An industrial safety event time series data synthesis method, device and related equipment

Abstract

The application provides an industrial safety event timing data synthesis method and device and related equipment, relates to the technical field of industrial data processing, and is used for improving the reliability and availability of industrial safety event synthesis data. The specific technical scheme is as follows: according to the multi-source data of a target industrial system, determining safety-related variables, safety boundary conditions, and the causal correlation relationship of each safety state, constructing a safety event evolution graph for representing the evolution process of a safety event; generating a rare safety event evolution path; based on the rare safety event evolution path and the safety boundary conditions, respectively generating variable trajectories, global safety state signal sequences, and global protection action timing chains; after time alignment, forming multi-variable synthesis timing data; and based on the multi-variable synthesis timing data, generating safety event timing data samples. The application is applied to the data synthesis scene of industrial safety early warning, interlocking logic testing, and digital twin safety simulation.

Description

Technical Field

[0001] This application relates to the field of industrial data processing technology, and in particular to a method, apparatus and related equipment for synthesizing time-series data of industrial security events. Background Technology

[0002] As industrial digital transformation deepens, industrial time-series data is becoming increasingly crucial in industrial AI tasks such as anomaly detection, fault diagnosis, and safety early warning. However, there is an extreme lack of samples of rare safety events such as high-altitude alarms, interlock triggers, and emergency shutdowns, which severely restricts the improvement of industrial safety intelligence.

[0003] In related technologies, data-driven time series generation methods generate new time series samples by learning the distribution of real data, which can alleviate the problem of industrial data scarcity to some extent.

[0004] However, due to the complex logic involved in industrial safety incidents, such as multivariate coupling, state duration, and protection action sequence chains, existing generation methods cannot guarantee the rationality of the synthesized samples at the industrial logic level, and cannot effectively cover the boundary scenarios that are difficult to collect in real industrial sites. As a result, the generated data cannot be directly used for downstream tasks such as safety early warning model training and interlocking logic testing. Summary of the Invention

[0005] This application provides a method, apparatus, and related equipment for synthesizing time-series data of industrial safety incidents, which can improve the reliability and usability of synthesized data of industrial safety incidents.

[0006] In a first aspect, embodiments of this application provide a method for synthesizing time-series data of industrial safety events, the method comprising:

[0007] Based on multi-source data from the target industrial system, determine at least one safety-related variable, the safety boundary conditions corresponding to each safety-related variable, and the causal relationships between each safety state.

[0008] Based on the security boundary conditions and the causal relationship, a security event evolution graph is constructed to characterize the evolution process of security events. The security event evolution graph contains at least two security state nodes and at least one evolution edge. The evolution edge is determined according to the causal relationship and is used to characterize the state transition direction and causal transition relationship between each security state node.

[0009] At least one rare security event evolution path is generated by sampling in the security event evolution graph, and each rare security event evolution path corresponds to a security event instance;

[0010] Based on the rare security event evolution path and the security boundary conditions, variable trajectories corresponding to each security-related variable are generated respectively, and a global security state signal sequence and a global protection action time sequence chain corresponding to the rare security event evolution path are generated according to the security boundary conditions. The variable trajectory is used to characterize the change process of the values ​​of the security-related variables approaching or moving away from their respective security boundaries in the time series.

[0011] The variable trajectory, the safety status signal sequence, and the protection action timing chain are time-aligned to form multivariate composite timing data;

[0012] Based on the multivariate synthesized time-series data, a security event time-series data sample is generated.

[0013] The technical solution provided in this application brings at least the following beneficial effects:

[0014] The industrial safety event time-series data synthesis method in this embodiment determines the causal relationships between safety-related variables, safety boundary conditions, and safety states based on multi-source data of the target industrial system. It then constructs a safety event evolution graph containing state nodes and evolution edges, transforming previously implicit and scattered safety knowledge into a structured graph model. This provides clear causal path constraints for time-series data generation, avoiding the problem of traditional black-box generation models easily generating unreasonable industrial samples due to a lack of causal guidance. Furthermore, by sampling rare safety event evolution paths in the safety event evolution graph, and based on these paths and safety boundary conditions, it generates trajectories of variables approaching or moving away from the boundaries, a global safety state signal sequence, and protective actions. The time-series chain, after time alignment, forms multivariate synthetic time-series data. This enables the final generated safety event time-series data samples to not only precisely control the evolution of each safety-related quantity relative to the safety boundary at each sampling moment at the variable level, but also to synchronously match the safety state transition logic and the execution order of protection actions at the system level. This effectively solves the problem of "statistical similarity but industrial incompatibility" caused by the lack of structured constraints on the dynamic relationship between variables and boundaries in existing technologies. At the same time, it overcomes the defect that the separation of safety logic and physical dynamics makes it impossible to generate rare but high-value scenarios such as interlock precursors and protection action boundaries. This significantly improves the usability and reliability of synthetic data in practical tasks such as industrial safety early warning, interlock testing, and control system verification.

[0015] One possible implementation involves constructing a security event evolution graph to characterize the security event evolution process based on the security boundary conditions and the causal relationships, including:

[0016] Based on the aforementioned safety boundary conditions, the plurality of safety state nodes are determined, each of which represents a safety state. The safety state includes at least a normal operation state, an alarm trigger state, an interlock trigger state, and a protection action execution state.

[0017] Based on the causal relationship, the state transition direction between the safe state nodes is determined, and an evolutionary edge connecting the safe state nodes is constructed based on the state transition direction.

[0018] Based on the security boundary conditions and the causal relationship, each evolutionary edge is assigned a corresponding edge weight, which represents the probability of transitioning from the current safe state node to the next safe state node along the evolutionary edge.

[0019] The security event evolution graph is constructed based on the security state nodes, the evolution edges, and the edge weights.

[0020] Another possible implementation, wherein the sampling and generation of at least one rare security event evolution path in the security event evolution graph includes:

[0021] Based on the security state nodes and the evolution edges, at least one candidate security event evolution path is determined in the security event evolution graph, and each candidate security event evolution path is formed by connecting the security state nodes sequentially according to the direction of the evolution edges;

[0022] Based on the security severity of each security state node in each candidate security event evolution path and the edge weight of each evolution edge, the path risk value of each candidate security event evolution path is determined.

[0023] Based on the path risk value of each candidate security event evolution path, the sampling generation probability of each candidate security event evolution path is determined, wherein the path risk value is proportional to the sampling generation probability.

[0024] Based on the sampling generation probability of each candidate security event evolution path, at least one rare security event evolution path is sampled and generated from each candidate security event evolution path.

[0025] Another possible implementation involves generating variable trajectories corresponding to each security-related variable based on the evolution path of the rare security event and the security boundary conditions, including:

[0026] Based on the evolution path of the rare security event, the target security state is determined. The target security state includes the target distance range between the real-time value of the security-related variable and the security boundary, and the target duration for which the security-related variable remains within the target distance range.

[0027] Based on the safety boundary conditions and the preset sampling period, determine the real-time safety distance of each of the safety-related variables relative to the safety boundary at each sampling time;

[0028] Based on the target safety status and constrained by the real-time safety distance, a distance evolution trajectory is generated for each of the safety-related variables, which is to approach the target distance range over time and maintain the target duration after entering the target distance range.

[0029] Based on the distance evolution trajectory of each of the aforementioned safety-related variables, the variable trajectory corresponding to each of the aforementioned safety-related variables is determined.

[0030] Another possible implementation is that the security boundary conditions include alarm boundary conditions and interlocking boundary conditions;

[0031] The alarm boundary conditions include alarm thresholds and alarm duration thresholds, and the interlocking boundary conditions include interlocking triggering conditions and interlocking duration thresholds.

[0032] Another possible implementation is that the safety status signal sequence includes an alarm status signal sequence and an interlock trigger signal sequence;

[0033] The step of generating the global security state signal sequence and global protection action timing chain corresponding to the rare security event evolution path based on the security boundary conditions includes:

[0034] Based on the variable trajectory, the alarm boundary conditions, and the interlocking boundary conditions, the values ​​of each safety-related variable at each time point are compared with the alarm threshold, and the comparison results are output.

[0035] The alarm status signal sequence is constituted based on the alarm status at each moment generated by the comparison result and the alarm duration threshold;

[0036] Determine whether the values ​​of each safety-related variable at each time point meet the interlocking trigger condition, and construct the interlocking trigger signal sequence based on the interlocking state at each time point generated by the duration of the interlocking trigger condition and the interlocking duration threshold.

[0037] When there is a moment in the interlocking trigger signal sequence that is in the triggering state, the corresponding protection action sequence is determined according to the interlocking boundary conditions, and an execution timestamp is configured for each protection action according to the time interval constraint between adjacent protection actions in the protection action sequence, thereby generating the protection action timing chain.

[0038] Another possible implementation involves generating security event time-series data samples based on the multivariate synthesized time-series data, including:

[0039] Perform a security consistency check on the multivariate synthesized time series data according to the security boundary conditions, and output the check result;

[0040] If the verification result indicates that the multivariate synthesized time-series data passes the security consistency verification, the multivariate synthesized time-series data will be output as the security event time-series data sample; or

[0041] If the verification result indicates that the multivariate synthesized time series data has failed the security consistency verification, the multivariate synthesized time series data is subjected to consistency repair processing to generate new multivariate synthesized time series data. The consistency repair processing includes correction and regeneration.

[0042] Another possible implementation involves performing a security consistency check on the multivariate synthesized time-series data based on the security boundary conditions and outputting the check result, including:

[0043] Based on the security boundary conditions, and according to multiple verification indicators including at least the boundary distance consistency verification indicator and the interlocking trigger condition consistency verification indicator, the corresponding types of data in the multivariate synthesized time series data are verified respectively to obtain the verification results of each verification indicator.

[0044] The verification results are fused to obtain the comprehensive quality score of the multivariate synthesized time series data.

[0045] The verification result is determined based on the comparison between the comprehensive quality score and the preset threshold.

[0046] Another possible implementation, after aligning the variable trajectories, the safety state signal sequence, and the protection action timing chain to form multivariate synthesized timing data, further includes:

[0047] Obtain intervention parameters, which include intervention actions, intervention values, and intervention times;

[0048] The multivariate composite time series data is processed according to the intervention parameters to generate counterfactual multivariate composite time series data, and counterfactual security event time series data samples are determined based on the counterfactual multivariate composite time series data.

[0049] Secondly, embodiments of this application provide an industrial safety event time-series data synthesis apparatus, comprising:

[0050] The determination module is used to determine at least one safety-related variable, the safety boundary conditions corresponding to each safety-related variable, and the causal relationships between each safety state based on multi-source data of the target industrial system.

[0051] A construction module is used to construct a security event evolution graph to characterize the evolution process of a security event based on the security boundary conditions and the causal relationship. The security event evolution graph includes at least two security state nodes and at least one evolution edge. The evolution edge is determined according to the causal relationship and is used to characterize the state transition direction and causal transition relationship between each security state node.

[0052] A generation module is used to sample and generate at least one rare security event evolution path in the security event evolution graph, and each rare security event evolution path corresponds to a security event instance;

[0053] The generation module is also used to generate variable trajectories corresponding to each security-related variable based on the rare security event evolution path and the security boundary conditions, and to generate a global security state signal sequence and a global protection action time sequence chain corresponding to the rare security event evolution path according to the security boundary conditions. The variable trajectory is used to characterize the change process of the values ​​of the security-related variables approaching or moving away from their respective security boundaries in the time series.

[0054] The alignment module is used to time-align the variable trajectory, the safety status signal sequence, and the protection action timing chain to form multivariate composite timing data.

[0055] The generation module is also used to generate security event time series data samples based on the multivariate synthesized time series data.

[0056] One possible implementation is that the building module is specifically used for:

[0057] Based on the aforementioned safety boundary conditions, the plurality of safety state nodes are determined, each of which represents a safety state. The safety state includes at least a normal operation state, an alarm trigger state, an interlock trigger state, and a protection action execution state.

[0058] Based on the causal relationship, the state transition direction between the safe state nodes is determined, and an evolutionary edge connecting the safe state nodes is constructed based on the state transition direction.

[0059] Based on the security boundary conditions and the causal relationship, each evolutionary edge is assigned a corresponding edge weight, which represents the probability of transitioning from the current safe state node to the next safe state node along the evolutionary edge.

[0060] The security event evolution graph is constructed based on the security state nodes, the evolution edges, and the edge weights.

[0061] Another possible implementation, the generation module, is specifically used for:

[0062] Based on the security state nodes and the evolution edges, at least one candidate security event evolution path is determined in the security event evolution graph, and each candidate security event evolution path is formed by connecting the security state nodes sequentially according to the direction of the evolution edges;

[0063] Based on the security severity of each security state node in each candidate security event evolution path and the edge weight of each evolution edge, the path risk value of each candidate security event evolution path is determined.

