Generative confrontation deduction method for safety events for industrial production

Abstract

The embodiment of the invention relates to the technical field of safety event deduction, in particular to an industrial production-oriented safety event generative adversarial deduction method, which comprises the following steps of: determining target equipment, and constructing a generative adversarial network-based deduction model adaptive to the target equipment; an equipment fault tree and a personnel behavior decision tree are obtained and coded into a directed acyclic graph as a causal graph constraint of the deduction model; a physical generation engine generates physical parameters under extreme working conditions based on the causal graph constraint, and then a language behavior generation engine generates an operator behavior sequence based on the causal graph constraint; performing data consistency verification by a multi-modal discriminator based on a cross-modal attention mechanism, and under the condition that the verification is passed, approving generation; and performing iterative training on the deduction model, deploying the deduction model after the training is completed, synchronizing the real digital twin model of the target equipment into the deduction model, and obtaining a deduction result of the security event in the future preset time output by the deduction model.

Description

Technical Field

[0001] Embodiments of the present application relate to the technical field of safety incident deduction, and particularly to a generative adversarial deduction method for safety incidents in industrial production. Background Art

[0002] In industrial production, the simulation and deduction of safety incidents are very important. Scientific simulation and prediction can enable safety management in industrial production to shift from "mending the fold after the sheep are lost" to "taking precautions", predicting which equipment is most likely to have problems at what time and in what link, so that weak points can be strengthened before an accident occurs, the high-risk operation plan can be adjusted, and the signs of an accident can be nipped in the bud.

[0003] However, the traditional simulation and deduction of safety incidents mainly rely on data augmentation methods, and data augmentation methods rely on shallow transformation means such as image rotation and noise addition, making it difficult to generate extreme risk scenarios with physical rationality and logical coherence (such as a chain accident caused by the superposition of equipment aging and operation errors), resulting in serious deficiencies in the generalization ability of the trained deduction model for complex and rare "black swan events". At the same time, most traditional methods focus on single-modal data and cannot achieve the dynamic coupling generation of multi-modal information such as sensor time-series data, operation log texts, and monitoring videos, which makes it difficult for the deduction model to capture the collaborative laws of cross-modal risk signals in complex accident chains. In addition, traditional deduction methods lack explicit modeling of causal relationships and often deduce false scenarios that violate physical laws or standard operation procedures (such as pressure exceeding the limit but no alarm being triggered), and such contradictions will directly reduce the reliability of the deduction model.

[0004] In summary, with the increasing complexity of industrial production, the accuracy requirements for risk prediction in high-risk industries such as petrochemical and nuclear power are becoming increasingly stringent. There is an urgent need to break through the technical bottlenecks of traditional technologies in physical simulation, multi-modal generation, and causal deduction, and build a new solution that can deduce the real risk evolution process. Summary of the Invention

[0005] To solve the above technical problems, embodiments of the present application propose a generative adversarial deduction method for safety incidents in industrial production, aiming to achieve a closed-loop deduction of generation, verification, and optimization, ensure the logical self-consistency of the deduction process, effectively improve the deduction ability for "black swan events", thereby formulating scientific and effective emergency plans, maximizing the protection of the lives of workers, and ensuring the continuous and stable operation of industrial production.

[0006] To achieve the above object, an embodiment of the present application proposes a generative adversarial deduction method for safety events in industrial production. The method includes: determining a target device for which safety event deduction is required, constructing a deduction model based on a generative adversarial network adapted to the target device, the deduction model being composed of a dual-channel generator and a multi-modal discriminator, and the dual-channel generator being composed of a physical scene generation engine and a language behavior generation engine; obtaining a device fault tree of the target device and a personnel behavior decision tree corresponding to the device fault tree, encoding the device fault tree and the personnel behavior decision tree into a directed acyclic graph as the causal graph constraint of the deduction model; the physical generation engine generates physical parameters that follow the basic physics laws of the target device under extreme working conditions as the physical scene based on the causal graph constraint, and then the language behavior generation engine generates an operator behavior sequence that matches the physical scene and conforms to the logic of the standard operation process based on the causal graph constraint by using a pre-trained language model; the multi-modal discriminator verifies the data consistency of the physical scene and the operator behavior sequence based on the cross-modal attention mechanism, and approves the generation of the dual-channel generator in the case of passing the verification; iteratively training the deduction model until convergence, deploying the trained deduction model, synchronizing the real digital twin model of the target device to the deployed deduction model, and obtaining the deduction result of the safety event within a future preset time output by the deduction model.

