Operation risk identification method and system based on edge cloud collaboration and edge end device

Abstract

The invention relates to an operation risk identification method and system based on edge cloud collaboration and an edge end device. The method comprises the steps that multi-source heterogeneous data generated through multi-dimensional risk perception is used as a drive, a risk identification model is constructed through a side cloud cooperation method, risk perception, risk analysis and risk early warning are achieved, a risk management and control system is constructed based on a knowledge graph, generated risks are effectively managed and controlled, and accidents are prevented. According to the method, a risk identification and edge cloud cooperation device is also constructed, and the device is an edge-side carrier for operation of the risk identification system. Edge cloud cooperation, risk perception, risk analysis and local early warning can be realized, early warning information can be fed back to a platform side for analysis, and disposal information downloaded by the platform is received.

Description

Technical Field

[0001] The present application relates to the field of industrial production safety technology, and in particular to an operation risk identification method, system and edge device based on edge-cloud collaboration. Background Art

[0002] With the current development of industrial informatization and intelligence, industry is facing new challenges in safe production. The unsafe factors of safe production are affected by the multiple spatial elements of physical, social and information. Traditional safety supervision methods have regulatory bottlenecks in the current industrial development process. Therefore, it is necessary to use a variety of new technologies such as the Internet of Things, cloud computing, and artificial intelligence to obtain safety information from the physical and social spaces of production activities, and reversely improve the facility safety of the physical space and the behavioral safety of the social space. The development of information technology provides a basis for the conversion of various risk factors into data. Through the digital integration of risk perception data and data analysis and calculation, the identification and control of various safety hazards of people, machines, objects, environment, and management, and the transformation of human defense, material defense, and technical defense are realized. This is the mainstream of the current development of safe production.

[0003] Existing methods for operational risk identification and control mostly process data by moving control balls or adding external sensor equipment, and uploading the pre-processed data to the platform for processing and early warning. This is a one-way risk control method with disadvantages such as high resource consumption, long delay, and single risk identification type, and it cannot adapt well to complex and changeable operational risk identification. Summary of the invention

[0004] Based on this, it is necessary to provide an edge-cloud collaborative operation risk identification method, system and edge device that can integrate multi-dimensional data to perform real-time early warning analysis on risk data in response to the above-mentioned technical problems.

[0005] A method for identifying operation risks based on edge-cloud collaboration, the method being implemented between the edge end of the operation site and the back-end cloud platform;

[0006] The method comprises implementing at the edge of the work site:

[0007] Acquire multi-source heterogeneous data at the work site, and fuse the multi-source heterogeneous data to obtain multi-source fused data;

[0008] Using a risk identification model to perform risk analysis based on the multi-source fusion data to extract risk factors, and judging whether there are hidden dangers in the current operation site based on the risk factors;

[0009] If there is a hidden danger, a risk warning and risk factors are sent to the backend cloud platform, and the multi-source fusion data of the hidden danger is sent to the backend cloud platform as a hidden danger sample;

[0010] The method includes implementing on the backend cloud platform:

[0011] Acquire the risk warning and risk factors, and when triggered by the risk warning, perform a secondary risk analysis on the risk factors to obtain analysis results;

[0012] The knowledge base of the knowledge graph is used to provide corresponding hidden danger management methods according to the analysis results;

[0013] Periodically evaluate the effectiveness of the risk warning and hidden danger control measures after implementation;

[0014] If the evaluation result indicates that optimization is required, the risk identification model and the knowledge graph are optimized online according to the hidden danger samples, and the optimized risk identification model is sent to the edge of the work site.

[0015] In one embodiment, fusing the multi-source heterogeneous data to obtain multi-source fused data includes:

[0016] Using a risk data extraction model to extract risk data from the multi-source heterogeneous data according to a preset risk data feature set table;

[0017] The risk data is classified according to the data structure type, and the classified data is processed accordingly and labeled with time series labels to obtain the multi-source fusion data.

[0018] In one embodiment, the risk data is classified according to the data structure type, and after the classified data is processed accordingly and labeled with a time series label, the multi-source fusion data obtained includes:

[0019] Classifying the risk data into structured data types, semi-structured data types, and unstructured data types;

[0020] For the risk data of the structured data type, a time series label is directly added;

[0021] For the risk data of the semi-structured data type, format conversion is performed or the original data is kept according to the requirements of the risk data feature set table, and time series labels are added;

[0022] The risk data of the unstructured data type is standardized according to the requirements of the risk data feature set table and is labeled with a time series label.

[0023] In one embodiment, using a risk identification model to perform risk analysis and extract risk factors based on the multi-source fusion data includes:

[0024] Extracting risk factors of multiple risk types from the multi-source fusion data using the risk identification model;

[0025] The preset weights of each risk type and each risk factor are calculated to obtain a total risk factor, and it is determined whether there are hidden dangers in the current work site based on the total risk factor.

