Laboratory safety detection method, electronic device and program product

Abstract

The invention discloses a safety detection method for a laboratory, electronic equipment and a program product, and relates to the field of computers.The method comprises the steps that in response to a safety detection request triggered for a target laboratory, experiment type information and safety level information of the target laboratory are obtained, determining detection item information of the target laboratory based on the experiment type information and the security level information; based on target laboratory information formed by the experiment type information, the safety level information and the detection item information, a target laboratory detection model matched with the target laboratory information is determined from the multiple laboratory detection models, and one laboratory detection model corresponds to one piece of laboratory information; and acquiring multi-modal detection data corresponding to the plurality of detection items, and inputting the multi-modal detection data into the target laboratory detection model for safety detection to obtain a target safety detection result of the target laboratory.

Description

Technical Field

[0001] This invention relates to the field of computers, and more specifically, to a laboratory safety testing method, electronic device, and program product. Background Technology

[0002] In the field of laboratory safety management, traditional safety detection methods are often limited to data analysis of fixed models and data acquisition of single modalities. This results in insufficient flexibility and adaptability when facing changes in the experimental environment and the diversity of experimental types. Furthermore, it fails to comprehensively capture multidimensional safety information within the laboratory, easily leading to the underreporting of complex risk events. Therefore, related technologies suffer from the technical problem of low accuracy in laboratory safety detection. Summary of the Invention

[0003] This invention provides a laboratory safety testing method, electronic device, and program product to at least address the technical problem of low accuracy in laboratory safety testing in related technologies.

[0004] To achieve the above objectives, according to one aspect of this application, a laboratory safety inspection method is provided, comprising: responding to a safety inspection request triggered on a target laboratory, acquiring experiment type information and safety level information of the target laboratory, and determining inspection item information of the target laboratory based on the experiment type information and safety level information, wherein the inspection item information is used to indicate multiple inspection items associated with the target laboratory; determining a target laboratory inspection model matching the target laboratory information from multiple laboratory inspection models based on the target laboratory information constituted by the experiment type information, safety level information, and inspection item information, wherein the multiple laboratory inspection models are pre-trained models that perform safety inspections on laboratories based on input multimodal data, and one laboratory inspection model corresponds to one laboratory information; acquiring multimodal inspection data corresponding to multiple inspection items, and inputting the multimodal inspection data into the target laboratory inspection model for safety inspection, thereby obtaining a target safety inspection result for the target laboratory.

[0005] According to another aspect of the present invention, a laboratory safety testing device is also provided, comprising: an acquisition unit, configured to acquire, in response to a safety testing request triggered on a target laboratory, experimental type information and safety level information of the target laboratory, and determine testing item information of the target laboratory based on the experimental type information and safety level information, wherein the testing item information is used to indicate multiple testing items associated with the target laboratory; a determination unit, configured to determine a target laboratory testing model matching the target laboratory information from multiple laboratory testing models based on the target laboratory information constituted by the experimental type information, safety level information, and testing item information, wherein the multiple laboratory testing models are pre-trained models that perform safety testing on laboratories based on input multimodal data, and one laboratory testing model corresponds to one laboratory information; and a testing unit, configured to acquire multimodal testing data corresponding to multiple testing items, and input the multimodal testing data into the target laboratory testing model for safety testing, thereby obtaining a target safety testing result for the target laboratory.

[0006] According to another aspect of the present invention, an electronic device is also provided, including one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the laboratory safety testing method of any of the above.

[0007] According to another aspect of the present invention, a computer program product is also provided, including a computer program that, when executed by a processor, implements the steps of the laboratory safety testing method described above.

[0008] The embodiments provided in this application, in response to a safety inspection request triggered by a target laboratory, not only are basic information about the target laboratory, such as the type of experiment and its safety level, obtained, but also specific inspection items are determined based on this information. This ensures the targetedness and accuracy of the safety inspection, that is, the safety inspection items are customized according to the specific attributes of the laboratory, avoiding the one-size-fits-all problem of traditional methods and reducing the possibility of false alarms and false negatives. Next, from multiple pre-trained laboratory inspection models, the model that best matches the target laboratory information is selected for safety inspection. Each laboratory inspection model is trained based on input multimodal data, aiming to cover multiple levels of information related to laboratory safety, such as visual, environmental, equipment status, and personnel behavior, thereby providing a more comprehensive and in-depth risk assessment. Through this method, laboratory safety inspection is no longer limited to fixed models or single modalities, but can integrate multimodal data such as video images, environmental parameters, equipment status, and personnel behavior, and perform data analysis based on the selected model that best matches the target laboratory information, significantly improving the comprehensiveness and accuracy of the inspection. In this way, by integrating multimodal data acquisition, personalized testing project formulation, and highly matched intelligent model testing, the technical effect of improving the accuracy of laboratory safety testing is achieved, solving the technical problem of low accuracy in traditional laboratory safety testing methods. Attached Figure Description

[0009] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0010] Figure 1 A hardware block diagram of a computer terminal (or mobile device) for implementing a laboratory safety testing method is shown.

[0011] Figure 2 This is a flowchart of an optional laboratory safety testing method according to an embodiment of the present invention.

[0012] Figure 3 This is a flowchart of an optional intelligent early warning method for campus experimental safety based on multimodal large model technology according to an embodiment of the present invention.

[0013] Figure 4 This is a framework diagram of an optional intelligent early warning system for campus experimental safety based on multimodal large model technology according to an embodiment of the present invention.

[0014] Figure 5 This is a schematic diagram of an optional laboratory safety testing device according to an embodiment of the present invention.

[0015] Figure 6This is a structural block diagram of an electronic device according to an embodiment of this application. Detailed Implementation

[0016] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0017] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0018] It should be noted that the laboratory safety testing methods and devices disclosed herein can be used in the computer field, or in any field other than the computer field. This disclosure does not limit the application field of the laboratory safety testing methods and devices.

[0019] It should be noted that the information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, and displayed data) collected in this public disclosure are information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of related data all comply with the relevant laws, regulations, and standards of the relevant regions, necessary confidentiality measures have been taken, and they do not violate public order and good morals. Corresponding operation entry points are provided for users to choose to authorize or refuse. For example, this system has interfaces with relevant users or organizations. Before obtaining relevant information, a request to obtain the information needs to be sent to the aforementioned user or organization through the interface, and the relevant information is obtained only after receiving consent from the aforementioned user or organization.

[0020] It should be noted that in this disclosure, customer information is collected and analyzed, and users are provided with corresponding operation entry points to choose whether to agree to or reject the automated decision results; if the user chooses to reject, the process will proceed to the expert decision-making process.

[0021] The present invention will now be described in detail with reference to various embodiments.

[0022] Example 1

[0023] According to an embodiment of the present invention, an embodiment of a laboratory safety testing method is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0024] The laboratory safety testing method embodiment provided in Embodiment 1 of this application can be executed on a mobile terminal, computer terminal, or similar computing device. Figure 1 A hardware block diagram of a computer terminal (or mobile device) for implementing a laboratory safety testing method is shown. Figure 1 As shown, computer terminal 10 (or mobile device) may include one or more ( Figure 1 (Illustrated as 102a, 102b, ..., 102n) Processor 102 (processor 102 may include, but is not limited to, a microprocessor MCU (Microcontroller Unit) or a programmable gate array (FPGA), etc.), memory 104 for storing data, and transmission device 106 for communication functions. In addition, it may include: a display, input / output interface (I / O interface), Universal Serial Bus (USB) port (which may be included as one of the ports of the BUS bus), network interface, power supply, and / or camera. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, computer terminal 10 may also include... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0025] It should be noted that the aforementioned one or more processors 102 and / or other data processing circuits are generally referred to herein as "data processing circuits". These data processing circuits may be embodied, in whole or in part, in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuits may be a single, independent processing module, or may be integrated, in whole or in part, into any other element within the computer terminal 10 (or mobile device). As involved in the embodiments of this application, the data processing circuits serve as a processor control mechanism (e.g., selection of a variable resistor termination path connected to an interface).

[0026] The memory 104 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the laboratory safety testing method in this embodiment. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, thereby realizing the aforementioned laboratory safety testing method. The memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the computer terminal 10 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0027] The transmission device 106 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of the computer terminal 10. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 may be a Radio Frequency (RF) module, used for wireless communication with the Internet.

[0028] The display can be, for example, a touchscreen liquid crystal display (LCD), which allows the user to interact with the user interface of the computer terminal 10 (or mobile device).

[0029] Under the aforementioned operating environment, this application provides the following: Figure 2 The laboratory safety testing methods shown are illustrated. Figure 2 This is a flowchart of an optional laboratory safety testing method according to an embodiment of the present invention, such as... Figure 2 As shown, the method includes the following steps:

[0030] S202, in response to a security testing request triggered on the target laboratory, obtain the experiment type information and security level information of the target laboratory, and determine the testing item information of the target laboratory based on the experiment type information and security level information, wherein the testing item information is used to indicate multiple testing items associated with the target laboratory;

[0031] S204. Based on the target laboratory information composed of experiment type information, safety level information and detection item information, determine the target laboratory detection model matching the target laboratory information from multiple laboratory detection models. Among them, multiple laboratory detection models are pre-trained models that perform safety detection on laboratories based on input multimodal data, and one laboratory detection model corresponds to one laboratory information.

[0032] S206, acquire multimodal detection data corresponding to multiple detection items, input the multimodal detection data into the target laboratory detection model for safety detection, and obtain the target safety detection results of the target laboratory.