[0007] To achieve the above object, an embodiment of the present application further provides a generative adversarial deduction system for safety events in industrial production. The system includes: a target determination module for determining a target device for which safety event deduction is required; a model construction module for constructing a deduction model based on a generative adversarial network adapted to the target device. The constructed deduction model consists of a dual-channel generator and a multi-modal discriminator. The dual-channel generator consists of a physical scene generation engine and a language behavior generation engine; a constraint generation module for obtaining the device fault tree of the target device and the corresponding human behavior decision tree of the device fault tree, encoding the obtained device fault tree and human behavior decision tree into a directed acyclic graph as the causal graph constraint of the constructed deduction model; a sample generation module for instructing the physical generation engine to generate physical parameters that follow the basic physical laws of the target device under extreme working conditions as the physical scene based on the causal graph constraint, then instructing the language behavior generation engine to generate an operator behavior sequence that matches the physical scene and conforms to the logic of the standard operation process based on the causal graph constraint using a pre-trained language model, and finally instructing the multi-modal discriminator to perform data consistency verification on the physical scene and the operator behavior sequence based on the cross-modal attention mechanism. In the case of passing the verification, the generation of the dual-channel generator is approved; a training and deployment module for iteratively training the constructed deduction model until convergence to obtain a trained deduction model, and deploying the trained deduction model; a safety event deduction execution module for synchronizing the real digital twin model of the target device into the deployed deduction model to obtain the deduction result of the safety event within a future preset time output by the deduction model.

[0008] To achieve the above object, an embodiment of the present application further provides an electronic device. The electronic device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute a generative adversarial deduction method for safety events in industrial production as described above.

[0009] To achieve the above object, an embodiment of the present application further provides a computer-readable storage medium storing a computer program, which when executed by a processor, can implement a generative adversarial deduction method for safety events in industrial production as described above.

[0010] A generative adversarial deduction method for safety events in industrial production proposed in this application. Facing the problem of insufficient data for "black swan events", this application constructs a deduction model composed of a dual-channel generator and a multi-modal discriminator based on the generative adversarial network. The physical scene generation engine in the dual-channel generator can well cover extreme scenarios, breaking through the limitation that traditional data augmentation can only fine-tune the existing data distribution. The physical scene generation engine and the language behavior generation engine work together, enabling the causal relationship between the generated physical scene and the operator behavior sequence to be explicitly modeled. To improve the authenticity of safety event deduction, this application designs a causal graph constraint for the deduction model based on the equipment fault tree of the target device and the corresponding personnel behavior decision tree. This makes the deduction of the deduction model have a causal premise, rather than being deduced randomly and isolatedly, and can avoid deducing obviously wrong and illogical results. After completing the training and deployment of the deduction model, the real digital twin model of the target device is synchronized with the deduction model to ensure that the deduction of the deduction model is aligned with the real situation, thereby increasing the rationality of the deduction results. All in all, this application well realizes the closed-loop deduction of generation, verification, and optimization, ensures the logical self-consistency of the safety event deduction process, effectively improves the deduction ability for "black swan events", facilitates the staff to formulate scientific and effective emergency plans, maximally protects the lives of the staff, and ensures the continuous and stable operation of industrial production.

[0011] Optionally, obtaining the equipment fault tree of the target device and the corresponding personnel behavior decision tree, and encoding the equipment fault tree and the personnel behavior decision tree into a directed acyclic graph as the causal graph constraint of the deduction model, includes: obtaining the equipment fault tree of the target device and the corresponding personnel behavior decision tree, where the equipment fault tree records several possible faults of the target device, and the personnel behavior decision tree records several behavior decisions that technicians may make for the target device; traversing each fault recorded in the equipment fault tree, and matching the current fault with each behavior decision recorded in the personnel behavior decision tree to determine the causal relationship between each fault and each behavior decision; adding mandatory constraints and prohibited constraints to the equipment fault tree and the personnel behavior decision tree respectively, where the mandatory constraint is used to represent that there is an inevitable association between two faults or two behavior decisions, and the prohibited constraint is used to represent that there should be no association between two faults or two behavior decisions; based on the causal relationship between each fault and each behavior decision, establishing a connection between the equipment fault tree and the personnel behavior decision tree with mandatory constraints and prohibited constraints added and the possible safety events, and encoding them into a directed acyclic graph as the causal graph constraint of the deduction model.

[0012] Optionally, denote the physical scene generated by the physical generation engine as , and denote the operator behavior sequence generated by the language behavior generation engine as , , , is the th element in is the th element in is the total number of elements in The total number of elements in is the same as the total number of elements in Based on the cross-modal attention mechanism, the multi-modal discriminator verifies the data consistency of the physical scene and the operator behavior sequence. In the case of successful verification, the generation of the dual-channel generator is approved, including: Calculate the consistency score between the physical scene and the operator behavior sequence through the following formula: ; ; ; ; ; ; ; ; ; where is the calculated consistency score, represents the similarity attention score, represents the content attention score, represents the structure attention score, , , are respectively , , corresponding balance weights, represents the cosine similarity score, represents the Euclidean distance similarity score, represents the Pearson correlation coefficient similarity score, , , are respectively , , corresponding balance weights, represents The mean of the elements in denotes The mean of the elements in denotes the numerical feature extractor, denotes performing word embedding, denotes the graph attention network, denotes the recurrent attention network; Judge whether the calculated consistency score is greater than the preset consistency score threshold; If it is determined that the calculated consistency score is greater than the consistency score threshold, approve the generation of the dual-channel generator, otherwise reject the generation of the dual-channel generator.