[0026] In one embodiment, a knowledge base using a knowledge graph provides corresponding hidden danger management methods according to the analysis results, including:

[0027] The hidden danger search agent performs semantic analysis on the analysis result to obtain multiple key semantics related to the risk hidden dangers in the analysis result;

[0028] Search the knowledge base according to the multiple key semantics to obtain relevant knowledge, wherein the knowledge base includes accident analysis and prediction, laws, regulations and standards, danger and controllable operability, operating procedures and abnormal handling, risk assessment and management knowledge graphs, and the industry language model currently formed in various industries;

[0029] The treatment fusion agent integrates the relevant knowledge, and the treatment method agent then arranges and processes the integrated relevant knowledge to obtain the corresponding hidden danger management method.

[0030] The present application also provides an edge device, which includes: a risk perception unit, a data processing unit, a risk warning unit, and a hidden danger sample extractor;

[0031] The risk perception unit is used to obtain multi-source heterogeneous data at the work site;

[0032] The data processing unit is used to fuse the multi-source heterogeneous data to obtain multi-source fused data, and use a risk identification model to perform risk analysis based on the multi-source fused data to extract risk factors, and determine whether there are hidden dangers in the current work site based on the risk factors;

[0033] If there is a hidden danger, the risk warning unit will send a risk warning and risk factors to the backend cloud platform;

[0034] The hidden danger sample extractor is used to extract multi-source fusion data with hidden dangers and send them to the back-end cloud platform as hidden danger samples.

[0035] In one embodiment, the risk perception unit includes an inherent perception module and a perception expansion module;

[0036] The inherent perception module is used to collect multi-source heterogeneous data related to vision, position, environment and status inside the work site;

[0037] The perception expansion module is used to collect multi-source heterogeneous data outside the work site.

[0038] In one embodiment, the edge device further includes: an edge-cloud coordination management engine, a converged communication unit, and a power supply unit;

[0039] The edge-cloud coordination management engine is used for model collaboration, data collaboration and application collaboration with the backend cloud platform;

[0040] The fusion communication unit is used to communicate with the backend cloud platform;

[0041] The power supply unit is used to supply power to the risk perception unit, data processing unit, risk warning unit, hidden danger sample extractor, edge-cloud coordination management engine and fusion communication unit.

[0042] The present application also provides a job risk identification system based on multi-source heterogeneous data edge-cloud collaboration, the system comprising a job site edge terminal deployed at the job site and a backend cloud platform, the job site edge terminal and the backend cloud platform are connected via a communication network, wherein the job site edge terminal is the above-mentioned edge terminal device, and the backend cloud platform comprises a disposal decision unit, a graph analysis unit, an effect evaluation unit and a model optimization unit;

[0043] The disposal decision unit is used to obtain the risk warning and risk factors, and when triggered by the risk warning, conduct a secondary risk analysis on the risk factors to obtain analysis results, and then use the knowledge base of the knowledge graph in the graph analysis unit to give corresponding hidden danger management measures according to the analysis results;

[0044] The effect evaluation unit is used to periodically evaluate the effects of the risk warning and hidden danger control measures after implementation;

[0045] If the evaluation result indicates that optimization is required, the model optimization unit is used to perform online optimization on the risk identification model and the knowledge graph according to the hidden danger samples, and the optimized risk identification model is sent to the edge of the work site.

[0046] The above-mentioned operation risk identification method, system and edge device based on edge-cloud collaboration are driven by multi-source heterogeneous data generated by multi-dimensional risk perception, and a risk identification model is constructed through the edge-cloud collaboration method to realize risk perception, risk analysis, and risk warning. A risk management system is constructed based on the knowledge graph to effectively manage the risks generated and prevent accidents. This method also constructs a risk identification and edge-cloud collaboration device, which is the edge-side carrier for the operation of the risk identification system. It can realize edge-cloud collaboration and risk perception, risk analysis and local warning, and can feed back warning information to the platform side for analysis, and receive disposal information downloaded by the platform. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a flowchart of a method for identifying operation risks based on edge-cloud collaboration in one embodiment;

[0048] Figure 2 It is a flowchart of a multi-source heterogeneous data fusion method based on time triggering in one embodiment;

[0049] Figure 3 A schematic diagram of risk factor extraction in one embodiment;

[0050] Figure 4 A schematic diagram of a process flow for generating a risk management method based on a knowledge graph in one embodiment;

[0051] Figure 5 It is a flowchart of a method for identifying operational risks based on edge-cloud collaboration in one embodiment;

[0052] Figure 6 A structural block diagram of an edge device in an embodiment;

[0053] Figure 7 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0055] In the existing technology, the methods for operation risk identification and control are mostly processed by moving the control ball or adding external sensor equipment to process the data, and uploading the pre-processed data to the platform for processing and early warning. This is a one-way risk control method with the disadvantages of large resource consumption, long delay, and single risk identification type. It cannot adapt well to the problem of complex and changeable operation risk identification, such as Figure 1 As shown, a method for identifying operation risks based on edge-cloud collaboration is provided. The method is implemented between the edge end of the operation site and the back-end cloud platform. The method includes implementing at the edge end of the operation site:

[0056] Step S100, acquiring multi-source heterogeneous data at the work site, and fusing the multi-source heterogeneous data to obtain multi-source fused data.

[0057] Step S110, using a risk identification model to perform risk analysis based on multi-source fusion data to extract risk factors, and judging whether there are hidden dangers in the current work site based on the risk factors.

