Metal smelting enterprise operation hidden danger identification method and device based on artificial intelligence

Abstract

The invention relates to the technical field of risk management and control, and provides a metal smelting enterprise operation hidden danger identification method and device based on artificial intelligence. The method comprises the following steps: the central service platform establishes a digital twinborn model of each grid according to sensing data of each grid; inputting the abnormal sensing data into an alarm analysis model to obtain an alarm and an alarm risk level of each grid, and performing color identification on a grid corresponding area in the digital twin model; whether a material carrying vehicle enters the material carrying area of each grid or not is detected; and when a material handling vehicle is driven in, the alarm risk level of each alarm in the grid is improved by one level. According to the method, the digital twinborn model is established, the digital twinborn model is subjected to color identification on the grid according to the alarm, and the dynamic link of material carrying is taken into consideration, so that when the material carrying action occurs, the alarm risk level of the grid is improved, and a user can conveniently and timely pay attention to an area with operation hidden dangers.

Description

Technical Field

[0001] The present invention relates to the field of risk management technology, and in particular to an artificial intelligence-based method and device for identifying hidden dangers in metal smelting operations. Background Art

[0002] With the continuous development of science and technology, people have put forward new demands for the monitoring and control of safety risks. They require that online monitoring be provided in addition to conventional control measures, and that the data generated by monitoring be integrated and analyzed so as to timely grasp the real-time status of production safety risks.

[0003] Metal smelting enterprises usually occupy a large area, have many equipment, and most of them are heavy industrial equipment. Therefore, people often have higher requirements for risk monitoring of metal smelting enterprises. In the existing technology, risk monitoring in metal smelting enterprises still has the following shortcomings:

[0004] (1) Existing security information still has gaps in meeting the requirements for security risk monitoring and early warning and information resource sharing and use. On-site real-time safety data is not effectively collected and applied, and the degree of integration between data and applications is not high.

[0005] (2) Each system has an "island" problem, and the overall data utilization is not high. For example, the system records the business data of each department of the enterprise in real time. There is a large amount of data in the database and various systems. These data are limited to various modules for the future planning of the enterprise, such as personnel management, dangerous operations, hidden danger investigation, training and education, etc. However, the data of a single system (such as a security system) can only express information within a certain range and scenario. It is impossible to combine all the structured and semi-structured data of the enterprise for analysis, and it is impossible to compare different dimensions and indicators and mine the inherent implicit information of the data through algorithms.

[0006] (3) Risk identification is not timely, tracing is not accurate, and resources are wasted. Some safety information management platforms have not conducted in-depth research and applied safety science principles to build effective and systematic accident prevention and control models. They do not have a deep understanding of the attributes and distribution of risks, resulting in a large variety of identified risks, consuming a large amount of data resources and hardware resources, resulting in low risk identification efficiency and failure to achieve accurate tracing.

[0007] (4) In addition, the existing risk classification and control system is unable to intuitively display the risk situation of each region, and is unable to dynamically warn and control major risk points; the system is also unable to display the collected data in an integrated manner, failing to meet the requirements of unified visualization.

[0008] In view of this, overcoming the defects of the prior art is an urgent problem to be solved in the field of this technology. Summary of the invention

[0009] The technical problem to be solved by the present invention is to provide a method for identifying hidden dangers in metal smelting operations based on artificial intelligence.

[0010] The present invention adopts the following technical solution:

[0011] In the first aspect, the present invention provides a method for identifying hidden dangers in metal smelting operations based on artificial intelligence, wherein a central service platform is set for the enterprise area, the enterprise area is divided into a plurality of grids, an intelligent sensor device is set at each monitoring position of each grid, and a centralized device is set for one or more grids, wherein the centralized device is used to collect sensor data of the intelligent sensor device and report the collected sensor data to the central service platform; a camera is set in each grid for real-time imaging in the grid, and the method comprises:

[0012] The central service platform establishes a digital twin model of each grid based on the sensor data of each grid; wherein the digital twin model is presented to the user in the form of image modeling combined with data presentation;

[0013] The central service platform also compares the sensor data with the corresponding monitoring threshold to determine whether the sensor data is abnormal; the sensor data with abnormalities is input into the pre-trained alarm analysis model to perform alarm analysis on each grid to obtain the alarm of each grid and the corresponding alarm risk level;

[0014] According to the alarm risk level of each alarm in the grid, the corresponding area of ​​the grid in the digital twin model is color-coded, so that users can intuitively confirm the risk status of the grid through the digital twin model;

[0015] The central service platform also receives real-time images of each grid, and uses a pre-trained image recognition model to perform image recognition on the real-time images to detect whether there are hidden danger behaviors in each grid; when there are hidden danger behaviors in the grid, the alarm risk level of each alarm in the grid is increased by one level;

[0016] The risk behaviors include: the entry of material handling vehicles into the material handling area of ​​the grid, irregular behaviors of the grid administrator, and the identification of risk devices or risk signs in the grid.

[0017] Preferably, a corresponding grid administrator is assigned to each grid, and the grid administrator is used to perform daily maintenance and alarm processing on the equipment in the grid. A grid captain is also designated for multiple grid administrators, and an interactive device is assigned to the grid captain, and the assigned interactive device is used to exchange information with the central service platform; the grid administrator and the grid captain are both trained and employed; the method further includes:

[0018] The central service platform conducts training effect tests on the corresponding grid administrators at corresponding intervals. The training effect tests specifically include:

[0019] The central service platform selects a first alarm item from the preset alarm table items, and reversely generates test data according to the first alarm item; the test data is sensor data that triggers the abnormality of the first alarm item;

[0020] The test data is sent to the centralized device so that the centralized device sends the test data to the corresponding intelligent sensor device, and the intelligent sensor device uses the test data to replace the actually collected sensor data and displays it on the dial, so that the grid administrator can observe the abnormal data through the dial of the intelligent sensor device and report the abnormal data to the grid leader, who reports the abnormal data to the central service platform through the interactive device;

[0021] The central service platform calculates the matching degree between the abnormal data received from the interactive device and the test data;

[0022] According to the matching degree, the training score of the grid administrator of the corresponding grid is calculated, and the training score is displayed in the digital twin model, so as to judge whether the grid administrator is suitable for the position or whether further education and training is needed through the training score.

[0023] Preferably, the first alarm item is an alarm item whose alarm processing priority is lower than a preset priority and whose alarm level is higher than a preset level.

[0024] Preferably, the central service platform calculates the matching degree between the abnormal data received from the interactive device and the test data, specifically including:

[0025] The matching degree is calculated using the first formula; the first formula is Among them, I t_n is the matching degree obtained by the nth training effect test of the grid administrator, Num(error_data) is the number of abnormal data received by the central service platform from the interactive device, and Num(test_data) is the number of generated test data.

[0026] Preferably, the calculating the training score of the grid administrator of the corresponding grid according to the matching degree specifically includes:

[0027] The test results of the grid administrator's training effectiveness test are calculated using the second formula;

[0028] Use the third formula to calculate the training score of the grid administrator after the training effect test;

[0029] The second formula is I't_n =k1I t_n +k2I' t_n-1 , the third formula is I g_n =k3I' t_n +k4I r ; k1, k2, k3 and k4 are corresponding weight values, where k1+k2=1, k3+k4=1, I′ t_n is the test result of the nth training effect test, I r is the evaluation score of the grid administrator after training, I g_n is the training score of the grid administrator after the nth training effect test, I t_n is the matching degree obtained from the nth training effect test of the grid administrator, I' t_n-1 This is the test result of the n-1th training effectiveness test for grid administrators.

[0030] Preferably, during the training effect test, the central service platform still receives the sensor data, and when an alarm is found in the grid where the training effect test is performed, the training effect test is terminated to display the sensor data collected by the intelligent sensor device on the dial;

[0031] The grid captain is informed of the time of the training effect test through the interactive device, so that the grid captain can notify the grid administrator to terminate the training effect test and carry out alarm processing in time.

[0032] Preferably, the central service platform conducts a training effect test on the corresponding grid administrators at corresponding intervals, specifically including:

[0033] At intervals of a first preset period, a random number is generated for the grid. If the generated random number is a prime number, statistics are then performed to determine whether the device configuration information of the grid has changed within a second preset period before the current moment, or whether an alarm has been generated by the grid within the second preset period before the current moment;

[0034] If the device configuration information of the grid has not changed within the second preset period before the current moment, and the grid has not generated an alarm, a training effect test is performed on the grid administrator of the grid;

[0035] The generating of random numbers specifically includes:

[0036] A seed number is obtained by adding the grid code to the current time, a random intermediate number is generated by using the seed number, and a modulus of M is obtained by using the random intermediate number to obtain the random number;

[0037] in, T0 is a preset value, T is a third preset period, times is the number of alarms in the grid in the third preset period, and when no alarm occurs in the grid in the third preset period before the current moment, M=T0×T.

