Fuel plant side intelligent supervision system based on deep learning
A deep learning-based system for fuel plants addresses low intelligence in fuel management by predicting and monitoring worker violations, enhancing fuel safety and stability in thermal power plants.
Patent Information
- Application Number
- CN202411135115.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-16
- Publication Date
- 2024-12-27
AI Technical Summary
Thermal power plants face challenges in fuel management due to low intelligence levels and inadequate supervision, leading to potential energy waste and safety hazards.
A deep learning-based intelligent supervision system for fuel plants that includes a business process module, deep learning module, violation evaluation module, and monitoring and alarm module to predict and monitor worker violations, set violation thresholds, and issue alarms.
Enhances fuel quality and safety by timely detection of violations, preventing accidents and ensuring stable power plant operations.
Smart Images
Figure CN119204650A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of fuel plant management technology, and more specifically, to a fuel plant-side intelligent supervision system based on deep learning. Background Art
[0002] Thermal power plants are one of the main sources of electricity supply in my country. The normal operation of fuel is directly related to the stability and reliability of power supply. Efficient fuel operation management can reduce the operating costs of thermal power plants and improve economic benefits. The fuel entry process includes four steps: acceptance, sampling, testing and unloading. The strict implementation of these steps can ensure the quality and safety of the fuel, thereby providing guarantee for the stable operation of thermal power plants.
[0003] However, in terms of fuel management, power plants pay too much attention to the control of fuel procurement and neglect the management work within the fuel plant. There are problems such as backward benchmarking of management indicators within the fuel plant, low level of intelligence in fuel sampling and testing processes. Some staff members lack professional capabilities and are unfamiliar with fuel management and combustion, resulting in waste of energy resources and, in serious cases, a series of major disasters. Summary of the invention
[0004] The present invention provides a deep learning-based intelligent supervision system for fuel plants, which is used to solve the problem of low intelligence level of management at the fuel plant side and inability to timely supervise violations on site in the prior art. It includes: The business process module is used to obtain the current fuel entry task information, determine the fuel business process based on the fuel entry task information, and set the violation coefficient tolerance threshold of each process based on the fuel business process; A deep learning module is used to collect business process image data on the fuel plant side according to the fuel business process, and determine and predict violations based on the deep learning model according to the business process image data on the fuel plant side; A violation evaluation module is used to determine a violation evaluation coefficient and a behavior forgiveness coefficient according to predicted violation behaviors, and to modify the violation evaluation coefficient according to the behavior forgiveness coefficient; The monitoring and alarm module is used to monitor the workers' violation behaviors according to the revised violation evaluation coefficient and the violation coefficient tolerance threshold.
[0005] Furthermore, the business process module is also used to: Obtain the fuel entry information, and extract the type data, specification data and quantity data of the fuel entering the factory based on the fuel entry information; Formulate fuel entry tasks based on the type, specification and quantity data of the fuel entering the factory, and generate corresponding fuel business processes based on the fuel entry tasks.
[0006] Furthermore, the business process module is also used to: According to the fuel business process, determine the number of subtasks in each process and the average completion time of the process, calculate the ratio of the number of subtasks to the average completion time, and obtain the allowable violation coefficient; Determine whether the difference between the current allowable violation coefficient and the preset allowable violation coefficient is greater than a first preset threshold value, and if the difference between the current allowable violation coefficient and the preset allowable violation coefficient is greater than the first preset threshold value, set the preset first violation coefficient allowable threshold value as the violation coefficient allowable threshold value; If the difference between the current allowable violation coefficient and the preset allowable violation coefficient is less than or equal to the first preset threshold, determining whether the difference between the current allowable violation coefficient and the preset allowable violation coefficient is greater than a second preset threshold; If the difference between the current permissible violation coefficient and the preset permissible violation coefficient is greater than the second preset threshold, setting the preset second permissible violation coefficient threshold as the permissible violation coefficient threshold; If the difference between the current allowable violation coefficient and the preset allowable violation coefficient is less than or equal to the second preset threshold, the preset third violation coefficient allowable threshold is set as the violation coefficient allowable threshold.