[0058] Step S120: If there are hidden dangers, a risk warning and risk factors are sent to the backend cloud platform, and the multi-source fusion data of the hidden dangers are sent to the backend cloud platform as hidden danger samples.

[0059] The method also includes implementing on a backend cloud platform:

[0060] Step S130, obtaining risk warnings and risk factors, and performing secondary risk analysis on the risk factors under the triggering of the risk warning to obtain analysis results.

[0061] Step S140, using the knowledge base of the knowledge graph to provide corresponding hidden danger management measures based on the analysis results.

[0062] Step S150, periodically evaluate the effects of risk warning and hidden danger control measures after implementation.

[0063] Step S160: If the evaluation result indicates that optimization is required, the risk identification model and the knowledge graph are optimized online according to the hidden danger samples, and the optimized risk identification model is sent to the edge end of the work site.

[0064] In this application, basic multi-source risk perception, edge-cloud collaboration and knowledge graph are used to identify risks and give corresponding governance measures. At the same time, a risk identification device that can adaptively learn is built, which can automatically replace the identification model according to the needs of the operation, automatically optimize and automatically optimize and update the model according to the operation process and risk identification results and disposal, automatically replace the deployment model according to the operation scenario, refine the supervision granularity, and automatically deploy and update the risk identification model to achieve scene adaptation.

[0065] In this embodiment, risk identification of industrial scenarios is performed through edge-cloud collaboration, where the edge end of the work site is set at the front end of the work to collect multi-source heterogeneous data, and perform data processing on its side to obtain preliminary results, and then the cloud platform performs secondary analysis based on the preliminary results and provides governance measures. When explaining the method part, the method implemented in the edge end of the work site is first explained, and then the method steps implemented in the cloud platform, namely steps S130 to S160, are explained.

[0066] In step S100, after obtaining multi-source heterogeneous data at the work site, it is necessary to fuse these multi-source heterogeneous data for subsequent risk identification. The specific data included in the multi-source heterogeneous data will be described in detail in the subsequent description of the edge device, and will not be repeated here.

[0067] In this embodiment, the multi-source heterogeneous data are fused to obtain the multi-source fused data, including: using a risk data extraction model to extract risk data from the multi-source heterogeneous data according to a preset risk data feature set table, classifying the risk data according to the data structure type, and performing corresponding processing on the classified data and adding time series labels to obtain the multi-source fused data.

[0068] Specifically, the risk data extraction model can extract data through protocol parsing, OCR, network data intelligent analysis, LLM model, etc. according to the type of heterogeneous data.

[0069] Specifically, the risk data feature set table is a key feature sent from the platform to the edge end, which is determined according to the input data type of the risk identification model and can be used to extract the name definition set of risk data, including but not limited to Chinese characters, letters, numbers, symbols and feature names composed of the above.

[0070] Furthermore, the risk data is classified according to the data structure type, and after corresponding processing and time series labeling are performed on the classified data, multi-source fusion data is obtained, including: the risk data is divided into structured data types, semi-structured data types and unstructured data types, wherein, for risk data of structured data types, time series labels are directly added, for risk data of semi-structured data types, format conversion is performed or the original data is kept according to the requirements of the risk data feature set table, and time series labels are added, and for risk data of unstructured data types, standardization processing is performed according to the requirements of the risk data feature set table, and time series labels are added.

[0071] In this embodiment, when the acquired multi-source heterogeneous data are fused, the data received during this period are fused after the preset time trigger is triggered.

[0072] Specifically, when fusing multi-source heterogeneous data, a multi-source heterogeneous data fusion method is used based on the risk extraction time series trigger, such as Figure 2 As shown. Since the safety risks in the operation process are usually associated in real time, for external multi-sensor sources and multi-type data, if they are out of time association, it may lead to risk identification errors, that is, missed judgment or misjudgment. Conventional data fusion methods usually ignore the time factor, and therefore cannot fully meet the needs of data fusion for operation risk identification. In this method, the proposed data fusion method uses time triggers to associate various types of risk data with time, providing time series data for risk identification. The method specifically includes multiple steps: risk data extraction, data classification, data processing, risk data feature set table, time trigger, and data storage.

[0073] Furthermore, multi-source heterogeneous data triggers risk data extraction under the control of time triggers. The provision of risk data requires the extraction of risk data through the risk data feature set table according to the configuration management of the platform. The risk data extraction is performed through the risk data extraction model to extract the required risk data in the risk data feature set table. The extracted risk data may have multiple data types. The extracted data needs to be classified into structured data, semi-structured data and / or unstructured data, and data processing is performed according to the requirements of the risk data feature set table. The processing includes but is not limited to structured conversion, standardization processing, numerical transformation, etc.

[0074] Specifically, for structured data, the data is stored in a structured database after time series labeling. For semi-structured data, it needs to be formatted or not converted according to the requirements of the risk feature set table. After the conversion, if the data is structured, it is stored in a structured database; if it is semi-structured data, it is stored in a semi-structured database. For unstructured data, it needs to be standardized according to the requirements of the risk feature set table, and stored in an unstructured database after time series labeling.

[0075] In this embodiment, various databases use current conventional databases. The risk data after standardized time series processing is used for the model of the subsequent risk identification link.