[0013] Optionally, the loss function used when iteratively training the deduction model is represented by the following formula: ; where denotes the physical scene generation engine, denotes the language behavior generation engine, denotes the multi-modal discriminator, denotes the expectation of the real data of denotes the real data distribution of denotes the discrimination result given by the multi-modal discriminator for the real data of denotes the expectation of the noise of denotes the noise latent distribution of denotes the physical scene generated by the physical scene generation engine based on the noise of denotes the discrimination result given by the multi-modal discriminator for of denotes the expectation of the noise of denotes the noise latent distribution of denotes the operator behavior sequence generated by the language behavior generation engine based on the noise of denotes the discrimination result given by the multi-modal discriminator for of

[0014] Optionally, when iteratively training the deduction model, the multi-modal discriminator also needs to calculate the risk severity score of the physical scenarios generated by the physical scenario generation engine based on the causal graph constraint, and assign different dynamic weights to different physical scenarios in the loss function based on the calculated risk severity score. Among them, the higher the risk severity score of a physical scenario, the greater the dynamic weight assigned to it.

[0015] Optionally, synchronize the true digital twin model of the target device to the deployed deduction model, and obtain the deduction result of the security event within the future preset time output by the deduction model, including: performing digital twin modeling on the target device to obtain the true digital twin model of the target device, or directly calling the true digital twin model of the target device from the factory digital twin model corresponding to the factory where the target device is located; synchronizing the true digital twin model of the target device with the physical scenario generation engine in the deduction model to ensure that the physical scenarios generated by the physical scenario generation engine are aligned with the true digital twin model of the target device; starting the deduction model, and performing security event deduction based on the true digital twin model of the target device to obtain the deduction result of the security event within the future preset time output by the deduction model.

[0016] Optionally, after obtaining the deduction result of the security event within the future preset time output by the deduction model, the method further includes: sequentially performing physical rationality verification, logical consistency verification, and expert knowledge verification on the deduction result. If all three verifications pass, allow targeted analysis of the deduction result; otherwise, use the deduction model to re-deduce. Among them, physical rationality verification is to reversely verify whether the deduction result conforms to the law of conservation of energy, the law of conservation of mass, and the law of conservation of momentum through numerical simulation. Logical consistency verification is to check whether the deduction result conforms to industry safety specifications based on a rule engine. Expert knowledge verification is to allow industry experts to check whether there are logical errors in the visualized deduction result through an interactive interface. Description of the Drawings

[0017] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the related art, the following will briefly introduce the drawings required for the description of the embodiments of the present application or the related art. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0018] Figure 1 is a flowchart of a generative adversarial deduction method for security events in industrial production provided in an embodiment of the present application; Figure 2 is a schematic structural diagram of a deduction model provided in an embodiment of the present application; Figure 3 It is provided in an embodiment of the present application. When simulating the leakage caused by corrosion of a chemical pipeline, it is a schematic diagram of the dynamic calculation of the relationship between the leakage rate and the pressure change by the physical scene generation engine; Figure 4 It is provided in an embodiment of the present application. It is a directed acyclic graph encoded based on an equipment fault tree and a human behavior decision tree; Figure 5 It is a schematic diagram of using a deduction model to deduce a safety event provided in an embodiment of the present application; Figure 6 It is a schematic structural diagram of a generative adversarial deduction system for safety events in industrial production provided in another embodiment of the present application; Figure 7 It is a schematic structural diagram of an electronic device provided in another embodiment of the present application. Detailed implementation manners

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the embodiments of the present application will be elaborated in detail below with reference to the accompanying drawings. However, those of ordinary skill in the art can understand that in the embodiments of the present application, many technical details are proposed to help readers better understand the present application. However, even without these technical details and various changes and modifications based on the following embodiments, the technical solutions claimed in the present application can still be implemented. The following division of each embodiment is for convenience of description and should not constitute any limitation to the specific implementation manner of the present application. Each embodiment can be combined and cross-referenced with each other on the premise of not being contradictory.

[0020] An embodiment of the present application proposes a generative adversarial deduction method for safety events in industrial production, which is applied to an electronic device. The electronic device can be a terminal or a server. In this embodiment and the following embodiments, the server is taken as an example for illustration. The implementation details of a generative adversarial deduction method for safety events in industrial production proposed in this embodiment are specifically described below. The following content is only implementation details provided for convenience of understanding and is not necessary for implementing the solution.

[0021] The specific process of a generative adversarial deduction method for safety events in industrial production proposed in this embodiment can be as Figure 1 shown and includes: Step 101: Determine the target device for which safety event deduction is required, and construct a deduction model based on a generative adversarial network adapted to the target device. The deduction model consists of a dual-channel generator and a multi-modal discriminator. The dual-channel generator consists of a physical scene generation engine and a language behavior generation engine.

[0022] In a specific implementation, the server first needs to select the target device for which security event deduction is to be performed. The target device can be a large combined device, such as a cooling system, or an independent small device, such as a chemical pipeline. After selecting the target device, the server can construct a deduction model based on a generative adversarial network that is adapted to the target device. The deduction model consists of a dual-channel generator and a multi-modal discriminator. The dual-channel generator consists of a physical scene generation engine and a language behavior generation engine.