[0064] Based on the path risk value of each candidate security event evolution path, the sampling generation probability of each candidate security event evolution path is determined, wherein the path risk value is proportional to the sampling generation probability.

[0065] Based on the sampling generation probability of each candidate security event evolution path, at least one rare security event evolution path is sampled and generated from each candidate security event evolution path.

[0066] Another possible implementation is that the generation module is specifically used for:

[0067] Based on the evolution path of the rare security event, the target security state is determined. The target security state includes the target distance range between the real-time value of the security-related variable and the security boundary, and the target duration for which the security-related variable remains within the target distance range.

[0068] Based on the safety boundary conditions and the preset sampling period, determine the real-time safety distance of each of the safety-related variables relative to the safety boundary at each sampling time;

[0069] Based on the target safety status and constrained by the real-time safety distance, a distance evolution trajectory is generated for each of the safety-related variables, which is to approach the target distance range over time and maintain the target duration after entering the target distance range.

[0070] Based on the distance evolution trajectory of each of the aforementioned safety-related variables, the variable trajectory corresponding to each of the aforementioned safety-related variables is determined.

[0071] Another possible implementation is that the security boundary conditions include alarm boundary conditions and interlocking boundary conditions;

[0072] The alarm boundary conditions include alarm thresholds and alarm duration thresholds, and the interlocking boundary conditions include interlocking triggering conditions and interlocking duration thresholds.

[0073] Another possible implementation is that the safety status signal sequence includes an alarm status signal sequence and an interlock trigger signal sequence;

[0074] The generation module is specifically used for:

[0075] Based on the variable trajectory, the alarm boundary conditions, and the interlocking boundary conditions, the values ​​of each safety-related variable at each time point are compared with the alarm threshold, and the comparison results are output.

[0076] The alarm status signal sequence is constituted based on the alarm status at each moment generated by the comparison result and the alarm duration threshold;

[0077] Determine whether the values ​​of each safety-related variable at each time point meet the interlocking trigger condition, and construct the interlocking trigger signal sequence based on the interlocking state at each time point generated by the duration of the interlocking trigger condition and the interlocking duration threshold.

[0078] When there is a moment in the interlocking trigger signal sequence that is in the triggering state, the corresponding protection action sequence is determined according to the interlocking boundary conditions, and an execution timestamp is configured for each protection action according to the time interval constraint between adjacent protection actions in the protection action sequence, thereby generating the protection action timing chain.

[0079] Another possible implementation is that the generation module is specifically used for:

[0080] Perform a security consistency check on the multivariate synthesized time series data according to the security boundary conditions, and output the check result;

[0081] If the verification result indicates that the multivariate synthesized time-series data passes the security consistency verification, the multivariate synthesized time-series data will be output as the security event time-series data sample; or

[0082] If the verification result indicates that the multivariate synthesized time series data has failed the security consistency verification, the multivariate synthesized time series data is subjected to consistency repair processing to generate new multivariate synthesized time series data. The consistency repair processing includes correction and regeneration.

[0083] Another possible implementation is that the generation module is specifically used for:

[0084] Based on the security boundary conditions, and according to multiple verification indicators including at least the boundary distance consistency verification indicator and the interlocking trigger condition consistency verification indicator, the corresponding types of data in the multivariate synthesized time series data are verified respectively to obtain the verification results of each verification indicator.

[0085] The verification results are fused to obtain the comprehensive quality score of the multivariate synthesized time series data.

[0086] The verification result is determined based on the comparison between the comprehensive quality score and the preset threshold.

[0087] Another possible implementation involves, after aligning the variable trajectory, the safety state signal sequence, and the protection action timing chain to form multivariate synthesized timing data, the generation module is further configured to:

[0088] Obtain intervention parameters, which include intervention actions, intervention values, and intervention times;

[0089] The multivariate composite time series data is processed according to the intervention parameters to generate counterfactual multivariate composite time series data, and counterfactual security event time series data samples are determined based on the counterfactual multivariate composite time series data.

[0090] Thirdly, this application provides an electronic device comprising: a processor and a memory; the memory stores a program or instructions executable on the processor, wherein the program or instructions, when executed by the processor, implement the method of the first aspect described above.

[0091] Fourthly, this application provides a readable storage medium on which a program or instructions are stored, which, when executed by a computer, implement the method of the first aspect described above.

[0092] Fifthly, this application provides a computer program product that is stored in a storage medium and implements the method of the first aspect described above when the computer program product is executed by a computer.

[0093] In a sixth aspect, embodiments of this application provide a chip including a processor and a communication interface, wherein the communication interface is coupled to the processor, and the processor is used to run programs or instructions to implement the method described in the first aspect.

[0094] The beneficial effects of the second to sixth aspects mentioned above are described in the corresponding description of the first aspect and will not be repeated here. Attached Figure Description

[0095] Figure 1 This application provides a network architecture for a method of synthesizing time-series data of industrial safety events.

[0096] Figure 2 This is a flowchart illustrating a method for synthesizing time-series data of industrial safety events, provided in an embodiment of this application.

[0097] Figure 3This is a flowchart illustrating a method for constructing a security event evolution graph, as provided in an embodiment of this application.

[0098] Figure 4 This is a flowchart illustrating a rare security event evolution path sampling method provided in an embodiment of this application.

[0099] Figure 5 This is a flowchart illustrating a variable trajectory generation method provided in an embodiment of this application.

[0100] Figure 6 This is a schematic diagram of a safety boundary distance provided for an embodiment of this application.

[0101] Figure 7 This is a flowchart illustrating a method for generating security event time-series data samples, provided in an embodiment of this application.

[0102] Figure 8 This is a flowchart illustrating a consistency verification method provided in an embodiment of this application.

[0103] Figure 9 This is a flowchart illustrating a method for generating counterfactual security event time-series data samples, provided in an embodiment of this application.

[0104] Figure 10 This is a schematic diagram of the structure of an industrial safety event time-series data synthesis device provided in an embodiment of this application.

[0105] Figure 11 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0106] The following section will describe in detail the method, apparatus, and related equipment for synthesizing time-series data of industrial safety events provided in this application, with reference to the accompanying drawings.

[0107] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0108] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0109] The terms "at least one," "at least one of," etc., used in the specification and claims of this application refer to any one, any two, or a combination of two or more of the included items. For example, at least one of a, b, and c can mean: "a," "b," "c," "a and b," "a and c," "b and c," and "a, b, and c," where a, b, and c can be single or multiple. Similarly, "at least two" refers to two or more items, and its meaning is similar to that of "at least one."

[0110] In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0111] The present application provides a method, apparatus, and related equipment for synthesizing time-series data of industrial safety events, which can be applied to data synthesis scenarios such as industrial safety early warning, interlocking logic testing, and digital twin security simulation.

[0112] As industrial digital transformation deepens, industrial time-series data is becoming increasingly crucial in industrial AI tasks such as anomaly detection, fault diagnosis, and safety early warning. However, there is an extreme scarcity of rare safety event samples, such as high-altitude alarms, interlock triggers, and emergency shutdowns, which severely restricts the improvement of industrial safety intelligence.

[0113] To alleviate the problem of data scarcity, various time series generation methods have been proposed. Rule-driven methods mainly include statistical signal synthesis and physical simulation. The former decomposes the time series into trend, periodic, and noise components for parametric modeling, while the latter generates sensor data by constructing physical models of devices, exhibiting strong interpretability. Data-driven methods mainly include GANs, VAEs, and diffusion models, which generate new samples by learning from real data distributions, demonstrating strong capabilities in representing complex distributions.

[0114] However, the existing technologies still have the following shortcomings: Industrial safety events are not triggered by a single instantaneous cross-boundary change in variables, but involve complex logic such as multi-variable coupling, state duration conditions, and causal chains of alarm-interlock-protection actions. Existing generation methods use the time series itself or simple condition labels as modeling objects, and take auxiliary information such as fault type and operating condition as overall condition inputs, making it difficult to model multiple factors separately, resulting in synthetic samples lacking rationality at the industrial logic level. At the same time, existing methods are unable to generate key scenarios that are difficult to collect in real-world situations, such as "approaching interlock but not triggered" or "protection action delayed or failed," around safety boundaries. They also lack a comprehensive quality verification mechanism oriented towards industrial physical boundaries, alarm rules, and business processes, which may produce statistically similar but industrially unreasonable samples. In addition, in cold-start scenarios where there are no fault samples for the target equipment, existing methods lack reliable sources of conditional information, making it difficult to generate highly reliable synthetic data for safety early warning model training and interlock logic testing.

[0115] To address the aforementioned technical problems, this application provides a method, apparatus, and related equipment for synthesizing time-series data of industrial safety events. By determining the causal relationships between safety-related variables, safety boundary conditions, and safety states based on multi-source data from the target industrial system, a safety event evolution graph containing state nodes and evolutionary edges is constructed. This transforms the originally implicit and scattered safety knowledge into a structured graph model, providing clear causal path constraints for time-series data generation and avoiding the problem of traditional black-box generation models easily generating unreasonable industrial samples due to a lack of causal guidance. Furthermore, by sampling rare safety event evolution paths in the safety event evolution graph, and based on these paths and safety boundary conditions, trajectories of each variable approaching or moving away from the boundary are generated, along with global safety data. The state signal sequence and protection action timing chain are then time-aligned to form multivariate synthetic timing data. This allows the final generated safety event timing data samples to not only accurately control the evolution of each safety-related quantity relative to the safety boundary at each sampling moment at the variable level, but also to synchronously match the safety state transition logic and the execution order of protection actions at the system level. This effectively solves the problem of "statistical similarity but industrial incompatibility" caused by the lack of structured constraints on the dynamic relationship between variables and boundaries in existing technologies. At the same time, it overcomes the defect that the separation of safety logic and physical dynamics makes it impossible to generate rare but high-value scenarios such as interlock precursors and protection action boundaries. This significantly improves the usability and reliability of synthetic data in practical tasks such as industrial safety early warning, interlock testing, and control system verification.

[0116] The following description, in conjunction with the accompanying drawings, details the industrial safety event time-series data synthesis method, apparatus, and related equipment provided in the embodiments of this application.

[0117] Figure 1This illustration shows the network architecture of an industrial safety event time-series data synthesis method provided in an embodiment of this application. For example... Figure 1 As shown, the network architecture includes a time-series data synthesis device 101 and a terminal device 102. The time-series data synthesis device 101 and the terminal device 102 are interconnected.

[0118] In some embodiments, the time-series data synthesis apparatus 101 may be a server, a computer, or a processor or processing unit within a server or computer. The server may be a single server or a server cluster consisting of multiple servers. It should be noted that the embodiments of this application do not limit the specific device form of the time-series data synthesis apparatus 101. Figure 1 The time-series data synthesis device 101 is used as an example of a single server.

[0119] In some embodiments, the terminal device may be a mobile phone, tablet computer, laptop computer, handheld computer, in-vehicle electronic device, mobile internet device (MID), augmented reality (AR) / virtual reality (VR) device, robot, wearable device, personal computer (PC), ultra-mobile personal computer (UMPC), netbook, or personal digital assistant (PDA), etc., and the embodiments of this application do not specifically limit it. Figure 1 The example shown is a mobile phone, with terminal device 102 as an example.

[0120] In some embodiments, the terminal device 102 is configured to acquire multi-source data of the target industrial system in response to a user's configuration operation, and send the multi-source data to the time-series data synthesis device 101. The multi-source data includes industrial historical time-series data, alarm records, interlock records, safety rules, and safety analysis documents. The time-series data synthesis device 101 is configured to receive the multi-source data sent by the terminal device 102, determine at least one safety-related variable, the safety boundary conditions corresponding to each safety-related variable, and the causal relationships between each safety state based on the multi-source data, and construct a safety event evolution diagram to characterize the evolution process of a safety event based on the safety boundary conditions and causal relationships. The safety event evolution diagram includes at least two safety state nodes and at least one evolution edge, which is determined according to the causal relationships and is used to characterize the state transition direction and causal transition relationship between each safety state node. The time-series data synthesis device 101 is further configured to sample and generate at least one rare security event evolution path in the security event evolution diagram, generate variable trajectories corresponding to each security-related variable based on the rare security event evolution path and security boundary conditions, and generate a global security state signal sequence and a global protection action time-series chain corresponding to the rare security event evolution path according to the security boundary conditions. The variable trajectory is used to characterize the change process of the values ​​of security-related variables approaching or moving away from their respective security boundaries in the time series. The time-series data synthesis device 101 is further configured to time-align the variable trajectory, security state signal sequence, and protection action time-series chain to form multivariate synthesized time-series data, generate a security event time-series data sample based on the multivariate synthesized time-series data, and send the security event time-series data sample to the terminal device 102. The terminal device 102 is further configured to receive and display the security event time-series data sample.