[0007] Furthermore, the deep learning module is also used to: Acquire historical worker motion image data on the fuel plant side, and establish a worker motion data set based on the historical worker motion image data on the fuel plant side; Establish a deep learning model and train it based on the worker action dataset to obtain a worker action prediction model; Determine the working area on the fuel plant side corresponding to the business process according to the fuel business process, and collect image data of the working area on the fuel plant side; Predicting worker actions of image data of the working area at the fuel plant side according to the worker action prediction model to obtain predicted worker actions, and extracting several predicted action key point data in the predicted worker actions; Obtain preset illegal action key point data, and calculate the correlation coefficient between the predicted action key point data and the illegal action key point data; Action image data whose correlation coefficient between predicted action key point data and illegal action key point data is less than a third preset threshold is screened out, and the action image data whose correlation coefficient is less than the third preset threshold is set as predicted illegal behavior.
[0008] Furthermore, the deep learning module is also used to: Determine the motion trajectory of the predicted motion key point data according to the predicted motion key point data, calculate the trajectory correction coefficient according to the motion trajectory of the predicted motion key point data, and correct the motion trajectory according to the trajectory correction coefficient; Calculate the area value of the corrected action trajectory and the illegal action trajectory corresponding to the key point data of the illegal action to obtain the difference area value; The difference area values of each key point are added together to obtain the difference area cumulative added value, and the difference area cumulative added values of the predicted worker's action are added together to obtain the correlation coefficient between the predicted action key point data and the illegal action key point data.
[0009] Furthermore, the deep learning module is also used to: Obtaining a motion trajectory of the predicted motion key point data within a preset period of time, and calculating a radian value of the motion trajectory according to the motion trajectory of the predicted motion key point data within the preset period of time; Determine the illegal action trajectory according to the illegal action key point data within a preset time period, calculate the arc value of the illegal action trajectory, and calculate the ratio of the arc value of the action trajectory to the arc value of the corresponding illegal action trajectory; The ratio of the arc value of the action trajectory to the arc value of the corresponding illegal action trajectory is normalized to obtain the trajectory correction coefficient.
[0010] Furthermore, the violation evaluation module is also used to: Acquire image data of predicted violation behaviors, determine the type of violation behaviors of the workers according to the action images of the violation behaviors of the workers, and set an initial violation index according to the type of violation behaviors; The correlation coefficient between the key point data of the action predicted by the violation and the key point data of the violation action is obtained, the initial violation index is corrected according to the correlation coefficient between the key point data of the action predicted by the violation and the key point data of the violation action, and the corrected violation index is set as the violation evaluation coefficient.
[0011] Furthermore, the violation evaluation module is also used to: Obtain the number of historical false alarms of the predicted violation, and determine a first violation forgiveness value according to the number of historical false alarms of the predicted violation; Obtain the number of historical violations of the worker corresponding to the predicted violation, and determine a second violation forgiveness value according to the number of historical violations of the worker corresponding to the predicted violation; The behavior forgiveness coefficient is determined according to the first violation forgiveness value and the second violation forgiveness value, and the behavior forgiveness coefficient is multiplied by the violation evaluation coefficient to obtain the corrected violation evaluation coefficient.
[0012] Furthermore, the violation evaluation module is also used to: The behavior forgiveness coefficient is calculated according to the behavior forgiveness coefficient calculation formula. The specific behavior forgiveness coefficient calculation formula is: , in, F is the behavior forgiveness coefficient, , are the first weight value and the second weight value respectively, , They are the first violation forgiveness value and the second violation forgiveness value respectively.
[0013] Furthermore, the monitoring module is used to: Determine the fuel business process corresponding to the predicted violation based on the predicted violation, and judge whether the violation evaluation coefficient after correction of the predicted violation exceeds the violation coefficient allowable threshold of the corresponding process; If the violation evaluation coefficient after correction of the predicted violation exceeds the violation coefficient allowable threshold of the corresponding process, an alarm reminder will be issued; If the corrected violation evaluation coefficient of the predicted violation does not exceed the violation coefficient allowable threshold of the corresponding process, the predicted violation will be recorded.