[0038] Preferably, the central service platform also compares the sensor data with the corresponding monitoring threshold to determine whether the sensor data is abnormal; the sensor data with abnormalities is input into a pre-trained alarm analysis model to perform alarm analysis on each grid to obtain the alarm of each grid and the corresponding alarm risk level; specifically including:

[0039] The monitoring thresholds include a high-level monitoring threshold, a medium-level monitoring threshold and a low-level monitoring threshold. The sensor data is compared with the high-level monitoring threshold, the medium-level monitoring threshold and the low-level monitoring threshold in sequence until the sensor data exceeds the range of the corresponding monitoring threshold. It is determined that the sensor data is abnormal, and the risk level of the exceeded monitoring threshold is used as the abnormal level of the sensor data;

[0040] Use all abnormal sensor data to generate an abnormal sensor vector; wherein each value in the abnormal sensor vector corresponds to a sensor data, and when there is no abnormality in the sensor data, the corresponding value in the abnormal sensor vector is set to 0, when the abnormal level of the sensor data is low, the corresponding value in the abnormal sensor vector is set to 1, when the abnormal level of the sensor data is medium, the corresponding value in the abnormal sensor vector is set to 2, and when the abnormal level of the sensor data is high, the corresponding value in the abnormal sensor vector is set to 3;

[0041] The abnormal sensing vector is input into a pre-trained alarm analysis model to obtain an alarm vector; wherein the alarm vector includes the triggering probability of alarms of each alarm risk level, and the alarm with a triggering probability higher than a preset probability is used as the alarm of each grid.

[0042] In a second aspect, the present invention further provides an artificial intelligence-based metal smelting enterprise operation hazard identification device, which is used to implement the artificial intelligence-based metal smelting enterprise operation hazard identification method described in the first aspect, and the device includes:

[0043] 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 processor to execute the method for identifying operational hazards in metal smelting enterprises based on artificial intelligence as described in the first aspect.

[0044] In a third aspect, the present invention further provides a non-volatile computer storage medium, wherein the computer storage medium stores computer executable instructions, and the computer executable instructions are executed by one or more processors to complete the method described in the first aspect.

[0045] In a fourth aspect, a chip is provided, comprising: a processor and an interface, for calling and running a computer program stored in the memory from a memory, and executing the method as described in the first aspect.

[0046] In a fifth aspect, a computer program product comprising instructions is provided, which, when executed on a computer or a processor, causes the computer or the processor to execute the method as described in the first aspect.

[0047] The present invention uses an alarm analysis model to perform alarm analysis, so that it can effectively use historical alarm data, analyze and process large-scale and highly complex alarm items, and establish a digital twin model. The digital twin model color-codes the grids according to the alarm, so that users can intuitively confirm the risk situation of each area, and take the dynamic link of material handling into consideration. When material handling occurs, the alarm risk level of the grid is increased, so that users can pay attention to areas with operational hazards in a timely manner. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments of the present invention. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0049] Figure 1 It is a schematic diagram of a first method for identifying hidden dangers in metal smelting operations based on artificial intelligence provided by an embodiment of the present invention;

[0050] Figure 2 This is a flow chart of a first method for identifying hidden dangers in metal smelting operations based on artificial intelligence provided by an embodiment of the present invention;

[0051] Figure 3 is a schematic diagram of a second method for identifying hidden dangers in metal smelting operations based on artificial intelligence provided by an embodiment of the present invention;

[0052] Figure 4 It is a schematic diagram of the architecture of an intelligent sensing device in a method for identifying hidden dangers in metal smelting operations based on artificial intelligence provided by an embodiment of the present invention;

[0053] Figure 5is a schematic diagram of a third method for identifying hidden dangers in metal smelting operations based on artificial intelligence provided by an embodiment of the present invention;

[0054] Figure 6 is a schematic diagram of a fourth method for identifying hidden dangers in metal smelting operations based on artificial intelligence provided by an embodiment of the present invention;

[0055] Figure 7 It is a flow chart of a second method for identifying hidden dangers in metal smelting operations based on artificial intelligence provided by an embodiment of the present invention;

[0056] Figure 8 is a schematic diagram of a fifth method for identifying hidden dangers in metal smelting operations based on artificial intelligence provided by an embodiment of the present invention;

[0057] Fig. 9 is a schematic diagram of a sixth method for identifying hidden dangers in metal smelting operations based on artificial intelligence provided by an embodiment of the present invention;

[0058] Fig.10 It is a schematic diagram of a first preset alarm table item in a method for identifying hidden dangers in metal smelting enterprises based on artificial intelligence provided by an embodiment of the present invention;

[0059] Fig.11 It is a flow chart of a third method for identifying hidden dangers in metal smelting operations based on artificial intelligence provided by an embodiment of the present invention;

[0060] Fig.12 It is a schematic diagram of test data in a method for identifying hidden dangers in metal smelting operations based on artificial intelligence provided by an embodiment of the present invention;

[0061] Fig.13 It is a flowchart of a fourth method for identifying hidden dangers in metal smelting operations based on artificial intelligence provided by an embodiment of the present invention;

[0062] Fig.14 It is a flowchart of a fifth method for identifying hidden dangers in metal smelting operations based on artificial intelligence provided by an embodiment of the present invention;

[0063] Fig.15 It is a flow chart of a sixth method for identifying hidden dangers in metal smelting operations based on artificial intelligence provided by an embodiment of the present invention;

[0064] Fig.16 It is a schematic diagram of the relationship between various regions in a method for identifying hidden dangers in metal smelting operations based on artificial intelligence provided by an embodiment of the present invention;

[0065] Fig.17It is a schematic diagram of equipment in a blast furnace workshop in a method for identifying hidden dangers in metal smelting operations based on artificial intelligence provided by an embodiment of the present invention;

[0066] Fig.18 It is a schematic diagram of a first preset alarm table item in another method for identifying hidden dangers in metal smelting enterprises based on artificial intelligence provided by an embodiment of the present invention;

[0067] Fig.19 It is a schematic diagram of a second preset alarm table item in a method for identifying hidden dangers in metal smelting enterprises based on artificial intelligence provided by an embodiment of the present invention;

[0068] Fig. 20 is a schematic diagram of a seventh method for identifying hidden dangers in metal smelting operations based on artificial intelligence provided by an embodiment of the present invention;

[0069] Fig.21 It is a flow chart of a seventh method for identifying hidden dangers in metal smelting operations based on artificial intelligence provided by an embodiment of the present invention;

[0070] Fig. 22 It is a flow chart of an eighth method for identifying hidden dangers in metal smelting operations based on artificial intelligence provided by an embodiment of the present invention;

[0071] Fig.23 It is a schematic diagram of a second preset alarm table item in another method for identifying hidden dangers in metal smelting enterprises based on artificial intelligence provided by an embodiment of the present invention;

[0072] Fig.24 It is a flow chart of a ninth method for identifying hidden dangers in metal smelting operations based on artificial intelligence provided by an embodiment of the present invention;

[0073] Fig.25 This is a flow chart of a tenth method for identifying hidden dangers in metal smelting operations based on artificial intelligence provided by an embodiment of the present invention;

[0074] Fig.26 It is a schematic diagram of a second preset alarm table item in another method for identifying hidden dangers in metal smelting enterprises based on artificial intelligence provided by an embodiment of the present invention;

[0075] Fig. 27 It is a schematic diagram of the architecture of an artificial intelligence-based metal smelting enterprise operation hazard identification device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0076] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention 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 invention and are not intended to limit the present invention.

[0077] Unless the context requires otherwise, throughout the specification and claims, the term "including" is to be interpreted as open inclusion, that is, "including, but not limited to". In the description of the specification, the terms "one embodiment", "some embodiments", "exemplary embodiments", "examples", "specific examples" or "some examples" and the like are intended to indicate that specific features, structures, materials or characteristics associated with the embodiment or example are included in at least one embodiment or example of the present disclosure. The schematic representation of the above terms does not necessarily refer to the same embodiment or example. In addition, the specific features, structures, materials or characteristics may be included in any one or more embodiments or examples in any appropriate manner, that is, although they may be carried in the embodiments or examples of the above terms due to reasons such as the order and position of appearance, it is not limited to that they can be carried in combination by one embodiment or example.

[0078] In the description of the present invention, the terms "first" and "second" are used only for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the embodiments of the present disclosure, unless otherwise specified, the meaning of "multiple" is two or more. In addition, for example, the same type of nouns may be described as two independent individuals by adding "A" and "B" at the end. In this case, the corresponding features defined as "A" and "B" are only used to distinguish the same type of individuals for description purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features.

[0079] In the description of the present invention, the expression "A and / or B" (where A and B are used to formally represent specific characteristic contents) will be involved, and the corresponding expressions include the following three combinations: only A, only B, and a combination of A and B.

[0080] As used herein, "about," "substantially," or "approximately" includes the stated value and an average value that is within an acceptable range of deviation from the particular value as determined by one of ordinary skill in the art taking into account the measurements in question and the errors associated with the measurement of the particular quantity (i.e., the limitations of the measurement system).

[0081] In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0082] Embodiment 1:

[0083] Embodiment 1 of the present invention provides a method for identifying hidden dangers in metal smelting operations based on artificial intelligence. A central service platform is set for the enterprise area, the enterprise area is divided into multiple grids, intelligent sensing devices are set at each monitoring position of each grid, and centralized devices are set for one or more grids. The centralized devices are used to collect sensor data of the intelligent sensing devices and report the collected sensor data to the central service platform. In a specific application scenario, one grid corresponds to one centralized device, such as Figure 1 As shown, the specific division method of the grid is obtained by a technician in the field according to the analysis of the degree of correlation between the areas in the enterprise area; a camera is set in the material handling area of ​​each grid to collect real-time images of the material handling area, such as Figure 2 As shown, the method includes:

[0084] In step 201, the central service platform establishes a digital twin model of each grid according to the sensor data of each grid; wherein the digital twin model is presented to the user in the form of image modeling combined with data presentation, such as Figure 3 As shown; the material handling area refers to the area in the grid where material handling vehicles may enter and exit.