[0033] Optionally, in this embodiment, the target laboratory refers to the specific laboratory required to conduct safety testing, and the specific attributes of the laboratory (such as the type of experiment and the safety level) determine the selection of subsequent testing items and testing models.

[0034] Optionally, in this embodiment, the experiment type information includes, but is not limited to, chemical, biological, radiation, and electromechanical experiments, each corresponding to different experimental operations and potential risks. Safety level information is typically categorized into different levels according to the laboratory's hazard level, such as Level I / Red, Level II / Orange, Level III / Yellow, and Level IV / Blue, reflecting the laboratory's safety management requirements and risk control level. The testing item information determines the specific safety items that need to be tested based on the nature of the target laboratory (experiment type and safety level), such as equipment status checks, compliance of flammable material storage, and personnel protective equipment wearing checks.

[0035] Optionally, in this embodiment, the target laboratory information integrates experiment type information, safety level information, and testing item information, forming the basis for a comprehensive understanding of the target laboratory and guiding the selection of subsequent testing models. The laboratory testing model is a model trained based on multimodal data, capable of comprehensively analyzing information from multiple aspects such as laboratory video streams, environmental parameters, equipment status, and personnel behavior to assess the laboratory's safety status.

[0036] Optionally, in this embodiment, the multimodal detection data includes various types of data such as laboratory video images, environmental change parameters, equipment operating parameters, and personnel behavior and voice recordings, which are input into the target laboratory detection model for analysis. The target safety detection result refers to the safety assessment result output by the laboratory detection model, including but not limited to the laboratory's current safety index, existing risks and hazards, and whether an early warning has been triggered.

[0037] Optionally, in this embodiment, a safety inspection process is initiated when a request for a safety inspection of a specific laboratory is received. This request may originate from the laboratory management department, the laboratory reservation system, or an automatic triggering mechanism (such as when environmental parameters within the laboratory reach a preset threshold).

[0038] The laboratory's experimental type and safety level information are obtained from the laboratory basic information management module to ensure the customization of subsequent testing projects and the accuracy of model selection.

[0039] The laboratory safety knowledge base module intelligently selects or automatically generates a list of suitable testing items based on the target laboratory's experimental type and safety level. This process utilizes the semantic understanding and structuring capabilities of multimodal large models to ensure that the testing items match the actual situation of the laboratory, avoiding unnecessary testing and resource waste.

[0040] Based on laboratory type, safety level, and testing item information, the optimal laboratory testing model is selected from a pre-built pool of intelligent risk assessment models. Each model undergoes automated training and optimization driven by a multimodal large model, specifically designed for safety risk assessment of a particular type of laboratory or experimental scenario, ensuring the model's professionalism and applicability.

[0041] The IoT sensing module collects various testing data from the laboratory in real time, including but not limited to multimodal data such as video images, environmental parameters, equipment status, and personnel behavior. This data will serve as input for the laboratory testing model.

[0042] The multimodal data processing module preprocesses and spatiotemporally aligns the collected testing data, transforming it into structured data and combining it with static laboratory information to form a "laboratory context snapshot." This "context snapshot" is then input into a selected laboratory testing model. The model performs a deep risk assessment based on the multimodal data, predicting the likelihood of risks occurring and thus deriving the safety testing results for the target laboratory. The results may include safety indices, potential risk warnings, and emergency response recommendations.

[0043] It should be noted that in this embodiment, suitable testing items are intelligently generated or selected based on the laboratory's experimental type and safety level, ensuring the targeting and effectiveness of the testing. From multiple pre-trained laboratory testing models, the model most suitable for the target laboratory's information is automatically matched, which greatly improves the accuracy and efficiency of risk assessment. Multiple data sources, such as collected laboratory videos, environmental changes, equipment status, and personnel behavior, are fused and processed to construct a current "situational snapshot" of the laboratory, providing comprehensive input for risk assessment. Deep multimodal cross-validation and risk causal inference based on the laboratory situational snapshot can not only identify explicit anomalies but also predict potential risks, enabling intelligent decision-making for early warning and emergency response.

[0044] Through the embodiments provided in this application, in response to a safety inspection request triggered by a target laboratory, not only is basic information about the target laboratory, such as the type of experiment and safety level, acquired, but specific inspection items are also determined based on this information. This ensures the targetedness and accuracy of the safety inspection, that is, the safety inspection items are customized according to the specific attributes of the laboratory, avoiding the one-size-fits-all problem in traditional methods and reducing the possibility of false alarms and false negatives. Next, from multiple pre-trained laboratory inspection models, the model that best matches the target laboratory information is selected for safety inspection. Each laboratory inspection model is trained based on input multimodal data, aiming to cover multiple levels of information related to laboratory safety, such as visual, environmental, equipment status, and personnel behavior, thereby providing a more comprehensive and in-depth risk assessment. Through this method, laboratory safety inspection is no longer limited to fixed models or single modalities, but can integrate multimodal data such as video images, environmental parameters, equipment status, and personnel behavior, and perform data analysis based on the selected model that best matches the target laboratory information, significantly improving the comprehensiveness and accuracy of the inspection. Thus, by integrating multimodal data acquisition, personalized inspection item formulation, and highly matched intelligent model inspection, the technical effect of improving the accuracy of laboratory safety inspection is achieved.

[0045] As an optional approach, multimodal detection data corresponding to multiple detection items can be obtained, including:

[0046] Acquire the detection data corresponding to each of the multiple detection projects, and integrate the detection data corresponding to each detection project to obtain multimodal detection data;

[0047] Multimodal detection data is input into the target laboratory detection model for security testing, resulting in target security testing results for the target laboratory, including:

[0048] Using the target laboratory testing model, a first safety test is performed on the testing data corresponding to each testing item to obtain the first safety test result of the target laboratory. The first safety test is used to test the impact of the testing data corresponding to each testing item on the occurrence of abnormal events.

[0049] Using the target laboratory detection model, a second security test is performed on the multimodal detection data corresponding to multiple detection items to obtain the second security test result of the target laboratory. The second security test is used to detect the temporal relationship between multiple detection data corresponding to multiple detection items and the degree of overlap between the expected temporal relationship and the expected detection data corresponding to the abnormal event.

[0050] The first security test result and the second security test result are combined to obtain the target security test result.

[0051] Optionally, in this embodiment, the first safety inspection refers to analyzing the data of each individual inspection item to assess whether it is abnormal and the degree of impact of the abnormality on laboratory safety. This focuses on detecting potential risk points in a single modality of data. The second safety inspection involves analyzing the temporal relationships between multiple inspection items to determine whether there are abnormal synergies or conflicts between the multimodal data, and the degree to which such abnormal relationships match the expected temporal relationships of known safety risk events. It emphasizes a comprehensive consideration of cross-modal data to identify complex risk events.

[0052] Optionally, in this embodiment, expected detection data refers to the detection data that the system predicts should be observed under normal circumstances, based on known risk paths and experimental specifications in the laboratory safety knowledge base. For example, for an oven heating experiment, the temperature rise at the moment of oven opening is considered normal expected detection data. Expected temporal relationships refer to the expected logical relationships of various detection data in the laboratory scenario snapshot over time, such as what change pattern environmental parameters should follow after a specific experimental operation, or the behavioral patterns of experimental personnel before and after operating the equipment.

[0053] Optionally, in this embodiment, detection data associated with each detection item is separated and extracted from the dynamic data stream of the laboratory. For example, for the detection of the operating status of ventilation equipment in a chemical experiment, the system will collect real-time current, voltage, and operating status data of the ventilation equipment; for the detection of equipment usage specifications, the system will acquire equipment usage footage and personnel behavior records from video surveillance.

[0054] These separate detection data were then integrated into a multimodal detection dataset, which included video images of the laboratory, environmental parameters (such as temperature and humidity), equipment status parameters (such as voltage and current), and personnel behavior and voice recordings, forming a complete overview of the laboratory's dynamic information.

[0055] Optionally, in this embodiment, the target laboratory testing model independently analyzes the testing data of each testing item according to pre-set evaluation rules, assesses whether the data deviates from the normal range and the potential impact of such deviation on laboratory safety.

[0056] The model uses the deep learning capabilities of a multimodal large model to identify and quantify anomalies in the detection data, such as sudden changes in device status or unauthorized operational behaviors, and generates a safety score for each detection item.

[0057] In the second safety testing phase, the target laboratory testing model analyzes the temporal correlation between different testing items to determine whether the interaction between multimodal data conforms to the expected temporal relationship of safe laboratory operations. For example, the system will assess whether changes in environmental parameters after a specific device is turned on are consistent with expectations, and the degree of matching between the operator's actions and the device status.

[0058] The results of the second safety inspection are reflected in the assessment of the consistency of the overall laboratory operating procedures and the identification of potential compound risk events, such as when a piece of equipment is operating abnormally and personnel have not taken appropriate safety measures.

[0059] The results of the first and second safety inspections are intelligently fused, taking into account the combined impact of individual anomaly detection data and complex risk events. A large-scale model weight adjustment and risk index calculation mechanism is used to reassess the laboratory's safety status. The fused target safety inspection results not only include the identification of overt anomalies but also the prediction of potential risks caused by latent or multi-factor factors. This makes the early warning mechanism more accurate and comprehensive, effectively addressing various safety challenges that may arise in complex laboratory environments.