[0121] It should be noted that the network architecture described in the embodiments of this application is for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and does not constitute a limitation on the technical solutions provided in the embodiments of this application. As network architectures evolve, the technical solutions provided in the embodiments of this application are also applicable to similar technical problems.

[0122] See Figure 2 This is a flowchart illustrating a method for synthesizing time-series data of industrial safety events provided in an embodiment of this application. Figure 2 As shown, the industrial safety event time-series data synthesis method provided in this application embodiment can be implemented by the above-mentioned time-series data synthesis device, specifically including the following steps 201 to 206.

[0123] Step 201: The time-series data synthesis device determines at least one safety-related variable, the safety boundary conditions corresponding to each safety-related variable, and the causal relationship between each safety state based on the multi-source data of the target industrial system.

[0124] Step 202: The time-series data synthesis device constructs a security event evolution diagram to characterize the evolution process of security events based on security boundary conditions and causal relationships. The security event evolution diagram contains at least two security state nodes and at least one evolution edge. The evolution edge is determined according to the causal relationships and is used to characterize the state transition direction and causal transition relationship between each security state node.

[0125] Step 203: The time-series data synthesis device samples and generates at least one rare security event evolution path in the security event evolution graph, and each rare security event evolution path corresponds to a security event instance.

[0126] Step 204: The time-series data synthesis device generates variable trajectories corresponding to each security-related variable based on the evolution path of rare security events and security boundary conditions, and generates a global security state signal sequence and a global protection action time-series chain corresponding to the evolution path of rare security events according to the security boundary conditions. The variable trajectory is used to characterize the change process of the values ​​of security-related variables approaching or moving away from their respective security boundaries in the time series.

[0127] Step 205: The timing data synthesis device aligns the variable trajectory, safety status signal sequence, and protection action timing chain in time to form multivariate synthesized timing data.

[0128] Step 206: The time-series data synthesis device generates security event time-series data samples based on multivariate synthesized time-series data.

[0129] First, based on the multi-source data of the target industrial system, determine at least one safety-related variable, the safety boundary conditions corresponding to each safety-related variable, and the causal relationships between each safety state.

[0130] In this embodiment, the target industrial system is an industrial system for which rare safety event time-series data samples need to be generated. This can be an industrial system with a clearly defined safety protection system, such as a petrochemical plant, nuclear power unit, rail transit system, wind farm, or precision manufacturing production line. Multi-source data refers to a heterogeneous data set from multiple data sources, which may include process variable time series, control variable time series, actuator feedback time series, alarm records, interlock records, PLC / DCS / SIS configuration files, HAZOP analysis documents, LOPA reports, FMEA reports, FTA documents, operation logs, and accident or near-accident reports.

[0131] Safety-related variables refer to variables whose values ​​are directly related to the system's safety status and are governed by safety protection logic. These can include process safety variables reflecting the system's physical state (e.g., pressure, temperature, liquid level, flow rate, speed, vibration, combustible gas concentration, etc.), actuator safety variables reflecting actuator status (e.g., valve opening, pump status, enabling status, etc.), alarm variables, interlock variables, and protection action variables. Safety-related variables can be identified through one or a combination of the following methods: matching variable name, workstation number, point table information, and unit with alarm and interlock configurations; marking process variables involved in deviations as safety-related based on HAZOP deviation node analysis; and identifying variables involved in alarm logic, interlock logic, or protection action logic based on SIS / PLC / DCS configuration files.

[0132] Safety boundary conditions are quantitative descriptions of safety-related variables that trigger specific safety state transitions. They include the threshold and duration conditions required to trigger an alarm, as well as the triggering conditions, duration conditions, and corresponding protection action configurations required to trigger interlocks and initiate protection actions. Safety boundary conditions can be read from SIS / PLC / DCS configuration files and alarm management systems, or supplemented based on HAZOP deviation analysis, LOPA protection layer analysis, or expert knowledge.

[0133] Causal relationships refer to the directed causal logic existing between different safety states, describing the pattern of a system transitioning from the current safety state to the next safety state when a specific condition is met. For example, the condition "process variables continuously exceed the high-high limit threshold" will trigger a transition from the "approaching interlock boundary" state to the "interlock triggered" state, demonstrating the causal relationship between the two. Causal relationships can be extracted from the cause-deviation-effect chain of HAZOP analysis, the failure mode-effect analysis of FMEA, the top-level event-gate-basic event tree of FTA, and historical alarm interlock records. By fusing multi-source safety knowledge, a more comprehensive causal relationship can be obtained. The fusion method can use a weighted summation approach to combine the correlation scores of different knowledge sources into a comprehensive score.

[0134] In one possible implementation, the aforementioned multi-source data may further include one or more of the following: FMEA Failure Mode and Effects Analysis (FMEA) reports, FTA Fault Tree Analysis (FTA) documents, safety specifications, safety PLC configuration documents, and risk assessment documents. The system then uses knowledge fusion to determine the aforementioned causal relationships, safety boundary conditions, and each safety state node. Specifically, the system can uniformly map HAZOP deviation analysis, LOPA protection layer analysis, FMEA failure mode analysis, FTA top-level event-gate-basic event tree, safety specification clauses, safety PLC configuration logic, risk assessment conclusions, and historical alarm interlock records into a safety event knowledge representation. A weighted summation method is then used to combine the correlation scores from different knowledge sources into a comprehensive score, thereby obtaining a more comprehensive causal relationship and safety boundary conditions. Thus, even when there are very few real interlock samples, near-accident samples, or protection failure samples, the system can still initialize or update the safety event evolution diagram, safety boundary conditions, and protection action chains based on prior knowledge from FMEA, FTA, and safety specifications, achieving reliable modeling in sparse sample scenarios.

[0135] Then, based on security boundary conditions and causal relationships, a security event evolution diagram is constructed to characterize the evolution process of security events.

[0136] In this embodiment, the safety event evolution diagram is used to characterize the evolution process of a safety event, that is, the complete evolution process of a safety event in the target industrial system from its initial deviation to its final outcome. The safety event evolution diagram contains at least two safety state nodes and at least one evolution edge. The evolution edge is determined according to the causal relationship and is used to characterize the state transition direction and causal transition relationship between each safety state node. The safety event evolution diagram can be represented by the following formula (1):

[0137] (Formula 1)

[0138] in, For a set of nodes in a safe state, For the set of evolutionary edges, For a set of relation types, Let be the set of edge weights.

[0139] Each safety state node represents a safety state, which includes at least normal operation, alarm triggering, interlock triggering, and protection action execution. In this embodiment, safety states may also include initial deviation, approaching alarm boundary, approaching interlock boundary, system recovery, and protection failure states, to describe the evolution of safety events with finer granularity. Evolutionary edges are determined by causal relationships. An evolutionary edge connects a safety state node representing the "cause" to a safety state node representing the "result," and its direction reflects the causal direction of the state transition. Evolutionary edges also have edge weights, which reflect the probability of transitioning from the current safety state node to the next safety state node along the evolutionary edge. These weights can be determined by combining the fusion knowledge score and historical transition frequency.

[0140] By using a safety event evolution graph, expert knowledge and historical data in the field of industrial safety can be organically combined. Even in cold start scenarios where there are few real and rare safety event samples, the graph structure can still be initialized based on prior safety knowledge. As historical alarm and interlocking data accumulate, the weights of each evolution edge in the graph are gradually calibrated, so that the evolution graph has both knowledge interpretability and data-driven adaptability.

[0141] In one optional embodiment, the industrial safety event time-series data synthesis method also supports cold start processing in scenarios with scarce data. Specifically, when there are very few or even no samples of real interlocking data, near-accident data, or protection failure data in the target industrial system, the system can first perform cold start processing based on prior safety knowledge from FMEA analysis reports, FTA fault trees, safety specification documents, safety PLC configuration information, expert safety rules, and risk assessment documents: initializing the structure and edge weights of the safety event evolution graph, determining the safety boundary conditions of each safety-related variable, and constructing the logical constraints of the protection action chain. Subsequently, as historical alarm records, interlocking records, or drill data are gradually accumulated during system operation, the system then calibrates and updates the edge weights, boundary distance model parameters, and protection action timing of the initialized safety event evolution graph based on this real data, thereby balancing the knowledge-driven reliability of the cold start phase and the data-driven adaptability of the subsequent operation phase.

[0142] After obtaining the security event evolution graph, at least one rare security event evolution path is sampled and generated from the security event evolution graph, and each rare security event evolution path corresponds to a security event instance.

[0143] In this embodiment, the rare security event evolution path is an ordered sequence of nodes in the security event evolution graph, starting from the initial security state node, passing through several evolution edges, connecting each intermediate security state node sequentially, and ending at the final security state node. This sequence represents a specific security event instance. The rare security event evolution path is not sampled directly according to historical frequency; instead, the sampling weight of low-probability, high-risk paths is increased to ensure sufficient coverage of high-risk paths that have historically occurred very rarely. Synthesizable rare security event evolution paths include: normal operation → initial deviation → approaching alarm boundary → no alarm triggered (approaching but not exceeding the boundary); normal operation → initial deviation → alarm triggered → approaching interlock boundary → interlock triggered → protection action execution → system recovery; normal operation → interlock triggered → protection action delay → shutdown, etc. Through diverse path sampling, various security event scenarios, ranging from early deviations to catastrophic failures, can be covered.

[0144] Subsequently, based on the evolution path of rare security events and security boundary conditions, variable trajectories corresponding to each security-related variable are generated, and global security state signal sequences and global protection action timing chains corresponding to the evolution path of rare security events are generated according to the security boundary conditions.

[0145] In this embodiment, the variable trajectory is a time-series curve describing the change process of the values ​​of safety-related variables as they approach or move away from their corresponding safety boundaries. Guided by the evolution path of rare safety events obtained from sampling and constrained by safety boundary conditions, a corresponding time-series trajectory is generated for each safety-related variable, ensuring that the distance of each safety-related variable relative to the safety boundary at each sampling time conforms to the semantic requirements of the corresponding safety state node in the path. For example, for the "interlock triggering" path segment, the value of the relevant variable needs to continuously exceed the interlock triggering threshold to reach the duration specified by the interlock duration threshold; while for the "approaching the alarm boundary but not triggered" path segment, the value of the relevant variable needs to approach but not exceed the alarm threshold.

[0146] The global safety status signal sequence includes an alarm status signal sequence generated by comparing variable values ​​at each time moment with alarm thresholds and combining this with alarm duration conditions, and an interlocking trigger signal sequence generated by determining whether variable values ​​at each time moment meet interlocking trigger conditions and combining this with interlocking duration thresholds. The global protection action timing chain is an ordered sequence of protection actions and its execution timestamps generated when a trigger state exists in the interlocking trigger signal sequence, based on the protection action configuration specified by the interlocking boundary conditions and the time interval constraints between adjacent protection actions.

[0147] Subsequently, the variable trajectories, safety status signal sequences, and protection action timing chains are time-aligned to form multivariate composite timing data.

[0148] In this embodiment, time alignment refers to uniformly sampling and splicing the variable trajectories of each safety-related variable, the alarm state signal sequence, the interlock trigger signal sequence, and the protection action timing chain on the same time axis. This ensures that all channels correspond to a consistent safety event evolution state at each sampling moment, thereby forming multivariate synthetic time-series data that can be directly used for model training or system verification. Multivariate synthetic time-series data can be represented as a multi-channel time-series matrix including safety-related variable time series, alarm state time series, interlock state time series, and protection action state time series.

[0149] Finally, based on multivariate synthesized time-series data, security event time-series data samples are generated.

[0150] In this embodiment, a security consistency check can be performed on the multivariate synthesized time-series data. If the check passes, it is output as a security event time-series data sample; if the check fails, the multivariate synthesized time-series data is corrected or regenerated. The security event time-series data sample may include tag information such as target security event type, security boundary distance label, alarm trigger label, interlock trigger label, protection action chain label, and comprehensive quality score.

[0151] In this embodiment, multi-source industrial safety knowledge is fused with historical data. A safety event evolution diagram serves as the framework to guide the sampling of rare safety scenarios. Safety boundary conditions constrain the generation of variable trajectories. Time alignment forms multivariate synthetic time-series data, and a safety consistency verification closed loop ensures sample quality. This allows for the systematic synthesis of rare safety event data covering the complete temporal logic chain from early deviation, alarm triggering, interlock intervention to protection action execution. Compared to existing synthesis methods such as GANs, VAEs, and diffusion models that only aim at statistical distribution similarity, the data synthesized in this embodiment meets industrial safety constraints and can be directly used for high-value safety scenarios such as safety early warning model training, alarm interlock system testing, control system verification, and operator training. This effectively solves the problem of an extreme lack of rare safety event samples in industrial settings.

