as temperature, pressure and component concentration at the same time, but a single sensor is easily affected by various factors and prone to errors. Such problems also exist in laboratories, for example, temperature sensors are easily affected by air flow, gas sensors have decreased sensitivity due to aging, and personnel operation records are not timely.

[0004] At present, in the laboratory hazard monitoring system based on IoT, gas, smoke, temperature and humidity sensors are used to detect risks in real time and alarm when the threshold is exceeded. However, such systems still rely on simple threshold judgment, and it is difficult to reliably determine the current safety state of the laboratory with only a single data source, and the correlation and relative reliability between multi-source sensor data are not analyzed in depth. SUMMARY

[0005] The present application provides a laboratory safety state evaluation method and system based on multi-source data fusion, a storage medium, a computer program product and an electronic device, to at least solve the problem that the laboratory safety monitoring in the prior art relies on single sensor data, the risk judgment accuracy is low, and the complex environment state cannot be comprehensively reflected.

[0006] In a first aspect, the embodiments of the present application provide a laboratory safety state evaluation method based on multi-source data fusion, which comprises: collecting multi-source sensor data corresponding to a laboratory environment based on a sensor module, and constructing a laboratory sensor data matrix under a unified time axis according to the multi-source sensor data; processing the laboratory sensor data matrix by a grey correlation analysis method to calculate a risk correlation degree corresponding to each sensor; the risk correlation degree is used to indicate the closeness between a current sensor sampling feature and a preset dangerous mode feature; obtaining baseline deviation information corresponding to each sensor, and determining a reliability weight corresponding to each sensor according to a historical stability of each sensor, the baseline deviation information, and a calibration period; the historical stability is defined as a standard deviation of data collected by a sensor in a historical normal state, and the baseline deviation information is defined as a deviation degree of a current measurement value of the sensor relative to a baseline mean value of a time period; multiplying the risk correlation degree of each sensor by the reliability weight and a risk severity coefficient to obtain a weighted risk energy of each sensor; each sensor has a unique corresponding risk severity coefficient; performing spatiotemporal coupling superposition on the weighted risk energies of all sensors to generate a laboratory risk field energy, and calculating a corresponding laboratory risk state level according to the laboratory risk field energy.

[0007] In a second aspect, the embodiments of the present application provide a laboratory safety state evaluation system based on multi-source data fusion, which comprises: a data acquisition unit configured to collect multi-source sensor data corresponding to a laboratory environment based on a sensor module, and construct a laboratory sensor data matrix under a unified time axis according to the multi-source sensor data; an associated risk analysis unit configured to process the laboratory sensor data matrix by a grey correlation analysis method to calculate a risk correlation degree corresponding to each sensor; the risk correlation degree is used to indicate the closeness between a current sensor sampling feature and a preset dangerous mode feature; a sensor reliability analysis unit configured to obtain baseline deviation information corresponding to each sensor, and determine a reliability weight corresponding to each sensor according to a historical stability of each sensor, the baseline deviation information, and a calibration period; the historical stability is defined as a standard deviation of data collected by a sensor in a historical normal state, and the baseline deviation information is defined as a deviation degree of a current measurement value of the sensor relative to a baseline mean value of a time period; a sensor risk energy analysis unit configured to multiply the risk correlation degree of each sensor by the reliability weight and a risk severity coefficient to obtain a weighted risk energy of each sensor; each sensor has a unique corresponding risk severity coefficient; a laboratory risk field energy generation unit configured to perform spatiotemporal coupling superposition on the weighted risk energies of all sensors to generate a laboratory risk field energy, and calculate a corresponding laboratory risk state level according to the laboratory risk field energy.

[0008] In a third aspect, an electronic device is provided, which includes at least one processor, and a memory connected with the at least one processor in communication, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the steps of the laboratory safety state evaluation method based on multi-source data fusion of any of the embodiments of the present application.

[0009] In a fourth aspect, the embodiments of the present application provide a storage medium having a computer program stored thereon, characterized in that the program, when executed by a processor, implements the steps of the laboratory safety state evaluation method based on multi-source data fusion of any of the embodiments of the present application.

[0010] In a fifth aspect, the embodiments of the present application provide a computer program product, which includes computer programs / instructions, and the computer programs / instructions, when executed by a processor, implement the steps of the laboratory safety state evaluation method based on multi-source data fusion of any of the embodiments of the present application.

[0011] Through the laboratory safety state evaluation method and system based on multi-source data fusion provided by the present application, at least the following technical effects can be achieved:

[0012] (1) By constructing a multi-source sensing data matrix under a unified time axis, the coordinated expression of environmental, equipment and personnel behavior data is realized, so that the dynamic relationship between different physical quantities can be quantitatively presented in the same time sequence framework. The gray correlation analysis method is introduced to calculate the relative change trend between multi-dimensional sensing features, and the correlation degree of each sensor and potential risk mode is obtained, so that the system is no longer dependent on a single threshold trigger, but makes a comprehensive judgment based on multi-dimensional trend coupling, so as to capture potential abnormal situations before the risk is explicit, and realize the change from passive alarm to active prediction. Therefore, the sensitivity and forward-looking of risk identification are improved, and the accuracy and continuity of the overall safety state evaluation of the laboratory are also significantly improved.

[0013] (2) By comprehensively considering the historical stability, baseline deviation information and calibration period of the sensor, the reliability weight is dynamically calculated, so that different sensors have different influence when contributing risk information, and the weighted mechanism is introduced to automatically reduce the interference degree of the sensors drifting, aging or abnormally signaling, and to maintain the stable response of the evaluation model to the actual risk change. In addition, the risk correlation degree, reliability weight and risk severity coefficient are fused and calculated to form a weighted risk energy, and time-space superposition is realized in the spatial dimension to construct a risk energy distribution field of the laboratory. Through the energy field model, the risk information is converted into a visualized and quantifiable continuous distribution feature, which can intuitively identify local high-risk areas and realize spatial analysis and fine management of the risk structure of the laboratory.

[0014] By the technical solution, multi-source heterogeneous sensor data fusion, trend correlation analysis and reliability weighting mechanism are organically combined to establish an evolvable risk energy field expression mechanism, which breaks through the discrete and single-point limitation of traditional monitoring methods, and makes the laboratory safety assessment form a closed loop from real-time fusion of multi-dimensional data, trend perception to spatial energy mapping. Therefore, dynamic quantification and hierarchical judgment of the overall safety state of the laboratory are realized, and the risk identification capability for complex laboratory environment is significantly improved. BRIEF DESCRIPTION OF DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, a brief introduction will be given below to the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0016] Figure 1 A flowchart of an example of a laboratory safety state evaluation method based on multi-source data fusion according to an embodiment of the present application is shown;

[0017] Figure 2 An operation flowchart of an example of calculating risk correlation degree according to an embodiment of the present application is shown;

[0018] Figure 3 An operation flowchart of an example of calculating reliability weight of a sensor according to an embodiment of the present application is shown;

[0019] Figure 4 An operation flowchart of an example of generating laboratory risk field energy according to an embodiment of the present application is shown;

[0020] Figure 5 An operation flowchart of an example of determining laboratory risk state level according to laboratory risk field energy according to an embodiment of the present application is shown;

[0021] Figure 6 Comparison results of risk score-time curves of different algorithms in a simulation environment are shown;

[0022] Figure 7 Performance comparison results of different algorithms in main risk determination indicators are shown;

[0023] Figure 8 A structural block diagram of an example of a laboratory safety state evaluation system based on multi-source data fusion according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0024] To make the purposes, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some but not all of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0025] It should be noted that multi-sensor data fusion technology has been widely studied in the fields of remote sensing images, aerospace and industrial process monitoring, etc. Algorithms can be divided into four types according to the fusion level, i.e. signal level, pixel level, feature level and decision level. Signal level fusion directly combines the original signals of different sensors to improve the signal-to-noise ratio; pixel level fusion fuses multispectral information at the pixel level of the image; feature level fusion extracts features of different source data and fuses them; decision level fusion makes a comprehensive decision based on independent decisions. However, these classic methods are mostly for visual images, and lack of consideration for multi-modal data (such as time series, event logs, device status, etc.) in laboratory environment. In addition, standard image fusion algorithms require high registration accuracy of input data, and are prone to introduce spectral distortion.