[0085] In step 202, the central service platform also compares the sensor data with the corresponding monitoring threshold to determine whether the sensor data is abnormal; the sensor data with abnormalities is input into the pre-trained alarm analysis model to perform alarm analysis on each grid to obtain the alarm of each grid and the corresponding alarm risk level; the monitoring threshold of each sensor data is obtained by a technician in this field based on empirical analysis. The intelligent sensor device includes a sensor, a control mainboard and a display dial. The input end of the control mainboard is connected to the output end of the sensor. The control mainboard is also connected to the corresponding centralized device through a network function. The sensor is used to collect the sensor data of the monitoring position and send the sensor data to the control mainboard. The control mainboard sends the sensor data to the centralized device, and the centralized device sends the sensor data to the central service platform. Among them, the centralized device and each intelligent sensor device can be connected by wired or local networking, such as using zigbee networking, and the centralized device and the central service platform are realized through TCP communication.

[0086] A control terminal of the control mainboard is also connected to the display dial to display the sensor data on the display dial. The control mainboard can also receive test data from the centralized equipment and display the test data on the display dial. Figure 4 shown.

[0087] In step 203, the corresponding area of ​​the grid in the digital twin model is color-coded according to the alarm risk level of each alarm in the grid, so that the user can intuitively confirm the risk situation of the grid through the digital twin model; wherein, when the highest alarm risk level in the corresponding grid is high, the corresponding area of ​​the grid in the digital twin model is marked as red, when the highest alarm risk level in the corresponding grid is medium, the corresponding area of ​​the grid in the digital twin model is marked as orange, when the highest alarm risk level in the corresponding grid is low, the corresponding area of ​​the grid in the digital twin model is marked as yellow, and when the highest alarm risk level in the corresponding grid is no risk, the corresponding area of ​​the grid in the digital twin model is marked as blue, that is, according to the distribution of various equipment and plant buildings in the blast furnace workshop, an image model is established, the corresponding color is marked for the corresponding grid in the image model, and when the user clicks on the grid, the sensor data of each device in the grid is displayed to the user.

[0088] In step 204, the central service platform also receives the real-time images of each grid, and uses the pre-trained image recognition model to perform image recognition on the real-time images to detect whether there are hidden danger behaviors in each grid; when there are hidden danger behaviors in the grid, the alarm risk level of each alarm in the grid is increased by one level, such as Figure 5 As shown; and the enhanced alarm risk level is displayed on the digital twin model.

[0089] The risk behaviors include: the entry of material handling vehicles into the material handling area of ​​the grid, irregular behaviors of the grid administrator, and the identification of risk devices or risk signs in the grid.

[0090] Since material handling vehicles in smelting enterprises often carry flammable or explosive materials, such as coking oil, when a security incident occurs in the corresponding area, if there are still material handling vehicles in the area, there is a high probability that the incident will be escalated. Therefore, when the material handling vehicle enters, the alarm risk level of each alarm in the grid is increased to enable the grid administrator to pay more attention to the grid work situation. Similarly, when the grid administrator behaves irregularly or identifies risk devices or risk signs, there may be hidden risks. Therefore, by increasing the alarm risk level, users can keep paying attention, and the corresponding real-time image screenshots are also presented to users as hidden danger details records, such as Figure 6 The image recognition model can be a deep learning model, such as a convolutional neural network model (CNN). The image recognition model is obtained by experts pre-collecting various hidden danger behavior images, collecting and updating the hidden danger behavior images to an online sample library, and then using the online sample library to train the image recognition model. Figure 8As shown, after a period of time, the online sample library is used to perform additional training on the alarm analysis model so that the model is continuously updated.

[0091] For some smelting enterprises that involve a wide range of fields and have relatively more correlations between equipment, the number of alarm items is usually large, and the relationship between each alarm item may be relatively more complex. The alarm analysis method in Example 1 and Example 2 may be difficult to cope with the analysis and processing of large-scale and highly complex alarm items. Therefore, this embodiment uses an alarm analysis model to perform alarm analysis, so that historical alarm data can be effectively used to analyze and process large-scale and highly complex alarm items, and a digital twin model is established. The digital twin model color-codes the grids according to the alarms, so that users can intuitively confirm the risk situation of each area, and take the dynamic link of material handling into consideration. When material handling occurs, the alarm risk level of the grid is increased, so that users can pay attention to areas with operational hazards in a timely manner.

[0092] The central service platform also compares the sensor data with the corresponding monitoring threshold to determine whether the sensor data is abnormal; the sensor data with abnormalities is input into the pre-trained alarm analysis model to perform alarm analysis on each grid to obtain the alarm of each grid and the corresponding alarm risk level; Figure 7 As shown, specifically including:

[0093] In step 301, the monitoring threshold includes a high-level monitoring threshold, an intermediate monitoring threshold and a low-level monitoring threshold. The sensor data is compared with the high-level monitoring threshold, the intermediate monitoring threshold and the low-level monitoring threshold in turn. When the comparison shows that the sensor data exceeds the range of the corresponding monitoring threshold, it is judged that the sensor data is abnormal, and the risk level of the exceeded monitoring threshold is used as the abnormal level of the sensor data; when the comparison shows that the sensor data does not exceed the range of each monitoring threshold, it is judged that the sensor data is not abnormal.

[0094] In step 302, all abnormal sensor data are used to generate an abnormal sensor vector; wherein each value in the abnormal sensor vector corresponds to a sensor data, and when there is no abnormality in the sensor data, the corresponding value in the abnormal sensor vector is set to 0, when the abnormal level of the sensor data is low, the corresponding value in the abnormal sensor vector is set to 1, when the abnormal level of the sensor data is medium, the corresponding value in the abnormal sensor vector is set to 2, and when the abnormal level of the sensor data is high, the corresponding value in the abnormal sensor vector is set to 3.

[0095] In step 303, the abnormal sensing vector is input into the pre-trained alarm analysis model to obtain an alarm vector; wherein the alarm vector includes the triggering probability of the alarms of each alarm risk level, and the alarm with a triggering probability higher than the preset probability is used as the alarm of each grid. For example, for the alarm item "low blast furnace fuel flow", there are three alarm risk levels, and there are three corresponding values ​​in the alarm vector, the first value is the probability of triggering the "low blast furnace fuel flow" alarm with a high alarm risk level, the second value is the probability of triggering the "low blast furnace fuel flow" alarm with a medium alarm risk level, and the third value is the probability of triggering the "low blast furnace fuel flow" alarm with a low alarm risk level. When the triggering probabilities of multiple alarm risk levels of the same alarm item are all higher than the preset probability, the alarm item with the highest alarm risk level is used as the alarm obtained by analysis.

[0096] The alarm analysis model may be a deep learning model, where experts generate a training data set based on historical sensor data and alarms, and input the training data set into the deep learning model for training to obtain a pre-trained alarm analysis model. The preset probability is obtained by those skilled in the art based on empirical analysis. The deep learning model may be as follows: Fig. 9 One or more of the models shown.

[0097] The training process is as follows: (1) Use bottom-up unsupervised learning (that is, start from the bottom and train layer by layer to the top): Use uncalibrated data (calibrated data is also acceptable) to train the parameters of each layer in layers. This step can be regarded as an unsupervised training process, which is the most different from the traditional neural network (this process can be regarded as a feature learning process); specifically, first use uncalibrated data to train the first layer, and learn the parameters of the first layer during training (this layer can be regarded as a hidden layer of a three-layer neural network that minimizes the difference between output and input). Due to the capacity limitation and sparsity constraints of the model, the obtained model can learn the structure of the data itself, thereby obtaining features that are more representative than the input; after learning the n-1th layer, use the output of the n-1th layer as the input of the nth layer, train the nth layer, and thus obtain the parameters of each layer. (2) Top-down supervised learning (that is, training through labeled data, the error is transmitted from top to bottom, and the network is fine-tuned). Based on the parameters of each layer obtained in the first step, the parameters of the entire multi-layer model are further fine-tuned. This step is a supervised training process. The first step is similar to the random initialization process of the neural network. Since the first step of DL is not random initialization, but obtained by learning the structure of the input data, this initial value is closer to the global optimum, thus achieving better results. Therefore, the good effect of deep learning is largely attributed to the feature learning process in the first step.

[0098] The monitoring threshold and the sensor data are managed using a first preset alarm table item in an actual application scenario. The first preset alarm table item is as follows: Fig.10 As shown, including sensor sequence number, sensor data, monitoring index, low-level monitoring threshold (including low-level monitoring threshold lower limit and low-level monitoring threshold upper limit), intermediate monitoring threshold (including intermediate monitoring threshold lower limit and intermediate monitoring threshold upper limit) and advanced monitoring threshold (including advanced monitoring threshold lower limit and advanced monitoring threshold upper limit), when the corresponding monitoring index only monitors the threshold upper limit or threshold upper limit, the party that does not need to be monitored is set to the default value (such as -1). The production abnormality sensor vector is set to the corresponding position of the abnormal sensor vector according to the order of the sensor sequence number.

[0099] In actual use, the grid is divided for better management, that is, a corresponding grid administrator is assigned to each grid, and the grid administrator is used to perform daily maintenance and alarm processing on the equipment in the grid. A grid captain is also designated for multiple grid administrators, and an interactive device is assigned to the grid captain, and the assigned interactive device is used to interact with the central service platform. The grid administrator and the grid captain are both trained and trained before taking up their posts. The method also includes:

[0100] The central service platform conducts training effectiveness tests on the corresponding grid administrators at corresponding intervals, such as Fig.11 As shown, the training effect test specifically includes:

[0101] In step 401, the central service platform selects the first alarm item from the preset alarm table items, and reversely generates test data based on the first alarm item; the test data is the abnormal sensor data that triggers the first alarm item; the central service platform selects the first alarm item from the preset alarm table items, and reversely generates test data based on the first alarm item, specifically including: selecting the first alarm item from the second preset alarm table items, and finding the corresponding sensor data (that is, the sensor data corresponding to the bit with a value of 1 in the alarm code) according to the alarm code of the first alarm item, and then finding the monitoring threshold corresponding to these sensor data from the first preset alarm table item, and generating a value outside the monitoring threshold range as test data.