[0152] In an optional embodiment, the model parameters in the above-described industrial safety event time-series data synthesis method can be determined by a phased training strategy and a joint fine-tuning mechanism.

[0153] Specifically, the system can adopt a phased training strategy, completing the following training stages sequentially: First, determine safety-related variables based on multi-source data to identify process safety variables, actuator safety variables, alarm variables, interlock variables, and protection action variables governed by safety protection logic; Second, determine safety boundary conditions based on safety rule documents and historical data; Third, fuse HAZOP, LOPA, FMEA, FTA, and historical alarm interlock data to obtain causal relationships; Fourth, construct a safety event evolution graph based on the fused causal relationships and safety boundary conditions, and update the edge weights of each evolution edge; Fifth, generate variable trajectories corresponding to each safety-related variable using safety boundary conditions as constraints; Sixth, generate a global safety state signal sequence based on alarm and interlock boundary conditions; Seventh, generate a global protection action time sequence chain based on interlock boundary conditions; Eighth, time-align the variable trajectories, safety state signal sequences, and protection action time sequence chains to form multivariate synthetic time-series data. After each stage of processing is completed, the system can also introduce the security consistency verification results for joint fine-tuning. By using feedback optimization, the processing parameters of each stage can be adjusted so that the final generated synthetic data satisfies both local constraints and global security logic consistency.

[0154] See Figure 3 This is a flowchart illustrating a method for constructing a security event evolution graph according to an embodiment of this application.

[0155] In an alternative embodiment, such as Figure 3 As shown, the method for constructing the security event evolution diagram includes the following steps 301 to 304.

[0156] Step 301: The timing data synthesis device determines multiple safety state nodes based on the safety boundary conditions. Each safety state node represents a safety state, which includes at least the normal operation state, alarm trigger state, interlock trigger state, and protection action execution state.

[0157] Step 302: The time-series data synthesis device determines the state transition direction between safe state nodes based on the causal relationship, and constructs the evolution edge connecting the safe state nodes based on the state transition direction.

[0158] Step 303: The time-series data synthesis device assigns corresponding edge weights to each evolution edge according to the security boundary conditions and causal relationships. The edge weights are used to represent the probability of transitioning from the current safe state node to the next safe state node along the evolution edge.

[0159] Step 304: The time-series data synthesis device constructs a security event evolution diagram based on the security state nodes, evolution edges, and edge weights.

[0160] First, based on the security boundary conditions, multiple security state nodes are determined.

[0161] In this embodiment, the semantics of the safety state node is determined by the safety state interval defined by the safety boundary conditions. For example, when the value of a process variable is lower than the alarm threshold, the system is in the state interval corresponding to the normal operation state node; when the value of the process variable exceeds the alarm threshold and continues to reach the alarm duration threshold, the system enters the state interval corresponding to the alarm trigger state node; when the interlock trigger condition continuously meets the duration specified by the interlock duration threshold, the system enters the state interval corresponding to the interlock trigger state node; when the protection action is activated after the interlock is triggered, the system enters the state interval corresponding to the protection action execution state node.

[0162] Each of the above safety status nodes represents a safety status. Safety status includes at least normal operation status, alarm trigger status, interlock trigger status, and protection action execution status. It may also include initial deviation status, near alarm boundary status (indicating that the variable value is close to but has not yet exceeded the alarm threshold), near interlock boundary status (indicating that the variable value has exceeded the alarm threshold but has not yet met the interlock trigger duration condition), system recovery status, and protection failure status, etc.

[0163] Then, based on the causal relationship, the state transition direction between safe state nodes is determined, and evolutionary edges connecting safe state nodes are constructed according to the state transition direction.

[0164] In this embodiment, the state transition direction can reflect the temporal causal logic of events. For example, when the variable value of a normal operating state node begins to deviate from the normal range, it transitions along the evolution edge to the initial deviation state node; when the variable value of an initial deviation state node continues to approach the alarm threshold, it transitions along the evolution edge to the near alarm boundary state node; when the variable value of a near alarm boundary state node exceeds the alarm threshold and continues to meet the alarm duration condition, it transitions along the evolution edge to the alarm trigger state node; and so on until the termination state node such as protection action execution or shutdown.

[0165] Then, based on the safety boundary conditions and causal relationships, corresponding edge weights are assigned to each evolutionary edge. The edge weight represents the probability of transitioning from the current safe state node to the next safe state node along that evolutionary edge.

[0166] In this embodiment, the edge weights can be obtained by normalizing the weighted fusion of HAZOP knowledge scores, FMEA knowledge scores, FTA knowledge scores, historical co-occurrence frequency scores, and expert prior scores using a normalization function. During the cold start phase when historical security event samples are scarce, the weight of the historical co-occurrence frequency score is relatively low, and the system mainly relies on the prior knowledge provided by security analysis documents such as HAZOP, LOPA, FMEA, and FTA to initialize the edge weights. As historical alarm chains and interlocking data are gradually accumulated, the weight of the historical co-occurrence frequency score can be gradually increased, allowing the evolution graph weights to balance prior knowledge with data-driven calibration.

[0167] In addition, each safety status node can be assigned a safety severity value to characterize the severity of damage to production safety and personnel safety after the system enters the state. The safety severity value can be obtained directly from the severity rating of FMEA or the risk matrix of HAZOP analysis, or it can be set by safety engineers based on their professional knowledge.

[0168] Finally, a security event evolution graph is constructed based on the security state nodes, evolution edges, and edge weights. The security event evolution graph constructed in this way integrates security boundary conditions, causal relationships, historical frequency information, and expert knowledge, providing a structured foundation for subsequent rare path sampling.

[0169] See Figure 4 This is a flowchart illustrating a rare security event evolution path sampling method provided in an embodiment of this application.

[0170] In an alternative embodiment, such as Figure 4 As shown, the rare security event evolution path sampling method includes the following steps 401 to 404.

[0171] Step 401: The time-series data synthesis device determines at least one candidate security event evolution path in the security event evolution graph based on the security state nodes and evolution edges. Each candidate security event evolution path is formed by connecting the security state nodes sequentially according to the direction of the evolution edges.

[0172] Step 402: The time-series data synthesis device determines the path risk value of each candidate security event evolution path based on the security severity of each security state node in each candidate security event evolution path and the edge weight of each evolution edge.

[0173] Step 403: The time-series data synthesis device determines the sampling generation probability of each candidate security event evolution path based on the path risk value of each candidate security event evolution path. The path risk value is proportional to the sampling generation probability.

[0174] Step 404: According to the sampling generation probability of each candidate security event evolution path, the time series data synthesis device samples and generates at least one rare security event evolution path from each candidate security event evolution path.

[0175] First, based on the security state nodes and evolution edges, at least one candidate security event evolution path is determined in the security event evolution graph.

[0176] In this embodiment, at least one candidate security event evolution path is determined in the security event evolution graph based on the security state nodes and evolution edges. The candidate security event evolution path is formed by sequentially connecting security state nodes according to the direction of the evolution edges, representing a complete security event evolution scenario. The candidate security event evolution path can be determined by performing directed path enumeration on the security event evolution graph or by using graph traversal methods such as depth-first search. Each candidate path starts from an initial node (e.g., a normal operation state node), passes through a series of evolution edges, and finally reaches a termination node (e.g., a system recovery state node, a protection failure state node, or a shutdown state node).

[0177] Then, based on the security severity of each security state node in each candidate security event evolution path and the edge weight of each evolution edge, the path risk value of each candidate security event evolution path is determined.

[0178] In this embodiment, the path risk value can comprehensively consider the security severity of each node traversed by the path and the conditional transition probability of each node, assigning a higher path risk value to paths with a low probability of occurrence in history but serious security consequences. Specifically, the path risk value can be defined as the negative of the security severity of each node in the path and the probability of occurrence of the corresponding node under the conditions of its parent node. The sum of the products of powers, where This is the tail enhancement coefficient, which takes a value greater than zero. It is used to amplify the contribution of low-probability nodes to the risk value, thereby increasing the weight of rare but high-risk paths being sampled.

[0179] Alternatively, the path risk value can be calculated using the following formula (2):

[0180] (Formula 2)

[0181] in, Let π be the path risk value of the π-th candidate security event evolution path. For safe state nodes The severity of the security situation It is a safe state node Along the evolutionary edge Transfer to a safe state node The conditional transition probability, This is the tail enhancement coefficient, which takes a value greater than zero and is used to amplify the contribution of low-probability nodes to the risk value.

[0182] Subsequently, based on the path risk value of each candidate security event evolution path, the sampling generation probability of each candidate security event evolution path is determined, and the path risk value is proportional to the sampling generation probability.

[0183] In this embodiment, the sampling generation probability can be obtained by analyzing the path risk value. The results are obtained by normalization and the introduction of a sampling temperature parameter. The larger the sampling temperature parameter, the more concentrated the sampling probability of high-risk paths is relative to low-risk paths; the smaller the sampling temperature parameter, the more uniform the sampling probability of each path tends to be.

[0184] Alternatively, the sampling generation probability can be calculated using the following formula (3):

[0185] (Formula 3)

[0186] in: For the first The sampling generation probability of candidate security event evolution paths. For the first The path risk value of each candidate security event evolution path. The sampling temperature parameter is used to adjust the concentration of the sampling probability distribution between high-risk paths and low-risk paths. The larger the value, the more concentrated the probability of high-risk paths being sampled is relative to low-risk paths.

[0187] Finally, based on the sampling generation probability of each candidate security event evolution path, at least one rare security event evolution path is sampled and generated from each candidate security event evolution path.

[0188] Through the above mechanism of driving sampling probability with path risk value, time series samples corresponding to paths that have occurred very rarely in history but have serious safety consequences can be generated in a targeted manner, thereby effectively covering high-value and rare scenarios such as "approaching interlock but not triggered", "just triggered interlock", "protection action successfully executed", and "protection action delayed or failed".

[0189] See Figure 5 This is a flowchart illustrating a variable trajectory generation method provided in an embodiment of this application.

[0190] In an alternative embodiment, such as Figure 5 As shown, the variable trajectory generation method includes the following steps 501 to 504.

[0191] Step 501: The time-series data synthesis device determines the target security state based on the evolution path of rare security events. The target security state includes the target distance range between the real-time values ​​of security-related variables and the security boundary, and the target duration during which the security-related variables remain within the target distance range.

[0192] Step 502: The time-series data synthesis device determines the real-time safe distance of each safety-related variable relative to the safety boundary at each sampling time based on the safety boundary conditions and the preset sampling period.

[0193] Step 503: The time-series data synthesis device generates, based on the target's security status and constrained by the real-time security distance, a distance evolution trajectory in which the real-time security distance of each security-related variable approaches the target's distance range over time and maintains the target's continuous duration after entering the target's distance range.

[0194] Step 504: The time-series data synthesis device determines the variable trajectory corresponding to each safety-related variable based on the distance evolution trajectory of each safety-related variable.

[0195] First, determine the target's security status based on the evolution path of rare security events.

[0196] In this embodiment, the target safety status includes the target distance range between the real-time values ​​of safety-related variables and the safety boundary, and the target duration for which the safety-related variables remain within the target distance range. For example, for a "just triggered interlock" node in the path, the target distance range is when the variable exceeds the interlock trigger threshold, and the duration of exceeding the threshold exactly meets the interlock duration threshold; for a "close to the interlock boundary but not triggered" node in the path, the target distance range is when the difference between the variable value and the interlock trigger threshold is within a near-boundary interval that is less than zero, and the duration of meeting the condition does not exceed the interlock duration threshold.

[0197] Then, based on the safety boundary conditions and the preset sampling period, the real-time safety distance of each safety-related variable relative to the safety boundary at each sampling time is determined.

[0198] In this embodiment, the preset sampling period can be set according to the dynamic characteristics of the actual industrial process and the acquisition frequency of the data acquisition system. The preset sampling period discretizes continuous time into a sequence of equally spaced sampling moments. For each sampling moment, the current sampled value of each safety-related variable is obtained, and the real-time safety distance of each safety-related variable relative to the safety boundary is determined using the safety boundary as a reference. Specifically, the real-time safety distance reflects the margin between the variable value at the current sampling moment and the safety boundary. For the upper boundary, the safety distance is defined as the difference between the boundary threshold and the current value; for the lower boundary, the safety distance is defined as the difference between the current value and the boundary threshold. A positive safety distance indicates that the variable value has not yet crossed the safety boundary, while a safety distance of zero or negative indicates that the variable value has crossed the safety boundary.

[0199] Subsequently, based on the target's safety status and constrained by the real-time safety distance, the distance evolution trajectory of each safety-related variable is generated, showing how the real-time safety distance approaches the target distance range over time and maintains the target's duration after entering the target distance range.