[0076] Furthermore, the risk data extraction model can be dynamically obtained from the cloud platform according to the risk data feature set table. Specifically, it can be obtained from the stored model library in the cloud platform and can be deployed and run at the edge to form a risk identification safety production industrial model.

[0077] In step S110, risk identification model is used to perform risk analysis based on multi-source fusion data to extract risk factors, including: using the risk identification model to extract risk factors of multiple risk types in the multi-source fusion data, calculating according to preset weights of each risk type and each risk factor to obtain a total risk factor, and judging whether there are hidden dangers in the current work site based on the total risk factor.

[0078] In this embodiment, risk identification is performed on the operating scenes monitored at the edge of the operating site. There are multiple types of risks in each monitoring scene, and each type of risk corresponds to a variety of different risk points. The presence of such a risk is determined by multiple risk points corresponding to a risk type. Risk identification is performed according to the risk data and the risk identification model to obtain a specific risk factor value. Assume that risk factors S1.1 to S1.n are identified by the risk identification model for a certain monitoring operation scene. These risk factors are all sources of risk and have certain hidden risks. The risk factors are calculated as follows:

[0079]

[0080] In the above formula, the ∑a 1.i =1.0, where a 1.i For S 1.i The preset weights of the risk factor should be determined according to the mandatory requirements of relevant laws and regulations. For major accident hazards, the coefficient of the risk factor should be set to 0.8, and the sum of other non-major accident hazards should be set to 0.2. In this way, the largest risk category is identified. 1.i If it is greater than 1.0, the mandatory value is 1.0. The major accident hazard is recorded as an additional attribute in the maximum hazard type.

[0081] In this embodiment, the risk analysis model refers to a risk identification model that matches risk points. It can be a mechanism model or an artificial intelligence model, and a small model runs on the edge. The technologies used include but are not limited to image analysis, deep learning, mathematical logic operations, convolutional networks, etc. The technology needs to be selected according to the actual risk points.

[0082] In step S130, the risk factor Dn.j (wherein the subscript n indicates the type of risk and j indicates the risk factor corresponding to a certain risk type) is uploaded to the cloud platform. The risk factor Dn.j indicates the risk value and a list of major accident hazards. All Dn.j are uploaded to the cloud platform for comprehensive risk assessment, and the risk impact index and risk type are calculated. The risk assessment method on the platform side adopts the industry large model method for assessment, such as Figure 3 shown.

[0083] In step S130, while uploading the risk warning and risk factors to the backend cloud platform, the hidden danger samples are also sent to the backend cloud platform to facilitate the subsequent optimization of the risk identification model.

[0084] Next, the method implemented on the backend cloud platform is explained.

[0085] In step S140, under the triggering of the risk warning, a secondary risk analysis is performed on the risk factors to obtain an analysis result.

[0086] In this embodiment, the knowledge base using the knowledge graph gives corresponding hidden danger control measures according to the analysis results, including: the hidden danger search agent performs semantic analysis on the analysis results to obtain multiple key semantics related to the risk hidden dangers in the analysis results, searches the knowledge base according to the multiple key semantics to obtain relevant knowledge, wherein the knowledge base includes accident analysis and prediction, laws, regulations and standards, danger and controllable operability, operating procedures and abnormal handling, risk assessment and management knowledge graphs, and the industry large language models currently formed in various industries, the handling fusion agent fuses the relevant knowledge, and the handling method agent arranges and processes the fused relevant knowledge to obtain corresponding hidden danger control measures.

[0087] Specifically, the hidden danger handling method is a hidden danger handling method based on the knowledge graph, specifically a hidden danger handling method automatically generated based on the knowledge base composed of the knowledge graph and the industry big model. Figure 4 shown.

[0088] The risks identified by the risk identification model are input into the hidden danger search agent, which performs semantic analysis through the agent and analyzes them into several key semantics for hidden danger search and enters the knowledge base for search. Among them, semantics include but are not limited to hidden danger causes, disposal methods, risk hazards, relevant standards and relevant regulations, etc. The knowledge base includes but is not limited to accident analysis and prediction, laws, regulations and standards, danger and controllable operability, operating procedures and abnormal disposal, risk assessment and management, and other knowledge graphs and industry large language models. Among them, the industry large language model refers to the large language model currently formed in various industries.

[0089] The knowledge generated through knowledge search is sent to the processing fusion agent for knowledge fusion. The fused output knowledge is processed by the processing method agent and the processing method that the processing personnel can understand is output.

[0090] Specifically, the output includes but is not limited to text, graphics, video and sound, and also includes a variety of mixed output forms, which are determined by the functions of the intelligent agent.

[0091] In step S150 and step S160, after periodically evaluating the effects of risk warning and hidden danger control measures, if the evaluation result shows that optimization is needed, the risk identification model and knowledge graph are optimized online according to the hidden danger samples, and the optimized risk identification model is sent to the edge end of the operation site. In this way, the risk identification model and knowledge graph can be dynamically updated and optimized, further improving the accuracy and real-time performance of monitoring.

[0092] like Figure 5 The figure shows a schematic diagram of the implementation process of the operation risk identification method based on multi-source heterogeneous data edge-cloud collaboration.