[0026] In recent years, some scholars and researchers have proposed laboratory safety systems based on IoT, such as the document “Laboratory Safety System using IOT”, which uses flame, gas and personnel detectors to build a smart laboratory, and can send an alarm to the responsible person when the measured parameters exceed the set value. However, this system mainly targets single sensor over-limit situations, does not consider the superposition of multiple risk factors or the correlation between sensor data, and does not provide a quantitative risk assessment model. Another study “IoT-Based Real-Time Monitoring System for Laboratory Hazards” proposes to use gas, smoke, temperature and humidity sensors combined with a cloud platform for real-time monitoring and notification. This document acknowledges that traditional manual supervision methods are inefficient and prone to error, but its data processing still mainly relies on setting thresholds, lacks modeling of sensor data quality differences, and does not consider the impact of laboratory daily activity patterns on thresholds.

[0027] In addition, the theoretical research of multi-sensor fusion in academia is mostly represented by Dempster-Shafer (DS) evidence theory, which allows the fusion of multiple uncertain information and obtains the belief distribution. However, studies have shown that when conflicting evidence is caused by sensor failure, traditional improved methods are difficult to remove false information, and in high conflict situations, it will lead to unreliable combination results. Laboratory environment sensors may have failures such as aging and drift, and if the sensor reliability is not distinguished, only relying on the DS algorithm will misjudge or fail to identify abnormalities.

[0028] More specifically, the system proposed in "IoT-Based Real-Time Monitoring System for Laboratory Hazards" uses gas, smoke, temperature and humidity sensors to collect environmental data, and sends data to the cloud platform in real time through ESP32 or NodeMCU, and sends an alarm to the mobile application or email when the threshold is exceeded. It points out that traditional manual inspection is inefficient and prone to missed detection, and that continuous collection and historical data logs can improve maintenance strategies. However, the fusion algorithm of this system is still a simple threshold fusion, although the paper mentions that "mathematical models support hazard threshold setting, sensor data fusion, anomaly detection, and probability estimation", but does not give specific algorithms or data quality processing methods. Experiments show that this system based on fixed thresholds has certain response ability to sudden events, but has weak fault tolerance to multi-factor complex scenarios, sensor failures and interference.

[0029] For chemical process monitoring, the paper "Monitoring Chemical Processes Using Judicious Fusion of Multi-Rate Sensor Data" proposes a multi-rate sensor fusion scheme based on Bayesian inference to improve the accuracy and reliability of process monitoring. The paper points out that laboratory analyzers have low sampling frequency but high precision, while online analyzers have high sampling frequency but limited accuracy. This study combines the measurement values of laboratory analyzers, online analyzers and soft sensors to achieve state estimation using Kalman filtering. In this method, it can effectively fuse multi-rate data, but assumes that the sampling rate of each sensor is fixed and reliable, while the sampling frequency and accuracy of actual laboratory sensors change with the environment; and the fusion algorithm is mainly aimed at variable estimation problems, without considering risk assessment and alarm requirements; in addition, when there is a sensor failure or drift, the Bayesian method is difficult to detect and remove abnormal data, which may lead to false estimates.

[0030] Data fusion (DS) evidence theory has advantages in solving multi-sensor data fusion problems, handling uncertain and incomplete information. However, researchers have found that standard DS fusion yields unreasonable results in cases of strong conflicts caused by sensor failures. To overcome this problem, scholars have proposed improved combination rules or preprocessing methods, such as weighted averaging based on evidence distance or entropy weighting. However, these improvements mainly weight conflicts at the combination rule level, rarely considering the interaction structure between multi-source data and lacking business implications specific to the scenario. In laboratory security scenarios, sensor failures often manifest as continuous offsets or signal jamming, which simple DS fusion struggles to identify.

[0031] It should be understood that the above description of the relevant technologies is intended only to help the public better understand the inventive spirit and motivation of this application, and is not intended to limit this application. Furthermore, the technical solutions described in the above-mentioned relevant technologies are not prior art, and may also be undisclosed technical solutions, such as those under research or in the laboratory stage.

[0032] The technical solutions in this application, including the collection, storage, use, processing, transmission, provision, and disclosure of users' personal information, comply with relevant laws and regulations and do not violate public order and good morals.

[0033] Figure 1 A flowchart illustrating an example of a laboratory safety status assessment method based on multi-source data fusion according to an embodiment of this application is shown.

[0034] Regarding the execution subject of the method in the embodiments of this application, it can be any controller or processor with computing or processing capabilities, such as a laboratory risk early warning platform controller, which transforms laboratory safety assessment from a single threshold judgment to a multi-dimensional and interpretable comprehensive assessment system, realizing collaborative judgment of multi-source information and dynamic risk perception of complex laboratory environments.

[0035] In some examples, it may be integrated into an electronic device or terminal through software, hardware, or a combination of both, and the type of terminal or electronic device may be diverse.

[0036] like Figure 1 As shown, in step S110, multi-source sensor data corresponding to the laboratory environment is collected based on the sensor module, and a laboratory sensor data matrix under a unified time axis is constructed based on the multi-source sensor data.

[0037] In some embodiments, multi-source data is synchronously collected based on sensor modules deployed in a laboratory environment, including environmental sensors, equipment state sensors, and personnel behavior sensors. Specifically, the laboratory macro-environmental state is collected by environmental sensors (such as temperature and humidity, gas concentration, smoke, and illumination sensors), the running state and energy consumption characteristics of key instruments are monitored by equipment state sensors (such as current, voltage, vibration, and pressure sensors), and the spatial distribution and activity frequency of operating behavior are perceived by personnel behavior sensors (such as infrared induction, ultrasonic ranging, access control recording, or video behavior analysis nodes).

[0038] In addition, due to the differences in sampling frequency and clock offset of different sensors, the system aligns various data streams on a unified time reference through timestamp synchronization and interpolation resampling algorithms to form a multi-dimensional time series matrix. Exemplarily, the laboratory sensor data matrix maintains equal interval sampling in the time dimension and contains multi-source sensor variables in the feature dimension, achieving a time series expression of the full state of the laboratory and avoiding misjudgments caused by sampling delay or asynchronization.

[0039] In some examples of embodiments of the present application, data preprocessing operations can also be performed before constructing the laboratory sensor data matrix. Specifically, time synchronization and interpolation completion are performed on sensor data of different sampling frequencies. It should be noted that the sampling period of environmental temperature and humidity sensors can reach several seconds, and electrical power monitoring or infrared action detection modules can sample at a millisecond level, and inconsistent sampling frequencies will cause alignment deviation of multi-source data in the time dimension. In some embodiments, the original sampling timestamp sequence of each type of sensor can be obtained, and a unified time axis is constructed at an integer multiple of the global minimum sampling period as the time reference of the laboratory sensor data matrix, ensuring consistent time scales.

[0040] In addition, a sliding window filtering algorithm is used to smooth high-frequency sensor data to remove transient noise interference. It should be noted that in the sampling process of high-frequency sensors (such as current, voltage, or noise intensity sensors), they are often affected by transient interference, electromagnetic noise, or human operation disturbances, resulting in spikes or mutations in the measured data. In some embodiments, the sensor data stream can be scanned in a sliding window manner on the time axis, and mean filtering, median filtering, or weighted moving average algorithms can be applied to the data sequence within the window to smooth local fluctuations and effectively weaken high-frequency noise caused by short-term disturbances.

[0041] In addition, when missing sensor data is detected, statistical reconstruction is performed using observation values of adjacent time periods and associated sensors to complete the missing samples, thereby ensuring the continuity and integrity of the laboratory sensor data matrix under the unified time axis. It should be noted that the laboratory monitoring equipment can have temporary communication interruptions, sensor aging, or network packet loss, etc., resulting in missing samples in the data stream. In some embodiments, missing sample points are detected based on the unified time axis, and the missing positions are marked; then, statistical reconstruction is performed based on the observation values of the same type or spatially adjacent sensors, for example, when certain temperature sensor data is missing, the average trend and spatial correlation of other temperature sensors in the same region can be used for compensation, thereby ensuring the time continuity and spatial integrity of the laboratory sensor data matrix.

[0042] In step S120, the laboratory sensor data matrix is processed by the grey correlation analysis method to calculate the risk correlation degree corresponding to each sensor, which is used to indicate the closeness between the current sensor sampling feature and the preset dangerous mode feature.