[0102] In step 402, the test data is sent to a centralized device so that the centralized device sends the test data to a corresponding intelligent sensor device, and the intelligent sensor device uses the test data to replace the actually collected sensor data and displays it on a dial, so that the grid administrator can observe abnormal data through the dial of the intelligent sensor device and report the abnormal data to the grid leader, who reports the abnormal data to the central service platform through an interactive device; wherein, while the intelligent sensor device displays the test data on the dial, the intelligent sensor device still collects sensor data and transmits the sensor data to the central service platform through the centralized device.

[0103] When the test data is sent to the centralized device, the data ID of the test data (i.e., the data ID of the sensor data) is also sent. The sensor serial number can be used directly as the data ID, or a sensor ID can be additionally determined for each sensor data, such as using the UUID of the corresponding smart sensor device as the sensor ID, and sending the sensor ID and the value of the test data together to the centralized device. In actual use, the data format sent to the centralized device is as follows: Fig.12 As shown, it includes dataId (i.e. sensor ID), value (i.e. the value of test data) and other data (such as enterprise code, timestamp, etc.). For example, the Data field in a message sent from the central service platform to the centralized device is as follows:

[0104] {"dataId":"90134c28c31b49ea85e17bb90ff32eef","enterpriseId":"123456789","gat ewayId":"123456789","collectTime":"20180615123456","isConnectDataSource":true,"reportType":"report","datas":[{"quotaId":"123","value":123.0},{"quotaId":"321","value":123.0}]}

[0105] In actual use, this field is also encrypted using ASE, resulting in the following content:

[0106] {"appId":"xxxxxxxxx","serviceId":"xxxxxxxxxx","dataId":"1563259577529","data":"BLnaeC X2a0t

[0107] This message represents the sending of a test data with a sensor ID of 90134c28c31b49ea85e17bb90ff32eef and a value of 123.0. The centralized device finds the corresponding smart sensor device based on the sensor ID and sends the test data to the smart sensor device so that the test data can be displayed on the dial.

[0108] In actual use, the reporting of sensor data is also done using Fig.12 The data format shown is implemented, that is, the data reported by the intelligent sensor device to the central service platform via the centralized device is the sensor data, and the data sent by the central service platform to the intelligent sensor device via the centralized device is the test data. In addition, the sending of test data is carried out in the form of continuous sending. Each time the intelligent sensor device receives a test data, it displays the test data on the dial. Once it detects that no new test data has been received within the corresponding time after the last test data was received, it switches to displaying the sensor data on the dial, so that the switching of the display mode is managed by the intelligent sensor device itself, so as to avoid the problem that when the upstream performs the display mode switching management through commands, the test data display mode cannot be switched back to the sensor data display mode due to network interruption.

[0109] In actual use, since the grid captain also needs to perform daily maintenance of the equipment in his own grid, in order to facilitate his movement, the interactive device is a portable device in actual use, such as a mobile terminal, etc. In actual use, the application can be installed in the mobile phone of the grid captain as the interactive device. The interactive device is used to receive pending tasks from the central service platform, that is, pending alarms of the grids under the jurisdiction of the grid captain, pending alarms of the grids under the jurisdiction of other grid administrators under the grid captain, and daily maintenance tasks. The abnormal data can be understood as the data on the dial observed by the grid administrator.

[0110] In step 403, the central service platform calculates the matching degree between the abnormal data received from the interactive device and the test data; wherein the central service platform calculates the matching degree between the abnormal data received from the interactive device and the test data, specifically including:

[0111] The matching degree is calculated using the first formula; the first formula is Among them, I t_n is the matching degree obtained by the grid administrator's nth training effect test, Num(error_data) is the number of abnormal data received by the central service platform from the interactive device, and Num(test_data) is the number of test data generated. In actual use, there is also an optional implementation method: combining the difference between the value of the abnormal data and the value of the test data, so that the difference is involved in the calculation of the matching degree. In this case, the first formula is Among them, k'1 and k'2 are weights obtained by those skilled in the art based on empirical analysis, error_data i is the i-th abnormal data, test_data i is the i-th test data. When there is no i-th abnormal data, let The value of is 1.

[0112] In actual use, the grid administrator also analyzes the cause of the abnormal data based on training and learning, and reports the abnormal cause obtained by analysis to the grid captain. The grid captain reports the abnormal cause to the central service platform through the interactive device. The central service platform also uses the pre-trained semantic analysis model to analyze the abnormal cause, obtains the corresponding semantic vector, and calculates the vector distance D between the semantic vector and the first alarm item. The first formula is Here, k'3 is a weight obtained by those skilled in the art based on empirical analysis.

[0113] In step 404, the training score of the grid administrator of the corresponding grid is calculated based on the matching degree, and the training score is displayed in the digital twin model, so as to judge whether the grid administrator is suitable for the position or whether further education and training is needed through the training score.

[0114] This embodiment tests the effectiveness of training grid administrators, thereby determining whether the grid administrators can effectively manage their grids, thereby effectively managing personnel and improving the security of grid area management.

[0115] In an optional implementation, the matching degree can be directly used as the training score of the corresponding grid administrator. The higher the matching degree, the better the training effect of the grid administrator and the more adaptable the grid administrator is to the position. There is also a preferred implementation:

[0116] The calculating the training score of the grid administrator of the corresponding grid according to the matching degree specifically includes:

[0117] The second formula is used to calculate the test results of the grid administrator's training effectiveness test; the third formula is used to calculate the training score of the grid administrator after the training effectiveness test.

[0118] The second formula is I' t_n =k1I t_n +k2I' t_n-1 , the third formula is I g_n =k3I' t_n +k4I r ; k1, k2, k3 and k4 are corresponding weight values, where k1+k2=1, k3+k4=1, I′ t_n is the test result of the nth training effect test, I r The evaluation score of the grid administrator after training is usually obtained by the trainer after the training. The score is directly entered into the central service platform. g_n is the training score of the grid administrator after the nth training effect test, I t_n is the matching degree obtained from the nth training effect test of the grid administrator, I' t_n-1 This is the test result of the n-1th training effectiveness test for grid administrators.

[0119] In some application scenarios, when an alarm occurs, for some locations that cannot be intelligently monitored, the grid administrator needs to use corresponding monitoring tools to reach the monitoring location for monitoring and recording. In order to further quantify the adaptability of the grid administrator to the position, this embodiment also provides a preferred implementation method, in which the processing action of the grid administrator after monitoring the abnormal data is also included in the evaluation. The specific preferred implementation method is:

[0120] The matching degree between the abnormal data received from the interactive device and the test data is calculated as the first matching degree.

[0121] A UWB tag is pre-assigned to each grid administrator; the central service platform also determines the movement trajectory of each grid administrator after arriving at the target intelligent sensor device based on the UWB tag; the movement trajectory is matched with a preset map to determine each monitoring location passed by the movement trajectory; a second matching degree between each monitoring location and a preset processing behavior of the first alarm item is calculated; wherein the target intelligent sensor device is an intelligent sensor device that displays test data.

[0122] A training score of a grid administrator of a corresponding grid is calculated according to the first matching degree and the second matching degree.

[0123] The calculating of the second matching degree between each monitoring location and the preset processing behavior of the first alarm item specifically includes:

[0124] The second matching degree is calculated using the second formula; the second formula is: Among them, I p_n is the second matching degree obtained from the nth training effect test of the grid administrator, Num(test_pos) is the number of first monitoring positions that the grid administrator needs to manually monitor as specified in the preset processing behavior, and Num(passed_test_pos) is the number of first monitoring positions passed by the grid administrator's movement trajectory.

[0125] The preset map is formed by technicians in this field by marking each monitoring location on a grid map, and the preset processing behavior is obtained by technicians in this field based on empirical analysis, and is specifically manifested as a collection of one or more monitoring locations on the preset map.

[0126] The calculating the training score of the grid administrator of the corresponding grid according to the first matching degree and the second matching degree specifically includes:

[0127] The third formula is used to calculate the test results of the grid administrator's training effectiveness test.

[0128] The fourth formula is used to calculate the training score of the grid administrator after the training effectiveness test.

[0129] The third formula is I' t_n =k1I t_n +k2I p_n +k3I' t_n-1 , the fourth formula is I g_n =k4I' t_n +k5I r ; k1, k2, k3, k4 and k5 are all corresponding weight values, where k1+k2+k3=1, k4+k5=1, I′ t_n is the test result of the nth training effect test, I r is the evaluation score of the grid administrator after training, I g_n is the training score of the grid administrator after the nth training effect test, I t_n is the first matching degree obtained from the nth training effect test of the grid administrator, I p_n is the second matching degree obtained from the nth training effect test of the grid administrator, I′ t_n-1 This is the test result of the n-1th training effectiveness test for grid administrators.