[0014] The beneficial effects of the present invention are: By applying the above technical solutions, the present invention collects image data of various business processes on the fuel plant side and accurately predicts the actions of staff based on the deep learning model, so as to timely discover the violations of staff, effectively prevent the occurrence of power plant accidents, ensure the quality and safety of fuel, and thus provide guarantee for the stable operation of thermal power plants. BRIEF DESCRIPTION OF THE DRAWINGS
[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.
[0016] Figure 1 A deep learning-based intelligent supervision system at the fuel plant side proposed in an embodiment of the present invention is shown. DETAILED DESCRIPTION
[0017] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0018] In the description of the present application, it should be understood that the terms "center", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on the present application.
[0019] The terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of this application, unless otherwise specified, "plurality" means two or more.
[0020] In the description of this application, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in this application can be understood according to specific circumstances.
[0021] The present application embodiment provides a fuel plant-side intelligent monitoring system based on deep learning, such as Figure 1 As shown, it includes: a business process module, which is used to obtain the current fuel entry task information, determine the fuel business process according to the fuel entry task information, and set the violation coefficient tolerance threshold of each process according to the fuel business process; a deep learning module, which is used to collect the business process image data on the fuel plant side according to the fuel business process, and determine the predicted violation behavior based on the deep learning model according to the business process image data on the fuel plant side; a violation evaluation module, which is used to determine the violation evaluation coefficient and the behavior forgiveness coefficient according to the predicted violation behavior, and correct the violation evaluation coefficient according to the behavior forgiveness coefficient; a monitoring and alarm module, which is used to supervise the workers' violation behavior according to the corrected violation evaluation coefficient and the violation coefficient tolerance threshold.
[0022] In this embodiment, when fuel is detected entering the factory, the fuel business process corresponding to the corresponding fuel is extracted through the business process module and the violation coefficient tolerance threshold of each process is set. The image data corresponding to each process is collected based on the deep learning module, the actions of workers in the business process area are monitored in real time and violations are predicted. The violation evaluation coefficient is calculated for the workers' violations through the violation evaluation module, and the workers' violations are supervised through the violation evaluation coefficients of each process to ensure the quality and safety of the fuel.
[0023] In some embodiments of the present application, the business process module is also used to: obtain fuel entry information, extract the type data, specification data and quantity data of the fuel entering the factory based on the fuel entry information; formulate fuel entry tasks based on the type data, specification data and quantity data of the fuel entering the factory, and generate corresponding fuel business processes based on the fuel entry tasks.
[0024] In this embodiment, after the fuel enters the factory, the type data, specification data and quantity data of the fuel are detected in real time and corresponding factory entry tasks are generated. The factory entry tasks include specific fuel acceptance, sampling, sample preparation and testing tasks. The fuel business process is the specific matters for each factory entry task.
[0025] In some embodiments of the present application, the business process module is also used to: determine the number of subtasks of each process and the average completion time of the process according to the fuel business process, calculate the ratio of the number of subtasks to the average completion time, and obtain the allowable violation coefficient; determine whether the difference between the current allowable violation coefficient and the preset allowable violation coefficient is greater than a first preset threshold value, if the difference between the current allowable violation coefficient and the preset allowable violation coefficient is greater than the first preset threshold value, then set the preset first violation coefficient allowable threshold value as the violation coefficient allowable threshold value; if the difference between the current allowable violation coefficient and the preset allowable violation coefficient is less than or equal to the first preset threshold value, then determine whether the difference between the current allowable violation coefficient and the preset allowable violation coefficient is greater than the second preset threshold value; if the difference between the current allowable violation coefficient and the preset allowable violation coefficient is greater than the second preset threshold value, then set the preset second violation coefficient allowable threshold value as the violation coefficient allowable threshold value; if the difference between the current allowable violation coefficient and the preset allowable violation coefficient is less than or equal to the second preset threshold value, then set the preset third violation coefficient allowable threshold value as the violation coefficient allowable threshold value.
[0026] In this embodiment, the subtasks of each process are specific operations required by workers to complete the fuel business process, and the average completion time of the process is the average value of the time required to complete the fuel business process in historical data.
[0027] In this embodiment, the first violation coefficient allowable threshold, the second violation coefficient allowable threshold, and the third violation coefficient allowable threshold decrease in sequence. The larger the ratio of the number of subtasks to the average completion time, the higher the set violation coefficient allowable threshold.