[0043] In some embodiments, after obtaining the multi-source data matrix under the unified time axis, the system calls the grey correlation analysis method (GRA) to perform correlation calculation on the laboratory multi-dimensional time sequence features. Specifically, the preset dangerous mode feature is used as the reference sequence, representing the feature change trajectory under the typical dangerous state (for example, the temperature rise-gas concentration rise-current fluctuation linkage mode); the real-time sampling sequence of each sensor is used as the comparison sequence, and then the relative closeness between the two is calculated based on the GRA model, thereby capturing the trend similarity between the preset dangerous mode feature. The laboratory sensor data matrix is analyzed by the grey correlation analysis method to quantitatively measure the closeness between the change trajectory of each sensor data and the known dangerous mode feature, thereby obtaining the risk correlation degree of each sensor.

[0044] It should be noted that the grey correlation analysis has robustness to small samples and uncertain data, and can adapt to the situation of incomplete data, noise interference and local anomalies in the laboratory scene.

[0045] In the embodiments of the present application, the closeness between the current environment state and the known dangerous mode is identified by the grey correlation analysis, instead of simply relying on the numerical threshold trigger mechanism, so that the risk assessment has the ability of pattern recognition and trend prediction. Even in the case of partial data anomaly or missing measurement, the overall risk change trend can still be stably reflected.

[0046] In step S130, the baseline deviation information corresponding to each sensor is obtained, and the reliable degree weight corresponding to each sensor is determined according to the historical stability, baseline deviation information and calibration period of each sensor.

[0047] It should be noted that different sensors will be affected by environmental drift, component aging, maintenance delay and other factors in long-term operation, thereby leading to inconsistent reliability of measurement data. To prevent data of low reliability sensors from excessively affecting the result in overall evaluation, the reliability weight of each sensor is dynamically determined by comprehensive calculation of its historical stability, current baseline deviation information and calibration period.

[0048] Here, the historical stability is defined as the standard deviation of data collected by the sensor in the historical normal state, and the baseline deviation information is defined as the deviation degree of the current measurement value of the sensor relative to the baseline mean value of the period. In some examples, the historical stability factor can be calculated by analyzing the output standard deviation of the sensor in the historical normal state, which reflects the degree of signal fluctuation, and the smaller the standard deviation, the higher the stability; the baseline deviation factor can be the deviation degree of the current period measurement value relative to the baseline mean value of the same period, and the greater the deviation amplitude, the more likely the sensor state is abnormal; the calibration period factor can be time decay adjustment of reliability according to the running time from the last calibration to the present, and the device not calibrated for a long time automatically reduces the corresponding reliability weight. Further, by integrating the above three factors, for example, by a nonlinear weighting function or a neural network process, the reliability weight of each sensor is dynamically determined, which can adaptively reflect the change of sensor health status.

[0049] Therefore, by introducing a dynamic reliability mechanism, the influence of abnormal sensors can be reduced, preventing a single abnormal sensor from causing distortion of the overall judgment, maintaining the stability and accuracy of the risk analysis model in long-term operation, and having self-adaptive suppression ability for sensor drift, aging and noise interference.

[0050] In step S140, the risk correlation degree of each sensor is multiplied by the reliability weight and the risk severity coefficient to obtain the weighted risk energy of each sensor.

[0051] Here, each sensor has a unique corresponding risk severity coefficient, and the risk severity coefficient represents the potential harm level that may be caused by the abnormality of the physical quantity monitored by the sensor, for example, the severity coefficient of the combustible gas leakage sensor is higher than that of the temperature and humidity sensor.

[0052] Specifically, the risk correlation degree, reliability weight and risk severity coefficient of each sensor are multiplied and fused to obtain the corresponding weighted risk energy, realizing the weighted fusion of multi-source risk information under the unified measurement system, and making the risk contributions of different types comparable. In addition, the calculation process can adopt a time window aggregation form, and the weighted risk energy is smoothed by a sliding time window to filter out transient interference.

[0053] By calculating the weighted risk energy, the system can measure the intensity and credibility of multiple risk sources at the same time, and realize the risk analysis from the dispersion monitoring of heterogeneous indicators to centralized evaluation by using the unified scale of energy measurement.

[0054] In step S150, the weighted risk energy of all sensors is spatiotemporally coupled and superimposed to generate a laboratory risk field energy, and a corresponding laboratory risk state level is calculated according to the laboratory risk field energy.

[0055] In some embodiments, according to the laboratory space layout information (such as sensor distribution coordinates, equipment area division) and the sampling time sequence, the weighted risk energy of each sensor is interpolated and convolved to form a two-dimensional or three-dimensional risk field distribution map. Then, by calculating the total energy or energy gradient of the risk field in a given time window, the overall risk intensity index of the laboratory is extracted. Further, according to the preset grading standard (such as quantile threshold or adaptive threshold based on historical data), the risk field energy is converted into a risk state level (such as "safe", "warning", "high risk", etc.), and output to the monitoring interface or the upper control system.

[0056] Here, the weighted risk energy of all sensors is spatiotemporally coupled and superimposed according to its spatial coordinate information to generate a risk field energy distribution of the laboratory. Exemplarily, a two-dimensional plane or a three-dimensional structure of the laboratory can be taken as a reference space, and a heat diffusion model or a Gaussian kernel function can be used to diffuse and superimpose the risk energy of each sensor in its neighborhood range to form a continuous energy distribution field. The risk field reflects the distribution density of risk in space and is dynamically updated through a sliding window in time. Further, the overall risk state level of the laboratory can be divided according to the total energy and local energy gradient change of the risk field to quantify the overall security situation.

[0057] As a preferred embodiment of the present application, a two-dimensional plane can be used as a reference space of the laboratory to reduce system resource consumption and state monitoring timeliness. Specifically, although the laboratory space essentially has a three-dimensional structure, considering that the distribution characteristics of most laboratory sensors and risk diffusion characteristics are mainly concentrated in the same working plane (such as the ground layer or the operation table plane), various environmental sensors (temperature and humidity, gas concentration), equipment state sensors, and personnel behavior sensors are mostly installed in a similar height range, and the difference in data change in the vertical direction (Z-axis) is much smaller than the distribution difference in the plane (X-Y direction). Therefore, when modeling the risk field, using a two-dimensional plane as a reference space for calculation is more reasonable and stable in engineering. The heat diffusion model or Gaussian kernel function based on the two-dimensional reference space can fully describe the diffusion and superposition process of risk energy in the laboratory plane, realize efficient spatiotemporal coupling calculation, and significantly reduce the modeling complexity and real-time operation load.

[0058] Figure 2 An operation flowchart of an example of calculating a risk correlation degree according to an embodiment of the present application is shown.

[0059] As shown in step S210, for each sensor, a dangerous reference feature sequence corresponding to a dangerous mode is defined. Figure 2

[0060] In the laboratory safety monitoring scenario, in order to measure the closeness of the current sampling state of each sensor to the potential dangerous state, a reference feature sequence representing a typical dangerous mode of the laboratory needs to be established .

[0061] In some embodiments, according to historical safety events or laboratory operation anomaly records, time series data of several key indicators (such as temperature rise rate, oxygen concentration drop, equipment power surge, smoke concentration rise, etc.) are extracted from the database; the sensor data of these abnormal periods are normalized and pattern clustered to extract common time series trends; and then, the mode sequence with the most significant trend is taken as the dangerous mode reference feature sequence , which can reflect the multi-dimensional time series features in the typical dangerous evolution process.

[0062] Thus, a unified time series reference for risk determination is established, so that the data of different sensors can be compared in the same feature space; in addition, by introducing the dangerous mode reference sequence, the grey correlation analysis is no longer dependent on the absolute value difference, but judges the risk similarity according to the change trend, enhancing the sensitivity and universality of anomaly identification.

[0063] In step S220, the grey correlation coefficient between the current sensor sampling sequence indicated by the laboratory sensor data matrix and the dangerous reference feature sequence is calculated.