[0060] The embodiments provided in this application achieve accurate assessment of laboratory safety status through in-depth analysis of multimodal detection data. Specifically, by separating and processing the detection data according to the detection items, a first safety detection is performed to identify individual risk points, and a second safety detection is performed to analyze the temporal relationship of cross-modal data. The results of these two detections are then fused to obtain the final target safety detection result. This enables dynamic monitoring and intelligent early warning of laboratory safety risks, effectively overcoming the limitations of single-dimensional assessment in existing technologies, improving the accuracy and response speed of early warnings, and providing more reliable safety assurance for laboratories.

[0061] As an optional approach, a target laboratory testing model is used to perform a first safety test on the testing data corresponding to each testing item, obtaining the first safety test results of the target laboratory, including:

[0062] Feature extraction is performed on the detection data corresponding to each detection item to obtain multiple first detection feature nodes for multiple detection items, and feature extraction is performed on the expected detection data to obtain multiple second detection feature nodes for abnormal events.

[0063] Obtain the first number of key detection feature nodes among multiple first detection feature nodes, wherein key detection feature nodes are detection feature nodes whose influence on the generation of abnormal events is greater than a preset threshold.

[0064] The result obtained by dividing the first number by the second number of multiple second detection feature nodes is determined as the first security detection result.

[0065] Optionally, in this embodiment, the feature nodes are: during security detection, the system extracts features from the detection data using deep learning methods, resulting in feature points that can be used for subsequent analysis. The "first detection feature node" corresponds to the features of the detection data; the "second detection feature node" is the feature that an abnormal event should exhibit under expected normal or safe conditions.

[0066] Optionally, in this embodiment, key detection feature nodes refer to those feature nodes that play an important role in triggering abnormal events, and their abnormal detection results have a significant impact on the overall security assessment.

[0067] Optionally, in this embodiment, the degree threshold is a pre-set numerical standard used to filter out key detection feature nodes. That is, only when the influence of a feature node on an abnormal event exceeds this threshold is it considered a key node.

[0068] Optionally, in this embodiment, for each testing item, the system uses the feature extraction module in the target laboratory testing model to perform in-depth analysis on the collected testing data and identify various safety-related feature points. These feature points may include, but are not limited to, abnormal equipment status indicators, abnormal fluctuations in environmental parameters, and abnormal personnel behavior.

[0069] The system also extracts features from normal operating conditions or expected abnormal data defined in the laboratory safety knowledge base to obtain a set of expected feature nodes associated with abnormal events, which serve as a baseline standard for assessing the degree of abnormality in the detection data.

[0070] By setting a threshold, the system filters out the number of feature nodes that have a significant impact on abnormal events in the actual detection data. This is based on an assessment of the importance of feature nodes, retaining only those key features that have a direct impact on security early warning.

[0071] Optionally, in this embodiment, the first security detection result is calculated as the ratio of the number of key detection feature nodes to the expected number of key feature nodes for the abnormal event. This ratio reflects the degree of matching between the actual detected abnormal features and the expected abnormal features, thereby quantifying the direct impact of a single detection item data anomaly on the occurrence of a security event.

[0072] The embodiments provided in this application achieve a quantitative assessment of laboratory safety status through three key steps: feature extraction, key feature node screening, and calculation of the first safety detection result. This series of operations highlights the accurate identification and quantitative analysis of abnormal features in the detection data, ensuring the accuracy and effectiveness of the first safety detection. Simultaneously, by setting threshold values ​​for key detection feature nodes, the system can focus on factors that truly pose a threat to laboratory safety, avoiding unnecessary interference from excessive alarms and improving the practicality and reliability of the intelligent early warning system.

[0073] As an optional approach, a second security test is performed on the multimodal test data corresponding to multiple test items using the target laboratory testing model, resulting in the second security test results for the target laboratory, including:

[0074] Obtain the third number of overlapping first-overlapping detection feature nodes among multiple first detection feature nodes and multiple second detection feature nodes;

[0075] The result of dividing the third quantity by the second quantity is determined as the first coincidence parameter;

[0076] Obtain the fourth number of overlapping second detection feature nodes among multiple first detection feature nodes and multiple second detection feature nodes, where the paths between nodes overlap;

[0077] The result of dividing the fourth quantity by the second quantity is determined as the second coincidence parameter;

[0078] The result obtained by weighted fusion of the first and second coincident parameters is determined as the second security detection result.

[0079] Optionally, in this embodiment, the first overlapping detection feature node refers to the node that matches the feature type between the first detection feature node and the second detection feature node. This means that the actual detected abnormal feature and the expected abnormal feature overlap in type, but the temporal relationship between the features may not be taken into account.

[0080] Optionally, in this embodiment, the first overlap parameter represents the proportion of the number of first overlap detected feature nodes to the total number of key feature nodes in the expected abnormal event, and is a measure to evaluate the correlation between a single feature node and the abnormal event.

[0081] Optionally, in this embodiment, the second overlapping detection feature node, on this basis, not only requires the feature types to overlap, but also requires that the time sequence logic between the feature nodes is consistent with the expectation, that is, the actual occurrence order and time interval of the detected features match the expected feature time sequence of the abnormal event.

[0082] Optionally, in this embodiment, the second overlap parameter is the ratio of the number of second overlap detected feature nodes to the total number of key feature nodes in the expected abnormal event, which is a quantification of the degree of matching of the temporal relationship of abnormal features.

[0083] Optionally, in this embodiment, weighted fusion refers to combining the importance of the first and second coincident parameters, assigning appropriate weights to both for fusion calculation, in order to comprehensively evaluate the degree of matching between the laboratory scenario and the expected abnormal event, thereby obtaining the second safety detection result.

[0084] Optionally, in this embodiment, the number of first overlapping detection feature nodes is first identified and counted as the number of key feature nodes (first detection feature nodes) that appear in the actual detection data and are completely identical to the nodes (second detection feature nodes) defined in the expected detection data.

[0085] The proportion of the number of first overlapping detected feature nodes (third quantity) to the total number of expected key feature nodes of the abnormal event (second quantity) is calculated. This proportion is used as the first overlap parameter to measure the degree of matching between the actual detected features and the expected features of the abnormal event when the temporal logic is ignored.

[0086] Further analysis not only considers the matching of feature types, but also ensures that the temporal relationship between these feature nodes (i.e., the path overlap between nodes) is consistent with expectations. The total number of such feature nodes is counted, which is the number of the second overlapping detection feature nodes (the fourth number).

[0087] Similarly, the second overlap parameter is obtained by dividing the number of second overlap detection feature nodes (fourth quantity) by the total number of key feature nodes of the expected abnormal event (second quantity). This parameter more comprehensively reflects the degree of consistency between the actual detected features and the expected abnormal event in terms of type and temporal logic.

[0088] Finally, the system assigns appropriate weights to the two overlapping parameters based on their different importance and influence, and then performs a fusion calculation to obtain the second security detection result. This result comprehensively considers both feature node type matching and temporal logic matching between nodes, providing a more rigorous and comprehensive basis for determining whether the laboratory is in an abnormal state.

[0089] The embodiments provided in this application assess whether a laboratory is facing or about to face a specific abnormal event (such as an experimental accident) by quantitatively analyzing the matching degree between the actual detection characteristics and expected abnormal characteristics of the laboratory, as well as the temporal relationship between the characteristics. By calculating the first and second coincidence parameters and then performing weighted fusion to obtain the second safety detection result, this process effectively incorporates feature type matching and complex temporal logic consistency into the safety detection and evaluation system, improving the accuracy and reliability of laboratory safety early warning.

[0090] As an optional approach, the first security detection result and the second security detection result are fused to obtain the target security detection result, including:

[0091] Given the first sub-result obtained by multiplying the first weight coefficient by the first security detection result and the second sub-result obtained by multiplying the second weight coefficient by the second security detection result, the sum of the first sub-result and the second sub-result is determined as the target security detection result.

[0092] If the target safety test result exceeds the preset safety threshold, the target laboratory is determined to be in an abnormal operating state.

[0093] If the target safety test result is less than or equal to the preset safety threshold, the target laboratory is determined to be in normal operating condition.

[0094] Optionally, in this embodiment, the first safety test result is weighted using a first weighting coefficient to obtain a first sub-result that reflects the impact of anomalies in a single test item on laboratory safety.

[0095] Then, the second safety test result is weighted using a second weighting coefficient to obtain a second sub-result that reflects the impact of the temporal logical relationship between multimodal data on laboratory safety.

[0096] Finally, the first sub-result is added to the second sub-result to obtain an evaluation value that integrates the consistency between individual risk point anomalies and complex risk events, which serves as the final target safety detection result.

[0097] If the final calculated target safety test result exceeds the preset safety threshold, it indicates that the current state of the laboratory poses a high safety risk. The system will determine that the target laboratory is in an abnormal operating state and may require immediate intervention and safety measures.

[0098] Conversely, if the target safety test result is below the preset safety threshold, the system considers the current safety status of the laboratory to be within a controllable range, determines that it is in normal operation, and the experimenters can continue to carry out the experiment according to the plan.

[0099] The embodiments provided in this application offer a complete workflow from the first safety test result and the second safety test result to the final determination of the laboratory's operational status. This process, through weighted fusion, effectively balances the importance of anomaly detection in individual test items with the identification of complex risk events, ensuring the comprehensiveness and accuracy of the target laboratory's safety test results.