[0093] In this application, an edge device that can implement this part of the method and steps S100 to S120 is also proposed for the edge end of the work site, such as Figure 6 As shown, the edge device includes: a risk perception unit, a data processing unit, a risk warning unit and a hidden danger sample extractor. Among them, the risk perception unit is used to obtain multi-source heterogeneous data at the work site, the data processing unit is used to fuse the multi-source heterogeneous data to obtain multi-source fusion data, and use the risk identification model to perform risk analysis based on the multi-source fusion data to extract risk factors, and judge whether there are hidden dangers in the current work site based on the risk factors. If there are hidden dangers, the risk warning unit sends the risk warning and risk factors to the back-end cloud platform, and the hidden danger sample extractor is used to extract the multi-source fusion data with hidden dangers as hidden danger samples and send them to the back-end cloud platform.

[0094] In this embodiment, the risk perception unit includes an inherent perception module and a perception expansion module. The inherent perception module is used to collect multi-source heterogeneous data related to vision, position, environment and status inside the work site, and the perception expansion module is used to collect multi-source heterogeneous data outside the work site.

[0095] Specifically, the risk perception unit consists of two parts: inherent perception and perception extension. It is mainly used to perceive risk data in the industrial production process through internal sensors or external interfaces, and provide risk factor data for risk analysis. Inherent perception includes four parts: vision, position, environment and status. Visual perception is mainly used to collect video and audio data, position perception is to obtain the current position of the device, environmental perception is to collect basic environmental data through built-in temperature, humidity, and combustible gas sensors, and status is to collect the current state of the device or the status indicator information of external devices collected through the video interface. The perception of these four types of data is integrated in the device, and basic risk data in the working environment can be collected with the device. However, the inherent risk perception of the device is of course not enough for safe production. The device can achieve multi-dimensional and comprehensive risk perception by accessing various external sensor data through perception extension. It can access external audio and video, personnel positioning system, DCS system, working environment and other related data.

[0096] Specifically, the data processing unit is composed of multi-source heterogeneous data fusion and an industrial model, i.e., the above-mentioned risk identification model operating environment. Multi-source heterogeneous data fusion is to process the risk perception data of structured data, unstructured data, and semi-structured data obtained from various sensors and business systems in the device using the multi-source heterogeneous data fusion method based on risk extraction timing triggering mentioned above, and store the processed data according to time sequence and type. The industrial model, i.e., the risk identification model operating environment, is used to run the mechanism model and the artificial intelligence model, and is managed in a Docker manner. After the heterogeneous data is fused and processed, it is input into the industrial model for risk analysis to identify the risk impact index, risk type, etc.

[0097] Specifically, the risk warning unit implements risk warning based on the results of risk analysis, including local warning of the device and push notification to the warning platform.

[0098] In this embodiment, the edge device further includes: an edge-cloud coordination management engine, a fusion communication unit, and a power supply unit. The edge-cloud coordination management engine is used for model collaboration, data collaboration, and application collaboration with the back-end cloud platform. The fusion communication unit is used to communicate with the back-end cloud platform. The power supply unit is used to power the risk perception unit, the data processing unit, the risk warning unit, the hidden danger sample extractor, the edge-cloud coordination management engine, and the fusion communication unit.

[0099] Specifically, the edge-cloud collaborative management engine is used for model collaboration, data collaboration, and application collaboration between the device and the cloud platform. It is the core unit for sharing between the device and the platform.

[0100] Specifically, the hidden danger sample extraction unit extracts and processes the data that constitutes hidden dangers in the risk analysis, and performs data collaboration through the edge-cloud collaborative management engine, and synchronizes the extracted data to the back-end cloud platform for model optimization and evolution.

[0101] Specifically, the fusion communication unit is the core of the device cloud-edge collaboration. The fusion communication uses wireless communication priority to communicate with the cloud. This allows for flexible adaptation to the operating environment. The fusion communication methods mainly include 5G, 4G, NB-IoT, WiFi, Beidou short messages, etc. At the same time, the fusion communication unit also has LoRa and Zigbee, which can realize sensor data access to the device.

[0102] Specifically, the power supply unit supports three power supply modes: AC220V, battery, and solar energy. Batteries and solar energy can be combined for power supply to achieve mobile and rapid deployment of edge devices.

[0103] Furthermore, the edge device uses an embedded chip as the main control chip. The chip has an ARM architecture, built-in NPU or GPU, and supports PCIE3.0 expansion of computing cards or graphics cards.

[0104] Furthermore, a binocular camera is built into the edge device, and the binocular camera supports 1080P pixels and 30fps video data acquisition, which can collect relevant environment at the site. The binocular camera also has a sound pickup function, which can collect audio data at the work site. The binocular camera has infrared night vision and thermal imaging functions, and has a pan-tilt function. Through the industrial model, it can support 360-degree target tracking and platform reverse control data collection functions.

[0105] Furthermore, the location data is the location of the edge device itself, which supports single Beidou positioning. When 4G, 5G, and No-IoT signals are normal, a positioning method combining base station positioning and satellite positioning can be realized. The positioning function can automatically obtain the operation location and control the operation. After the location is uploaded to the platform, the risk analysis of the external impact of the operation can be performed based on the location and related database information around the operation.