[0064] Here, to depict the similarity of each sensor sampling sequence relative to the dangerous reference sequence, the grey correlation coefficient model in the grey system theory is used for calculation. For each sensor in the laboratory sensor data matrix, the time series difference between the sensor sampling sequence and the dangerous mode sequence is calculated within a sliding time window , and then the grey correlation coefficient formula is used:

[0065] , formula (1)

[0066] The numerator part in the right side of formula (1) is the weighted combination of the global minimum difference and the maximum difference, and the denominator is the actual difference between the current sensor and the reference mode. When is smaller, that is, the sensor sampling trend is closer to the dangerous reference sequence, the corresponding​ The closer the value is to 1, the better; conversely, if the difference is large, The value is close to 0. Therefore, it can effectively reflect the dynamic similarity of different sensors over time series.

[0067] Unlike traditional correlation analysis based on mean squared error, the grey relational analysis method does not require assumptions about data distribution characteristics and can handle common nonlinear, short-sequence, and small-sample data in laboratories. This allows the system to identify potential risk trends in the early stages, rather than only issuing an alarm after the anomaly has fully manifested, thus significantly improving the foresight of early warning.

[0068] In step S230, the mean value of the grey relational coefficient is calculated within the time window to obtain the risk relational degree.

[0069] Because the laboratory environment is highly volatile, transient anomalies may originate from temporary disturbances (such as opening doors for ventilation, starting up equipment, etc.). Therefore, by smoothing the risk characteristics within a time window, a stable risk measurement can be obtained.

[0070] Specifically, in the sliding time window (length is...) Within this range, the mean value of the grey relational coefficient for each sensor is calculated:

[0071] Equation (2)

[0072] In the formula, and For sensor indexing, , This represents the total number of sensors in the sensor module. and For discrete-time indexes within a time window, where and , The window length is the time window length. and These are the reference sequences for the dangerous mode at time [time]. and time The value, For the first Each sensor at time The sampled values, For the resolution coefficient, satisfying ; For the first Each sensor at time The sampled values; For the first Each sensor at time The grey relational coefficient; For the first Risk correlation of individual sensors.

[0073] In formula (2), through the mean value calculation process, the short-term fluctuation signal (noise) is smoothed, and the persistent deviation trend is amplified, effectively filtering out the interference of transient disturbance on risk judgment, and retaining the risk information reflecting the systematic change trend.

[0074] In some embodiments, the window length can also be dynamically adjusted according to the sensor sampling frequency and the laboratory response characteristics: for sensors with fast response speed (such as electrical equipment current, voltage), a shorter window is used to quickly capture risks; for environmental variables (such as temperature and humidity, gas concentration), a longer window is used to stabilize the calculation results.

[0075] In some examples of the embodiments of the present application, the dangerous reference feature sequence can also be used to describe the semantics of the dangerous experimental situation, so that the system has situation awareness capability. Specifically, the dangerous reference feature sequence not only serves as a benchmark for numerical feature patterns, but also extends to a dangerous situation vector model with semantic level description capability. In the training stage, historical laboratory accident reports, sensor abnormal data and operation logs are jointly encoded, and the dangerous event is abstracted as a semantic label (such as "solvent leakage", "high temperature overload", "combustible gas diffusion", etc.), and is mapped into the feature space in a multi-modal manner. Thus, the dangerous reference feature sequence can accommodate both numerical trends and semantic situations, so that the risk judgment process is extended from simple feature similarity to situation understanding and reasoning.

[0076] As a further preferred embodiment, in the case of detecting that the current sensor state has a high similarity to a certain semantic dangerous pattern, the system not only outputs the risk correlation degree score, but also triggers the situation deduction operation. Specifically, through the predefined causal relationship chain in the knowledge graph, such as "solvent leakage → vapor accumulation → electric spark ignition → deflagration risk", the potential chain reaction process is automatically deduced. In this process, by dynamically querying the state information of other sensor nodes (such as temperature, humidity, current, smoke concentration, etc.), it is verified whether there are signs consistent with the deduced path, thereby improving the prediction accuracy of complex risks.

[0077] For example, when detecting the dangerous pattern of flammable solvent leakage, the secondary risk sources that may be related to it will be automatically deduced, and it is checked whether there are high-temperature equipment or live devices in the adjacent area. If such potential trigger sources are detected, the vapor cloud diffusion range can be further estimated based on the air flow model and the Gaussian diffusion function, and the affected area is marked in real time on the laboratory floor plan. Thus, the administrator can intuitively see the risk propagation trend on the interface, so as to take disposal measures such as isolation, cooling or power cut in advance.

[0078] By introducing the aforementioned semantic inference and multi-layer perception mechanisms, the function of the hazard reference feature sequence is transformed from passive pattern matching to proactive risk evolution prediction. This not only enhances the system's contextual awareness and adaptive decision-making capabilities but also achieves a leap from "event detection" to "risk reasoning." Ultimately, the laboratory safety monitoring system can identify risk sources, predict risk chains, and locate key triggering nodes at an early stage, thereby achieving more intelligent laboratory safety control.

[0079] It should be noted that the risk correlation values ​​of each sensor... This reflects the degree of proximity between the monitoring dimensions represented by the sensor (temperature, humidity, gas, equipment power, behavioral activity, etc.) and potential hazard patterns. Through risk correlation, multi-source heterogeneous sensor data can be fused within the same risk measurement framework, enabling laboratory safety assessments to possess quantitative, dynamic, and intelligent characteristics. In some preferred embodiments, for high-risk sensor areas (i.e.... If a value remains high for an extended period, the system can automatically mark it as a key monitoring point, triggering enhanced sampling and local anomaly tracking mechanisms.

[0080] Figure 3 A flowchart illustrating an example of calculating the reliability weights of a sensor according to an embodiment of this application is shown.

[0081] like Figure 3 As shown, in step S310, the standardized deviation of each sensor is calculated.

[0082] During laboratory operations, the physical quantities collected by different sensors exhibit differences in dimensions and distribution. To achieve a unified risk assessment standard across sensors, it is necessary to standardize the real-time sampling values ​​of each sensor.

[0083] Specifically, for each sensor Based on historical normal operation data, a corresponding baseline period statistical model is established, including the baseline mean. and baseline standard deviation Collect the sensor's measurement value at the current moment. Calculate its standardized deviation relative to the baseline:

[0084] Equation (3)

[0085] In the formula, For the first The current measurement value of each sensor, For the first The baseline mean of each sensor during its respective time period. For the first The baseline standard deviation of each sensor over its respective time period; For the first The standardized deviation of each sensor represents baseline deviation information.

[0086] Specifically, when A large deviation indicates a significant deviance between the current measurement and the normal baseline, which is considered a potential sign of sensor anomaly. Introducing standardized deviation allows for comparison of the degree of anomaly in different physical quantities (such as temperature, humidity, current, vibration, etc.) on a uniform scale. Simultaneously, it introduces... This reflects the historical stability of the sensor, allowing noisier sensors to automatically reduce their sensitivity and improve the robustness of risk analysis.

[0087] In step S320, the historical stability of each sensor is calculated.

[0088] Specifically, the standard deviation of each sensor during the stable operation phase is calculated in advance. and with a certain reference standard deviation As a benchmark, the basic reliability is calculated using an exponential decay method:

[0089] Equation (4)

[0090] In the formula, For the first The basic reliability of a sensor represents its historical stability. For the first The stability and sensitivity coefficient of each sensor This is a reference stability constant.

[0091] Specifically, the smaller the historical fluctuations (i.e.) The lower the value), the corresponding The closer the value is to 1, the more stable and reliable the sensor is in the long term; for sensors with high noise or long-term instability, its... This significantly reduces the risk and enables adaptive hierarchical adjustment of reliability, effectively suppressing the misleading influence of abnormal sensors on the overall risk assessment.

[0092] In step S330, the calibration cycle attenuation factor of each sensor is calculated.

[0093] After long-term operation, the measurement accuracy of the sensor will gradually deviate from the factory standard. Therefore, the system should consider the impact of the sensor's last calibration time and calibration cycle on reliability.

[0094] Specifically, record the moment when each sensor last completed calibration. and its nominal calibration cycle Then, the calibration overdue ratio at the current moment is calculated:

[0095] Equation (5)

[0096] Equation (6)

[0097] In the formula, Indicates the current assessment time. For the first The calibration cycle of each sensor, For the first Calibration overdue ratio of individual sensors For the first The most recent calibration completion time for each sensor To calibrate the overdue sensitivity coefficient, Indicates the first Calibration cycle attenuation factor for each sensor.