[0100] As an optional approach, before acquiring multimodal detection data corresponding to multiple detection items, inputting the multimodal detection data into the target laboratory's detection model for safety testing, and obtaining the target laboratory's target safety testing results, the method further includes:

[0101] A first safety assessment was conducted on the basic environmental information of the target laboratory, and the first safety assessment results of the target laboratory were obtained.

[0102] Acquire laboratory test data collected when performing multiple tests;

[0103] A second safety assessment is conducted on the laboratory testing data to obtain the second safety assessment results for the target laboratory.

[0104] After acquiring multimodal detection data corresponding to multiple detection items, inputting the multimodal detection data into the target laboratory detection model for safety detection, and obtaining the target safety detection results of the target laboratory, the method further includes:

[0105] The results of the first safety assessment, the second safety assessment, and the target safety test results are integrated to obtain a safety test report for the target laboratory.

[0106] Optionally, in this embodiment, the first safety assessment is a preliminary safety risk assessment based on the basic environmental information of the target laboratory. This assessment does not involve dynamic data during experimental activities and mainly focuses on static safety conditions, such as the structural safety of the laboratory, equipment status and maintenance records, laboratory type and safety level, etc. The result of the first safety assessment reflects the assessment result of the static safety status of the target laboratory and is the basic assessment data of the laboratory safety early warning system, used to preemptively eliminate potential risks under static conditions.

[0107] Optionally, in this embodiment, the second safety assessment is a dynamic safety risk assessment based on laboratory testing data. This assessment focuses on real-time data during the experimental activity, including changes in environmental parameters, equipment operating status, personnel behavior, and abnormal operations. The result of the second safety assessment is a safety assessment result based on real-time laboratory testing data, used to identify safety risks during the dynamic process of the experimental activity, and provides immediate feedback on the current operating status of the laboratory.

[0108] Optionally, in this embodiment, the safety inspection report is a comprehensive report reflecting the safety status of the target laboratory, generated by integrating the first safety assessment results, the second safety assessment results, and the target safety inspection results. It includes static and dynamic safety risk assessments, warning levels, and corresponding safety recommendations for the laboratory.

[0109] Optionally, in this embodiment, before formally conducting multimodal data testing, the system performs a preliminary safety assessment based on the static information of the target laboratory, such as building structure, safety equipment configuration, laboratory type and level, and generates a first safety assessment result. This result provides a basis for subsequent dynamic safety testing, ensuring that the laboratory meets safety standards at the structural and equipment levels.

[0110] During the experiment, the system collects various monitoring data in the laboratory in real time or on demand, including video images, environmental parameters, equipment status parameters, personnel behavior and voice recordings, for subsequent dynamic safety assessments.

[0111] The system performs in-depth analysis of laboratory testing data to identify potential safety risks and generate a second safety assessment result. This result focuses on risk monitoring in dynamic environments and can promptly reflect any abnormal situations that occur during experimental activities.

[0112] After completing the security testing of multimodal data and obtaining the target security testing results, the system integrates and analyzes all the results of the first security assessment, the second security assessment, and the target security testing to generate a comprehensive security testing report. The report includes not only static and dynamic security risk assessments but may also include risk levels, early warning information, and emergency recommendations, providing detailed safety status feedback and guidance for laboratory administrators and personnel.

[0113] The embodiments provided in this application encompass a series of assessment processes, from a first safety assessment of static environmental information, to a second safety assessment of dynamic data during experimental activities, and then to comprehensive target safety detection using multimodal large model technology, ultimately generating a safety detection report through data integration. This series of assessment processes comprehensively covers both static and dynamic aspects, as well as overall and local aspects of laboratory safety, ensuring the comprehensiveness and accuracy of the assessment.

[0114] As an optional approach, multimodal detection data corresponding to multiple detection items can be obtained, including:

[0115] Acquire laboratory test data collected when performing multiple tests;

[0116] The laboratory test data is spatiotemporally aligned to obtain dynamic time-series data corresponding to multiple test items;

[0117] Acquire static environmental data corresponding to multiple testing items. The static environmental data includes target laboratory information and test information of experiments being conducted in the target laboratory.

[0118] Dynamic time-series data and static environmental data are fused to obtain multimodal detection data.

[0119] Optionally, in this embodiment, laboratory testing data refers to data collected from various monitoring devices in the laboratory during various experiments and operations, including but not limited to video images, environmental parameters (such as temperature, humidity, and gas concentration), equipment operating status parameters (such as current and voltage), personnel behavior, and voice recordings.

[0120] Optionally, in this embodiment, dynamic time-series data refers to real-time data in laboratory testing data that changes over time. This data can reflect the immediate changes in the laboratory environment, equipment status, and personnel activities, and is an important basis for assessing the dynamic status of laboratory safety.

[0121] Optionally, in this embodiment, static environmental data mainly includes basic laboratory information (such as type, safety level, hazard sources, etc.) and detailed information on ongoing experiments (such as experiment type, experiment time, experiment personnel, etc.). These data change little over a certain period of time, but have an important impact on the assessment of laboratory safety.

[0122] Optionally, in this embodiment, data from various devices and sensors in the laboratory can be collected in real time or on demand through an IoT sensing module. This data covers various types such as laboratory video surveillance, environmental monitoring, equipment status monitoring, and personnel behavior monitoring.

[0123] All collected dynamic detection data are unified to the same timeline and timestamped to ensure accurate temporal correspondence between video streams, environmental parameter changes, equipment status, and personnel activity data, forming a dynamic time-series data sequence. This process is typically performed in the multimodal data processing module, utilizing multimodal large model technology to achieve spatiotemporal alignment of the data.

[0124] The laboratory basic information management module retrieves basic static information about the target laboratory, as well as relevant experimental information for ongoing experiments, including experiment type, personnel, and equipment list. This information provides background knowledge and contextual understanding for laboratory safety assessment.

[0125] In the multimodal data processing module, dynamic time-series data is combined with static environmental data, and a structured "laboratory scenario snapshot" is generated using the experiment type and safety management classification information provided by the laboratory safety knowledge base module. This process not only ensures the comprehensiveness and timeliness of the data, but also utilizes the laboratory's basic information and historical data, providing rich multimodal input for subsequent risk assessment and early warning.

[0126] The embodiments provided in this application utilize an IoT sensing module to collect various dynamic monitoring data from the laboratory in real time. Subsequently, in the multimodal data processing module, multimodal large model technology is used to align the data spatiotemporally, ensuring accurate temporal correspondence between video images, environmental parameters, equipment status, and personnel activity records, generating a dynamic time-series data sequence. This serves as the real-time data foundation for assessing the laboratory's safety status. Simultaneously, the system obtains static information about the laboratory and detailed information about the current experiment through the laboratory basic information management module, providing necessary background knowledge for the laboratory context. Finally, the dynamic time-series data and static environmental data are fused in the multimodal data processing module to form a "laboratory context snapshot." This snapshot contains comprehensive information about the laboratory's current status and serves as a multimodal dataset for the intelligent early warning system to conduct risk assessment and dynamic monitoring. Through this rigorous and comprehensive data preparation process, the system can accurately assess the laboratory's safety status, promptly identify potential risks and issue early warnings, effectively improving the intelligence and efficiency of laboratory safety management.

[0127] As an optional approach, before obtaining information on the target laboratory's experimental type and safety level, and before determining the target laboratory's testing items based on the experimental type and safety level information, the method further includes:

[0128] Create a first set of mapping relationships between the type and level information of each laboratory and the project information of each laboratory, wherein the first set of mapping relationships includes a first target mapping relationship, which is used to indicate the mapping relationship between the laboratory type information and safety level information and the testing project information;

[0129] Before identifying the target laboratory testing model that matches the target laboratory information from multiple laboratory testing models, the method also includes:

[0130] Create a second set of mapping relationships between laboratory information for each laboratory and a testing model for each laboratory. The second set of mapping relationships includes a second target mapping relationship, which is used to indicate the mapping relationship between target laboratory information and target laboratory testing model.

[0131] Optionally, in this embodiment, the first mapping relationship set is a set of data that defines the correspondence between experimental type information, safety level information, and testing item information, used to intelligently determine the specific testing items that need to be tested for a specific type of laboratory. The first target mapping relationship specifically refers to the specific mapping relationship between the type and level of a laboratory and the testing items that the laboratory needs to perform, and is an instance in the first mapping relationship set, used to guide the safety testing work of the target laboratory.

[0132] Optionally, in this embodiment, the second mapping relationship set is a set of data that associates laboratory information with laboratory testing models, used to clarify which testing model should be selected for safety risk assessment for different laboratory information. The second target mapping relationship specifies the mapping between detailed information of the target laboratory and the most suitable laboratory testing model; it is a specific instance in the second mapping relationship set, ensuring that the selection of laboratory testing models is targeted and efficient.

[0133] Optionally, in this embodiment, in the laboratory basic information management module, the system administrator or laboratory head enters the type and safety level information of each laboratory, and combines the management specifications and safety requirements extracted from the laboratory safety knowledge base module to intelligently generate or manually set the testing item list for each laboratory, forming a set of first mapping relationships.

[0134] These primary mapping relationships provide a foundational data architecture for laboratory safety management and risk assessment, ensuring that the selection of laboratory testing items is highly correlated with laboratory type and safety level, thereby improving the relevance and effectiveness of testing.

[0135] In the risk assessment model pool management module, the system automatically creates matching laboratory testing models based on the testing items defined in the first mapping relationship set. Each model is designed specifically to handle information and testing items from a particular laboratory.