[0023] In one example, the specific structure of the deduction model can be as Figure 2 shown. The physical scene generation engine and the language behavior generation engine form the dual-channel generator. The physical scene generation engine is connected to the language behavior generation engine, and both generation engines are bi-directionally connected to the multi-modal discriminator integrated with a cross-modal attention mechanism.

[0024] In one example, the physical scene generation engine is used to generate a physical scene, and the language behavior generation engine is used to generate an operator behavior sequence that matches the physical scene. In the case where the selected target device is a chemical pipeline, the physical scene generated by the physical scene generation engine can be as Figure 3 shown.

[0025] In one example, the physical scene generated by the physical scene generation engine is essentially in numerical modality, and the operator behavior sequence generated by the language behavior generation engine is in text modality. The multi-modal discriminator needs to simultaneously discriminate the physical scene in numerical modality and the operator behavior sequence in text modality, so it is "multi-modal".

[0026] Step 102: Obtain the device fault tree of the target device and the personnel behavior decision tree corresponding to the device fault tree, and encode the device fault tree and the personnel behavior decision tree into a directed acyclic graph as the causal graph constraint of the deduction model.

[0027] In a specific implementation, the deduction model needs to perform security event deduction under certain constraints to ensure that the security deduction process is scientific, reasonable, and logical. The server needs to obtain the device fault tree of the target device and the personnel behavior decision tree corresponding to the device fault tree, and encode the device fault tree and the personnel behavior decision tree into a directed acyclic graph, so as to use the directed acyclic graph as the causal graph constraint of the deduction model.

[0028] In one example, when the server constructs the causal graph constraint, it first needs to obtain the device fault tree of the target device and the personnel behavior decision tree corresponding to the device fault tree. The device fault tree records several possible faults of the target device, and the personnel behavior decision tree records several behavior decisions that technicians may make for the target device.

[0029] Next, the server traverses each fault recorded in the equipment fault tree and matches the current fault with each behavior decision recorded in the personnel behavior decision tree, so as to determine the causal relationship between each fault and each behavior decision.

[0030] After that, the server adds mandatory constraints and prohibited constraints to the equipment fault tree and the personnel behavior decision tree respectively. The mandatory constraint is used to represent an inevitable association between two faults or two behavior decisions, and the prohibited constraint is used to represent that there should be no association between two faults or two behavior decisions.

[0031] For example, in the equipment fault tree, "pump failure → flow rate decrease → reactant ratio imbalance → explosion" is a mandatory causal chain and needs to be added to the equipment fault tree as a mandatory constraint. "Flow rate decrease → normal reactant ratio" is an impossible contradictory event and therefore needs to be added to the equipment fault tree as a prohibited constraint.

[0032] Finally, based on the causal relationship between each fault and each behavior decision, the server needs to establish a connection between the equipment fault tree and the personnel behavior decision tree with mandatory constraints and prohibited constraints added and the possible safety events, and encode them into a directed acyclic graph as the causal graph constraint of the deduction model.

[0033] In an example, the directed acyclic graph encoded based on the equipment fault tree and the personnel behavior decision tree of a chemical pipeline can be as Figure 4 shown.

[0034] Step 103: The physical generation engine generates physical parameters that follow the basic laws of physics for the target device under extreme working conditions as the physical scenario based on the causal graph constraint. Then, the language behavior generation engine generates an operator behavior sequence that matches the physical scenario and conforms to the logic of the standard operation process based on the causal graph constraint using a pre-trained language model.

[0035] In a specific implementation, both the physical generation engine and the language behavior generation engine need to work under the causal graph constraint. The physical generation engine generates physical parameters that follow the basic laws of physics for the target device under extreme working conditions as the physical scenario based on the causal graph constraint, and the language behavior generation engine generates an operator behavior sequence that matches the physical scenario and conforms to the logic of the standard operation process based on the causal graph constraint using a pre-trained language model.

[0036] In an example, the target device is a chemical pipeline. The physical parameters generated by the physical generation engine include but are not limited to temperature, pressure, vibration spectrum, etc., and these physical parameters need to follow the basic laws of physics such as thermodynamics and fluid mechanics. The operator behavior sequence generated by the language behavior generation engine is stored in the form of log text. For example, "The operator did not close the valve in time" is an operator behavior.

[0037] Step 104: The multi-modal discriminator verifies the data consistency of the physical scene and the operator's behavior sequence based on the cross-modal attention mechanism. If the verification passes, the generation of the dual-channel generator is approved.

[0038] In a specific implementation, after the physical generation engine generates the physical scene and the language behavior generation engine generates the operator's behavior sequence, the multi-modal discriminator will start working. The multi-modal discriminator verifies the data consistency of the physical scene and the operator's behavior sequence based on the cross-modal attention mechanism. Only when the verification passes can the generation of the dual-channel generator be approved. If the verification fails, the generation of the dual-channel generator is rejected.