[0098] In equation (5), when When this occurs, it indicates that the sensor is within its effective calibration period and the calibration attenuation factor is equal to 1; when When the sensor has exceeded its nominal calibration period, its reliability declines exponentially. Therefore, ensuring the system automatically identifies expired sensors and reduces their weight improves the overall system's long-term reliability and security.

[0099] In step S340, the sensor reliability weight of the corresponding sensor is determined by combining the deviation threshold and the modulation coefficient.

[0100] In some implementations, to comprehensively consider the current deviation, historical stability, and the impact of the calibration cycle, these three factors are coupled through a nonlinear function to obtain the final reliability weight. Specifically, a deviation threshold is introduced. This is used to distinguish between slight fluctuations and significant anomalies, while the standardized deviation... At that time, reliability will rapidly decline:

[0101] Equation (7)

[0102] In the formula, For deviation from the threshold, This is the attenuation modulation coefficient, used to control the sensitivity of weight descent. For the first Reliability weights for each sensor.

[0103] when When the sensor is considered to be operating within the normal range, the reliability weight remains at a high level. This weighting mechanism enables dynamic control based on a three-dimensional perspective of sensor physical deviation, historical fluctuations, and time aging. It can automatically suppress the negative impact of abnormal, drifting, and expired sensors on the system risk analysis results.

[0104] Figure 4 A flowchart illustrating an example of generating laboratory risk field energy according to an embodiment of this application is shown.

[0105] like Figure 4 As shown, in step S410, for each sensor within the time window The weighted risk energy within is integrated to obtain the time aggregated energy of a single sensor.

[0106] In a laboratory environment, the instantaneous weighted risk energy collected by various sensors (such as temperature and humidity sensors, gas concentration sensors, power sensors, smoke sensors, etc.) changes dynamically over time. To reflect the accumulation and sustained effect of risk energy over time, it is necessary to perform integral calculations on it within a certain time window to obtain the time-aggregated energy of a single sensor.

[0107] In some implementations, the system at each time step Push forward a length of time window This ensures that energy calculations include both historical risk evolution information and retain the response characteristics at the current moment. Specifically, for the first... Instantaneous weighted risk energy of individual sensors Integrate within the time window:

[0108] Equation (8)

[0109] Equation (9)

[0110] In the formula, Represents the time integral variable; Indicates the first Each sensor at time Instantaneous weighted risk energy, Indicates the first Risk severity coefficient of each sensor Indicates the first Each sensor at time Time aggregates energy.

[0111] Here, the time integration process can be implemented through sliding window accumulation to improve the real-time responsiveness of the system. The time-aggregated energy comprehensively considers risk trends, signal credibility, and risk severity, realizing dynamic energy accumulation in the time-series dimension; through integration calculation, it can smooth out false anomalies caused by instantaneous fluctuations and improve the stability of risk assessment.

[0112] In step S420, based on the spatial positional relationship of the sensors, nuclear diffusion is carried out with each sensor position as the center to construct a spatial energy field with the sensors as field points.

[0113] It should be noted that sensors in the laboratory are usually distributed in multiple areas (such as workbenches, vents, around equipment, near medicine cabinets, etc.), and risk signals from different areas need to be fused in a spatial dimension.

[0114] In some implementations, a spatial energy field is established using a nuclear diffusion method based on the spatial location relationships of the sensors. Specifically, a neighborhood set is constructed centered on the sensor's location coordinates in the laboratory space. The spatial diffusion effect of risk energy is modeled using a spatial Gaussian kernel function:

[0115] Equation (10)

[0116] Spatial kernel function Choose the Gaussian kernel function:

[0117] Equation (11)

[0118] In the formula, Sensor indexes for spatial field sampling points. ; For sensors The spatial neighborhood set centered on, This represents the sensor index within the spatial neighborhood set. Indicates the first Each sensor at time Time-based energy aggregation and They represent the first The spatial coordinates of the first sensor and the first The spatial coordinates of each sensor; It is a Gaussian kernel bandwidth (or spatial diffusion bandwidth), which can be adaptively adjusted according to the laboratory area and sensor spacing to support transferable modeling of spaces of different sizes; For distance Let Gaussian kernel function be the independent variable. For in the sensor The spatial energy field value at the field point.

[0119] By means of nuclear diffusion, a smooth energy distribution surface can be constructed in continuous space, so that the risk naturally attenuates in the spatial dimension rather than being simply divided into regions. Thus, continuous modeling of the laboratory risk in space is realized, so that the risk energy of different sensors can form a global distribution situation; in addition, the smoothing property of the Gaussian kernel effectively avoids the local risk value mutation caused by the difference in sensor density.

[0120] In step S430, the spatial energy of all field points is summed to obtain the laboratory risk field energy.

[0121] In order to comprehensively reflect the overall risk level of the laboratory, the energy field values of all spatial sampling points are summed to obtain the real-time total energy of the laboratory risk field.

[0122] Specifically, the spatial energy fields of all field points are summed as follows:

[0123] , Equation (12)

[0124] In the equation, E (t) represents the laboratory risk field energy at time t.

[0125] In Equation (12), the summation process can be regarded as global energy integration in the laboratory space to obtain the total amount of risk energy at the current time. By analyzing the time series of E (t), the risk energy change trend can be further monitored, so as to identify the risk rising, stable or decay stage. Thus, a quantitative index of the overall risk energy of the laboratory is obtained, which can be directly used for risk level division and state determination. For example, when E (t) exceeds the safety threshold, the system can automatically issue an alarm and lock the high-energy aggregation area to realize rapid tracing and emergency response.

[0126] In the embodiments of the present application, a complete laboratory risk field energy modeling mechanism is constructed by means of time integration, spatial diffusion and global aggregation, which preserves the risk evolution trend in the time domain, realizes energy smooth diffusion in the spatial domain, realizes time-space fusion risk field modeling on the basis of multi-dimensional monitoring signals, and enhances the comprehensive perception ability of the system. In addition, by means of sliding time window and continuous field modeling, the system can reflect the laboratory running state in real time, while providing a risk intensity index with clear physical meaning.

[0127] Figure 5 An operation flowchart of an example of determining the laboratory risk state level according to the laboratory risk field energy according to an embodiment of the present application is shown.

[0128] As shown in Figure 5 ​​​​As shown, in step S510, a laboratory activity baseline model is constructed based on historical data of laboratory risk field energy in a preset historical time period.

[0129] In the laboratory safety assessment system, the historical risk field energy data can reflect the typical energy distribution law of different experimental stages. For example, a historical sample set of laboratory risk field energy collected in a preset historical time period (for example, the past 30 days or the last several experimental periods) is counted, and the samples are derived from the fusion results of multiple source sensors, covering multi-dimensional information such as environmental temperature and humidity, gas concentration, equipment current, voltage, and personnel behavior. Through statistical analysis and normalization processing, a time-ordered risk energy feature set is formed, which is input into the laboratory activity baseline model to depict the typical energy state distribution of experimental operation. Thus, multi-dimensional feature induction of the historical running state of the laboratory is realized, and subjective bias caused by the experience-based judgment of the traditional threshold setting is avoided.

[0130] In step S520, the laboratory activity baseline model is divided into several activity clusters by an unsupervised clustering method, and each activity cluster corresponds to a typical experimental running period.

[0131] In some embodiments, on the basis of the constructed baseline model, an unsupervised clustering algorithm (such as the K-means algorithm or the Gaussian mixture model) is used for clustering analysis of the historical risk field energy. Further, each clustering center represents a typical experimental running mode, such as the device startup phase, the stable running phase, the reaction peak phase, or the cooling and ending phase. The clustering process automatically divides the time periods with similar risk energy into the same activity cluster according to the risk energy distribution density and fluctuation characteristics in each time window.

[0132] Thus, adaptive classification of experimental state can be realized, and the key stages in the experimental process can be automatically identified without human intervention, thereby improving the universality and migratability of the model for complex experimental scenarios.

[0133] In step S530, the target activity cluster to which the current time belongs is determined according to the similarity between the multi-source sensor state vector of the current time and the center vector of each activity cluster.

[0134] In some embodiments, in the real-time running phase, the similarity of the current time to each cluster is calculated according to the Euclidean distance or cosine similarity between the multi-source sensor state vector (such as various environmental parameters, electrical parameters, and behavior feature vectors) of the current time and the center vector of each activity cluster. When the cluster with the highest similarity is determined, the current time is attributed to the corresponding target activity cluster.