[0136] System administrators or experts further optimize these models to ensure they can accurately assess safety risks based on the type, level, and specific testing items of the laboratory, thereby constructing a second set of mapping relationships.

[0137] The second mapping relationship set clarifies the correspondence between laboratory information and laboratory testing models, ensuring that the system can intelligently select the most suitable testing model for safety testing and risk assessment based on the actual needs of the target laboratory.

[0138] Through the embodiments provided in this application, by creating a first mapping relationship set, the system can intelligently determine corresponding testing items for laboratories of different types and safety levels. These items are closely related to the characteristics of the laboratories, effectively covering potential hazards and safety risks. This process fully utilizes the capabilities of the laboratory safety knowledge base module, achieving project customization that combines automation and manual review, thereby improving the level of intelligence in laboratory safety management. Creating a second mapping relationship set aims to ensure that each testing item has a matching laboratory testing model. The system can automatically select the most suitable model for safety risk assessment based on the specific information of the target laboratory. This not only simplifies the complexity of model selection but also improves the accuracy of risk assessment, because each model is customized based on a specific laboratory type, safety level, and testing item, enabling more precise identification and assessment of the laboratory's safety status.

[0139] As an alternative approach, the aforementioned laboratory safety testing methods can be applied to intelligent early warning scenarios for campus laboratory safety based on multimodal large model technology. In this scenario, the types of university laboratories are constantly increasing, their scale is expanding, and their usage frequency is significantly rising. The interdisciplinary integration of chemistry, biology, materials science, and electronics is making the experimental environment increasingly complex, with a significant increase in the types and number of hazardous sources involved, such as flammable and explosive chemicals, high-voltage equipment, and biological agents.

[0140] At the same time, the difficulty of campus laboratory safety management has also increased dramatically. On the one hand, the limited professionalism of laboratory safety managers and the limited number of safety officers result in insufficient regulatory coverage and difficulty in grasping the dynamic risks of laboratories in real time. On the other hand, factors such as long-term operation of experimental equipment, frequent use of hazardous chemicals, and high personnel turnover further exacerbate the uncontrollability of safety hazards.

[0141] To overcome the above-mentioned shortcomings, this embodiment proposes an intelligent early warning method for campus experimental safety based on multimodal large model technology, as shown in the flowchart below. Figure 3 As shown, the steps include:

[0142] S1. Static information is constructed for the laboratory, and basic information about the laboratory is labeled, including laboratory type, laboratory safety level, and experiments.

[0143] Specifically, in this embodiment, the laboratory constructs static information. The laboratory management department or system administrator performs static information modeling for each laboratory, including: registration of basic laboratory information: laboratory number, name, building, floor, area, affiliated department / safety officer, and contact information; registration of laboratory type (chemical, biological, radiation, electromechanical, etc.); registration of laboratory safety level (Level I / Red, Level II / Orange, Level III / Yellow, Level IV / Blue), referring to the "Management Measures for the Classification and Grading of Laboratory Safety in Higher Education Institutions" issued by the Ministry of Education; registration of laboratory hazard sources (such as hazardous gases, toxic chemicals, high-temperature and high-pressure equipment, etc.); and registration of laboratory IoT devices (environmental monitors, equipment status monitors, video surveillance equipment, microphones). The devices connected through the IoT sensing module are then associated with the laboratories in the system.

[0144] S2, Laboratory Safety Knowledge Base Construction: Utilizing multimodal large-scale model semantic understanding and structuring capabilities, the imported laboratory safety management documents are deeply analyzed and structurally processed. Based on the analysis results, a hierarchical index system is constructed with experiment type, safety management category, and inspection item as the main framework, forming a laboratory safety knowledge graph.

[0145] Specifically, in this embodiment, the laboratory safety knowledge base is constructed by using a multimodal large model to perform semantic understanding and content parsing on existing safety management documents. This involves identifying elements such as paragraph themes, key safety clauses, responsible parties, operating procedures, hazard sources, and protective measures; transforming unstructured text into structured semantic triples (experiment type, involved hazard source, and preventive measures); forming a three-level tree-like index structure of "experiment type (type established in S1) → safety management classification → inspection items"; and storing the structured extracted safety knowledge in the form of a graph database (Neo4j).

[0146] S3, the intelligent risk assessment model pool, is built upon the laboratory safety knowledge base of S2. Driven by a multimodal large model, it automatically performs intelligent batch matching of "inspection items" with available detection / recognition algorithms to construct risk assessment models for different experimental scenarios. Furthermore, for detection items without readily available algorithm support, the multimodal large model's few-sample learning and transfer learning capabilities allow it to perform associative retrieval based on existing models, finding 1-3 relatively close models. Gap analysis is then conducted, and the system transforms these gaps into specific training tasks, guiding managers to provide supplementary textual and graphical knowledge to complete the training and orchestration of the new model.

[0147] Specifically, in this embodiment, the construction of the intelligent risk assessment model pool and the design of the risk assessment model architecture are composed of dimensions such as laboratory type, safety management classification, inspection items, detection sub-tasks, potential risk paths, and risk index.

[0148] The automated inspection item generation system identifies laboratory types and safety management categories based on user-input text requirements through large-scale model semantic understanding analysis, retrieves similar inspection item management requirements from the safety knowledge base, and generates inspection items.

[0149] The automated inspection subtask generation process involves semantic parsing of the inspection items, employing a multi-head attention mechanism to extract key entities and safety requirements, and transforming abstract requirements into executable inspection tasks. Based on the task entities and task objectives, it searches existing inspection algorithm libraries to match the inspection methods. For example, if the inspection item uses open flame equipment such as a resistance furnace, human supervision is required. Key entity extracted: resistance furnace; safety requirement: continuous personnel presence during equipment operation. Subtask 1 is generated: inspect the operating status of the resistance furnace equipment, selecting thermal imaging detection equipment for temperature + visible light detection of open flame; subtask 2: personnel presence detection, selecting visual area for people counting.

[0150] Potential risk path deduction involves examining projects using a large-scale model's semantic understanding to infer potential risks and extract risk nodes. A risk evolution path is constructed based on a Bayesian network. The causal strength of each node is calculated through counterfactual simulation to verify and update the risk path. The Risk Index (RI) is calculated based on the probability of occurrence (P) and the causal strength of the risk (C), using the formula: RI = P Norm ^α×C Norm ^β; where α (0.6) and β (0.4) are used as weight adjustment parameters; P Norm and C Norm To convert the probability of occurrence (P) and the causal strength of risk (C) after standard data dimension transformation, the transformation formula is as follows:

[0151]

[0152] The probability of occurrence (P) is initially set by experts and divided into five levels: extremely low (3%), low (12%), medium (35%), high (65%), and extremely high (90%). During system operation, it is dynamically calculated and updated based on experimental data.

[0153] The calculation method for risk causality strength (C) is as follows:

[0154]

[0155] Among them, c x→y Let be the total causal strength of target node X for the final risk outcome Y; n be the total number of independent causal paths from node X to outcome Y; K represent the sequence number of the independent causal path; m k C represents the number of causal nodes in the Kth path; i,K Let be the causal strength of the i-th node on the K-th path; Let X represent the target node X under the Kth path, and let X represent the risk causality strength, which is obtained by multiplying the strengths of all causal nodes on this path.

[0156] The formula calculation logic first goes through Calculate the strength of target node X under each independent path; then through... The total causal strength of the target node X for the final risk outcome Y is obtained by summing the strengths of all paths.

[0157] S4 is a laboratory dynamic multimodal data acquisition system that collects data in real time from IoT devices (environmental monitors, equipment status monitors, video surveillance equipment, microphones), including laboratory images, environmental parameters (humidity, gas concentration, smoke / open flame detection), equipment operating status (current, voltage, operating status), personnel behavior, and voice recordings.

[0158] Specifically, in this embodiment, the laboratory dynamically acquires multimodal data in real time, collecting data generated by IoT devices (environmental monitors, equipment status monitors, video surveillance equipment, and microphones), including laboratory images, environmental parameters (humidity, gas concentration, smoke / open flame detection), equipment operating status (current, voltage, operating status), personnel behavior, and voice recordings. Specifically, edge computing nodes (such as industrial control computers or AI boxes) are deployed in each laboratory to handle real-time data acquisition and initial caching. Video stream acquisition involves pulling camera video streams via the RTSP protocol and sampling continuous video streams. Environmental parameter acquisition uses LoRa / Zigbee / WiFi connections to environmental sensors to collect temperature, humidity, toxic and harmful gas concentrations, and combustible gas concentrations, with a sampling rate of once per second. Equipment operating status acquisition uses object modeling protocols to access and read equipment operating status (current, voltage, operating status), with a sampling rate of 10 times per second. Voice acquisition uses a microphone array to sample continuous audio streams. Personnel behavior record acquisition collects personnel behavior records identified by front-end intelligent analysis, including behavior type and occurrence time.

[0159] S5, multimodal data fusion preprocessing, preprocesses the dynamic data collected in S4 using a multimodal large model, performs spatiotemporal alignment, and timestamps and synchronizes laboratory video streams, environmental change parameters (humidity, gas concentration, smoke / open flame detection), equipment operating parameters (current, voltage, operating status), personnel behavior, and voice recordings to form a laboratory dynamic data sequence timeline; further, it integrates laboratory static information (laboratory type, laboratory safety level, hazard sources) as well as experimental information such as current time, current laboratory usage status, current experiment content, experiment duration, and current number of experimenters; forming a rich and structured "laboratory scenario snapshot" as input for risk assessment.