[0106] Furthermore, the perception expansion function is to connect to third-party sensor data through an external interface. The third-party sensor data includes but is not limited to camera data, personnel positioning system data, DCS system data, environmental monitoring data, safe production resource library data, etc. The data is connected by wirelessly returning data to the network platform to improve the convenience and adaptability of device deployment.

[0107] Furthermore, the perception extension part in the edge device has industrial protocol management and configuration functions, which can automatically retrieve and adapt the protocol of the platform or subsystem according to the model of the docking device, and automatically realize data docking. The industrial protocol management can be imported through the platform in an edge-cloud collaborative manner to expand the protocol.

[0108] Furthermore, in the data processing unit, various data of the perception end can be processed according to the data format required by the model to form a unified data format. Data processing includes but is not limited to deduplication, completion, cleaning, conversion, etc., and the data format includes but is not limited to json, xml, avi, txt, md, etc.

[0109] Furthermore, in the data processing unit, the industrial model, i.e., the risk identification model operating environment, is run in a docker environment. The industrial model supports both mechanism models and artificial intelligence models. The artificial intelligence model supports NPU accelerated computing and can also expand PCIE computing power cards to achieve reasoning acceleration. The industrial model operating environment has the function of industrial model edge-cloud collaborative management.

[0110] Furthermore, the industrial model can collaborate with the back-end cloud platform to convert data on potential warning hazards into small samples, which are uploaded to the platform through the potential warning sample extraction unit. After the platform operation effect is evaluated, retraining and optimization can be performed based on the newly added samples, and automatically sent to the edge for deployment.

[0111] Furthermore, the risk warning unit can issue warnings based on the risk analysis results of the industrial model, and push warnings for potential accidents. The push features include event platform push and event local warnings. Local warnings mainly include hidden danger warnings in the form of sound and light.

[0112] Furthermore, the edge-cloud collaborative management engine mainly realizes the collaborative management of data, models, and applications between the edge side where the device is located and the platform backend. The engine can work in collaboration with the intelligent edge on the cloud side to achieve efficient management and normal operation of the platform's edge, and has functions such as task orchestration, distribution management, upload management, and exception management.

[0113] Furthermore, the fusion communication unit realizes the communication between the edge and the platform side, mainly using 4G, 5G, WiFi, NB-IoT, LoRa, Zigbee, Beidou short message and other wireless communications, and the wired Ethernet supports the expansion of other CPEs. The fusion communication unit can also form an internal communication network through WiFi, LoRa, Zigbee and other methods to receive various risk perception data.

[0114] Furthermore, the integrated communication unit has a built-in transmission adaptive algorithm that can automatically select the signal channel and automatically detect the bandwidth of the channel. For low-speed channels or signal-free environments, the device can automatically run detection through the algorithm to achieve local early warning and store the early warning information. When the signal is restored, it can automatically upload to the platform. In the same time period, only one public network communication is selected for communication. The priority is 5G>4G>No-IoT>WiFi>Beidou short message. When there is no signal in the above, you can use the Ethernet port to expand the access emergency communication channel or manually intervene to set up emergency communication extension communication.

[0115] Furthermore, the hidden danger sample extractor is used to extract hidden danger samples and realize the small sample data collection for model optimization through edge-cloud collaboration. The extractor has the functions of offline storage and online re-transmission, which effectively ensures the platform synchronization of hidden danger samples.

[0116] In the present application, a job risk identification system based on multi-source heterogeneous data edge-cloud collaboration is also provided. The system includes a job site edge terminal deployed at the job site and a back-end cloud platform. The job site edge terminal and the back-end cloud platform are connected through a communication network, wherein the job site edge terminal is the above-mentioned edge terminal device, and the back-end cloud platform mainly implements the part of the above-mentioned job risk identification method based on multi-source heterogeneous data edge-cloud collaboration implemented on the back-end cloud platform.

[0117] In one embodiment, the backend cloud platform includes a disposal decision unit, an effect evaluation unit, a sample extraction unit and a model optimization unit.

[0118] In this embodiment, the disposal decision unit obtains risk warnings and risk factors, responds to the risk warnings, and performs secondary analysis based on the risk factors to obtain analysis results. The knowledge base of the knowledge graph is then used to give corresponding hidden danger control measures based on the analysis results, and provide auxiliary decision-making for hidden danger control.

[0119] In this embodiment, when performing secondary analysis based on risk factors, a risk-weighted scoring method is used for evaluation, and the risk-weighted coefficients are graded and assigned according to major accident hazards and general risk hazards.

[0120] In this embodiment, the effect evaluation unit periodically evaluates the effect of risk warning and hidden danger control measures after implementation, and confirms the effect of the operation risk identification system with the help of external expert system or human evaluation, providing means for self-optimization of the system.

[0121] In this embodiment, the sample extraction unit extracts the characteristic samples of warnings or accidents based on the secondary analysis of warnings and the retrieval analysis of knowledge graphs, and stores the sample data for correction. Further, sample extraction refers to the data operation samples when the warning occurs during risk warning, or the data generated during correction, from the collection end, such as sensors, cameras, etc.

[0122] In this embodiment, if the evaluation result indicates that optimization is required, the model optimization unit optimizes the risk identification model and the knowledge graph online based on the hidden danger samples, and sends the optimized risk identification model to the edge of the work site.