[0130] Considering that in actual use, different alarms are handled differently, for alarms with lower alarm levels, grid administrators often only need to handle them by themselves, without reporting and analyzing. Similarly, there are some alarms that are very likely to cause security incidents if they are not handled in time. Therefore, once data anomalies are detected, they need to be processed immediately and reported after the processing is completed. Once these alarms appear, they will directly affect the working condition of the equipment. In order to avoid the impact of training effect tests on the working condition of the equipment, the first alarm item is an alarm item with an alarm processing priority lower than the preset priority and an alarm level higher than the preset level. The preset priority and preset level are obtained by technical personnel in this field based on empirical analysis. It can be considered that: when the alarm processing priority is higher than or equal to the preset priority, the grid administrator needs to handle it immediately, and when the alarm level is lower than or equal to the preset level, the grid administrator does not need to report.

[0131] In some application scenarios, during the training effect test, the central service platform still receives sensor data. When analysis shows that an alarm occurs in the grid where the training effect test is being conducted, the training effect test is terminated to display the sensor data collected by the smart sensor device on the dial.

[0132] The grid captain is informed of the time of the training effect test through the interactive device, so that the grid captain can notify the grid administrator to terminate the training effect test and carry out alarm processing in time.

[0133] In order to ensure that the training effect test does not affect the conventional data monitoring, this embodiment also provides the following preferred implementation mode, that is, the central service platform performs a training effect test on the corresponding grid administrator at a corresponding interval, such as Fig.13 As shown, specifically including:

[0134] In step 501, a random number is generated for the grid at intervals of the first preset period. If the generated random number is a prime number, statistics are then made on whether the device configuration information of the grid has changed in the second preset period before the current moment, or whether the grid has generated an alarm in the second preset period before the current moment; wherein the first preset period and the second preset period are obtained by those skilled in the art based on empirical analysis. The central service platform allocates a respective timer to each grid, and successively starts the timer to time the first preset period, so that the training effect test time of each grid is staggered.

[0135] In step 502, if the device configuration information of the grid has not changed within the second preset period before the current moment, and the grid has not generated an alarm, a training effect test is performed on the grid administrator of the grid;

[0136] When the generated random number is a prime number, it is considered that the time has come to conduct a random training effect test, and further judgment is made as to whether the conditions for the training effect test are met at present, that is, whether the devices in the grid are running stably. The judgment standard for stable operation is: if the device runs for the second preset period under a set of configuration information conditions without generating an alarm, then it is considered that the devices in the grid are running stably.

[0137] Wherein, the random number is generated, such as Fig.14 As shown, specifically including:

[0138] In step 601, a seed number is obtained by adding the grid code to the current time, and a random intermediate number is generated using the seed number.

[0139] In step 602, the random intermediate number is used to modulo M to obtain the random number.

[0140] in, T0 is a preset value, T is a third preset period, times is the number of alarms in the grid in the third preset period, and when no alarm occurs in the grid in the third preset period before the current moment, M=T0×T.

[0141] The third preset period is obtained by those skilled in the art based on empirical analysis, and the third preset period is much longer than the second preset period and the first preset period. In actual use, the third preset period is usually one week, one month or multiple months. The preset value is obtained by those skilled in the art based on demand analysis.

[0142] Among them, when the number of alarms in the third preset period is more, the M value is smaller, and the corresponding probability that the generated random number is a prime number is greater. In this way, the number of times the grid administrators in the grid with a large number of alarms conduct training effect tests is increased.

[0143] Embodiment 2:

[0144] On the basis of Example 1, this embodiment also provides another method for identifying hidden dangers in metal smelting enterprises based on artificial intelligence, setting a central service platform for the enterprise area, dividing the enterprise area into multiple grids, setting intelligent sensing equipment at each monitoring position of each grid, and setting centralized equipment for one or more grids, such as Figure 1 As shown, the centralized device is used to collect sensor data from intelligent sensor devices and report the collected sensor data to the central service platform; Fig.15 As shown, the method includes:

[0145] In step 701, the central service platform compares the sensor data with the corresponding monitoring threshold to determine whether there is any abnormality in the sensor data.

[0146] In step 702, firstly, for each grid, an alarm analysis is performed on abnormal sensor data in the corresponding grid to obtain an alarm for each grid.

[0147] In step 703, the alarms of the downstream grid are filtered out in combination with the alarms of the upstream grid to obtain the final alarm analysis result.

[0148] Among them, the steps 702-703 can be understood as first analyzing each grid to determine the possible alarms in the grid (i.e., the reasons within the grid that cause the abnormal sensor data). However, in actual use, due to the mutual correlation between grids, the alarm of the downstream grid may also be caused by the alarm in the upstream grid. For example, when the output material of the upstream grid enters the downstream grid for processing, if the output material of the upstream is abnormal due to the corresponding alarm, an alarm will also be generated in the downstream grid. At this time, the source of the alarm of the downstream grid is the alarm of the upstream grid. Therefore, the above step 703 is used to filter out such alarms of the downstream grid, thereby reducing the number of duplicate alarms and presenting clear alarm analysis results to the user. Taking some areas of steel smelting in a smelting enterprise as an example, if Fig.16As shown, it includes sintering area, pelletizing area, ironmaking area, steelmaking area, steel rolling area and coking area. There are upstream and downstream relationships between these multiple areas. For example, the molten iron produced in the ironmaking area is used as the input of the steelmaking area for steelmaking operations. For the internal area, each workshop also has an upstream and downstream relationship. For example, the upstream workshops of the blast furnace workshop include the blast equipment workshop, the hot blast furnace workshop, etc., and the output of these upstream workshops will affect the downstream workshops. For example, when the blast equipment workshop generates an alarm, resulting in a decrease in the cold air output flow, it will cause the cold air inlet of the blast furnace workshop (such as Fig.17 Insufficient cold air flow may also cause the outlet air temperature of the hot air outlet to be too high.

[0149] This embodiment divides the grids, first performs alarm analysis on each grid, and then filters the alarms of the downstream grids in combination with the alarms of the upstream grids, thereby effectively eliminating the number of duplicate alarms and determining the grid where the actual alarm is located, thereby assigning responsibility to the specific grid and facilitating alarm processing by grid managers.

[0150] In an optional application scenario, each sensor data and each alarm are managed using corresponding entries, such as using a first preset alarm entry to manage each sensor data, and using a second preset alarm entry to manage each alarm item. The first preset alarm entry and the second preset alarm entry are collectively referred to as preset alarm entries. Fig.18 As shown, it includes a sensor serial number, sensor data (also called sensor data name), monitoring indicators and monitoring thresholds (including a monitoring threshold lower limit and a monitoring threshold upper limit), wherein one sensor data can correspond to one or more monitoring indicators, and each monitoring indicator corresponds to a unique sensor serial number. When the corresponding monitoring indicator only monitors the threshold upper limit or the threshold lower limit, the party that does not need to be monitored is set to a default value (such as -1). For example, for CO concentration, only the threshold upper limit is monitored, and the threshold lower limit is set to -1. When the sensor data is subsequently compared with the corresponding monitoring threshold, there is no need to compare the threshold upper limit with the default value and the threshold lower limit with the default value in the first preset alarm table item.

[0151] The second preset alarm table entry is as follows: Fig.19As shown, it includes a grid number, an alarm number, an alarm item, an alarm code and an associated alarm exception code; a grid number represents a grid, which contains multiple alarm items, each alarm item corresponds to a unique alarm number, and each alarm item includes an alarm code, which is obtained by technical personnel in this field based on empirical analysis. Specifically: find multiple sensor data that trigger the corresponding alarm item, and set the bits corresponding to the positions of the multiple sensor data in the alarm code of the alarm item to 1, and set the other bits to 0. For example, assuming that there are M monitoring indicators in the first preset alarm table item, the length of each alarm code is M bits. When the sensor numbers of the multiple sensor data that trigger the corresponding alarm item are m1, m2, ..., m i When the alarm code of the alarm item is m1+1, m2+1, ..., m i +1 bit is set to 1, the other bits are set to 0.

[0152] Similarly, the associated alarm exception code is also obtained by those skilled in the art based on empirical analysis. Specifically, find the alarm item that triggers the corresponding alarm (usually the alarm item of the upstream grid), set the bit corresponding to the alarm item to 1, and set the other bits to 0. For example, assuming that there are N alarm items in the second preset alarm table item, the length of each associated alarm exception code is N bits, and the alarm item (for distinction, hereinafter referred to as the second alarm item) that triggers the alarm of the corresponding alarm item (for distinction, hereinafter referred to as the first alarm item) has an alarm sequence number of n1, n2, ..., n respectively. j , then the n1+1, n2+1, ..., nth j +1 bit is set to 1, the other bits are set to 0.

[0153] Sensor data can be understood as the surface manifestations directly brought about by sensor data, and alarm items can be understood as the deep-seated reasons that lead to the abnormality of corresponding sensor data.

[0154] First, for each grid, an alarm analysis is performed on the abnormal sensor data in the corresponding grid to obtain the alarm of each grid, and then the alarm of the downstream grid is filtered out in combination with the alarm of the upstream grid to obtain the final alarm analysis result, which specifically includes:

[0155] All abnormal sensor data are used to generate an abnormal sensor code; wherein each bit in the abnormal sensor code corresponds to a sensor data, and when the corresponding sensor data is abnormal, the corresponding bit in the abnormal sensor code is set to 1, otherwise, the corresponding bit is set to 0 to form the abnormal sensor code; the bit corresponding to each sensor data is determined by the sensor sequence number.

[0156] The abnormal sensor code is ANDed with each alarm code in the preset alarm table item to obtain an operation result. If the operation result is greater than 0, the alarm corresponding to the alarm code is analyzed to obtain the alarm; when the operation result is greater than 0, it means that all sensor data corresponding to the alarm code are abnormal, then it is considered that these abnormalities are caused by the alarm item corresponding to the alarm code, triggering the alarm of the alarm item.