[0135] Through the matching operation, the system can quickly identify which running mode (e.g., high-energy reaction or regular monitoring stage) the current experiment is in, and realize automatic identification of the context semantics of the current environment. Compared with the static model, through the embodiments of the present application, the determination standard can be flexibly switched according to the risk characteristics of different experimental stages, and the sensitivity and accuracy of risk identification are significantly improved.

[0136] In step S540, the risk field energy distribution parameters of the target activity cluster are calculated to dynamically set the risk level threshold.

[0137] In some embodiments, for a determined target activity cluster, the system calculates the risk field energy distribution parameters, including the mean and standard deviation, in its historical samples. Based on the two parameters, the system dynamically generates multi-level risk thresholds:

[0138] , Equation (13)

[0139] , Equation (14)

[0140] , Equation (15)

[0141] In the equation, , and respectively represent the safety threshold, the warning threshold, and the danger threshold, is a preset coefficient; and respectively represent the mean of the risk energy and the standard deviation of the risk energy of the current activity cluster.

[0142] By introducing statistical distribution parameters to calculate dynamic thresholds, the risk determination threshold is adapted to the energy fluctuation level of the current experimental activity, effectively solving the problem of high false positive rate of traditional fixed threshold models under different experimental types or device states, and realizing adaptive adjustment of risk level determination.

[0143] In step S550, the risk field energy at the current time is compared with the dynamically set risk level threshold, and the corresponding laboratory risk state level is output according to the comparison result.

[0144] Here, the laboratory risk state level includes any one of the following: safe state, low-risk state, warning state, and danger state. More specifically, the risk field energy value calculated at the current time is compared with the above dynamic threshold, when the energy value is lower than , it is determined to be a safe state; when the energy value is between and , it is determined to be a low-risk state; when the energy value is between and When the energy value is between 0 and 1, it is determined as a warning state; when the energy value exceeds 1, it is determined as a dangerous state. Preferably, the final determination result is output to a laboratory risk management interface or an alarm module.

[0145] Through the embodiments of the present application, the activity baseline model and clustering mechanism are introduced to realize adaptive identification of the experiment running stage; then, the risk energy distribution parameters are used to obtain activity clusters by unsupervised clustering and set dynamic threshold according to the cluster statistics, so that the determination is adaptively matched with the current running stage, and the robustness and adaptability of risk level determination are enhanced.

[0146] Regarding details of laboratory risk positioning using laboratory risk field energy, in some examples of the embodiments of the present application, the contribution degree of each sensor to the total risk field energy is calculated.

[0147] Due to different sensor types and arrangement positions, their contributions to the overall risk energy are not consistent. Therefore, the time-aggregated energy of each sensor is calculated at the current time, and it is normalized with the sum of the time-aggregated energies of all sensors to obtain the risk energy contribution degree.

[0148] , Equation (16)

[0149] wherein, and are sensor indexes, is the risk energy contribution degree of the i-th sensor, and respectively represent the time-aggregated energy of the i-th sensor at time t and the time-aggregated energy of the i-th sensor at time t. In Equation (16), the relative importance of each sensor in the global risk energy is quantified by normalization, for example, if a certain environmental sensor is continuously in a high-risk correlation state, its value will be significantly higher than that of other sensors, thereby forming a dominant influence on the regional risk in the spatial distribution. Therefore, the energy imbalance problem caused by the difference in dimension or sampling between different sensors is avoided. The normalized risk energy contribution degree makes the laboratory risk assessment result more comparable and overall consistent, ensures the fair response of the model to multi-source heterogeneous data, and reasonably and dynamically quantifies the influence of each sensor.

[0150] Based on the spatial coordinates of each sensor, a laboratory risk thermal field is constructed.

[0151] Therefore, the energy imbalance problem caused by the difference in dimension or sampling between different sensors is avoided. The normalized risk energy contribution degree makes the laboratory risk assessment result more comparable and overall consistent, ensures the fair response of the model to multi-source heterogeneous data, and reasonably and dynamically quantifies the influence of each sensor.

[0152] Based on the spatial coordinates of each sensor, a laboratory risk thermal field is constructed.​​​​

[0153] Specifically, after obtaining the risk energy contribution of each sensor, a risk thermal field is constructed based on the spatial distribution coordinates of the sensors in the laboratory plane, which spreads the risk energy of each sensor to its surrounding area through a two-dimensional Gaussian diffusion function:

[0154] , Equation (17)

[0155] wherein, represents the risk intensity value at the laboratory plane coordinate , and represents the spatial coordinate of the th sensor in the laboratory plane; is a thermal diffusion radius coefficient for controlling the diffusion range of the two-dimensional Gaussian kernel function in space.

[0156] It should be noted that the risk in the laboratory environment usually has spatial continuity, for example, the risk of chemical gas leakage, high temperature equipment failure or personnel gathering behavior is not limited to a single point, but has a spatial diffusion feature. By modeling with a Gaussian kernel function, the spatial decay effect of the risk can be naturally reflected: the closer the area to the sensor, the greater the influence of its energy, and the farther the area, the influence gradually weakens. At the same time, a two-dimensional plane rather than a three-dimensional space model is used based on the actual situation that most risk monitoring points (environment, equipment, personnel behavior sensors) in the laboratory are arranged at similar heights, and the vertical direction has little difference, so the plane modeling can accurately reflect the main spatial distribution characteristics. Thus, the purpose of converting discrete sensor signals into continuous spatial risk distribution map is achieved, so that the system can identify risk hotspots in the global range.

[0157] The laboratory risk thermal field is displayed to indicate the main risk area in the laboratory.

[0158] In some embodiments, after the laboratory risk thermal field is constructed, the thermal distribution is presented in the form of color gradient or contour map, so that the administrator can intuitively understand the laboratory risk distribution through the graphical interface. Thus, the multi-source sensor data is converted from numerical signal to spatial risk energy distribution, realizing the leap of laboratory risk from point monitoring to field cognition, and the visualized results support rapid tracing and regional control, greatly improving the operability and response efficiency of the laboratory safety monitoring system.

[0159] In order to verify the effectiveness of the method proposed in the embodiments of the present application, a simulation-based experiment is designed in this paper. Since it is difficult to obtain long-term recorded data in an actual laboratory, sensor data in a simulated laboratory environment is used for comparative analysis.

[0160] In terms of simulation data generation, the simulation environment contains four types of sensors: temperature, gas concentration, people flow, and humidity. Each type of sensor follows a normal distribution in the normal state, with a mean and standard deviation of 、 、 (relative concentration unit), 、 (persons per unit time), 、 、 . The sampling time length is 1000 steps, simulating a day of operation. Three types of anomalies are introduced in the simulation:

[0161] Gas leakage event: superimpose a concentration of 0.8 on the gas sensor at steps 200-250 to simulate a leakage accident;

[0162] High temperature and dryness: raise the temperature to about at steps 600-650, and reduce the humidity by 20% to simulate device overheating;

[0163] Personnel congestion: increase the personnel flow by 10 people at steps 800-830 to simulate excessive gathering in the laboratory.

[0164] In addition, simulate a gas sensor failure at steps 500-550, making its reading 0.7 higher, but there is no real danger at this time.

[0165] Regarding the algorithm settings of the comparative experiment, the baseline algorithm uses linear weighted summation of each sensor data as the comprehensive risk score:

[0166] , equation (18)

[0167] , equation (19)

[0168] where represents the time , represents the th sensor at time , which is used to eliminate the differences in dimension and amplitude of different sensor data; represents the original measurement value of the th sensor at time , represents the mean value of the th sensor in the baseline stage (e.g., take the first 100 samples); represents the standard deviation of the measurement value of the th sensor in the baseline stage; 、 、 and The standardized risk features of temperature class sensors, gas concentration sensors, electrical load sensors, and personnel behavior sensors are represented respectively, and the weights are selected by experience. The mean and standard deviation are calculated according to the first 100 samples before the baseline, and the warning threshold and danger threshold are set. When the score exceeds the warning threshold, an alarm is given.