[0123] In this embodiment, the backend cloud platform also includes a risk identification model library, an intelligent edge unit, and a graph analysis unit.

[0124] In this embodiment, the intelligent edge module is mainly used to achieve data collaboration, model collaboration and application collaboration with the edge end, and to achieve edge end management, automatic model deployment, model operation control and switching, etc.

[0125] In this embodiment, the graph analysis module constructs the ontology of hidden danger disposal based on the knowledge graph construction engine, forms a hidden danger management knowledge graph, and provides graph services for disposal decisions.

[0126] In this embodiment, the risk identification model library imports the optimized model into the industrial model, namely the risk identification model management library, and cooperates with the intelligent edge to realize that the model can be automatically deployed on the edge.

[0127] In this embodiment, the communication between the edge end of the work site and the back-end cloud platform uses public network communication as the main communication method. The public network communication methods include but are not limited to 5G, 4G, NB-IoT, Beidou short messages, emergency communications, WiFi, and Ethernet in wired communications. It is used to pull data from the edge end of the work site and the back-end cloud platform. During the low-speed passage of Beidou short messages or emergency communications, the edge and the platform only realize data coordination to determine the timeliness of risk identification. For video data, analysis is only performed on the edge side, the warning results are uploaded to the platform, and the sample data is stored locally.

[0128] In the above-mentioned operation risk identification method, system and ambition based on edge-cloud collaboration, the multi-source heterogeneous data generated by multi-dimensional risk perception is used as the driving force, and a risk identification model is constructed through the edge-cloud collaboration method to realize risk perception, risk analysis, and risk warning. A risk management system is constructed based on the knowledge graph to effectively manage the risks generated and prevent accidents. This method also constructs a risk identification and edge-cloud collaboration device, which is the edge-side carrier for the operation of the risk identification system. It can realize edge-cloud collaboration and risk perception, risk analysis and local warning, and can feed back warning information to the platform side for analysis, and receive disposal information downloaded by the platform.

[0129] It should be understood that although Figure 1 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 1 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0130] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 7As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected via a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for identifying job risks based on edge-cloud collaboration is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0131] Those skilled in the art will understand that Figure 7 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0132] In one embodiment, a computer device applied to an edge end of a work site is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

[0133] Acquire multi-source heterogeneous data at the work site, and fuse the multi-source heterogeneous data to obtain multi-source fused data;

[0134] Using a risk identification model to perform risk analysis based on the multi-source fusion data to extract risk factors, and judging whether there are hidden dangers in the current operation site based on the risk factors;

[0135] If there are hidden dangers, risk warnings and risk factors are sent to the back-end cloud platform, and multi-source fusion data of hidden dangers are sent to the back-end cloud platform as hidden danger samples.

[0136] In one embodiment, a computer device applied to a backend cloud platform is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

[0137] Acquire the risk warning and risk factors, and when triggered by the risk warning, perform a secondary risk analysis on the risk factors to obtain analysis results;

[0138] The knowledge base of the knowledge graph is used to provide corresponding hidden danger management methods according to the analysis results;

[0139] Periodically evaluate the effectiveness of the risk warning and hidden danger control measures after implementation;

[0140] If the evaluation result indicates that optimization is required, the risk identification model and the knowledge graph are optimized online according to the hidden danger samples, and the optimized risk identification model is sent to the edge of the work site.

[0141] In one embodiment, a computer-readable storage medium applied to an edge end of a work site is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0142] Acquire multi-source heterogeneous data at the work site, and fuse the multi-source heterogeneous data to obtain multi-source fused data;

[0143] Using a risk identification model to perform risk analysis based on the multi-source fusion data to extract risk factors, and judging whether there are hidden dangers in the current operation site based on the risk factors;

[0144] If there are hidden dangers, risk warnings and risk factors are sent to the back-end cloud platform, and multi-source fusion data of hidden dangers are sent to the back-end cloud platform as hidden danger samples.

[0145] In one embodiment, a computer-readable storage medium applied to a backend cloud platform is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0146] Acquire the risk warning and risk factors, and when triggered by the risk warning, perform a secondary risk analysis on the risk factors to obtain analysis results;

[0147] The knowledge base of the knowledge graph is used to provide corresponding hidden danger management methods according to the analysis results;

[0148] Periodically evaluate the effectiveness of the risk warning and hidden danger control measures after implementation;

[0149] If the evaluation result indicates that optimization is required, the risk identification model and the knowledge graph are optimized online according to the hidden danger samples, and the optimized risk identification model is sent to the edge of the work site.

[0150] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0151] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0152] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A method for identifying operational risks based on edge-cloud collaboration, characterized in that: The method is implemented between the edge end at the work site and the back-end cloud platform; The method comprises implementing at the edge of the work site: Acquire multi-source heterogeneous data at the work site, and fuse the multi-source heterogeneous data to obtain multi-source fused data; Using a risk identification model to perform risk analysis based on the multi-source fusion data to extract risk factors, and judging whether there are hidden dangers in the current operation site based on the risk factors; If there is a hidden danger, a risk warning and risk factors are sent to the backend cloud platform, and the multi-source fusion data of the hidden danger is sent to the backend cloud platform as a hidden danger sample; The method includes implementing on the backend cloud platform: Acquire the risk warning and risk factors, and when triggered by the risk warning, perform a secondary risk analysis on the risk factors to obtain analysis results; The knowledge base of the knowledge graph is used to provide corresponding hidden danger management methods according to the analysis results; Periodically evaluate the effectiveness of the risk warning and hidden danger control measures after implementation; If the evaluation result indicates that optimization is required, the risk identification model and the knowledge graph are optimized online according to the hidden danger samples, and the optimized risk identification model is sent to the edge of the work site.