[0157] Use all the alarms to generate abnormal alarm codes, and perform AND operations on the abnormal alarm codes and the associated alarm abnormal codes corresponding to each alarm in a preset order. If the operation result is greater than 0, the alarm corresponding to the associated alarm abnormal code is filtered out; the preset order can be expressed as an alarm sequence number. When the AND operation result is greater than 0, it is considered that there is an alarm in the upstream grid that triggers the alarm, and the alarm in the upstream grid is the real cause. The alarm in the downstream grid is only the chain effect caused by the alarm in the upstream grid, so the alarm in the downstream grid is filtered out.

[0158] Among them, the preset alarm table items include the monitoring thresholds corresponding to each sensor data, the alarm codes of each alarm item and the associated alarm exception codes of each alarm item; among them, in the alarm code of the corresponding alarm item, the bits corresponding to all the sensor data that cause the alarm item to be abnormal are 1, and the other bits are 0; in the associated alarm exception code of the corresponding alarm item, the bits corresponding to all other alarms that cause the alarm are 1, and the other bits are 0.

[0159] In combination with the method described in Example 1, this embodiment further provides a preferred implementation method, that is, a corresponding grid administrator is assigned to each grid, the grid administrator is used to perform daily maintenance and alarm processing on the equipment in the grid, and grid captains are designated for multiple grid administrators, and interactive devices are assigned to the grid captains, and the assigned interactive devices are used to exchange information with the central service platform; the grid administrators and grid captains are both trained and employed; the method further includes:

[0160] The central service platform conducts training effect tests on the corresponding grid administrators at corresponding intervals. The training effect tests specifically include:

[0161] The central service platform selects a first alarm item from the preset alarm table items, and reversely generates test data according to the first alarm item; the test data is the sensor data that causes the first alarm item to be abnormal.

[0162] The test data is sent to a centralized device so that the centralized device sends the test data to a corresponding intelligent sensor device. The intelligent sensor device uses the test data to replace the actually collected sensor data and displays it on a dial, so that the grid administrator can observe abnormal data through the dial of the intelligent sensor device and report the abnormal data to the grid team leader, who then reports the abnormal data to the central service platform through an interactive device.

[0163] The central service platform calculates the matching degree between the abnormal data received from the interactive device and the test data.

[0164] According to the matching degree, the training score of the grid administrator of the corresponding grid is calculated, and the training score is displayed on the central service platform, so as to judge whether the grid administrator is suitable for the position or whether further education and training is needed through the training score.

[0165] The calculation methods of the matching degree and the training score are implemented based on the same concept as those in Example 1. The methods described in Example 1 are applicable to this embodiment and will not be described in detail here.

[0166] Embodiment 3:

[0167] On the basis of Example 1, this embodiment also provides a third method for identifying hidden dangers in metal smelting enterprises based on artificial intelligence, setting a central service platform for the entire blast furnace area, dividing the blast furnace workshop into multiple grids, and setting intelligent sensing equipment at each monitoring position of each grid, such as Fig. 20 As shown, the intelligent sensor device is used to collect sensor data of the monitoring location and report the sensor data to the central service platform; the method described in this embodiment is suitable for scenarios where the enterprise is small in scale or the distance between the intelligent sensor device and the central service platform is relatively close, such as Fig.21 As shown, the method includes:

[0168] In step 801, the central service platform receives the sensor data of each intelligent sensor device, and establishes a digital twin model of each grid according to the sensor data of each grid; wherein the digital twin model is presented to the user in the form of image modeling combined with data presentation, such as Figure 3 shown.

[0169] In step 802, the central service platform also compares the sensor data with the corresponding monitoring threshold to determine whether the sensor data is abnormal; and performs alarm analysis on the abnormal sensor data in the corresponding grid to obtain the alarm of each grid and the corresponding alarm risk level.

[0170] In step 803, the corresponding areas of the grid in the digital twin model are color-coded according to the alarm risk level of each alarm in the grid, so that the user can intuitively confirm the risk situation of the grid through the digital twin model; among them, the alarm risk levels include no risk, low level, medium level and high level.

[0171] In step 804, when the highest alarm risk level in the corresponding grid is high, the corresponding area of ​​the grid in the digital twin model is marked as red; when the highest alarm risk level in the corresponding grid is medium, the corresponding area of ​​the grid in the digital twin model is marked as orange; when the highest alarm risk level in the corresponding grid is low, the corresponding area of ​​the grid in the digital twin model is marked as yellow; when the highest alarm risk level in the corresponding grid is no risk, the corresponding area of ​​the grid in the digital twin model is marked as blue.

[0172] This embodiment sets multiple alarm risk levels and establishes a digital twin model. The alarm risk level of each grid is identified by color in the digital twin model, so that the user can intuitively observe the risk status of each grid.

[0173] In an optional implementation, the central service platform also compares the sensor data with the corresponding monitoring threshold to determine whether the sensor data is abnormal; performs alarm analysis on the abnormal sensor data in the corresponding grid to obtain the alarm of each grid and the corresponding alarm risk level, such as Fig. 22 As shown, specifically including:

[0174] In step 901, the monitoring threshold includes a high-level monitoring threshold, an intermediate monitoring threshold and a low-level monitoring threshold. The sensor data is compared with the high-level monitoring threshold, the intermediate monitoring threshold and the low-level monitoring threshold in turn. When the comparison shows that the sensor data exceeds the range of the corresponding monitoring threshold, it is determined that the sensor data is abnormal, and the risk level of the exceeded monitoring threshold is used as the abnormality level of the sensor data.

[0175] In step 902, all abnormal sensor data are used to generate an abnormal sensor array; wherein each value in the abnormal sensor array corresponds to a sensor data, and when there is no abnormality in the sensor data, the corresponding value in the abnormal sensor array is set to 0, when the abnormal level of the sensor data is low, the corresponding value in the abnormal sensor array is set to 1, when the abnormal level of the sensor data is medium, the corresponding value in the abnormal sensor array is set to 2, and when the abnormal level of the sensor data is high, the corresponding value in the abnormal sensor array is set to 3.

[0176] In step 903, the abnormal sensor array is used to subtract each alarm code array in the preset alarm table item to obtain a result array. If all arrays in the result array are greater than or equal to 0, it is analyzed that the alarm corresponding to the alarm code array appears, and the alarm risk level of the alarm is the alarm risk level corresponding to the alarm code array; wherein, the abnormal sensor array minus the alarm code array is specifically: each value in the abnormal sensor array is subtracted from the corresponding value in the alarm code array, that is, the nth value in the abnormal sensor array is subtracted from the nth value in the alarm code array to obtain the nth value in the result array.

[0177] The preset alarm table entry includes a first preset alarm table entry and a second preset alarm table entry. The first preset alarm table entry is as follows: Fig.10 As shown, including sensor sequence number, sensor data, monitoring indicators, low-level monitoring thresholds (including low-level monitoring threshold lower limit and low-level monitoring threshold upper limit), intermediate monitoring thresholds (including intermediate monitoring threshold lower limit and intermediate monitoring threshold upper limit) and advanced monitoring thresholds (including advanced monitoring threshold lower limit and advanced monitoring threshold upper limit). When the corresponding monitoring indicator only monitors the upper limit of the threshold or the upper limit of the threshold, the party that does not need to be monitored is set to the default value (such as set to -1).

[0178] The second preset alarm table entry is as follows: Fig.23 As shown, it includes a grid number, an alarm number, an alarm risk level and an alarm code array, wherein each alarm item corresponds to 3 alarm risk levels and 3 alarm code arrays, each alarm code array is obtained in advance by technical personnel in this field based on empirical analysis, specifically: in the order of the sensor number, the numerical value corresponding to the abnormal level (i.e., 0, 1, 2 and 3) of the sensor data that triggers the corresponding alarm is used as the corresponding numerical value in the alarm code array.

[0179] In combination with the method described in Example 1, this embodiment further provides a preferred implementation method, that is, a corresponding grid administrator is assigned to each grid, the grid administrator is used to perform daily maintenance and alarm processing on the equipment in the grid, and grid captains are designated for multiple grid administrators, and interactive devices are assigned to the grid captains, and the assigned interactive devices are used to exchange information with the central service platform; the grid administrators and grid captains are both trained and employed; the method further includes:

[0180] The central service platform conducts training effectiveness tests on the corresponding grid administrators at corresponding intervals, such as Fig.24 As shown, the training effect test specifically includes:

[0181] In step 1001, the central service platform selects a first alarm item from a preset alarm table item, and reversely generates test data based on the first alarm item; the test data is the sensor data that causes the first alarm item to be abnormal.

[0182] In step 1002, the test data is sent to the intelligent sensor device, and the intelligent sensor device uses the test data to replace the actually collected sensor data and displays it on the dial, so that the grid administrator can observe abnormal data through the dial of the intelligent sensor device and report the abnormal data to the grid team leader, who then reports the abnormal data to the central service platform through the interactive device.

[0183] In step 1003, the central service platform calculates the matching degree between the abnormal data received from the interactive device and the test data.

[0184] In step 1004, the training score of the grid administrator of the corresponding grid is calculated based on the matching degree, and the training score is displayed in the digital twin model, so as to judge whether the grid administrator is suitable for the position or whether further education and training is needed through the training score.

[0185] The calculation methods of the matching degree and the training score are implemented based on the same concept as those in Example 1. The methods described in Example 1 are applicable to this embodiment and will not be described in detail here.