[0160] Specifically, in this embodiment, multimodal data fusion preprocessing is performed by preprocessing the dynamic data collected in S4 using a multimodal large model. Spatiotemporal alignment is performed, and the laboratory video stream, environmental change parameters (humidity, gas concentration, smoke / open flame detection), equipment operating parameters (current, voltage, operating status), personnel behavior, and voice recordings are timestamped to form a laboratory dynamic data sequence timeline. Furthermore, the laboratory static information (laboratory type, laboratory safety level, hazard source) and experimental information such as current time, current laboratory usage status, current experiment content, experiment duration, and current number of experimenters are fused to form a rich and structured "laboratory scenario snapshot" as input for risk assessment. Specifically, based on the association between equipment and laboratory in S1, the collected data sources are spatially aligned; the timestamps of each data stream are converted to Unix timestamps (millisecond level) and the time zone is unified (UTC+8) for timestamp alignment; and a laboratory dynamic data sequence timeline is constructed with a step size of 1 second. The laboratory's static information and reservation information are obtained through the laboratory basic information module; laboratory snapshots are defined and generated, with each "laboratory situation snapshot" representing a complete state snapshot with a 1-second granularity.

[0161] S6, Risk Assessment and Early Warning, is based on the "Laboratory Context Snapshot" generated in S5. The system intelligently matches the most suitable risk assessment model from the "Model Pool" in S3 for the current situation using a large model. Hierarchical Task Technology (HTN) is applied to intelligently decompose and schedule the inspection tasks in the matched risk assessment model. Furthermore, for false alarm scenarios identified in the risk assessment model, the system automatically performs cross-validation, supplementing verification by integrating laboratory footage, personnel behavior, environmental parameters, and equipment parameters to increase credibility and reduce false alarms. Moreover, the system automatically incorporates the current round of risk assessment results and the current "Laboratory Context Snapshot" into the next round of analysis, continuously iteratively monitoring laboratory risks, performing causal inference based on a multimodal large model, predicting the likelihood of risk occurrence, and providing early warnings.

[0162] Specifically, in this embodiment, the initial risk assessment involves joint analysis of dynamic and static data from laboratory scene snapshots to achieve a deep semantic understanding of the laboratory scenario. For obvious anomalies (such as abnormal temperature and humidity, abnormal gas concentration, smoke detection, equipment malfunction, and personnel falls), risk weights are dynamically adjusted based on multimodal cross-validation. Based on the assessment results, a decision is made on whether to generate an alarm. (For example, a false alarm due to a sudden temperature rise during furnace heating experiments; when a high temperature is detected, the large model interprets and judges the laboratory footage and personnel behavior recordings before and after that period, confirming it as normal operation, and ultimately assessing it as a false alarm, thus not generating an alarm).

[0163] Deep risk assessment intelligently selects inspection tasks based on the experimental scenario. Based on an understanding of the current laboratory scenario, it dynamically selects the most suitable security inspection tasks from the risk assessment model library for the current scenario through semantic similarity weight (cosine similarity calculation). It applies hierarchical task technology (HTN) to intelligently decompose and schedule the inspection tasks in the matched risk assessment model. When resource conflicts occur, they are sorted and executed according to the risk index of the inspection items. Based on the inspection results, it decides whether to generate an alarm.

[0164] The system dynamically predicts potential risks. Based on the current laboratory environment, it intelligently retrieves the causal network of potential risks from the risk assessment model pool, identifies possible risk paths (paths containing risk nodes in the current situation) and critical risk nodes (nodes in the path with a high causal strength impact on the eventual safety incident). It continuously monitors real-time dynamic data changes to verify risk evolution trends. The system calculates an early warning index by weighting and summing risk path overlap and critical node hit rate, with a weight of 0.5 for both risk path overlap and critical node hit rate. A potential risk alert is issued when the risk value reaches 0.3, and an early warning is issued when the risk value reaches 0.5.

[0165] The method for calculating the risk variable path overlap is: Overlap=α·|Pcurrent∩Pref| / k+b·Length(LCS(Pcurrent,Pref)) / k.

[0166] Pcurrent is the actual set of detected risk nodes (p1...pk), Pref is the set of nodes in the reference risk evolution path; K is the total length of the current reference risk evolution path Length(Pref) (total number of risk nodes); |Pcurrent|Pref| is the number of overlapping nodes; LCS(Pcurrent,Pref) is the number of nodes with the same order in the traversal path; Length(LCS(Pcurrent,Pref)) calculates the number of nodes with the same order; a(0.5) and b(0.5) are the weighting weights.

[0167] The critical node hit rate is calculated as the number of critical nodes in the Pcurrent path / the number of nodes in the Pref path.

[0168] S7, Intelligent Emergency Response: When a risk assessment triggers an early warning, it intelligently matches the emergency plan and automatically generates dispatch instructions based on the plan's measures, such as: generating alarm text prompts, issuing early warning reminder broadcasts, guiding laboratory personnel to implement emergency measures; linking the ventilation system and fire protection system; and notifying on-duty safety management personnel, etc.

[0169] Specifically, in this embodiment, the emergency response intelligent decision-making is as follows: when the risk assessment triggers an early warning, the emergency plan is intelligently matched, and dispatch instructions are automatically generated based on the plan measures, such as: generating alarm text prompts, issuing early warning reminder broadcasts, guiding experimental personnel to implement emergency measures; linking the ventilation system and fire protection system; and notifying on-duty safety management personnel, etc. Specifically, based on the risk assessment model, an emergency plan is set when a corresponding risk event is discovered. The setting can be completed through automatic matching from the safety knowledge base and manual review and supplementation. When an alarm is triggered, the pre-existing emergency plan instructions are automatically matched based on the alarm event, and the surrounding systems are called through API to complete the implementation of the emergency instructions.

[0170] S8 features a historical early warning reflection and intelligent model optimization system. The system automatically records all events that trigger early warnings and their corresponding complete laboratory scenario snapshot data (laboratory data, matching evaluation model, evaluation conclusions, and causal inference chain of conclusions) to form a case set. Periodically or when the model performs poorly, the system initiates "early warning result reflection". Using a multimodal large model based on historical alarm events, it retrieves case sets of similar laboratory parameters, conducts comparative review analysis, evaluates the rationality of the original conclusions, diagnoses the root cause of the problem, and proposes optimization solutions for management decision-making.

[0171] Specifically, in this embodiment, the system performs intelligent optimization based on historical early warning reflection. It automatically records all events that trigger early warnings and their corresponding complete laboratory scenario snapshot data (early warning event, triggering event, laboratory data, matching evaluation model, evaluation conclusion, and causal inference chain of conclusions), forming a case set. It periodically executes automatic (e.g., weekly automatic analysis of high-frequency alarms and alarms marked as false alarms) or manually executes on a specified model, reflecting on the early warning results. Using a multimodal large model based on historical alarm events, it retrieves case sets of similar laboratory parameters, conducts comparative review analysis, evaluates the rationality of the original conclusions, diagnoses the root cause of the problem, proposes optimization schemes, and assists managers in decision-making. Specifically, through the large model... The current reflection event is characterized by feature understanding, and a query vector is constructed. Multiple sets of cases with similar semantics and context are retrieved from the historical case library, including similar experiment types, snapshot data of similar experiment contexts, and similar early warning trigger paths (causal inference chains). A review is performed, and similar cases are compared horizontally to generate a causal inference chain difference matrix. Causal strength is calculated based on the frequency of causal occurrences in each inference step. For example: if it is detected that the water tank is about to dry out and the equipment is not stopped, the inference conclusion is: 70% chance of the water drying up, 30% chance of a fire. If it is detected that the water tank has dried out, the equipment is not stopped, and no one notices, the inference conclusion is: 50% chance of a fire, 50% chance of equipment failure. For inference steps that significantly impact the final early warning comment, root cause diagnosis is performed, considering dimensions such as missing inference data, deficiencies in the early warning model rules, noise interference, and novelty. Based on the root cause diagnosis results, optimization solutions are output. These include adjusting early warning rules, adjusting early warning values, adding verification parameters, adding delayed early warning mechanisms, and adjusting the early warning inference chain. Through human-computer interaction guidance, managers are gradually assisted in completing the optimization.

[0172] This embodiment also provides a campus experimental safety intelligent early warning system based on multimodal large model technology, the framework diagram of which is shown below. Figure 4 As shown, it specifically includes:

[0173] Laboratory Basic Information Management Module: Used to register laboratory types; register laboratory safety levels; register experimental risk sources; register IoT devices associated with laboratories; manage laboratory reservation records; manage laboratory early warning records; and manage risk and hazard handling records.

[0174] IoT sensing module: responsible for collecting and recording data uploaded to the system from IoT sensing devices such as environmental sensing devices, equipment status sensing devices, video surveillance, and microphones, including laboratory images, environmental parameters, equipment operating status, personnel behavior, and voice data.

[0175] Multimodal data processing module: responsible for preprocessing heterogeneous data from the sensing module, performing spatiotemporal alignment, and forming structured laboratory state parameters.

[0176] Laboratory Safety Knowledge Base Module: Responsible for importing and managing laboratory safety management knowledge documents, and intelligently parsing and structuring them to establish a hierarchical indexing system for the knowledge graph.