[0028] In some embodiments of the present application, the deep learning module is also used to: obtain historical worker motion image data on the fuel plant side, and establish a worker motion data set based on the historical worker motion image data on the fuel plant side; establish a deep learning model, and train the deep learning model based on the worker motion data set to obtain a worker motion prediction model; determine the working area on the fuel plant side corresponding to the business process according to the fuel business process, and collect image data of the working area on the fuel plant side; predict the worker movements of the image data of the working area on the fuel plant side according to the worker motion prediction model to obtain predicted worker movements, and extract several predicted action key point data from the predicted worker movements; obtain preset violation action key point data, and calculate the correlation coefficient between the predicted action key point data and the violation action key point data; screen out action image data whose correlation coefficient between the predicted action key point data and the violation action key point data is less than a third preset threshold, and set the action image data whose correlation coefficient is less than the third preset threshold as predicted violation behavior.
[0029] In this embodiment, the YOLO model is used to perform motion detection on the image data of the working area on the fuel plant side, and deep learning models such as DNN and CNN are selected as training models. The detected motion images are input into the trained deep learning model, and the predicted worker movements are output to achieve worker motion prediction.
[0030] In this embodiment, the key point data include the worker's head, left shoulder, right shoulder, left elbow, right elbow, left wrist, right wrist, left hip, right hip, left knee, right knee, left ankle and right ankle key points. By calculating the correlation coefficient between the predicted action key point data and the pre-set violation action key point data, it is detected whether there is a violation.
[0031] In some embodiments of the present application, the deep learning module is also used to: determine the motion trajectory of the predicted action key point data based on the predicted action key point data, calculate the trajectory correction coefficient based on the motion trajectory of the predicted action key point data, and correct the motion trajectory based on the trajectory correction coefficient; calculate the area value of the corrected motion trajectory and the illegal action trajectory of the corresponding illegal action key point data to obtain the difference area value; add the difference area values of each key point to obtain the difference area accumulation value, add the difference area accumulation value of the predicted worker's action to obtain the correlation coefficient between the predicted action key point data and the illegal action key point data.
[0032] In this embodiment, the starting point coordinates and the end point coordinates of the two action trajectories are connected in sequence through the corrected action trajectory coordinates of the predicted action key point data and the corresponding illegal action trajectory coordinates to form a closed polygon, and the area value of the corrected action trajectory and the illegal action trajectory of the corresponding illegal action key point data is obtained. By splicing multiple action trajectories in the predicted worker's action and adding the corresponding difference area accumulation values, the correlation coefficient between the predicted action key point data and the illegal action key point data is obtained. The larger the value, the less correlated the predicted action key point data is with the illegal action key point data.
[0033] In some embodiments of the present application, the deep learning module is also used to: obtain the motion trajectory of the predicted action key point data within a preset time period, and calculate the radian value of the motion trajectory based on the motion trajectory of the predicted action key point data within the preset time period; determine the illegal action trajectory based on the illegal action key point data within the preset time period, calculate the radian value of the illegal action trajectory, and calculate the ratio of the radian value of the action trajectory to the radian value of the corresponding illegal action trajectory; normalize the ratio of the radian value of the action trajectory to the radian value of the corresponding illegal action trajectory to obtain a trajectory correction coefficient.
[0034] In this embodiment, the proportional error of the two motion trajectories is obtained by calculating the radian value of the motion trajectory and the radian value of the corresponding illegal motion trajectory, and the motion trajectory is corrected by the proportional error to obtain a more accurate motion trajectory.
[0035] In some embodiments of the present application, the violation evaluation module is also used to: obtain image data of predicted violation behavior, determine the type of violation behavior of the worker based on the action image of the worker's violation behavior, and set an initial violation index based on the type of violation behavior; obtain the correlation coefficient between the action key point data of the predicted violation behavior and the violation action key point data, correct the initial violation index based on the correlation coefficient between the action key point data of the predicted violation behavior and the violation action key point data, and set the corrected violation index as the violation evaluation coefficient.