[0186] Embodiment 4:

[0187] On the basis of Example 1, this embodiment further provides a fourth method for identifying hidden dangers in metal smelting enterprises based on artificial intelligence, which includes setting a central service platform for the enterprise area, dividing the enterprise area into multiple grids, setting intelligent sensing equipment at each monitoring position of each grid, and setting centralized equipment for one or more grids, wherein the centralized equipment is used to collect sensing data of the intelligent sensing equipment and report the collected sensing data to the central service platform; setting cameras in the material handling area of ​​each grid to collect real-time images of the material handling area; Fig.25 As shown, the method includes:

[0188] In step 1101, the central service platform receives the sensor data of each intelligent sensor device, and establishes a digital twin model of each grid based on the sensor data of each grid; wherein the digital twin model is presented to the user in the form of image modeling combined with data presentation.

[0189] In step 1102, the central service platform also compares the sensor data with the corresponding monitoring threshold to determine whether there is any abnormality in the sensor data; and first performs alarm analysis on the abnormal sensor data in the corresponding grid for each grid, obtains the alarm of each grid and the corresponding alarm risk level, and then combines the alarm of the upstream grid to filter out the alarm of the downstream grid to obtain the final alarm analysis result.

[0190] In step 1103, the corresponding area of ​​the grid in the digital twin model is color-coded according to the alarm risk level of each alarm in the grid, so that the user can intuitively confirm the risk status of the grid through the digital twin model.

[0191] In step 1104, the central service platform also receives real-time images of the material handling areas of each grid, and uses a pre-trained image recognition model to perform image recognition on the real-time images to detect whether there are material handling vehicles entering the material handling areas of each grid.

[0192] In step 1105, when a material handling vehicle enters, the alarm risk level of each alarm in the grid is increased by one level; wherein the alarm risk levels include no risk, low level, medium level and high level; when the highest alarm risk level in the corresponding grid is high level, the corresponding area of ​​the grid in the digital twin model is marked as red, when the highest alarm risk level in the corresponding grid is medium level, the corresponding area of ​​the grid in the digital twin model is marked as orange, when the highest alarm risk level in the corresponding grid is low level, the corresponding area of ​​the grid in the digital twin model is marked as yellow, and when the highest alarm risk level in the corresponding grid is no risk, the corresponding area of ​​the grid in the digital twin model is marked as blue.

[0193] The method further includes: assigning a corresponding grid administrator to each grid, the grid administrator is used to perform daily maintenance and alarm processing on the equipment in the grid, and also designating grid captains for multiple grid administrators, and assigning interactive devices to the grid captains, the assigned interactive devices are used to interact with the central service platform; the grid administrators and grid captains are both trained and employed; UWB tags are assigned to each grid administrator in advance; the method further includes:

[0194] The central service platform conducts training effect tests on the corresponding grid administrators at corresponding intervals. The training effect tests specifically include:

[0195] The central service platform selects a first alarm item from the preset alarm table items, and reversely generates test data according to the first alarm item; the test data is abnormal sensor data that triggers the first alarm item.

[0196] The test data is sent to a centralized device so that the centralized device sends the test data to a corresponding intelligent sensor device. The intelligent sensor device uses the test data to replace the actually collected sensor data and displays it on a dial, so that the grid administrator can observe abnormal data through the dial of the intelligent sensor device and report the abnormal data to the grid team leader, who then reports the abnormal data to the central service platform through an interactive device.

[0197] The central service platform calculates a first matching degree between the abnormal data received from the interactive device and the test data.

[0198] The central service platform also determines the movement trajectory of each grid administrator after arriving at the target intelligent sensor device based on the UWB tag; matches the movement trajectory with the preset map to determine the monitoring locations passed by the movement trajectory; calculates the second matching degree between each monitoring location and the preset processing behavior of the first alarm item; wherein the target intelligent sensor device is an intelligent sensor device that displays test data.

[0199] Based on the first matching degree and the second matching degree, the training score of the grid administrator of the corresponding grid is calculated, and the training score is displayed in the digital twin model, so as to judge whether the grid administrator is suitable for the position or whether further education and training is needed through the training score.

[0200] In an optional implementation, the central service platform also compares the sensor data with the corresponding monitoring threshold to determine whether the sensor data is abnormal; based on the sensor data with abnormalities, an alarm analysis is performed on each grid to obtain the alarm of each grid and the corresponding alarm risk level, and then combined with the alarm of the upstream grid, the alarm of the downstream grid is filtered out to obtain the final alarm analysis result, which specifically includes:

[0201] The monitoring threshold includes a high-level monitoring threshold, an intermediate monitoring threshold and a low-level monitoring threshold. The sensor data is compared with the high-level monitoring threshold, the intermediate monitoring threshold and the low-level monitoring threshold in turn until the comparison shows that the sensor data exceeds the range of the corresponding monitoring threshold. In this case, it is determined that the sensor data is abnormal, and the risk level of the exceeded monitoring threshold is used as the abnormal level of the sensor data.

[0202] Use all abnormal sensor data to generate an abnormal sensor array; wherein each value in the abnormal sensor array corresponds to a sensor data, when there is no abnormality in the sensor data, the corresponding value in the abnormal sensor array is set to 0, when the abnormal level of the sensor data is low, the corresponding value in the abnormal sensor array is set to 1, when the abnormal level of the sensor data is medium, the corresponding value in the abnormal sensor array is set to 2, and when the abnormal level of the sensor data is high, the corresponding value in the abnormal sensor array is set to 3.

[0203] Use the abnormal sensor array to subtract each alarm code array in the preset alarm table item to obtain a result array. If all arrays in the result array are greater than or equal to 0, the analysis results in the alarm corresponding to the alarm code array, and the alarm risk level of the alarm is the alarm risk level corresponding to the alarm code array; wherein, the abnormal sensor array minus the alarm code array is specifically: each numerical value in the abnormal sensor array minus the corresponding numerical value in the alarm code array.

[0204] Use all alarms to generate abnormal alarm codes, and perform AND operations on the abnormal alarm codes and the associated alarm abnormal codes corresponding to each alarm in a preset order. If the operation result is greater than 0, the alarm corresponding to the associated alarm abnormal code is filtered out.

[0205] The preset alarm table entry includes a first preset alarm table entry and a second preset alarm table entry. The first preset alarm table entry is as follows: Fig.10 As shown, including sensor sequence number, sensor data, monitoring indicators, low-level monitoring thresholds (including low-level monitoring threshold lower limit and low-level monitoring threshold upper limit), intermediate monitoring thresholds (including intermediate monitoring threshold lower limit and intermediate monitoring threshold upper limit) and advanced monitoring thresholds (including advanced monitoring threshold lower limit and advanced monitoring threshold upper limit). When the corresponding monitoring indicator only monitors the upper limit of the threshold or the upper limit of the threshold, the party that does not need to be monitored is set to the default value (such as set to -1).

[0206] The second preset alarm table entry is as follows: Fig.26 As shown, it includes a grid number, an alarm number, an associated alarm exception code, an alarm risk level and an alarm code array, wherein each alarm item corresponds to an associated alarm exception code, and corresponds to 3 alarm risk levels and 3 alarm code arrays, and each associated alarm exception code and alarm code array are obtained in advance by technical personnel in this field based on empirical analysis, specifically: in the order of sensor numbers, the numerical values ​​corresponding to the abnormal levels (i.e., 0, 1, 2 and 3) of the sensor data that triggers the corresponding alarm are used as the corresponding numerical values ​​in the alarm code array; the bits corresponding to other alarms that trigger the corresponding alarm are set to 1, and the other bits are set to 0, to obtain the associated alarm exception code, and each alarm item corresponds to a bit in the associated alarm exception code.

[0207] The central service platform calculates a first matching degree between the abnormal data received from the interactive device and the test data, specifically including:

[0208] The first matching degree is calculated using the first formula; the first formula is: Among them, I t_nis the first matching degree obtained by the nth training effect test of the grid administrator, Num(error_data) is the number of abnormal data received by the central service platform from the interactive device, and Num(test_data) is the number of generated test data.

[0209] Calculating a second matching degree between each monitoring location and a preset processing behavior of the first alarm item specifically includes:

[0210] The second matching degree is calculated using the second formula; the second formula is: Among them, I p_n is the second matching degree obtained from the nth training effect test of the grid administrator, Num(test_pos) is the number of first monitoring positions that the grid administrator needs to manually monitor as specified in the preset processing behavior, and Num(passed_test_pos) is the number of first monitoring positions passed by the grid administrator's movement trajectory.

[0211] The calculating the training score of the grid administrator of the corresponding grid according to the first matching degree and the second matching degree specifically includes:

[0212] The third formula is used to calculate the test results of the grid administrator's training effectiveness test.

[0213] The fourth formula is used to calculate the training score of the grid administrator after the training effectiveness test.

[0214] The third formula is I' t_n =k1I t_n +k2I p_n +k3I' t_n-1 , the fourth formula is I g_n =k4I' t_n +k5I r ; k1, k2, k3, k4 and k5 are all corresponding weight values, where k1+k2+k3=1, k4+k5=1, I′ t_n is the test result of the nth training effect test, I r is the evaluation score of the grid administrator after training, I g_n is the training score of the grid administrator after the nth training effect test, I t_n is the first matching degree obtained from the nth training effect test of the grid administrator, I p_n is the second matching degree obtained from the nth training effect test of the grid administrator, I′ t_n-1 This is the test result of the n-1th training effectiveness test for grid administrators.

[0215] Among them, the method described in Example 1 is applicable to this embodiment and will not be described in detail here.