[0177] The risk assessment model management module is used for orchestrating early warning strategies and emergency measures, as well as training and optimizing risk identification algorithms.

[0178] Risk assessment and early warning module: responsible for carrying out risk assessment tasks for experimental potential inspections and issuing safety warnings.

[0179] Intelligent Emergency Module: Used to configure emergency plans, automatically respond after triggering laboratory safety warnings, intelligently match emergency plans, automatically generate emergency commands, and link emergency systems such as fire protection, ventilation, and broadcasting.

[0180] The agent task execution traceability module records the agent's decision log and thought chain content, so that workers can check the agent's operation and perform error backtracking.

[0181] The embodiments provided in this application offer a multimodal risk assessment mechanism based on laboratory scenario understanding. Through cross-modal joint analysis of dynamic and static laboratory data, a deep semantic understanding of the laboratory scenario is achieved, overcoming the limitations of traditional single-modal detection. This effectively identifies potential risk scenarios (scenarios that appear safe but are actually dangerous; independent modal detection reports safety, while cross-modal joint analysis generates dangerous semantics, increasing risk weight), reducing false alarms (scenarios that appear dangerous but are actually safe; independent modal detection reports danger, while cross-modal joint analysis generates normal semantics, reducing pre-risk weight). It can dynamically adjust inspection tasks according to the experimental scenario; based on the current understanding of the laboratory scenario, it dynamically matches the most suitable safety inspection task from the risk assessment model library for the current scenario through semantic similarity weight (cosine similarity calculation), thereby solving the problems of poor adaptability and low inspection efficiency of traditional fixed detection modes in complex laboratory environments. It possesses the ability to infer risk causal chains and can dynamically predict risk evolution trends. By integrating dynamic and static laboratory data, multimodal large-scale model technology is used to dynamically construct risk evolution paths, and key risk nodes are identified and continuously monitored and verified based on the risk assessment model. Based on real-time feedback data, the system can dynamically adjust the risk path model and early warning parameter weights, effectively overcoming the lag inherent in traditional static threshold early warning models and achieving proactive early warning and intervention for risk events. It features an intelligent risk assessment model pool construction mode, automatically generating risk assessment models for different experimental scenarios through a safety knowledge base and intelligently matching corresponding safety inspection items and testing technologies. Simultaneously, it provides a knowledge transfer intelligent assisted training mode, training new experimental scenario risk assessment models based on existing models, significantly reducing the training difficulty of new scenario models and enabling non-professionals to efficiently complete model orchestration and training. It also offers an intelligent reflective model optimization mode, utilizing a multimodal large model based on historical alarm events to retrieve case sets of similar laboratory parameters, conduct comparative analysis, propose optimization solutions, and guide managers through gradual optimization via human-computer interaction. Furthermore, it possesses intelligent emergency response scheduling capabilities, automatically generating emergency instructions, such as linking ventilation and fire protection systems, generating alarm prompts to inform experimental personnel, and notifying experimental safety administrators, through the multimodal large model as the decision-making center.

[0182] Example 2

[0183] The laboratory safety testing device provided in this embodiment includes multiple implementation units, each of which corresponds to a specific implementation step in the above embodiment one. The specific implementation method and beneficial effects can be referred to the aforementioned method embodiment, and will not be repeated here.

[0184] Figure 5 This is a schematic diagram of an optional laboratory safety testing device according to an embodiment of the present invention, such as... Figure 5As shown, the safety detection device for this laboratory may include:

[0185] The acquisition unit 502 is used to acquire, in response to a safety testing request triggered by a target laboratory, experimental type information and safety level information of the target laboratory, and to determine the testing item information of the target laboratory based on the experimental type information and safety level information, wherein the testing item information is used to indicate multiple testing items associated with the target laboratory.

[0186] The determining unit 504 is used to determine the target laboratory detection model matching the target laboratory information from multiple laboratory detection models based on the target laboratory information composed of experiment type information, safety level information and detection item information. The multiple laboratory detection models are pre-trained models that perform safety detection on the laboratory based on input multimodal data, and one laboratory detection model corresponds to one laboratory information.

[0187] The detection unit 506 is used to acquire multimodal detection data corresponding to multiple detection items, and input the multimodal detection data into the target laboratory detection model for safety detection, so as to obtain the target safety detection result of the target laboratory.

[0188] As an optional solution, the detection unit 506 includes:

[0189] The first acquisition module is used to acquire the detection data corresponding to each of the multiple detection projects, and integrate the detection data corresponding to each detection project to obtain multimodal detection data.

[0190] Detection unit 506 includes:

[0191] The first detection module is used to perform a first safety test on the detection data corresponding to each detection item using the target laboratory detection model, and obtain the first safety test result of the target laboratory. The first safety test is used to detect the degree of impact of the detection data corresponding to each detection item on the occurrence of abnormal events.

[0192] The second detection module is used to perform a second security detection on the multimodal detection data corresponding to multiple detection items using the target laboratory detection model, and obtain the second security detection result of the target laboratory. The second security detection is used to detect the degree of overlap between the temporal relationship between multiple detection data corresponding to multiple detection items and the expected temporal relationship between the expected detection data corresponding to the abnormal event.

[0193] The first fusion module is used to fuse the first security detection result and the second security detection result to obtain the target security detection result.

[0194] As an optional solution, the first detection module includes:

[0195] The extraction submodule is used to extract features from the detection data corresponding to each detection item to obtain multiple first detection feature nodes for multiple detection items, and to extract features from the expected detection data to obtain multiple second detection feature nodes for abnormal events.

[0196] The first acquisition submodule is used to acquire the first number of key detection feature nodes among multiple first detection feature nodes, wherein key detection feature nodes are detection feature nodes whose influence on the generation of abnormal events is greater than a preset threshold.

[0197] The first calculation submodule is used to determine the first security detection result by dividing the first quantity by the second quantity of multiple second detection feature nodes.

[0198] As an optional solution, the second detection module includes:

[0199] The second acquisition submodule is used to acquire the third number of overlapping first detection feature nodes among multiple first detection feature nodes and multiple second detection feature nodes.

[0200] The second calculation submodule is used to determine the result of dividing the third quantity by the second quantity as the first overlapping parameter;

[0201] The third acquisition submodule is used to acquire the fourth number of overlapping second overlapping detection feature nodes among multiple first detection feature nodes and multiple second detection feature nodes, where the paths between nodes overlap.

[0202] The third calculation submodule is used to determine the result of dividing the fourth quantity by the second quantity as the second coincidence parameter;

[0203] The fourth calculation submodule is used to weight and fuse the first and second overlapping parameters to determine the result as the second security detection result.

[0204] As an optional solution, the first fusion module includes:

[0205] The fifth calculation submodule is used to determine the sum of the first sub-result and the second sub-result as the target security detection result when the first sub-result is obtained by multiplying the first weight coefficient by the first security detection result and the second sub-result is obtained by multiplying the second weight coefficient by the second security detection result.

[0206] The first determination submodule is used to determine that the target laboratory is in an abnormal operating state when the target safety detection result is greater than the preset safety threshold.

[0207] The second determination submodule is used to determine that the target laboratory is in normal operating condition when the target safety detection result is less than or equal to a preset safety threshold.

[0208] As an optional solution, the device also includes:

[0209] The first assessment module is used to conduct a first safety assessment on the basic environmental information of the target laboratory before acquiring multimodal detection data corresponding to multiple detection items, inputting the multimodal detection data into the target laboratory detection model for safety detection, and obtaining the target safety detection result of the target laboratory.

[0210] The second acquisition module is used to acquire laboratory test data collected when performing multiple test items before acquiring multimodal test data corresponding to multiple test items, inputting the multimodal test data into the target laboratory test model for safety test, and obtaining the target safety test result of the target laboratory.

[0211] The second evaluation module is used to perform a second safety evaluation on the laboratory testing data before acquiring multimodal testing data corresponding to multiple testing items, inputting the multimodal testing data into the target laboratory testing model for safety testing, and obtaining the target safety testing result of the target laboratory.

[0212] The integration module is used to acquire multimodal detection data corresponding to multiple detection items, input the multimodal detection data into the target laboratory detection model for safety detection, obtain the target safety detection results of the target laboratory, and then integrate the first safety assessment results, the second safety assessment results, and the target safety detection results to obtain the target laboratory safety detection report.

[0213] As an optional solution, the detection unit 506 includes:

[0214] The third acquisition module is used to acquire laboratory test data collected when performing multiple test items;

[0215] The alignment module is used to perform spatiotemporal alignment processing on laboratory test data to obtain dynamic time-series data corresponding to multiple test items.

[0216] The fourth acquisition module is used to acquire static environmental data corresponding to multiple testing items. The static environmental data includes target laboratory information and experimental information of experiments being conducted in the target laboratory.

[0217] The second fusion module is used to fuse dynamic time-series data and static environmental data to obtain multimodal detection data.

[0218] As an optional solution, the device also includes:

[0219] The first creation module is used to create a first mapping relationship set between the type information and level information of each laboratory and the project information of each laboratory before acquiring the experimental type information and safety level information of the target laboratory and determining the testing project information of the target laboratory based on the experimental type information and safety level information. The first mapping relationship set includes a first target mapping relationship, which is used to indicate the mapping relationship between the experimental type information and safety level information and the testing project information.

[0220] The device also includes:

[0221] The second creation module is used to create laboratory information for each laboratory and a second mapping relationship set between each laboratory and each laboratory testing model before determining the target laboratory testing model that matches the target laboratory information from multiple laboratory testing models. The second mapping relationship set includes a second target mapping relationship, which is used to indicate the mapping relationship between the target laboratory information and the target laboratory testing model.