[0200] In this embodiment, the distance evolution trajectory can be generated in the following way: a parameterized function (such as an exponential decay function) is used to describe the overall trend of the real-time safe distance of the variable gradually decreasing from the initial value to approach the target distance range over time, and a random disturbance term is superimposed to simulate the measurement noise and process fluctuations in the real industrial process. At the same time, distance boundary constraints are applied throughout the generation process to ensure that the distance evolution trajectory meets the requirements of the target distance range and the target duration within the final time window.

[0201] Finally, based on the distance evolution trajectory of each safety-related variable, the variable trajectory corresponding to each safety-related variable is determined.

[0202] In this embodiment, the distance evolution trajectory can be restored to the time series value of the original variable domain based on the relationship between the safety boundary threshold and the distance evolution trajectory, thereby obtaining the specific numerical trajectory of safety-related variables on the time axis.

[0203] By using the above-mentioned variable trajectory generation method with real-time safety distance as the intermediate representation, the trajectory of each safety-related variable can naturally reflect the process of "gradually approaching or deviating from the safety boundary from the initial normal state" without directly modeling the high-dimensional multivariate time series space, thereby reducing the generation difficulty and improving the consistency between the trajectory and the safety boundary constraints.

[0204] In one optional embodiment, the safety boundary conditions include alarm boundary conditions and interlock boundary conditions. The alarm boundary conditions include an alarm threshold and an alarm duration threshold, while the interlock boundary conditions include an interlock trigger condition and an interlock duration threshold.

[0205] In this embodiment, alarm boundary conditions are quantified conditions that must be met to trigger the safety system to issue an alarm signal, including alarm thresholds and alarm duration thresholds. Interlock boundary conditions are quantified conditions that must be met to trigger the safety system to perform protective actions, including interlock trigger conditions and interlock duration thresholds.

[0206] Specifically, the alarm threshold limits the range of values ​​for safety-related variables. It can be a single variable's upper limit, lower limit, high-high limit, or low-low limit, or a combination of multiple variables. The alarm duration threshold specifies the minimum duration a variable value exceeding the alarm threshold must remain continuously to prevent false alarms caused by momentary fluctuations; an alarm state is only determined when the variable value exceeds the alarm threshold and remains continuously for the duration specified by the alarm duration threshold. The interlock trigger condition is the logical condition that triggers the protection action. It can be a single variable exceeding the limit condition or a logical combination of multiple variables (e.g., variable A exceeds the threshold while variable B is below the threshold). The interlock duration threshold specifies the minimum duration an interlock trigger condition must be continuously satisfied; the corresponding protection action is only executed when the interlock trigger condition is continuously satisfied for the duration specified by the interlock duration threshold. The protection actions corresponding to the interlock trigger condition can include one or more sequential actions such as closing the feed valve, cutting off heating power, initiating emergency relief, and stopping pump operation.

[0207] Figure 6 This is a schematic diagram of a safety boundary distance provided for an embodiment of this application.

[0208] like Figure 6 As shown, the relationship between process variables and alarm boundaries and interlock boundaries is illustrated. By modeling boundary distances, the synthetic sample can cover different safety boundary areas, including states such as far from the boundary, proximity alarm, triggered alarm, proximity interlock, and triggered interlock.

[0209] By clearly distinguishing between alarm boundary conditions and interlock boundary conditions, the dual-layer protection structure in the industrial safety protection system can be accurately modeled. The alarm layer provides early warning, while the interlock layer triggers automatic protection actions when the alarm is not responded to in a timely manner. This provides a strict logical constraint basis for the subsequent generation of safety status signal sequences and protection action timing chains.

[0210] In one optional embodiment, the safety status signal sequence includes an alarm status signal sequence and an interlock trigger signal sequence.

[0211] In this embodiment, the alarm status signal sequence is generated as follows: based on the variable trajectory, alarm boundary conditions, and interlocking boundary conditions, the values ​​of each safety-related variable at each time point are compared with the alarm threshold, and the comparison results are output; the alarm status signal sequence is formed based on the alarm status at each time point generated by the comparison results and the alarm duration threshold.

[0212] Specifically, for high-limit alarms, when the value of a safety-related variable exceeds the alarm threshold at a certain moment and this exceeding state has continued for a duration equal to or exceeding the alarm duration threshold, the alarm state at that moment is a triggered state; otherwise, it is a non-triggered state. The alarm state signal sequence can also include visible alarm states in alarm suppression scenarios, that is, when the alarm is suppressed, even if the value of the safety-related variable exceeds the alarm threshold, the visible alarm state is still a non-triggered state; it can also include visible alarm states in false alarm scenarios, that is, even if the value of the safety-related variable does not exceed the alarm threshold, the visible alarm state may be a triggered state due to the influence of false triggering signals.

[0213] In an optional embodiment, the generation process of the above-mentioned safety status signal sequence may further include the synthesis of sensor false triggering, missed triggering, bypassing, suppression, or detection delay scenarios. Specifically, for false triggering scenarios of safety input devices, false alarm samples with a visible alarm state as the triggered state can be synthesized when the variable trajectory is normal but a trigger state signal is still generated; for missed triggering scenarios, missed alarm samples where the alarm or interlock is not detected can be synthesized when the variable trajectory meets the alarm or interlock triggering conditions but a non-triggered state signal is still generated; for suppression scenarios, suppression samples where the visible alarm state is the non-triggered state but the theoretical alarm state is the triggered state can be synthesized when the variable trajectory meets the alarm threshold but the alarm is suppressed by the operator or logic; for detection delay scenarios, detection delay samples can be synthesized when the variable trajectory meets the triggering conditions but the signal state changes to the triggered state after a certain period of time. By synthesizing the above-mentioned sensor false triggering, missed triggering, suppression, and detection delay scenarios, synthetic time-series data covering more abnormal operating conditions of safety input devices can be generated to improve the robustness of safety early warning models and interlock tests to abnormal signals.

[0214] In this embodiment, the process of generating the interlocking trigger signal sequence is as follows: determine whether the values ​​of each safety-related variable at each time point meet the interlocking trigger conditions, and generate the interlocking state at each time point based on the duration of the interlocking trigger conditions and the interlocking duration threshold. The interlocking states at each time point together constitute the interlocking trigger signal sequence.

[0215] Specifically, when the interlocking trigger condition is met at a certain moment and the duration of continuous satisfaction reaches or exceeds the interlocking duration threshold, the interlocking state at the corresponding moment and its subsequent moments (until the interlocking trigger condition is no longer met or the system is reset) is the triggered state; otherwise, it is the non-triggered state.

[0216] In this embodiment, the generation process of the global protection action sequence chain is as follows: when there is a moment in the interlocking trigger signal sequence that is in the triggering state, the corresponding protection action sequence is determined according to the interlocking boundary conditions, and an execution timestamp is configured for each protection action according to the time interval constraint between adjacent protection actions in the protection action sequence, thereby generating the protection action sequence chain.

[0217] Specifically, time interval constraints can specify the minimum and maximum values ​​of the execution time difference between adjacent protection actions, reflecting the sequential dependency and time constraints of protection action execution. For example, the action of closing the feed valve must be completed within a few milliseconds after the interlock is triggered, while the action of initiating emergency relief must be executed after a specified delay following the completion of the feed valve closing action. The protection action sequence chain is expressed by an ordered set of "action-execution timestamp" pairs. The state value of each protection action at the time corresponding to its execution timestamp changes from untriggered to triggered, and then remains in the triggered state until the system recovers or is reset.

[0218] In an optional embodiment, the process of generating the global protection action sequence chain described above can also distinguish and synthesize different security event scenarios such as successful protection action, delayed protection action, failed protection action, or partial execution of protection action.

[0219] Specifically, for successful protection actions, the protection action sequence chain reflects the normal execution of each protection action within its specified time interval constraints. For delayed protection actions, a delay parameter can be introduced into the protection action sequence chain to make the execution timestamp of a certain protection action later than its specified time interval upper limit, thereby synthesizing a delayed protection action sample. For failed protection actions, a certain protection action can be marked as not executed or failed in the protection action sequence chain, thereby synthesizing a failed protection action sample. For partially executed protection actions, some protection actions can be marked as executed while the rest are not executed, thereby synthesizing a partially executed sample. By distinguishing and synthesizing the above different protection action scenarios, diverse synthetic time-series data covering the entire lifecycle of protection actions can be generated.

[0220] Using the above method, based on variable trajectories, alarm status signal sequences, interlock trigger signal sequences, and protection action timing chains are gradually deduced according to alarm rules and interlocking rules, ensuring logical consistency among the three: interlock triggering is preceded by alarm status, and protection action is triggered by interlock triggering. The causal logic chain between signals at each level is clear and complete, providing inherently consistent security semantics for the multivariate synthetic timing data formed after time alignment.

[0221] See Figure 7 This is a flowchart illustrating a method for generating security event time-series data samples according to an embodiment of this application.

[0222] In an alternative embodiment, such as Figure 7 As shown, the method for generating time-series data samples of security events includes the following steps 701 to 703.

[0223] Step 701: The time series data synthesis device performs a security consistency check on the multivariate synthesized time series data according to the security boundary conditions and outputs the check results.

[0224] Step 702: If the verification result indicates that the multivariate synthesized time series data has passed the security consistency verification, the time series data synthesis device outputs the multivariate synthesized time series data as a security event time series data sample.

[0225] Step 703: If the verification result indicates that the multivariate composite time series data has failed the security consistency verification, perform consistency repair processing on the multivariate composite time series data to generate new multivariate composite time series data. The consistency repair processing includes correction and regeneration.

[0226] In this embodiment, if the verification result indicates that the multivariate synthesized time-series data passes the security consistency check, the multivariate synthesized time-series data is output as a security event time-series data sample. If the verification result indicates that the multivariate synthesized time-series data fails the security consistency check, the multivariate synthesized time-series data is corrected and / or regenerated to generate new multivariate synthesized time-series data, and the security consistency check is performed again on the new multivariate synthesized time-series data. This process is repeated until the check is passed or the maximum number of iterations is reached. Correction operations may include: adjusting the approximation speed of the boundary distance trajectory; adjusting the alarm duration; adjusting the interlock trigger time; correcting the execution timestamp of the protection action chain; or resampling for time periods that do not meet the constraints. In practical applications, when there are multiple multivariate synthesized time-series data, each multivariate synthesized time-series data can be corrected and / or regenerated according to the actual consistency check results.

[0227] In an optional embodiment, the aforementioned security event time-series data samples can also output structured labels to facilitate subsequent processing and analysis. Specifically, each generated security event time-series data sample can be accompanied by at least one of the following label information: equipment type label, used to identify the equipment type of the target industrial system; security event type label, used to identify the type of security event; security path label, used to identify the complete evolution path of the security event; boundary type label, used to identify the type of security boundary involved; alarm trigger label, used to identify whether an alarm was triggered; interlock trigger label, used to identify whether an interlock was triggered; protection action chain label, used to identify the sequence of protection actions executed; counterfactual intervention type label, used to identify the type of intervention in the counterfactual scenario; near-accident level label, used to assess the severity of the security event; and comprehensive quality score label, used to quantify the degree of safety consistency of the sample. By providing the above-mentioned structured labels to the synthetic samples, the application value of the synthetic data in downstream tasks can be enhanced.

[0228] See Figure 8 This is a flowchart illustrating a consistency verification method provided in an embodiment of this application.

[0229] In an alternative embodiment, such as Figure 8 As shown, the consistency verification method includes the following steps 801 to 803.

[0230] Step 801: The time series data synthesis device performs verification on the corresponding type of data in the multivariate synthesized time series data according to the safety boundary conditions and multiple verification indicators, including at least the boundary distance consistency verification indicator and the interlocking trigger condition consistency verification indicator, to obtain the verification results of each verification indicator.

[0231] Step 802: The time series data synthesis device merges the verification results to obtain the comprehensive quality score of the multivariate synthesized time series data.

[0232] Step 803: The time-series data synthesis device determines the verification result based on the comparison result of the comprehensive quality score and the preset threshold.

[0233] In this embodiment, the boundary distance consistency verification index is used to verify whether the deviation between the real-time safe distance and the target distance range of each safety-related variable in the synthesis time series is within an acceptable range. It can be measured by calculating the sum of the absolute values ​​of the difference between the safe distance and the target safe distance in the synthesis time series at each time. The smaller the deviation, the higher the degree of consistency between the variable trajectory and the target safety state.

[0234] The interlocking trigger condition consistency verification index is used to verify whether the generation of interlocking trigger signals in the synthesized timing strictly conforms to the interlocking trigger conditions and interlocking duration thresholds. It can be measured by calculating the difference between the synthesized interlocking state and the theoretical interlocking state derived from the variable trajectory according to the interlocking rules.