2. The method for identifying operational risks based on edge-cloud collaboration according to claim 1 is characterized in that: The fusing the multi-source heterogeneous data to obtain multi-source fused data comprises: Using a risk data extraction model to extract risk data from the multi-source heterogeneous data according to a preset risk data feature set table; The risk data is classified according to the data structure type, and the classified data is processed accordingly and labeled with time series labels to obtain the multi-source fusion data.

3. The method for identifying operational risks based on edge-cloud collaboration according to claim 2 is characterized in that: The risk data is classified according to the data structure type, and after corresponding processing and time series labeling, the multi-source fusion data obtained includes: Classifying the risk data into structured data types, semi-structured data types, and unstructured data types; For the risk data of the structured data type, a time series label is directly added; For the risk data of the semi-structured data type, format conversion is performed or the original data is kept according to the requirements of the risk data feature set table, and time series labels are added; The risk data of the unstructured data type is standardized according to the requirements of the risk data feature set table and is labeled with a time series label.

4. The method for identifying operational risks based on edge-cloud collaboration according to claim 3 is characterized in that: Using the risk identification model to perform risk analysis based on the multi-source fusion data to extract risk factors includes: Extracting risk factors of multiple risk types from the multi-source fusion data using the risk identification model; The preset weights of each risk type and each risk factor are calculated to obtain a total risk factor, and it is determined whether there are hidden dangers in the current work site based on the total risk factor.

5. The method for identifying operational risks based on edge-cloud collaboration according to claim 4 is characterized in that: The knowledge base using the knowledge graph provides corresponding hidden danger management methods according to the analysis results, including: The hidden danger search agent performs semantic analysis on the analysis result to obtain multiple key semantics related to the risk hidden dangers in the analysis result; Search the knowledge base according to the multiple key semantics to obtain relevant knowledge, wherein the knowledge base includes accident analysis and prediction, laws, regulations and standards, danger and controllable operability, operating procedures and abnormal handling, risk assessment and management knowledge graphs, and the industry language model currently formed in various industries; The treatment fusion agent integrates the relevant knowledge, and the treatment method agent then arranges and processes the integrated relevant knowledge to obtain the corresponding hidden danger management method.

6. An edge device, characterized in that: The edge device includes: a risk perception unit, a data processing unit, a risk warning unit and a hidden danger sample extractor; The risk perception unit is used to obtain multi-source heterogeneous data at the work site; The data processing unit is used to fuse the multi-source heterogeneous data to obtain multi-source fused data, and use a risk identification model to perform risk analysis based on the multi-source fused data to extract risk factors, and determine whether there are hidden dangers in the current work site based on the risk factors; If there is a hidden danger, the risk warning unit will send a risk warning and risk factors to the backend cloud platform; The hidden danger sample extractor is used to extract multi-source fusion data with hidden dangers and send them to the back-end cloud platform as hidden danger samples.

7. The edge device according to claim 6, characterized in that: The risk perception unit includes an inherent perception module and a perception expansion module; The inherent perception module is used to collect multi-source heterogeneous data related to vision, position, environment and status inside the work site; The perception expansion module is used to collect multi-source heterogeneous data outside the work site.

8. The edge device according to claim 7, characterized in that: The edge device also includes: an edge-cloud coordination management engine, a converged communication unit, and a power supply unit; The edge-cloud coordination management engine is used for model collaboration, data collaboration and application collaboration with the backend cloud platform; The fusion communication unit is used to communicate with the backend cloud platform; The power supply unit is used to supply power to the risk perception unit, data processing unit, risk warning unit, hidden danger sample extractor, edge-cloud coordination management engine and fusion communication unit.

9. An operation risk identification system based on multi-source heterogeneous data edge-cloud collaboration, characterized in that: The system includes a work site edge terminal deployed at the work site and a backend cloud platform, wherein the work site edge terminal is connected to the backend cloud platform via a communication network, wherein the work site edge terminal is the edge terminal device according to any one of claims 6 to 8, and the backend cloud platform includes a disposal decision unit, a graph analysis unit, an effect evaluation unit and a model optimization unit; The disposal decision unit is used to obtain the risk warning and risk factors, and when triggered by the risk warning, conduct a secondary risk analysis on the risk factors to obtain analysis results, and then use the knowledge base of the knowledge graph in the graph analysis unit to give corresponding hidden danger management measures according to the analysis results; The effect evaluation unit is used to periodically evaluate the effects of the risk warning and hidden danger control measures after implementation; If the evaluation result indicates that optimization is required, the model optimization unit is used to perform online optimization on the risk identification model and the knowledge graph according to the hidden danger samples, and the optimized risk identification model is sent to the edge of the work site.

Images