[0222] The aforementioned laboratory safety detection device may also include a processor and a memory. The aforementioned acquisition unit 502, determination unit 504, detection determination unit 506, etc., are all stored in the memory as program units, and the processor executes the aforementioned program units stored in the memory to realize the corresponding functions.

[0223] The processor described above contains a kernel, which retrieves the corresponding program units from memory. One or more kernels can be configured, and laboratory safety testing can be achieved by adjusting kernel parameters.

[0224] The aforementioned memory may include non-permanent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0225] Example 3

[0226] Embodiments of this application may provide an electronic device. Figure 6 This is a structural block diagram of an electronic device for performing a laboratory safety testing method according to an embodiment of this application. Figure 6 As shown, the electronic device may include: one or more ( Figure 6 (Only one is shown) processor 602, memory 604, memory controller, and peripheral interface, wherein the peripheral interface is connected to the radio frequency module, audio module and display.

[0227] The memory can be used to store software programs and modules, such as the program instructions / modules corresponding to the laboratory safety testing method and apparatus in this embodiment. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, thereby realizing the aforementioned laboratory safety testing method. The memory may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include memory remotely located relative to the processor, and these remote memories can be connected to the terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0228] The processor can invoke information and applications stored in the memory via the transmission device to perform the following steps: in response to a first backup request triggered by a first resource of a first shared client, verifying the resource type of the first resource; if the resource type of the first resource is a shared resource type, determining the target source resource corresponding to the first resource on the shared server associated with the first shared client, wherein the shared server is used to share at least one source resource with at least two clients, the at least one source resource includes the target source resource, and the at least two clients include the first shared client; and copying the target source resource from the shared server to the backup server.

[0229] Those skilled in the art will understand that Figure 6 The structure shown is for illustrative purposes only. Electronic devices can also be smartphones, tablets, handheld computers, mobile internet devices (MIDs), PADs, and other terminal devices. Figure 6 This does not limit the structure of the aforementioned electronic device. For example, electronic devices may also include components that are more... Figure 6 The more or fewer components shown (such as network interfaces, display devices, etc.), or having the same Figure 6 The different configurations shown.

[0230] Those skilled in the art will understand that all or part of the steps in the various laboratory safety testing methods of the above embodiments can be implemented by a program instructing the hardware of the terminal device. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0231] Example 4

[0232] Embodiments of this application also provide a storage medium. Optionally, in this embodiment, the storage medium can be used to store the program code executed by the laboratory safety testing method provided in Embodiment 1.

[0233] According to another aspect of the present invention, a computer-readable storage medium is also provided, the computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform any of the laboratory safety testing methods in the first embodiment described above.

[0234] Optionally, in this embodiment, the storage medium may be located in any computer terminal in a group of computer terminals in a computer network, or in any mobile terminal in a group of mobile terminals.

[0235] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the laboratory safety testing method in various embodiments of this application.

[0236] This application also provides a computer program product, including a non-volatile computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the laboratory safety testing method in various embodiments of this application.

[0237] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0238] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0239] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some interfaces; indirect couplings or communication connections between units or modules may be electrical or other forms.

[0240] The units described as separate components may or may not be physically separate. Similarly, the components shown as units may or may not be physical units; they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0241] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0242] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0243] The above are merely preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A laboratory safety testing method, characterized in that, include: In response to a security testing request triggered by a target laboratory, the system acquires the experiment type information and security level information of the target laboratory, and determines the testing item information of the target laboratory based on the experiment type information and the security level information, wherein the testing item information is used to indicate multiple testing items associated with the target laboratory; Based on the target laboratory information constituted by the experiment type information, the safety level information, and the detection item information, a target laboratory detection model matching the target laboratory information is determined from multiple laboratory detection models. The multiple laboratory detection models are pre-trained models that perform safety detection on laboratories based on input multimodal data, and one laboratory detection model corresponds to one laboratory information. The multimodal detection data corresponding to the multiple detection items are obtained, and the multimodal detection data is input into the target laboratory detection model for safety detection to obtain the target safety detection result of the target laboratory.

2. The method according to claim 1, characterized in that, The acquisition of multimodal detection data corresponding to the multiple detection items includes: Obtain the detection data corresponding to each of the multiple detection items, and integrate the detection data corresponding to each detection item to obtain the multimodal detection data; The step of inputting the multimodal detection data into the target laboratory detection model for security detection, and obtaining the target security detection result of the target laboratory, includes: Using the target laboratory testing model, a first security test is performed on the testing data corresponding to each testing item to obtain the first security test result of the target laboratory. The first security test is used to detect the degree of impact of the testing data corresponding to each testing item on the occurrence of abnormal events. Using the target laboratory detection model, a second security detection is performed on the multimodal detection data corresponding to the multiple detection items to obtain the second security detection result of the target laboratory. The second security detection is used to detect the degree of overlap between the temporal relationship between the multiple detection data corresponding to the multiple detection items and the expected temporal relationship between the expected detection data corresponding to the abnormal event. The first security detection result and the second security detection result are fused to obtain the target security detection result.

3. The method according to claim 2, characterized in that, The step of using the target laboratory testing model to perform a first safety test on the testing data corresponding to each testing item, and obtaining the first safety test result of the target laboratory, includes: Feature extraction is performed on the detection data corresponding to each detection item to obtain multiple first detection feature nodes of the multiple detection items, and feature extraction is performed on the expected detection data to obtain multiple second detection feature nodes of the abnormal event; Obtain the first number of key detection feature nodes among the plurality of first detection feature nodes, wherein the key detection feature nodes are detection feature nodes whose influence on the generation of the abnormal event is greater than a preset threshold. The result obtained by dividing the first quantity by the second quantity of the plurality of second detection feature nodes is determined as the first security detection result.

4. The method according to claim 3, characterized in that, The step of using the target laboratory testing model to perform a second security test on the multimodal testing data corresponding to the multiple testing items, and obtaining the second security test result of the target laboratory, includes: Obtain the third number of overlapping first detection feature nodes among the plurality of first detection feature nodes and the plurality of second detection feature nodes; The result of dividing the third quantity by the second quantity is determined as the first coincidence parameter; Obtain the fourth number of overlapping second overlapping detection feature nodes among the plurality of first detection feature nodes and the plurality of second detection feature nodes, where the paths between the nodes overlap; The result of dividing the fourth quantity by the second quantity is determined as the second coincidence parameter; The result obtained by weighted fusion of the first and second overlapping parameters is determined as the second security detection result.

5. The method according to claim 2, characterized in that, The process of fusing the first security detection result and the second security detection result to obtain the target security detection result includes: In the case of obtaining a first sub-result obtained by multiplying the first weight coefficient by the first security detection result and a second sub-result obtained by multiplying the second weight coefficient by the second security detection result, the sum of the first sub-result and the second sub-result is determined as the target security detection result; If the target safety detection result is greater than a preset safety threshold, the target laboratory is determined to be in an abnormal operating state. If the target safety test result is less than or equal to the preset safety threshold, the target laboratory is determined to be in normal operating condition.

6. The method according to any one of claims 1 to 5, characterized in that, Before acquiring the multimodal detection data corresponding to the multiple detection items, and inputting the multimodal detection data into the target laboratory detection model for safety detection to obtain the target safety detection result of the target laboratory, the method further includes: A first security assessment is performed on the basic environmental information of the target laboratory to obtain the first security assessment result of the target laboratory; Acquire laboratory test data collected during the execution of the multiple test items; A second safety assessment is performed on the laboratory testing data to obtain the second safety assessment result of the target laboratory. After acquiring the multimodal detection data corresponding to the multiple detection items, and inputting the multimodal detection data into the target laboratory detection model for security detection to obtain the target security detection result of the target laboratory, the method further includes: The first safety assessment result, the second safety assessment result, and the target safety test result are integrated to obtain the safety test report of the target laboratory.

7. The method according to any one of claims 1 to 5, characterized in that, The acquisition of multimodal detection data corresponding to the multiple detection items includes: Acquire laboratory test data collected during the execution of the multiple test items; The laboratory test data is spatiotemporally aligned to obtain dynamic time-series data corresponding to the multiple test items. Obtain static environmental data corresponding to the multiple detection items, wherein the static environmental data includes the target laboratory information and the experimental information of the experiments being conducted in the target laboratory; The dynamic time-series data and the static environmental data are fused to obtain the multimodal detection data.

8. The method according to any one of claims 1 to 5, characterized in that, Before acquiring the experiment type information and safety level information of the target laboratory, and determining the testing item information of the target laboratory based on the experiment type information and safety level information, the method further includes: Create a first mapping relationship set between the type information and level information of each laboratory and the project information of each laboratory, wherein the first mapping relationship set includes a first target mapping relationship, which is used to indicate the mapping relationship between the experimental type information and the safety level information and the testing project information; Before determining the target laboratory testing model that matches the target laboratory information from multiple laboratory testing models, the method further includes: Create a second set of mapping relationships between laboratory information for each laboratory and a testing model for each laboratory, wherein the second set of mapping relationships includes a second target mapping relationship, which is used to indicate the mapping relationship between the target laboratory information and the target laboratory testing model.

9. An electronic device, characterized in that, It includes one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the method of any one of claims 1 to 8.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1 to 8.

Images