[0216] Embodiment 5:

[0217] like Fig. 27 , is a schematic diagram of the architecture of a metal smelting enterprise operation hidden danger identification device based on artificial intelligence according to an embodiment of the present invention. The metal smelting enterprise operation hidden danger identification device based on artificial intelligence according to this embodiment includes one or more processors 21 and a memory 22. Fig. 27 A processor 21 is taken as an example.

[0218] The processor 21 and the memory 22 may be connected via a bus or other means. Fig. 27 The example of connecting through bus is taken in the following.

[0219] The memory 22 is a non-volatile computer-readable storage medium that can be used to store non-volatile software programs and non-volatile computer executable programs, such as the artificial intelligence-based method for identifying hidden dangers in metal smelting operations described in any one of Embodiments 1 to 4. The processor 21 executes the artificial intelligence-based method for identifying hidden dangers in metal smelting operations by running the non-volatile software programs and instructions stored in the memory 22.

[0220] The memory 22 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage devices. In some embodiments, the memory 22 may optionally include a memory remotely arranged relative to the processor 21, and these remote memories may be connected to the processor 21 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0221] The program instructions / modules are stored in the memory 22, and when executed by the one or more processors 21, the method for identifying operational hazards in metal smelting enterprises based on artificial intelligence described in any one of the above-mentioned embodiments 1 to 4 is executed.

[0222] It is worth noting that the information interaction, execution process, etc. between the modules and units within the above-mentioned devices and systems are based on the same concept as the processing method embodiment of the present invention. The specific contents can be found in the description of the method embodiment of the present invention and will not be repeated here.

[0223] A person skilled in the art may understand that all or part of the steps in the various methods of the embodiments may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, and the storage medium may include: a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk, etc.

[0224] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for identifying hidden dangers in metal smelting operations based on artificial intelligence, characterized in that: A central service platform is set up for the enterprise area, the enterprise area is divided into multiple grids, intelligent sensor equipment is set up at each monitoring position of each grid, and a centralized device is set up for one or more grids, the centralized device is used to collect sensor data of the intelligent sensor equipment, and the collected sensor data is reported to the central service platform; A camera is set in each grid for real-time imaging in the grid, and the method includes: The central service platform establishes a digital twin model of each grid based on the sensor data of each grid; wherein the digital twin model is presented to the user in the form of image modeling combined with data presentation; The central service platform also compares the sensor data with the corresponding monitoring threshold to determine whether the sensor data is abnormal; the sensor data with abnormalities is input into the pre-trained alarm analysis model to perform alarm analysis on each grid to obtain the alarm of each grid and the corresponding alarm risk level; According to the alarm risk level of each alarm in the grid, the corresponding area of ​​the grid in the digital twin model is color-coded, so that users can intuitively confirm the risk status of the grid through the digital twin model; The central service platform also receives real-time images of each grid, and uses a pre-trained image recognition model to perform image recognition on the real-time images to detect whether there are hidden danger behaviors in each grid; when there are hidden danger behaviors in the grid, the alarm risk level of each alarm in the grid is increased by one level; The risk behaviors include: the entry of material handling vehicles into the material handling area of ​​the grid, irregular behaviors of the grid administrator, and the identification of risk devices or risk signs in the grid.

2. The method for identifying hidden dangers in metal smelting operations based on artificial intelligence according to claim 1 is characterized in that: A corresponding grid administrator is assigned to each grid. The grid administrator is responsible for daily maintenance and alarm processing of the equipment in the grid. Grid captains are also assigned to multiple grid administrators. Interactive devices are assigned to grid captains. The interactive devices are used to exchange information with the central service platform. Grid administrators and grid captains are trained before taking up their posts. The method also includes: The central service platform conducts training effect tests on the corresponding grid administrators at corresponding intervals. The training effect tests specifically include: The central service platform selects a first alarm item from the preset alarm table items, and reversely generates test data according to the first alarm item; the test data is sensor data that triggers the abnormality of the first alarm item; The test data is sent to the centralized device so that the centralized device sends the test data to the corresponding intelligent sensor device, and the intelligent sensor device uses the test data to replace the actually collected sensor data and displays it on the dial, so that the grid administrator can observe the abnormal data through the dial of the intelligent sensor device and report the abnormal data to the grid leader, who reports the abnormal data to the central service platform through the interactive device; The central service platform calculates the matching degree between the abnormal data received from the interactive device and the test data; According to the matching degree, the training score of the grid administrator of the corresponding grid is calculated, and the training score is displayed in the digital twin model, so as to judge whether the grid administrator is suitable for the position or whether further education and training is needed through the training score.

3. The method for identifying hidden dangers in metal smelting operations based on artificial intelligence according to claim 2 is characterized in that: The first alarm item is an alarm item whose alarm processing priority is lower than the preset priority and whose alarm level is higher than the preset level.

4. The method for identifying hidden dangers in metal smelting operations based on artificial intelligence according to claim 2 is characterized in that: The central service platform calculates the matching degree between the abnormal data received from the interactive device and the test data, specifically including: The matching degree is calculated using the first formula; the first formula is Among them, I t_n is the matching degree obtained by the nth training effect test of the grid administrator, Num(error_data) is the number of abnormal data received by the central service platform from the interactive device, and Num(test_data) is the number of generated test data.

5. The method for identifying hidden dangers in metal smelting operations based on artificial intelligence according to claim 2 is characterized in that: The calculating the training score of the grid administrator of the corresponding grid according to the matching degree specifically includes: The test results of the grid administrator's training effectiveness test are calculated using the second formula; Use the third formula to calculate the training score of the grid administrator after the training effect test; The second formula is I' t_n =k1I t_n +k2I' t_n-1 , the third formula is I g_n =k3I' t_n +k4I r ; k1, k2, k3 and k4 are corresponding weight values, where k1+k2=1, k3+k4=1, I′ t_n is the test result of the nth training effect test, I r is the evaluation score of the grid administrator after training, I g_n is the training score of the grid administrator after the nth training effect test, I t_n is the matching degree obtained from the nth training effect test of the grid administrator, I' t_n-1 This is the test result of the n-1th training effectiveness test for grid administrators.

6. The method for identifying hidden dangers in metal smelting operations based on artificial intelligence according to claim 2 is characterized in that: During the training effect test, the central service platform still receives sensor data. When an alarm occurs in the grid where the training effect test is conducted, the training effect test is terminated to display the sensor data collected by the intelligent sensor device on the dial. The grid captain is informed of the time of the training effect test through the interactive device, so that the grid captain can notify the grid administrator to terminate the training effect test and carry out alarm processing in time.

7. The method for identifying hidden dangers in metal smelting operations based on artificial intelligence according to claim 2 is characterized in that: The central service platform conducts training effect tests on the corresponding grid administrators at corresponding intervals, specifically including: At intervals of a first preset period, a random number is generated for the grid. If the generated random number is a prime number, statistics are then performed to determine whether the device configuration information of the grid has changed within a second preset period before the current moment, or whether an alarm has been generated by the grid within the second preset period before the current moment; If the device configuration information of the grid has not changed within the second preset period before the current moment, and the grid has not generated an alarm, a training effect test is performed on the grid administrator of the grid; The generating of random numbers specifically includes: A seed number is obtained by adding the grid code to the current time, a random intermediate number is generated by using the seed number, and a modulus of M is obtained by using the random intermediate number to obtain the random number; in, T0 is a preset value, T is a third preset period, times is the number of alarms in the grid in the third preset period, and when no alarm occurs in the grid in the third preset period before the current moment, M=T0×T.

8. The method for identifying hidden dangers in metal smelting operations based on artificial intelligence according to claim 1 is characterized in that: The central service platform also compares the sensor data with the corresponding monitoring threshold to determine whether there is an abnormality in the sensor data; The abnormal sensor data is input into the pre-trained alarm analysis model to perform alarm analysis on each grid to obtain the alarm of each grid and the corresponding alarm risk level; specifically including: The monitoring thresholds include a high-level monitoring threshold, a medium-level monitoring threshold and a low-level monitoring threshold. The sensor data is compared with the high-level monitoring threshold, the medium-level monitoring threshold and the low-level monitoring threshold in sequence until the sensor data exceeds the range of the corresponding monitoring threshold. It is determined that the sensor data is abnormal, and the risk level of the exceeded monitoring threshold is used as the abnormal level of the sensor data; Use all abnormal sensor data to generate an abnormal sensor vector; wherein each value in the abnormal sensor vector corresponds to a sensor data, and when there is no abnormality in the sensor data, the corresponding value in the abnormal sensor vector is set to 0, when the abnormal level of the sensor data is low, the corresponding value in the abnormal sensor vector is set to 1, when the abnormal level of the sensor data is medium, the corresponding value in the abnormal sensor vector is set to 2, and when the abnormal level of the sensor data is high, the corresponding value in the abnormal sensor vector is set to 3; The abnormal sensing vector is input into a pre-trained alarm analysis model to obtain an alarm vector; wherein the alarm vector includes the triggering probability of alarms of each alarm risk level, and the alarm with a triggering probability higher than a preset probability is used as the alarm of each grid.

9. A non-volatile computer storage medium, characterized in that: The computer storage medium stores computer executable instructions, which are executed by one or more processors to complete the method for identifying operational hazards in metal smelting enterprises based on artificial intelligence as described in any one of claims 1-8.

10. A device for identifying hidden dangers in metal smelting operations based on artificial intelligence, characterized in that: The device comprises: At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the processor to execute the method for identifying operational hazards in metal smelting enterprises based on artificial intelligence as described in any one of claims 1-8.

Images