[0235] In addition, multiple verification indicators may include alarm duration consistency verification indicators (used to verify the degree of consistency between the duration of the alarm status signal and the alarm duration threshold), protection action chain consistency verification indicators (used to verify whether the execution timestamp of the protection action meets the adjacent action interval constraint), and physical feasibility verification indicators (used to verify whether each safety-related variable exceeds its physical design range), etc.

[0236] The verification results of each verification indicator are fused to obtain a comprehensive quality score for the multivariate synthesized time series data. The fusion method can be a weighted sum of the verification results of each indicator according to preset weights. A higher comprehensive quality score indicates a higher degree of satisfaction with the security constraints across all dimensions of the multivariate synthesized time series data. The final verification result is determined based on a comparison between the comprehensive quality score and a preset threshold: if the comprehensive quality score is not lower than the preset threshold, the verification result is passed; if the comprehensive quality score is lower than the preset threshold, the verification result is failed.

[0237] Through a closed-loop mechanism of multi-dimensional security consistency verification and feedback correction, synthetic samples that are statistically reasonable but industrially unreasonable can be effectively filtered out. For example, variable values ​​may have exceeded the interlocking threshold but the interlocking signal has not been triggered, or the interlocking signal has been triggered but the protection action execution timestamp is missing. This ensures that the final output security event time series data samples meet industrial security constraints and have the quality assurance to serve as highly reliable training data.

[0238] In an optional embodiment, the aforementioned security consistency verification step can also be combined with a feedback optimization mechanism to further improve the quality of the synthesized data. Specifically, the security consistency verification results can be used for feedback optimization, rather than just as post-processing filtering conditions. For multivariate synthesized time-series data that fails the security consistency verification, the specific verification indicators that failed can be analyzed, and targeted adjustments can be made to these indicators: for example, adjusting the approach speed of the boundary distance trajectory, adjusting the alarm duration, adjusting the interlock trigger time, correcting the execution timestamp of the protection action chain, adjusting the rate of change of safety-related variables, or regenerating for time periods that do not meet the constraints locally. In addition, the sampling generation probability of rare security event evolution paths can be adjusted according to the proportion of samples that failed the verification, so that high-risk paths that are difficult to generate qualified samples can get more sampling opportunities. Alternatively, the multivariate synthesized time-series data that failed the verification can be corrected or regenerated, and the corrected data can be subjected to security consistency verification again, and so on until the verification is passed or the maximum number of iterations is reached. By introducing the above-mentioned feedback optimization mechanism, the overall quality of the synthesized data can be continuously improved during the iteration process.

[0239] See Figure 9 This is a flowchart illustrating a method for generating counterfactual security event time-series data samples, provided in an embodiment of this application.

[0240] In an alternative embodiment, such as Figure 9 As shown, the method for generating time-series data samples of counterfactual security events includes the following steps 901 to 903.

[0241] Step 901: The time series data synthesis device acquires intervention parameters, which include intervention actions, intervention values, and intervention times.

[0242] Step 902: The time series data synthesis device processes the multivariate synthesized time series data according to the intervention parameters to generate counterfactual multivariate synthesized time series data, and determines the counterfactual security event time series data sample based on the counterfactual multivariate synthesized time series data.

[0243] First, obtain the intervention parameters, which include the intervention action, intervention value, and intervention time.

[0244] In this embodiment, an intervention action refers to a human or hypothetical operation performed on the system in a counterfactual scenario, such as prematurely closing the feed valve at a certain time, adjusting the operator's given value from its current value to a specific value at a certain time, or restoring a suppressed alarm state at a certain time. The intervention value is the specific numerical value assigned to the corresponding variable after the intervention action is executed. The intervention time is the point in time on the timeline where the intervention action occurs. Time series data before the intervention time remains unchanged, while time series data after the intervention time is recalculated based on the intervention parameters.

[0245] Then, the multivariate composite time series data are processed according to the intervention parameters to generate counterfactual multivariate composite time series data, so as to determine the counterfactual security event time series data samples based on the counterfactual multivariate composite time series data.

[0246] In this embodiment, the counterfactual multivariate composite time-series data is composite time-series data under the hypothetical scenario of "how the system will evolve if a certain intervention action occurs at a certain time". By comparing the original multivariate composite time-series data with the counterfactual multivariate composite time-series data, the impact of specific intervention measures on the evolution path of the system's safety state can be quantitatively analyzed: for example, if the feed rate is reduced in advance, can the variable value be prevented from exceeding the interlock trigger threshold, thereby avoiding interlock intervention; if the alarm is not suppressed, can the operator respond in advance, thereby preventing the system from entering the interlock trigger state; if the protection action is delayed, how will the system's safety state further deteriorate?

[0247] The counterfactual intervention safety trajectory can also support multiple interventions, that is, to apply multiple intervention actions at different times simultaneously and calculate the corresponding counterfactual time-series evolution results for each, so as to cover more intervention combination scenarios.

[0248] Through the aforementioned counterfactual intervention mechanism, this embodiment can associate several counterfactual samples with each sample while synthesizing rare security event time series data, providing high-quality paired training data for causal inference model training, safety operation procedure optimization, and counterfactual security simulation, thereby further enhancing the application value of synthetic data in industrial safety intelligent systems.

[0249] The following specific embodiments illustrate the method for synthesizing time-series data of industrial safety events according to this application.

[0250] In this embodiment, the above-described method for synthesizing time-series data of industrial safety events is applied to rare safety event scenarios for collaborative industrial robots to illustrate the complete implementation process of the method.

[0251] I. Application Scenarios and Data Sources

[0252] A high-end manufacturing production line has deployed multiple six-axis collaborative industrial robots for precision assembly, loading and unloading, grinding, tightening, and handling operations. The robots are equipped with servo drive systems, joint torque sensors, end effector six-dimensional force sensors, safety laser scanners, safety light curtains, safety doors, emergency stop buttons, safety PLCs, brakes, and robot controllers.

[0253] In real production, the probability of safety incidents such as personnel approaching danger zones, end effectors approaching collision boundaries, and safety doors opening and triggering interlocks is extremely low. It is impossible to collect such incidents by artificially creating dangerous scenarios. However, these samples are crucial for robot safety early warning, interlock testing, and operator training.

[0254] The system integrates multi-source data from collaborative industrial robots to form multivariate time-series samples, specifically defined as follows:

[0255]

[0256]

[0257] in, Represents the joint angle vector. Represents the joint velocity vector. Represents the joint torque vector. Indicates the end position. Indicates the terminal velocity. Indicates that the end is subjected to force. Indicates servo current. This represents the minimum distance between the robot and the person. Indicates the minimum distance from the end to the fixture or tooling. Indicates the status of the safety light curtain. Indicates the status of the safety door. Indicates an emergency stop. Indicates the safety PLC interlock status. Indicates the servo enable status. Indicates the brake status. This indicates the robot's operating mode. In this embodiment, the sampling period is 10 ms, and the length of each sample is 30 s.

[0258] II. Identification of Safety-Related Variables and Determination of Safety Boundary Conditions

[0259] Based on multi-source data from collaborative industrial robots, the causal relationships between safety-related variables, safety boundary conditions, and safety states are determined.

[0260] The system identifies a set of safety-related variables based on the robot controller point table, safety PLC configuration, and risk assessment document:

[0261]

[0262]

[0263] And they are divided according to function as follows:

[0264] Process safety variables: ;

[0265] Executor variables: ;

[0266] Safety input alarm variables: ;

[0267] Interlocking variables: ;

[0268] Protection action variables: .

[0269] This embodiment defines three types of target safety events corresponding to safety boundary conditions: personnel approach safety boundary events, collision boundary events, and safety door / emergency stop interlocking events.

[0270] For events involving personnel approaching the safety boundary, the alarm boundary conditions are as follows:

[0271] Alarm threshold: ;

[0272] Alarm duration threshold: ;

[0273] The interlocking boundary conditions are:

[0274] Interlocking trigger conditions: And the duration is not less than Tstop;

[0275] in, , The corresponding protection actions are: protection stop → servo disconnect enable → brake hold.

[0276] For collision boundary events, the alarm boundary condition is: distance from the end effector to the clamp. Less than the collision warning threshold Or the absolute value of the joint torque exceeds the collision torque warning threshold. ;

[0277] The interlocking boundary condition is: end force Or the joint torque continuously exceeds the interlocking threshold Duration not less than The corresponding protection action is collision protection stop → servo disconnect enable.

[0278] For safety door / emergency stop interlocking events, the interlocking boundary condition is: safety door open signal. Duration not less than or emergency stop signal The corresponding protection actions are: immediate stop → servo deactivation → brake engagement.

[0279] The system extracts causal relationships from FMEA, FTA, and security specifications, mapping them to security event triples, mainly including: personnel approach → leading to → decline; Below And continue to exceed →Trigger→Safe speed reduction; Below And continue to exceed →Trigger→Protection Stop; Protection Stop→Trigger→Servo Disconnect Enable; Servo Disconnect Enable→Trigger→Brake Engagement; Brake Response Delay→Causes→Increased Stop Distance; End Contact Fixture→Causes→ or Rise; torque continuously exceeds threshold → trigger → collision protection stops; safety door opens → trigger → safety PLC interlock; emergency stop pressed → trigger → immediate protection stops.

[0280] III. Construction of Security Incident Evolution Diagram

[0281] Based on the aforementioned security boundary conditions and causal relationships, perform the second step of the above method to construct a security event evolution graph. ,in For a set of nodes in a safe state, For the set of evolutionary edges, For a set of relation types, Let be the set of edge weights.

[0282] The safety status nodes in this embodiment include: v0 (normal operation), v1 (personnel enter the remote detection area), v2 (personnel approach the warning area), v3 (safety deceleration trigger), v4 (personnel approach the danger zone), v5 (protection stop trigger), v6 (servo de-enabled), v7 (brake engaged), v8 (robot stops within a safe distance), v9 (stopping distance approaches the safety boundary), and v10 (protection action failure or delay). Among these, v0 is the normal operation status node, v2 and v3 are alarm trigger status nodes, v4 and v5 are interlock trigger status nodes, and v6 and v7 are protection action execution status nodes.

[0283] Based on causal relationships, determine the state transition directions between safe state nodes and construct evolutionary edges. For example, a complete evolution path of a personnel approach security event is represented as:

[0284]

[0285] A protection action delay evolution path is represented as:

[0286]

[0287] The edge weight of each evolutionary edge is determined by the FMEA knowledge score, FTA or security specification score, historical security event co-occurrence score, and expert prior score. After normalization function mapping, it reflects the possibility of transitioning from the current security state node to the next security state node along the evolutionary edge.

[0288] IV. Sampling of the Evolution Path of Rare Security Events

[0289] Based on the security event evolution diagram, perform the third step in the above method to sample and generate rare security event evolution paths.

[0290] The system enumerates candidate security event evolution paths on the security event evolution graph. Each candidate path is formed by sequentially connecting security state nodes according to the direction of the evolution edges. For each candidate path, the path risk value is calculated based on the security severity of each security state node in the path and the edge weight of each evolution edge.

[0291]

[0292] in, Indicates the severity of the security status node. Represents the conditional transition probability. This is the tail enhancement factor, used to amplify the weight of rare, high-risk paths. It is applied based on the path risk value. Normalization determines the sampling generation probability of each candidate path. The path risk value is proportional to the sampling generation probability, thus prioritizing the sampling of rare paths with low probability but high value.

[0293] This embodiment samples the following seven rare security event evolution paths to cover high-value security scenarios:

[0294] Path A: Personnel approaching the warning zone → safe deceleration → no shutdown triggered (personnel approaching but remaining outside the stop boundary); Path B: Personnel approaching the danger zone → protection stop triggered → successful shutdown; Path C: Personnel approaching the danger zone → brake response delay → stop distance approaching the safety boundary; Path D: End-effector approaching the gripper → torque approaching the threshold → collision stop not triggered; Path E: End-effector contacting the gripper → collision protection triggered → servo de-enabled; Path F: Safety door opening → safety PLC interlock → servo de-enabled → brake holding; Path G: Safety light curtain briefly mistriggered → robot mis-decelerated or mis-stopped.

[0295] V. Generation of Variable Trajectories

[0296] Based on the rare security event evolution paths and security boundary conditions obtained from sampling, the fourth step in the above method is executed to generate the variable trajectories corresponding to each security-related variable.

[0297] Taking path B (personnel approaching the danger zone → protection is triggered and the system stops successfully) as an example, the process of generating the variable trajectory is explained.

[0298] First, determine the target's safety status based on path B: It needs to drop to the target time. The following, and the duration meets the requirements. The goal is exist Internal continuous satisfaction And the time exceeding the limit shall not exceed the allowable error δt.