[0169] Regarding the algorithm setting details of the Spatio-Temporal Energy Fusion Risk Mapping (STEFRM) algorithm proposed in the embodiments of the present application, according to the foregoing method, the gray correlation degree and risk energy of each sensor are calculated first, and then the weighted risk energy is calculated according to the reliability weight function, and the risk field energy is obtained by summation . The threshold is set using the mean and standard deviation of the baseline segment. In the experiment, the resolution coefficient is taken, and the risk severity parameters of various sensors are , , , .

[0170] Figure 6 The comparison results of the risk scores of different algorithms in the simulation environment with time are shown, the horizontal axis represents the time step, and the vertical axis represents the risk score. Among them, the blue curve corresponds to the risk score calculated by the STEFRM algorithm proposed in this paper, the orange curve corresponds to the risk score of the linear weighted baseline algorithm; the red shaded area represents the period when the dangerous event actually occurs in the laboratory (including gas leakage, high temperature drying and personnel gathering), and the blue shaded area represents the period when the gas sensor signal drifts or intermittently fails but the environment has no actual risk.

[0171] As Figure 6 shown, the risk score of the baseline algorithm increases significantly during the red period when the dangerous event occurs, which can reflect the abnormal trend to some extent. However, during the sensor failure period shown by the blue shaded area, the score of the baseline algorithm also fluctuates significantly, leading to false positives, because the baseline algorithm uses a fixed weight linear weighting method (the gas weight accounts for the highest proportion), and lacks reliability constraints when a single sensor output is abnormal, thus amplifying the influence of sensor drift on the overall risk.

[0172] In contrast, the STEFRM algorithm proposed in the embodiments of the present application shows a smooth and targeted risk rising trend during the dangerous event period, which can accurately correspond to the real risk event; and during the sensor failure period, the risk score remains at a low level without significant fluctuations. Specifically, the STEFRM introduces a sensor reliability weight and a spatiotemporal energy aggregation mechanism in risk modeling: on the one hand, it suppresses the interference of abnormal signals by evaluating the historical stability and baseline deviation of the sensor, and on the other hand, it enhances the multi-source consistency judgment through time integration and spatial diffusion, so that the algorithm can distinguish between real dangerous events and false abnormal signals.

[0173] Therefore, the STEFRM algorithm has more robust risk perception and judgment ability in a multi-source heterogeneous sensor environment, significantly reduces the false positive rate caused by single point failure, and maintains a high response sensitivity during the real risk period. As shown in the simulation results, Figure 6 which verify the robustness and reliability improvement effect of the proposed STEFRM algorithm in identifying risk states in complex laboratory scenarios.

[0174] Figure 7 The performance comparison results of different algorithms in the main risk judgment indicators are shown, the horizontal axis represents different evaluation indicators, which include FPR (false positive rate), FNR (false negative rate) and ACC (accuracy), and the vertical axis represents the numerical size of the corresponding indicators.

[0175] As shown in Figure 7 , the STEFRM algorithm significantly improves the accuracy compared to the baseline algorithm, with an average accuracy of 0.70 to 0.90, an overall recognition accuracy of about 20%; the false positive rate is reduced from 0.30 to 0.10, with a decrease of more than two-thirds; the false negative rate is reduced from 0.25 to 0.15, showing higher event detection sensitivity.

[0176] Specifically, the STEFRM introduces a sensor reliability weight mechanism and a spatiotemporal energy aggregation mechanism in risk modeling, which dynamically adjusts the contribution weight of each sensor through historical stability and baseline deviation, thereby suppressing false positives caused by single point drift. In addition, through the time integration and spatial diffusion process, the multi-source consistency judgment is strengthened, so that the algorithm can remain sensitive to real risk events. Therefore, as Figure 7 the results verify that the STEFRM algorithm has higher accuracy and robustness than the linear weighted model in a multi-source heterogeneous sensor environment, which can effectively reduce the false positive rate while ensuring the risk detection sensitivity.

[0177] For the laboratory safety monitoring scenario, this paper proposes a highly creative multi-source data fusion evaluation method. By introducing the grey correlation degree, risk energy and adaptive reliability weight, a laboratory risk field energy model is constructed, which can comprehensively consider the relationship between various sensor data and their reliability differences. Experimental results show that compared with the baseline method of linear weighting, this method can effectively reduce false positives caused by sensor failures and provide risk source explanation capability.

[0178] Future work can be carried out in the following directions: 1) Collect long-period multi-source data, train and verify the model; 2) Explore more accurate reliability evaluation strategies, such as combining sensor health diagnosis models and cross-validation mechanisms; 3) Combine the proposed risk field model with machine learning methods, use deep learning to extract more complex features to improve anomaly detection capability. 4) Consider data privacy and security, design cross-laboratory data sharing and model federated learning while ensuring that sensitive laboratory information is not leaked.

[0179] It should be noted that for the foregoing method embodiments, in order to simply describe, they are all expressed as a series of actions, but those skilled in the art should know that the present application is not limited by the order of the described actions, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to the present application. In the above embodiments, the description of each embodiment is focused on, and the parts not described in detail in a certain embodiment can be referred to the related description of other embodiments.

[0180] Figure 8 A structural block diagram of an example of a laboratory safety state evaluation system based on multi-source data fusion according to an embodiment of the present application is shown.

[0181] As Figure 8 shown, the laboratory safety state evaluation system based on multi-source data fusion 800 includes a data acquisition unit 810, a correlation risk analysis unit 820, a sensor reliability analysis unit 830, a sensor risk energy analysis unit 840, and a laboratory risk field energy generation unit 850.

[0182] The data acquisition unit 810 is configured to acquire multi-source sensor data corresponding to the laboratory environment based on the sensor module, and construct a laboratory sensor data matrix under a unified time axis according to the multi-source sensor data.

[0183] The correlation risk analysis unit 820 is configured to process the laboratory sensor data matrix by a grey correlation analysis method to calculate a risk correlation degree corresponding to each of the sensors, wherein the risk correlation degree is used to indicate a closeness between a current sensor sampling feature and a preset dangerous mode feature.

[0184] The sensor reliability analysis unit 830 is configured to acquire baseline deviation information corresponding to each of the sensors, and determine a reliability weight according to a historical stability of each of the sensors, the baseline deviation information and a calibration period, wherein the historical stability is defined as a standard deviation of data collected by the sensor in a historical normal state, and the baseline deviation information is defined as a deviation degree of a current measurement value of the sensor relative to a baseline mean value of a time period.

[0185] The sensor risk energy analysis unit 840 is configured to multiply the risk correlation degree of each of the sensors by the reliability weight and a risk severity coefficient to obtain a weighted risk energy of each of the sensors, wherein each of the sensors has a unique corresponding risk severity coefficient.

[0186] The laboratory risk field energy generation unit 850 is configured to perform spatiotemporal coupling superposition on the weighted risk energies of all the sensors to generate a laboratory risk field energy, and calculate a corresponding laboratory risk state level according to the laboratory risk field energy.

[0187] In some embodiments, the embodiments of the present application provide a non-volatile computer readable storage medium, wherein the storage medium stores one or more programs including execution instructions, the execution instructions can be read and executed by an electronic device (including but not limited to a computer, a server, or a network device, etc.) to perform the steps of any one of the above laboratory safety state evaluation methods based on multi-source data fusion.

[0188] In some embodiments, the embodiments of the present application also provide a computer program product, which comprises a computer program stored on a non-volatile computer readable storage medium, the computer program comprising program instructions, when the program instructions are executed by a computer, the computer program instructions cause the computer to perform the steps of any one of the above laboratory safety state evaluation methods based on multi-source data fusion.

[0189] In some embodiments, the embodiments of the present application also provide an electronic device, which comprises at least one processor and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the steps of the laboratory safety state evaluation method based on multi-source data fusion.

[0190] The product can execute the method provided by the embodiments of the application, has the corresponding function modules and beneficial effects of executing the method. Technical details not described in detail in the embodiments can be referred to the method provided by the embodiments of the application.

[0191] The electronic device of the embodiments of the application exists in various forms, including but not limited to: a mobile communication device, an ultra-mobile personal computer device, a portable entertainment device, or other onboard electronic devices with data interaction functions.

[0192] The device embodiments described above are only schematic, and the units described as separate components can or can not be physically separate, and the components displayed as units can or can not be physical units, i.e. can be located in one place, or can be distributed to multiple network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the embodiments of the application.

[0193] From the above description of the embodiments, those skilled in the art can clearly understand that the embodiments can be implemented by means of software plus a general hardware platform, and of course can also be implemented by hardware. Based on such understanding, the above technical solutions can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in each embodiment or some parts of the embodiments.