[0039] In an example, the target device is a reaction boiler. If the physical scene generated by the physical generation engine is "boiler explosion", then the operator's behavior sequence generated by the language behavior generation engine must contain "safety valve failure" to meet the data consistency. If "safety valve failure" does not exist in the operator's behavior sequence, the multi-modal discriminator will reject the generation of the dual-channel generator.

[0040] In an example, let the physical scene generated by the physical generation engine be , and let the operator's behavior sequence generated by the language behavior generation engine be , , , be the th element in , be the th element in , be the total number of elements in . The total number of elements in is the same as the total number of elements in . When the multi-modal discriminator verifies the data consistency of the physical scene and the operator's behavior sequence , it needs to calculate the consistency score between the physical scene and the operator's behavior sequence through the following formula: ; ; ; ; ; ; ; ; ; wherein, is the calculated consistency score, represents the similarity attention score, represents the content attention score, represents the structure attention score, , , are respectively , , corresponding balance weights, represents the cosine similarity score, represents the Euclidean distance similarity score, represents the Pearson correlation coefficient similarity score, , , are respectively , , corresponding balance weights, represents the mean of the elements in represents the mean of the elements in represents the numerical feature extractor, represents performing word embedding, represents the graph attention network, represents the recurrent attention network.

[0041] Next, the multi-modal discriminator needs to determine whether the calculated consistency score between the physical scenario and the operator behavior sequence is greater than a preset consistency score threshold. If it is determined that the calculated consistency score is greater than the preset consistency score threshold, the generation of the dual-channel generator is approved; otherwise, the generation of the dual-channel generator is rejected. Among them, the preset consistency score threshold can be set by those skilled in the art according to actual needs.

[0042] In an example, the loss function used for iterative training of the deduction model is represented by the following formula: ; wherein, represents the physical scenario generation engine, represents the language behavior generation engine, represents the multi-modal discriminator, represents for real data expectation represents the real data distribution represents the discrimination result given by the multi-modal discriminator for the real data gives the discrimination result represents the expectation for the noise expectation represents the noise latent distribution represents the physical scene generated by the physical scene generation engine based on the noise generated physical scene represents the discrimination result given by the multi-modal discriminator for gives the discrimination result represents the expectation for the noise expectation represents the noise latent distribution represents the operator behavior sequence generated by the language behavior generation engine based on the noise generated operator behavior sequence represents the discrimination result given by the multi-modal discriminator for gives the discrimination result

[0043] In one example, when iteratively training the deduction model, the multi-modal discriminator also needs to calculate the risk severity score of the physical scene generated by the physical scene generation engine based on the causal graph constraint, and assign different dynamic weights to different physical scenes in the loss function based on the calculated risk severity score. Among them, the higher the risk severity score of the physical scene, the greater the dynamic weight assigned, and correspondingly, the lower the risk severity score of the physical scene, the smaller the dynamic weight assigned. This can force the dual-channel generator to prioritize optimizing the generation quality of such samples.

[0044] In one example, the target device is a chemical pipeline. When the multi-modal discriminator performs a risk severity score on the physical scene, it needs to consider the leakage amount, chain reaction level, etc. For a physical scene with multiple-level fault superposition, a higher risk severity score needs to be assigned.

[0045] Step 105, iteratively train the deduction model until convergence, deploy the trained deduction model, synchronize the real digital twin model of the target device to the deployed deduction model, and obtain the deduction result of the safety event within the future preset time output by the deduction model.

[0046] In a specific implementation, after the server constructs the deduction model, it can start training the deduction model, iteratively train the deduction model until convergence, deploy the trained deduction model, and then synchronize the real digital twin model of the target device to the deployed deduction model to obtain the deduction result of the safety event within the future preset time output by the deduction model.

[0047] In one example, the process of the deduction model performing security event deduction is as Figure 5 shown.

[0048] In one example, when the server uses the deduction model to perform security event deduction, it is necessary to perform digital twin modeling on the target device to obtain the true digital twin model of the target device, or directly call (through an interface) the true digital twin model of the target device from the factory digital twin model corresponding to the factory where the target device is located. Subsequently, synchronize the true digital twin model of the target device with the physical scene generation engine in the deduction model to ensure that the physical scene generated by the physical scene generation engine is aligned with the true digital twin model of the target device. Finally, start the deduction model and perform security event deduction based on the true digital twin model of the target device to obtain the deduction results of security events within a future preset time output by the deduction model.

[0049] In one example, the target device is a chemical pipeline, and the true digital twin model of the chemical plant shows that the corrosion rate of the chemical pipeline is 0.1 mm per year. The deduction model will simulate security events in the next 5 years based on this physical parameter.