[0299] Subsequently, the real-time safe distances of safety-related variables relative to the safety boundary were calculated at each time point. For the personnel distance variable... The real-time safe distance to the warning boundary is The real-time safe distance to the stop boundary is For joint torque variables The real-time safe distance to the collision threshold is .

[0300] Then, with real-time safe distance as a constraint, generate The distance evolution trajectory. The personnel approach process is generated using the following trajectory format:

[0301]

[0302] in, Indicates the speed at which people approach. This indicates the detection of noise or walking disturbance. Simultaneous constraints: Satisfy within the target time period And the duration exceeds the threshold for triggering the condition, i.e. And exceeding the amount .

[0303] For robot end-effector speed Generate velocity trajectory based on velocity response model: when At that time, the target speed is ;when At that time, the target speed is switched to ;when At that time, the target velocity drops to 0. The velocity response model is: ,in For speed adjustment coefficient, For velocity perturbation.

[0304] After determining the distance evolution trajectory of each safety-related variable, it is converted into a corresponding variable trajectory. This variable trajectory represents the process by which the values ​​of each safety-related variable approach or move away from the corresponding safety boundary in a time series.

[0305] VI. Generation of Global Safety Status Signal Sequence and Global Protection Action Timing Chain

[0306] Based on the above variable trajectories and security boundary conditions, a global security state signal sequence and a global protection action timing chain are generated corresponding to the evolution path of rare security events.

[0307] First, at each time point Values ​​and alarm thresholds The alarm status is compared to generate the alarm status at each time point, and then determined based on the alarm duration threshold. Determine whether the alarm conditions are continuously met, thereby forming an alarm status signal sequence. :

[0308] ,like And duration ;otherwise

[0309] Then, determine each moment. Does the interlocking trigger condition meet? ), and based on the duration and interlock duration threshold. Generate the interlocking states at each time point to form an interlocking trigger signal sequence. :

[0310] ,like And duration ;otherwise

[0311] For safety door opening or emergency stop signals, corresponding interlocking trigger signals are also generated:

[0312] ,like And duration ; ,like

[0313] The overall safety interlock status is determined by the logical OR of the above interlock signals:

[0314]

[0315] When there is a moment in the interlocking trigger signal sequence that is in the triggering state, the corresponding protection action sequence is determined according to the interlocking boundary conditions. Where a1 indicates the robot enters protection stop mode, a2 indicates the servo speed command drops to zero, a3 indicates the servo enable is cut off, and a4 indicates the brake is engaged. The time interval between adjacent protection actions is constrained. Configure execution timestamps for each protection action and generate a global protection action sequence chain.

[0316] Specifically, the servo enable state generation rule is as follows: ,like Where Ten is the delay from the safety PLC to the servo drive disconnection enable; the brake state generation rule is... ,like Where Tbr is the brake holding delay. The robot operation mode generation rule is: ,like and ; ,like and ; ,like .

[0317] VII. Time alignment to form multivariate composite time series data

[0318] The variable trajectory generated above ( etc.), safety status signal sequence ( The timing of the protection action sequence chain (ens(t), br(t), mode(t), etc.) is aligned to a time axis with a sampling period of 10ms, forming multivariate composite timing data. The multivariate composite timing data synthesized from each path covers the complete timing process from normal operation to approaching the safety boundary, alarm triggering, interlocking triggering, and execution of protection actions.

[0319] VIII. Security Consistency Verification

[0320] Security and consistency checks are performed on multivariate synthesized time-series data. Checks are performed on the corresponding data types according to multiple dimensions, including boundary distance consistency check index, safe deceleration consistency check index, protection stop consistency check index, protection action chain consistency check index, stop distance consistency check index, and collision boundary consistency check index, yielding the check results for each index.

[0321] Personnel distance boundary consistency check:

[0322]

[0323] Safety slowdown consistency check:

[0324]

[0325] Protection Stop Consistency Check:

[0326]

[0327] Protection action chain consistency check:

[0328]

[0329] Stop distance consistency check:

[0330]

[0331] Collision boundary consistency check:

[0332]

[0333] The results of each verification are weighted and fused to obtain a comprehensive quality score:

[0334]

[0335] like If the condition is met, the synthesized sample is output; otherwise, the multivariate synthesized time series data is modified according to the specific validation dimension that is not met—for example, adjusting the approach speed of personnel. Adjust the servo enable delay Adjust the brake holding delay Correct the timing of the protection action chain—or resample the evolution path of rare security events—until multivariate synthetic timing data that meets security consistency requirements is generated.

[0336] IX. Counterfactual safety trajectory generation

[0337] Based on generating multivariate composite time series data that meets safety and consistency requirements, counterfactual multivariate composite time series data can be further generated. Taking path C (brake response delay → stopping distance approaching the safety boundary) as an example, the intervention parameters are set as follows: the intervention action is "shorten the brake holding delay Tbr", and the intervention value is... Intervention time is .

[0338] Based on the above intervention parameters, the multivariate composite time series data are processed, and the counterfactual stopping distance is calculated using the following formula:

[0339]

[0340] If counterfactual intervention successfully prevents the stopping distance from exceeding the limit, then the condition is satisfied. This means that the stopping distance safety margin meets the requirements, thereby determining the counterfactual safety event time series data sample. By comparing the counterfactual sample with the original sample, the impact of brake response delay on the safety margin can be quantified.

[0341] 10. Output security event time sequence data samples

[0342] Through the above steps, the security event time-series data samples generated in this embodiment cover seven rare security event evolution paths, and each path contains a complete multivariate composite time series:

[0343]

[0344]

[0345] Simultaneously, structured labels are output, including: equipment type (collaborative industrial robot), safety event type (personnel approach / collision boundary / safety door opening / emergency stop), safety path (normal → approaching the warning zone → safe deceleration → approaching the danger zone → protective stop), boundary type (warning boundary / stop boundary / collision boundary), whether an alarm is triggered, whether an interlock is triggered, protection action chain (deceleration → disengagement of enable → braking), counterfactual intervention type (early deceleration / shortening of braking delay / cancellation of safety input suppression), near-accident level (low / medium / high), and comprehensive quality score Qtotal.

[0346] The above samples can be directly used for training robot safety early warning models, testing safety PLC interlocking logic, verifying collaborative robot safety area strategies, training collision detection models, operator training systems, digital twin safety simulation, verifying brake response and stopping distance, and testing safety input mis-triggering, missed triggering, and suppression scenarios, effectively solving the problem of scarce safety event data for collaborative industrial robots.

[0347] It should be noted that the above-described method embodiments, or the various possible implementations of the method embodiments, can be executed individually, or, provided there is no conflict, they can be combined with each other. The specific implementation can be determined according to actual usage requirements, and this application embodiment does not impose any restrictions on this.

[0348] As can be seen, the above mainly describes the solutions provided by the embodiments of this application from a methodological perspective. To achieve the above functions, the embodiments of this application provide corresponding hardware structures and / or software modules for executing each function. Those skilled in the art should readily recognize that, in conjunction with the modules and algorithm steps of the various examples described in the embodiments disclosed herein, the embodiments of this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0349] This application embodiment can divide the industrial safety event time-series data synthesis device into functional modules according to the above method example. For example, each function can be divided into its own functional modules, or two or more functions can be integrated into one processing module. The integrated module can be implemented in hardware or as a software functional module. Optionally, the module division in this application embodiment is illustrative and only represents one logical functional division; other division methods may be used in actual implementation.

[0350] In some embodiments, this application also provides an apparatus for synthesizing time-series data of industrial safety events. This apparatus may include one or more functional modules for implementing the industrial safety event time-series data synthesis method described in the above embodiments.

[0351] For example, Figure 10 This is a schematic diagram of the structure of an industrial safety event time-series data synthesis device provided in an embodiment of this application. Figure 10 As shown, the industrial safety event time-series data synthesis device 1000 includes: a determination module 1001, a construction module 1002, a generation module 1003, and an alignment module 1004.

[0352] The determination module 1001 is used to determine at least one safety-related variable, the safety boundary conditions corresponding to each safety-related variable, and the causal relationship between each safety state based on the multi-source data of the target industrial system.

[0353] Module 1002 is used to construct a security event evolution graph to characterize the evolution process of security events based on security boundary conditions and causal relationships. The security event evolution graph contains at least two security state nodes and at least one evolution edge. The evolution edge is determined according to the causal relationship and is used to characterize the state transition direction and causal transition relationship between each security state node.

[0354] The generation module 1003 is used to sample and generate at least one rare security event evolution path in the security event evolution graph, and each rare security event evolution path corresponds to a security event instance;

[0355] The generation module 1003 is also used to generate variable trajectories corresponding to each security-related variable based on the evolution path of rare security events and security boundary conditions, and to generate global security state signal sequences and global protection action timing chains corresponding to the evolution path of rare security events according to security boundary conditions. The variable trajectory is used to characterize the change process of the values ​​of security-related variables approaching or moving away from their respective security boundaries in the time series.

[0356] Alignment module 1004 is used to time-align variable trajectories, safety status signal sequences and protection action timing chains to form multivariate composite timing data;

[0357] The generation module 1003 is also used to generate security event time series data samples based on multivariate synthesized time series data.

[0358] In an optional embodiment, the construction module 1002 is specifically used for:

[0359] Based on the safety boundary conditions, multiple safety state nodes are determined. Each safety state node represents a safety state, and the safety state includes at least the normal operation state, the alarm trigger state, the interlock trigger state, and the protection action execution state.

[0360] Based on the causal relationship, determine the state transition direction between safe state nodes, and construct the evolution edge connecting the safe state nodes according to the state transition direction;

[0361] Based on the safety boundary conditions and causal relationships, each evolution edge is assigned a corresponding edge weight. The edge weight is used to represent the probability of transitioning from the current safe state node to the next safe state node along the evolution edge.

[0362] The security event evolution graph is constructed based on the security state nodes, evolution edges, and edge weights.

[0363] In an optional embodiment, the generation module 1003 is specifically used for:

[0364] Based on the security state nodes and evolution edges, at least one candidate security event evolution path is determined in the security event evolution graph. Each candidate security event evolution path is formed by connecting the security state nodes sequentially according to the direction of the evolution edges.

[0365] Based on the security severity of each security state node in each candidate security event evolution path and the edge weight of each evolution edge, the path risk value of each candidate security event evolution path is determined.

[0366] Based on the path risk value of each candidate security event evolution path, the sampling generation probability of each candidate security event evolution path is determined, and the path risk value is proportional to the sampling generation probability.

[0367] Based on the sampling generation probability of each candidate security event evolution path, at least one rare security event evolution path is sampled and generated from each candidate security event evolution path.

[0368] In an optional embodiment, the generation module 1003 is specifically used for:

[0369] Based on the evolution path of rare security events, the target security state is determined. The target security state includes the target distance range between the real-time values ​​of security-related variables and the security boundary, as well as the target duration during which the security-related variables remain within the target distance range.

[0370] Based on the safety boundary conditions and the preset sampling period, determine the real-time safe distance of each safety-related variable relative to the safety boundary at each sampling time;

[0371] Based on the target's safety status and constrained by the real-time safety distance, the distance evolution trajectory of each safety-related variable is generated, which shows that the real-time safety distance approaches the target distance range over time and maintains the target's duration after entering the target distance range.

[0372] Based on the distance evolution trajectory of each safety-related variable, the variable trajectory corresponding to each safety-related variable is determined.

[0373] In one optional embodiment, the safety boundary conditions include alarm boundary conditions and interlock boundary conditions;

[0374] Among them, alarm boundary conditions include alarm threshold and alarm duration threshold, and interlock boundary conditions include interlock trigger conditions and interlock duration threshold.

[0375] In one optional embodiment, the safety status signal sequence includes an alarm status signal sequence and an interlock trigger signal sequence;

[0376] Module 1003 is generated, specifically for:

[0377] Based on the variable trajectory, alarm boundary conditions, and interlock boundary conditions, the values ​​of each safety-related variable at each time point are compared with the alarm threshold, and the comparison results are output.

[0378] The alarm status signal sequence is formed by generating the alarm status at each moment based on the comparison results and the alarm duration threshold;

[0379] Determine whether the values ​​of each safety-related variable at each time point meet the interlocking triggering conditions, and construct an interlocking triggering signal sequence based on the interlocking status at each time point generated by the duration of the interlocking triggering conditions and the interlocking duration threshold.

[0380] When there is a moment in the interlocking trigger signal sequence that is in the triggering state, the corresponding protection action sequence is determined according to the interlocking boundary conditions, and an execution timestamp is configured for each protection action according to the time interval constraint between adjacent protection actions in the protection action sequence, thereby generating a protection action timing chain.