[0194] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the application, and not to limit them; although the application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements to some technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the application.

Claims

1. A laboratory safety state evaluation method based on multi-source data fusion, characterized in that, The method comprises: Based on the sensor module, collect the multi-source sensor data corresponding to the laboratory environment, and construct a laboratory sensor data matrix under a unified time axis according to the multi-source sensor data; The laboratory sensor data matrix is processed by a grey correlation analysis method to calculate the risk correlation degree corresponding to each sensor; the risk correlation degree is used to indicate the closeness between the current sensor sampling characteristics and the preset dangerous mode characteristics; Obtain the baseline deviation information corresponding to each sensor, and determine the corresponding reliability weight according to the historical stability, baseline deviation information and calibration period of each sensor; the historical stability is defined as the standard deviation of the data collected by the sensor in the historical normal state, and the baseline deviation information is defined as the deviation degree of the current measurement value of the sensor relative to the baseline mean value of the period; The risk correlation degree of each sensor is multiplied by the reliability weight and the risk severity coefficient to obtain the weighted risk energy of each sensor; each sensor has a unique corresponding risk severity coefficient; The weighted risk energy of all sensors is coupled and superimposed in space and time to generate a laboratory risk field energy, and the corresponding laboratory risk state level is calculated according to the laboratory risk field energy.

2. The method of claim 1, wherein, Before constructing the laboratory sensor data matrix under the unified time axis according to the multi-source sensor data, the method further comprises data preprocessing, specifically Comprise the following operations: Time synchronization and interpolation of sensor data with different sampling frequencies are performed; Smooth the high-frequency sensor data by using a sliding window filtering algorithm to remove transient noise interference; When detecting missing sensor data, use the observation values of adjacent periods and associated sensors to perform statistical reconstruction to complete the missing samples, so as to ensure the continuity and integrity of the laboratory sensor data matrix under the unified time axis.

3. The method of claim 1, wherein, The laboratory sensor data matrix is processed by a grey correlation analysis method to calculate the risk correlation degree corresponding to each sensor, comprising: For each sensor, define a dangerous reference feature sequence corresponding to the dangerous mode; Calculate the grey correlation coefficient between the current sensor sampling sequence indicated by the laboratory sensor data matrix and the dangerous reference feature sequence: ， Calculate the mean value of the grey correlation coefficient in the time window to obtain the risk correlation degree: ， wherein and is a sensor index, , is the total number of sensors in the sensor module; and is a discrete time index within a time window, wherein and , is the window length of the time window; and are the values of the hazard pattern reference sequence at time and time , is the sampling value of the th sensor at time , is a resolution coefficient satisfying ; is the sampling value of the th sensor at time ; is the grey correlation coefficient of the th sensor at time ; is the risk correlation degree of the th sensor.

4. The method of claim 1, wherein, The corresponding reliability weight is determined according to the historical stability, baseline deviation information and calibration period of each sensor, comprising: Calculate the normalized deviation degree of each sensor: ， wherein is the current measurement value of the th sensor, is the baseline mean of the th sensor for the time period in question, is the baseline standard deviation of the th sensor for the time period in question; is the normalized deviation of the th sensor, representing baseline deviation information; Calculate the historical stability of each sensor: ， In the formula, For the first The basic reliability of a sensor represents its historical stability. For the first The stability and sensitivity coefficient of each sensor For reference stability constants; Calculate the calibration period attenuation factor of each sensor: ， ， In the formula, Indicates the current assessment time. For the first The calibration cycle of each sensor, For the first Calibration overdue ratio of individual sensors For the first The most recent calibration completion time for each sensor To calibrate the overdue sensitivity coefficient, Indicates the first Calibration cycle attenuation factor for each sensor; Determine the sensor reliability weight of the corresponding sensor by combining the deviation threshold and the modulation coefficient: ， wherein is a deviation threshold, is a decay modulation coefficient, is a reliability weight of the th sensor.

5. The method of claim 4, wherein, The weighted risk energy of all sensors is coupled and superimposed in space and time to generate a laboratory risk field energy, comprising: For each sensor in the time window Integrating the weighted risk energy within the time frame yields the single-sensor time aggregated energy: ， ， wherein denotes the time integral variable; denotes the instantaneous weighted risk energy of the th sensor at time denotes the risk severity coefficient of the th sensor, denotes the time aggregated energy of the th sensor at time ;​ According to the spatial position relationship of the sensors, perform nuclear diffusion with each sensor position as the center to construct a spatial energy field with the sensors as the field points: ， Spatial kernel function A Gaussian kernel function is chosen: ， wherein is the sensor index of the spatial field sample point, is the set of spatial neighbors centered at sensor , is the sensor index in the set of spatial neighbors, is the time-aggregated energy of the th sensor at time , and are the spatial coordinates of the th sensor and the spatial coordinates of the th sensor, respectively; is the Gaussian kernel bandwidth, is the Gaussian kernel function with distance as the argument, is the spatial energy field value at the field point of sensor ;​ Sum the spatial energy of all field points to obtain the laboratory risk field energy: ， wherein represents the laboratory risk field energy at time t.

6. The method of claim 5, wherein, The method comprises the following steps: a laboratory activity baseline model is constructed based on historical data of laboratory risk field energy in a preset historical time period; the laboratory activity baseline model is divided into several activity clusters by an unsupervised clustering method, each activity cluster corresponding to a typical experimental running period; a target activity cluster to which the current time belongs is determined according to the similarity between the current time's multi-source sensor state vector and each activity cluster center vector; calculating a risk field energy distribution parameter of the target activity cluster dynamically setting a risk level threshold ， ， ， wherein, , and respectively represent a safety threshold, a warning threshold and a danger threshold, is a preset coefficient; and respectively represent a risk energy mean value and a risk energy standard deviation of the current active cluster. the risk field energy of the current time is compared with the dynamically set risk level threshold, and a corresponding laboratory risk state level is output according to the comparison result, the laboratory risk state level including any one of the following: a safe state, a low-risk state, an alert state and a dangerous state.

7. The method of claim 5, wherein, After the weighted risk energy of all sensors is coupled and superimposed in space and time to generate the laboratory risk field energy, the method further comprises: calculating the contribution degree of each sensor to the total risk field energy; ， wherein, and is a sensor index, is a risk energy contribution of the th sensor, and denote the time-aggregated energy of the th sensor at time and the time-aggregated energy of the th sensor at time , respectively. constructing a laboratory risk thermal field based on the spatial coordinates of each sensor; ， wherein, represents the risk intensity value at the laboratory planar coordinate represents the risk intensity value at the laboratory planar coordinate represents the spatial coordinate of the sensor in the laboratory planar coordinate; is a thermal diffusion radius coefficient for controlling the diffusion range of the two-dimensional Gaussian kernel function in space; displaying the laboratory risk thermal field to indicate the main risk area in the laboratory.

8. A laboratory safety state evaluation system based on multi-source data fusion, characterized in that, The system comprises: a data acquisition unit configured to acquire multi-source sensor data corresponding to a laboratory environment based on a sensor module, and to construct a laboratory sensor data matrix under a unified time axis according to the multi-source sensor data; an associated risk analysis unit configured to process the laboratory sensor data matrix by a grey correlation analysis method to calculate a risk correlation degree corresponding to each sensor; the risk correlation degree is used to indicate the closeness between the current sensor sampling feature and a preset dangerous mode feature; a sensor reliability analysis unit configured to obtain baseline deviation information corresponding to each sensor, and to determine a corresponding reliability weight according to the historical stability of each sensor, the baseline deviation information and a calibration period; the historical stability is defined as the standard deviation of data collected by the sensor in a historical normal state, and the baseline deviation information is defined as the deviation degree of the current measurement value of the sensor relative to the baseline mean value of the period; a sensor risk energy analysis unit configured to multiply the risk correlation degree of each sensor by the reliability weight and a risk severity coefficient to obtain the weighted risk energy of each sensor; each sensor has a unique corresponding risk severity coefficient; a laboratory risk field energy generation unit configured to couple and superimpose the weighted risk energy of all sensors in space and time to generate laboratory risk field energy, and to calculate a corresponding laboratory risk state level based on the laboratory risk field energy.

Images