[0050] In one example, after the server obtains the deduction results of security events within a future preset time output by the deduction model, it is also necessary to perform physical rationality verification, logical consistency verification, and expert knowledge verification on the deduction results in sequence. When all three verifications pass, targeted analysis of the deduction results is allowed; otherwise, the deduction model is used to perform deduction again. Among them, physical rationality verification is to reverse verify whether the deduction results conform to the law of conservation of energy, the law of conservation of mass, and the law of conservation of momentum through numerical simulation. Logical consistency verification is to check whether the deduction results conform to industry safety specifications based on a rule engine. Expert knowledge verification is to allow industry experts to check whether there are logical errors in the visualized deduction results through an interactive interface. The design of physical rationality verification, logical consistency verification, and expert knowledge verification can further enhance the credibility of the deduction results, thereby providing a solid data foundation for subsequent applications.

[0051] A generative adversarial deduction method for safety events in industrial production proposed in this embodiment addresses the problem of insufficient data for "black swan events". This application constructs a deduction model consisting of a dual-channel generator and a multi-modal discriminator based on a generative adversarial network. The physical scene generation engine in the dual-channel generator can well cover extreme scenarios, breaking through the limitation that traditional data augmentation can only fine-tune the existing data distribution. The physical scene generation engine and the language behavior generation engine work together, enabling the explicit modeling of the causal relationship between the generated physical scene and the operator's behavior sequence. To enhance the authenticity of safety event deduction, this application designs a causal graph constraint for the deduction model based on the equipment fault tree of the target device and the corresponding human behavior decision tree. This makes the deduction of the deduction model have a causal premise, rather than randomly and isolatedly deducing, and can avoid deducing obviously wrong and illogical results. After completing the training and deployment of the deduction model, the real digital twin model of the target device is synchronized with the deduction model to ensure that the deduction of the deduction model is aligned with the real situation, thereby increasing the rationality of the deduction result. All in all, this application well realizes the closed-loop deduction of generation, verification, and optimization, ensures the logical self-consistency of the safety event deduction process, effectively improves the deduction ability for "black swan events", facilitates the staff to formulate scientific and effective emergency plans, maximally protects the lives of the staff, and ensures the continuous and stable operation of industrial production.

[0052] The step division of the above various methods is only for clear description. When implemented, they can be combined into one step, or some steps can be split into multiple steps. As long as the same logical relationship is included, it is within the protection scope of this application. Adding insignificant modifications to the algorithm or process or introducing insignificant designs, but not changing the core design of its algorithm and process, are all within the protection scope of this application.

[0053] Correspondingly, another embodiment of this application proposes a generative adversarial deduction system for safety events in industrial production. The details of a generative adversarial deduction system for safety events in industrial production proposed in this embodiment are specifically described below. The following content is only implementation details provided for convenient understanding and is not necessary for this embodiment.

[0054] Figure 6 It is a schematic structural diagram of a generative adversarial deduction system for safety events in industrial production proposed in this embodiment, specifically including: a target determination module 201, a model construction module 202, a constraint generation module 203, a sample generation module 204, a training and deployment module 205, and a safety event deduction execution module 206.

[0055] The target determination module 201 is used to determine the target device for which safety event deduction needs to be performed.

[0056] The model construction module 202 is used to construct a deduction model based on a generative adversarial network adapted to the target device. The constructed deduction model consists of a dual-channel generator and a multi-modal discriminator. The dual-channel generator consists of a physical scene generation engine and a language behavior generation engine.

[0057] The constraint generation module 203 is used to obtain the device fault tree of the target device and the personnel behavior decision tree corresponding to the device fault tree, and encode the obtained device fault tree and personnel behavior decision tree into a directed acyclic graph as the causal graph constraint of the constructed deduction model.

[0058] The sample generation module 204 is used to instruct the physical generation engine to generate physical parameters that follow the basic physical laws of the target device under extreme working conditions as the physical scene based on the causal graph constraint, and then instruct the language behavior generation engine to generate an operator behavior sequence that matches the physical scene and conforms to the logic of the standard operation process based on the causal graph constraint using a pre-trained language model. Finally, it instructs the multi-modal discriminator to perform data consistency verification on the physical scene and the operator behavior sequence based on the cross-modal attention mechanism, and approves the generation of the dual-channel generator in the case of successful verification.

[0059] The training and deployment module 205 is used to iteratively train the constructed deduction model until convergence to obtain the trained deduction model, and deploy the trained deduction model.

[0060] The safety event deduction execution module 206 is used to synchronize the real digital twin model of the target device to the deployed deduction model, and obtain the deduction result of the safety event within a preset future time output by the deduction model.

[0061] It is worth mentioning that each module involved in this embodiment is a logical module. In practical applications, a logical unit can be a physical unit, a part of a physical unit, or a combination of multiple physical units. In addition, in order to highlight the innovative part of this application, units that are not closely related to solving the technical problems proposed in this application are not introduced in this embodiment, but this does not mean that there are no other units in this embodiment.

[0062] It is not difficult to find that this embodiment is a system embodiment corresponding to the above method embodiment, and this embodiment can be implemented in cooperation with the above method embodiment. The relevant technical details and technical effects mentioned in the above method embodiment are still valid in this embodiment, and in order to reduce repetition, they will not be elaborated here. Correspondingly, the relevant technical details mentioned in this embodiment can also be applied to the above method embodiment.