[0381] In an optional embodiment, the generation module 1003 is specifically used for:

[0382] Perform a security consistency check on the multivariate synthesized time series data based on the security boundary conditions, and output the check results;

[0383] If the verification results indicate that the multivariate synthesized time series data passes the security consistency verification, the multivariate synthesized time series data will be output as a security event time series data sample; or

[0384] If the verification result indicates that the multivariate synthesized time series data has failed the security consistency verification, the multivariate synthesized time series data is subjected to consistency repair processing to generate new multivariate synthesized time series data. The consistency repair processing includes correction and regeneration.

[0385] In an optional embodiment, the generation module 1003 is specifically used for:

[0386] Based on the safety boundary conditions, and according to multiple verification indicators, including at least the boundary distance consistency verification indicator and the interlocking trigger condition consistency verification indicator, the corresponding data in the multivariate synthesized time series data are verified respectively, and the verification results of each verification indicator are obtained.

[0387] The results of each verification are merged to obtain the comprehensive quality score of the multivariate synthesized time series data.

[0388] The verification result is determined based on the comparison between the comprehensive quality score and the preset threshold.

[0389] In an optional embodiment, after time-aligning the variable trajectories, safety status signal sequences, and protection action timing chains to form multivariate composite timing data, the generation module 1003 is further configured to:

[0390] Obtain intervention parameters, which include intervention actions, intervention values, and intervention times;

[0391] The multivariate composite time series data are processed according to the intervention parameters to generate counterfactual multivariate composite time series data, and counterfactual security event time series data samples are determined based on the counterfactual multivariate composite time series data.

[0392] It should be noted that the industrial safety incident time-series data synthesis device can implement all the processes implemented in the above method embodiments and achieve the same beneficial effects. To avoid repetition, it will not be described again here.

[0393] In the case where the functions of the integrated modules described above are implemented in hardware, this application provides a possible structural schematic diagram of the electronic device involved in the above embodiments. For example... Figure 11 As shown, the electronic device 1100 includes: a processor 1102, a communication interface 1103, and a bus 1104. Optionally, the electronic device 1100 may also include a memory 1101.

[0394] Processor 1102 may implement or execute various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 1102 may be a central processing unit, a general-purpose processor, a digital signal processor, an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It may implement or execute various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 1102 may also be a combination that implements computing functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0395] Communication interface 1103 is used to connect with other devices via a communication network. This communication network can be Ethernet, wireless access network, wireless local area network (WLAN), etc.

[0396] The memory 1101 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but is not limited thereto.

[0397] As one possible implementation, the memory 1101 can exist independently of the processor 1102. The memory 1101 can be connected to the processor 1102 via a bus 1104 and is used to store instructions or program code. When the processor 1102 calls and executes the instructions or program code stored in the memory 1101, it can implement the industrial safety event timing data synthesis method provided in this application embodiment.

[0398] In another possible implementation, the memory 1101 can also be integrated with the processor 1102.

[0399] Bus 1104 can be an Extended Industry Standard Architecture (EISA) bus, etc. Bus 1104 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 11The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0400] Through the above description of the implementation methods, those skilled in the art can clearly understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the industrial safety event time sequence data synthesis device can be divided into different functional modules to complete all or part of the functions described above.

[0401] This application embodiment also provides a chip, which includes a processor and a communication interface. The communication interface is coupled to the processor. The processor is used to run programs or instructions to implement the various processes of the above-described industrial safety event time-series data synthesis method embodiment, and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0402] It should be understood that the chip mentioned in the embodiments of this application may also be referred to as a system-on-a-chip, system chip, chip system, or system-on-a-chip, etc.

[0403] This application embodiment also provides a readable storage medium storing a program or instructions. When executed by a computer, the program or instructions implement the industrial security event time-series data synthesis method provided in the above embodiments. It is understood that all or part of the processes in the above method embodiments can be executed by computer instructions instructing related hardware; the readable storage medium can be any of the foregoing embodiments or memory; the readable storage medium can also be an external storage device of the industrial security event time-series data synthesis apparatus, such as a pluggable hard drive, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the industrial security event time-series data synthesis apparatus. Further, the readable storage medium can include both internal storage units of the industrial security event time-series data synthesis apparatus and external storage devices. The readable storage medium is used to store the computer program and other programs and data required by the industrial security event time-series data synthesis apparatus. The readable storage medium can also be used to temporarily store data that has been output or will be output.

[0404] This application also provides a computer program product, which is stored in a storage medium and, when executed by a computer, implements the industrial safety event time-series data synthesis method provided in the above embodiments.

[0405] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0406] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a computer software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0407] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A method for synthesizing time-series data of industrial safety incidents, characterized in that, include: Based on multi-source data from the target industrial system, determine at least one safety-related variable, the safety boundary conditions corresponding to each safety-related variable, and the causal relationships between each safety state. Based on the security boundary conditions and the causal relationship, a security event evolution graph is constructed to characterize the evolution process of security events. The security event evolution graph contains at least two security state nodes and at least one evolution edge. The evolution edge is determined according to the causal relationship and is used to characterize the state transition direction and causal transition relationship between each security state node. At least one rare security event evolution path is generated by sampling in the security event evolution graph, and each rare security event evolution path corresponds to a security event instance; Based on the rare security event evolution path and the security boundary conditions, variable trajectories corresponding to each security-related variable are generated respectively, and a global security state signal sequence and a global protection action time sequence chain corresponding to the rare security event evolution path are generated according to the security boundary conditions. The variable trajectory is used to characterize the change process of the values ​​of the security-related variables approaching or moving away from their respective security boundaries in the time series. The variable trajectory, the safety status signal sequence, and the protection action timing chain are time-aligned to form multivariate composite timing data; Based on the multivariate synthesized time-series data, a security event time-series data sample is generated.

2. The method according to claim 1, characterized in that, The step of constructing a security event evolution diagram to characterize the evolution process of security events based on the security boundary conditions and the causal relationships includes: Based on the aforementioned safety boundary conditions, the plurality of safety state nodes are determined, each of which represents a safety state. The safety state includes at least a normal operation state, an alarm trigger state, an interlock trigger state, and a protection action execution state. Based on the causal relationship, the state transition direction between the safe state nodes is determined, and an evolutionary edge connecting the safe state nodes is constructed based on the state transition direction. Based on the security boundary conditions and the causal relationship, each evolutionary edge is assigned a corresponding edge weight, which represents the probability of transitioning from the current safe state node to the next safe state node along the evolutionary edge. The security event evolution graph is constructed based on the security state nodes, the evolution edges, and the edge weights.

3. The method according to claim 1, characterized in that, The step of sampling and generating at least one rare security event evolution path in the security event evolution graph includes: Based on the security state nodes and the evolution edges, at least one candidate security event evolution path is determined in the security event evolution graph. Each candidate security event evolution path is formed by connecting the security state nodes sequentially according to the direction of the evolution edges. Based on the security severity of each security state node in each candidate security event evolution path and the edge weight of each evolution edge, the path risk value of each candidate security event evolution path is determined. Based on the path risk value of each candidate security event evolution path, the sampling generation probability of each candidate security event evolution path is determined, wherein the path risk value is proportional to the sampling generation probability. Based on the sampling generation probability of each candidate security event evolution path, at least one rare security event evolution path is sampled and generated from each candidate security event evolution path.

4. The method according to claim 1, characterized in that, The process of generating variable trajectories corresponding to each security-related variable based on the evolution path of the rare security event and the security boundary conditions includes: Based on the evolution path of the rare security event, the target security state is determined. The target security state includes the target distance range between the real-time value of the security-related variable and the security boundary, and the target duration for which the security-related variable remains within the target distance range. Based on the safety boundary conditions and the preset sampling period, determine the real-time safety distance of each of the safety-related variables relative to the safety boundary at each sampling time; Based on the target safety status and constrained by the real-time safety distance, a distance evolution trajectory is generated for each of the safety-related variables, which is to approach the target distance range over time and maintain the target duration after entering the target distance range. Based on the distance evolution trajectory of each of the aforementioned safety-related variables, the variable trajectory corresponding to each of the aforementioned safety-related variables is determined.

5. The method according to claim 1, characterized in that, The safety boundary conditions include alarm boundary conditions and interlock boundary conditions; The alarm boundary conditions include alarm thresholds and alarm duration thresholds, and the interlocking boundary conditions include interlocking triggering conditions and interlocking duration thresholds.

6. The method according to claim 5, characterized in that, The safety status signal sequence includes an alarm status signal sequence and an interlock trigger signal sequence; The step of generating the global security state signal sequence and global protection action timing chain corresponding to the rare security event evolution path based on the security boundary conditions includes: Based on the variable trajectory, the alarm boundary conditions, and the interlocking boundary conditions, the values ​​of each safety-related variable at each time point are compared with the alarm threshold, and the comparison results are output. The alarm status signal sequence is constituted based on the alarm status at each moment generated by the comparison result and the alarm duration threshold; Determine whether the values ​​of each safety-related variable at each time point meet the interlocking trigger condition, and construct the interlocking trigger signal sequence based on the interlocking state at each time point generated by the duration of the interlocking trigger condition and the interlocking duration threshold. When there is a moment in the interlocking trigger signal sequence that is in the triggering state, the corresponding protection action sequence is determined according to the interlocking boundary conditions, and an execution timestamp is configured for each protection action according to the time interval constraint between adjacent protection actions in the protection action sequence, thereby generating the protection action timing chain.

7. The method according to claim 1, characterized in that, The process of generating security event time series data samples based on the multivariate synthesized time series data includes: Perform a security consistency check on the multivariate synthesized time series data according to the security boundary conditions, and output the check result; If the verification result indicates that the multivariate synthesized time-series data passes the security consistency verification, the multivariate synthesized time-series data will be output as the security event time-series data sample; or If the verification result indicates that the multivariate synthesized time series data has failed the security consistency verification, the multivariate synthesized time series data is subjected to consistency repair processing to generate new multivariate synthesized time series data. The consistency repair processing includes correction and regeneration.

8. The method according to claim 7, characterized in that, The step of performing a security consistency check on the multivariate synthesized time series data according to the security boundary conditions and outputting the check result includes: Based on the security boundary conditions, and according to multiple verification indicators including at least the boundary distance consistency verification indicator and the interlocking trigger condition consistency verification indicator, the corresponding types of data in the multivariate synthesized time series data are verified respectively to obtain the verification results of each verification indicator. The verification results are fused to obtain the comprehensive quality score of the multivariate synthesized time series data. The verification result is determined based on the comparison between the comprehensive quality score and the preset threshold.

9. The method according to claim 1, characterized in that, After aligning the variable trajectories, the safety status signal sequence, and the protection action timing chain to form multivariate composite time-series data, the method further includes: Obtain intervention parameters, which include intervention actions, intervention values, and intervention times; The multivariate composite time series data is processed according to the intervention parameters to generate counterfactual multivariate composite time series data, and counterfactual security event time series data samples are determined based on the counterfactual multivariate composite time series data.

10. An industrial safety incident time-series data synthesis device, characterized in that, include: The determination module is used to determine at least one safety-related variable, the safety boundary conditions corresponding to each safety-related variable, and the causal relationships between each safety state based on multi-source data of the target industrial system. A construction module is used to construct a security event evolution graph to characterize the evolution process of a security event based on the security boundary conditions and the causal relationship. The security event evolution graph contains at least two security state nodes and at least one evolution edge. The evolution edge is determined according to the causal relationship and is used to characterize the state transition direction and causal transition relationship between each security state node. A generation module is used to sample and generate at least one rare security event evolution path in the security event evolution graph, and each rare security event evolution path corresponds to a security event instance; The generation module is also used to generate variable trajectories corresponding to each security-related variable based on the rare security event evolution path and the security boundary conditions, and to generate a global security state signal sequence and a global protection action time sequence chain corresponding to the rare security event evolution path according to the security boundary conditions. The variable trajectory is used to characterize the change process of the values ​​of the security-related variables approaching or moving away from their respective security boundaries in the time series. The alignment module is used to time-align the variable trajectory, the safety status signal sequence, and the protection action timing chain to form multivariate composite timing data. The generation module is also used to generate security event time series data samples based on the multivariate synthesized time series data.

11. An electronic device, characterized in that, It includes a processor and a memory, the memory storing programs or instructions that can run on the processor, the programs or instructions being executed by the processor to implement the industrial security event time-series data synthesis method as described in any one of claims 1-9.

12. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions, which, when executed by a computer, implement the industrial safety event time-series data synthesis method as described in any one of claims 1-9.

13. A computer program product, characterized in that, The computer program product is stored in a storage medium, and when the computer program product is executed by a computer, it implements the industrial safety event time-series data synthesis method as described in any one of claims 1-9.

Images