[0063] Another embodiment of the present application provides an electronic device, whose specific structure can be as Figure 7 shown, including: at least one processor 301; and a memory 302 communicatively connected to the at least one processor 301; wherein, the memory 302 stores instructions executable by the at least one processor 301, and the instructions are executed by the at least one processor 301 to enable the at least one processor 301 to execute a generative adversarial deduction method for security events in industrial production as described in the above method embodiment.

[0064] Among them, the memory and the processor are connected in a bus manner. The bus may include any number of interconnected buses and bridges, and the bus connects various circuits of one or more processors and the memory together. The bus may also connect various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art, so they will not be further described herein. The bus interface provides an interface between the bus and the transceiver. The transceiver may be an element or multiple elements, such as multiple receivers and transmitters, and provides a unit for communicating with various other devices on the transmission medium. The data processed by the processor is transmitted over the wireless medium through the antenna. Further, the antenna also receives data and transmits the data to the processor.

[0065] The processor is responsible for managing the bus and general processing, and can also provide various functions, including timing, peripheral interface, voltage regulation, power management, and other control functions. The memory can be used to store the data used by the processor when executing operations.

[0066] Another embodiment of the present application provides a computer-readable storage medium storing a computer program, which when executed by a processor, can implement a generative adversarial deduction method for security events in industrial production as described in the above method embodiment.

[0067] That is, those skilled in the art can understand that all or part of the steps in the above method embodiment can be completed by instructing relevant hardware through a program. The program is stored in a storage medium, including several instructions to enable a device (which can be a single-chip microcomputer, a chip, etc.) or a processor to execute all or part of the steps of a generative adversarial deduction method for security events in industrial production as described in the method embodiment of the present application. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories, random access memories, magnetic disks, or optical discs that can store program codes.

[0068] Those of ordinary skill in the art can understand that the above embodiments are specific embodiments for implementing the present application. In actual applications, various changes can be made to them in form and details without departing from the spirit and scope of the present application.

Claims

1. A generative adversarial deduction method for safety events in industrial production, characterized in that Including: Determine the target device for which security event deduction is required, and construct a deduction model based on a generative adversarial network adapted to the target device. The deduction model consists of a dual-channel generator and a multi-modal discriminator. The dual-channel generator consists of a physical scene generation engine and a language behavior generation engine; Obtain the device fault tree of the target device and the corresponding human behavior decision tree of the device fault tree, and encode the device fault tree and the human behavior decision tree into a directed acyclic graph as the causal graph constraint of the deduction model; Based on the causal graph constraint, the physical generation engine generates physical parameters that follow the basic physics laws of the target device under extreme working conditions as the physical scene. Then, based on the causal graph constraint, the language behavior generation engine uses a pre-trained language model to generate an operator behavior sequence that matches the physical scene and conforms to the logic of the standard operation process; Based on the cross-modal attention mechanism, the multi-modal discriminator verifies the data consistency of the physical scene and the operator behavior sequence. If the verification passes, the generation of the dual-channel generator is approved; Iteratively train the deduction model until convergence, deploy the trained deduction model, synchronize the real digital twin model of the target device to the deployed deduction model, and obtain the deduction result of the security event within the future preset time output by the deduction model.

2. The generative adversarial deduction method for safety events in industrial production according to claim 1, characterized in that Obtain the device fault tree of the target device and the corresponding human behavior decision tree of the device fault tree, and encode the device fault tree and the human behavior decision tree into a directed acyclic graph as the causal graph constraint of the deduction model, including: Obtain the device fault tree of the target device and the corresponding human behavior decision tree of the device fault tree. The device fault tree records several possible faults of the target device, and the human behavior decision tree records several behavior decisions that technicians may make for the target device; Traverse each fault recorded in the device fault tree, and match the current fault with each behavior decision recorded in the human behavior decision tree to determine the causal relationship between each fault and each behavior decision; Add mandatory constraints and prohibited constraints to the device fault tree and the human behavior decision tree respectively. The mandatory constraint is used to represent that there is an inevitable association between two faults or two behavior decisions, and the prohibited constraint is used to represent that there should be no association between two faults or two behavior decisions; Based on the causal relationship between each fault and each behavior decision, establish a connection between the device fault tree and the human behavior decision tree with mandatory constraints and prohibited constraints added and the possible security events, and encode them into a directed acyclic graph as the causal graph constraint of the deduction model.

3. A generative adversarial deduction method for safety events in industrial production according to claim 1, characterized in that Let the physical scene generated by the physical generation engine be and the operator behavior sequence generated by the language behavior generation engine be , , , be the th element in , be the th element in , be the total number of elements in ; the total number of elements in is the same as the total number of elements in ;​​ Based on the cross-modal attention mechanism, the multi-modal discriminator verifies the data consistency of the physical scene and the operator behavior sequence. If the verification passes, the generation of the dual-channel generator is approved, including: Calculate the physical scenario through the following formula and the operator behavior sequence to obtain the consistency score ； ； ； ； ； ； ； ； ； wherein, is the calculated consistency score, represents the similarity attention score, represents the content attention score, represents the structure attention score, 、 、 are respectively 、 、 corresponding balanced weights, represents the cosine similarity score, represents the Euclidean distance similarity score, represents the Pearson correlation coefficient similarity score, 、 、 are respectively 、 、 corresponding balanced weights, represents the mean of the elements in represents the mean of the elements in represents the numerical feature extractor, represents performing word embedding, represents the graph attention network, represents the recursive attention network; Judge whether the calculated consistency score is greater than the preset consistency score threshold; If it is determined that the calculated consistency score is greater than the consistency score threshold, approve the generation of the dual-channel generator, otherwise reject the generation of the dual-channel generator.

4. A generative adversarial deduction method for safety events in industrial production according to claim 1, characterized in that The loss function used when iteratively training the deduction model is represented by the following formula: ； Among them, represents the physical scene generation engine, represents the language behavior generation engine, represents the multi-modal discriminator, represents the expectation of real data ; represents the distribution of real data ; represents the discrimination result given by the multi-modal discriminator for real data ; represents the expectation of noise ; represents the latent distribution of noise ; represents the physical scene generated by the physical scene generation engine based on noise ; represents the discrimination result given by the multi-modal discriminator for ; represents the expectation of noise ; represents the latent distribution of noise ; represents the operator behavior sequence generated by the language behavior generation engine based on noise ; represents the discrimination result given by the multi-modal discriminator for ; 5. A generative adversarial deduction method for safety events in industrial production according to claim 4, characterized in that When iteratively training the deduction model, the multi-modal discriminator also needs to calculate the risk severity score of the physical scenarios generated by the physical scenario generation engine based on the causal graph constraint, and assign different dynamic weights to different physical scenarios in the loss function based on the calculated risk severity score. Among them, the higher the risk severity score, the greater the dynamic weight assigned to the physical scenario.

6. A generative adversarial deduction method for safety events in industrial production according to any one of claims 1 to 5, characterized in that Synchronize the true digital twin model of the target device to the deployed deduction model, and obtain the deduction results of the security events within the future preset time output by the deduction model, including: Perform digital twin modeling on the target device to obtain the true digital twin model of the target device, or directly call the true digital twin model of the target device from the factory digital twin model corresponding to the factory where the target device is located; Synchronize the true digital twin model of the target device with the physical scenario generation engine in the deduction model to ensure that the physical scenarios generated by the physical scenario generation engine are aligned with the true digital twin model of the target device; Start the deduction model, perform security event deduction based on the true digital twin model of the target device, and obtain the deduction results of the security events within the future preset time output by the deduction model.

7. A generative adversarial deduction method for safety events in industrial production according to any one of claims 1 to 5, characterized in that After obtaining the deduction results of the security events within the future preset time output by the deduction model, the method further includes: Perform physical rationality verification, logical consistency verification, and expert knowledge verification on the deduction results in sequence. If all three verifications pass, allow targeted analysis of the deduction results; otherwise, use the deduction model to deduce again; Among them, physical rationality verification is to reversely verify whether the deduction results conform to the laws of conservation of energy, conservation of mass, and conservation of momentum through numerical simulation. Logical consistency verification is to check whether the deduction results conform to industry safety specifications based on a rule engine. Expert knowledge verification is to allow industry experts to check whether there are logical errors in the visualized deduction results through an interactive interface.

8. A generative adversarial deduction system for safety events in industrial production, characterized in that, including: A target determination module for determining the target device that needs to perform security event deduction; A model construction module for constructing a deduction model based on a generative adversarial network adapted to the target device. The constructed deduction model consists of a dual-channel generator and a multi-modal discriminator. The dual-channel generator consists of a physical scenario generation engine and a language behavior generation engine; A constraint generation module for obtaining the device fault tree of the target device and the corresponding human behavior decision tree of the device fault tree, and encoding the obtained device fault tree and human behavior decision tree into a directed acyclic graph as the causal graph constraint of the constructed deduction model; A sample generation module for instructing the physical generation engine to generate physical parameters that follow the basic laws of physics of the target device under extreme working conditions as physical scenarios based on the causal graph constraint, then instructing the language behavior generation engine to generate an operator behavior sequence that matches the physical scenario and conforms to the logic of the standard operation process based on the causal graph constraint using a pre-trained language model, and finally instructing the multi-modal discriminator to perform data consistency verification on the physical scenario and the operator behavior sequence based on the cross-modal attention mechanism. If the verification passes, approve the generation of the dual-channel generator; A training and deployment module for iteratively training the constructed deduction model until convergence to obtain a trained deduction model, and deploying the trained deduction model. A security event deduction execution module for synchronizing the real digital twin model of the target device into the deployed deduction model to obtain the deduction results of security events within a future preset time output by the deduction model.

9. An electronic device, characterized in that, Including: At least one processor; And a memory communicatively connected to the at least one processor; Wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute a generative adversarial deduction method for security events in industrial production as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it can implement a generative adversarial deduction method for security events in industrial production as described in any one of claims 1 to 7.

Images