[0036] In this embodiment, a neural network model can be established to identify the action images of workers' illegal behaviors, thereby identifying the types of workers' illegal behaviors. Each type of illegal behavior corresponds to an initial violation index. A mapping table of correlation coefficients and initial violation index correction values is set. The corresponding initial violation index correction value is selected through the mapping table. The smaller the correlation coefficient, the higher the correction value of the initial violation index.
[0037] In some embodiments of the present application, the violation evaluation module is also used to: obtain the historical number of false alarms of predicted violation behaviors, and determine a first violation forgiveness value based on the historical number of false alarms of predicted violation behaviors; obtain the historical number of violations of workers corresponding to the predicted violation behaviors, and determine a second violation forgiveness value based on the historical number of violations of workers corresponding to the predicted violation behaviors; determine a behavior forgiveness coefficient based on the first violation forgiveness value and the second violation forgiveness value, and multiply the behavior forgiveness coefficient by the violation evaluation coefficient to obtain a corrected violation evaluation coefficient.
[0038] In some embodiments of the present application, the violation evaluation module is further used to: calculate the behavior forgiveness coefficient according to the behavior forgiveness coefficient calculation formula, and the behavior forgiveness coefficient calculation formula is specifically: , in, F is the behavior forgiveness coefficient, , are the first weight value and the second weight value respectively, , are the first violation forgiveness value and the second violation forgiveness value respectively, and e is a natural constant.
[0039] In this embodiment, the number of incorrect predictions for each predicted violation and the total number of violations of each worker are obtained, and the behavior forgiveness coefficient of each worker for a certain violation is calculated using a behavior forgiveness coefficient calculation formula.
[0040] In some embodiments of the present application, the monitoring module is used to: determine the fuel business process corresponding to the predicted violation based on the predicted violation, and determine whether the violation evaluation coefficient of the predicted violation after correction exceeds the allowable threshold of the violation coefficient of the corresponding process; if the violation evaluation coefficient of the predicted violation after correction exceeds the allowable threshold of the violation coefficient of the corresponding process, issue an alarm; if the violation evaluation coefficient of the predicted violation after correction does not exceed the allowable threshold of the violation coefficient of the corresponding process, record the predicted violation.
[0041] In this embodiment, by judging whether the corrected violation evaluation coefficient of the predicted violation exceeds the allowable threshold of the violation coefficient of the corresponding process, the workers' violation behavior is supervised in time, and an alarm is issued when the violation evaluation coefficient is too high. When the violation evaluation coefficient does not exceed the allowable threshold of the violation coefficient of the corresponding process, the workers' violation behavior is recorded and reported, so that the staff's violation behavior can be discovered in time and the occurrence of power plant accidents can be effectively prevented.
[0042] By applying the above technical scheme, the present invention uses a business process module to obtain the current fuel entry task information, determine the fuel business process according to the fuel entry task information, and set the violation coefficient tolerance threshold of each process according to the fuel business process; a deep learning module is used to collect the business process image data of the fuel plant side according to the fuel business process, and determine the predicted violation behavior based on the deep learning model according to the business process image data of the fuel plant side; a violation evaluation module is used to determine the violation evaluation coefficient and the behavior forgiveness coefficient according to the predicted violation behavior, and correct the violation evaluation coefficient according to the behavior forgiveness coefficient; a monitoring and alarm module is used to supervise the workers' violation behavior according to the corrected violation evaluation coefficient and the violation coefficient tolerance threshold. By collecting image data on the fuel plant side, violations on the fuel plant side can be discovered and prevented in a timely manner.
[0043] Through the description of the above implementation methods, those skilled in the art can clearly understand that the present invention can be implemented by hardware, or by software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present invention can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.), including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each implementation scenario of the present invention.
[0044] Those skilled in the art will appreciate that the modules in the device in the implementation scenario can be distributed in the device in the implementation scenario according to the implementation scenario description, or can be changed accordingly and located in one or more devices different from the implementation scenario. The modules in the above implementation scenario can be combined into one module, or can be further split into multiple sub-modules.
[0045] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.
Claims
1. A fuel plant-side intelligent supervision system based on deep learning, characterized in that: include: The business process module is used to obtain the current fuel entry task information, determine the fuel business process based on the fuel entry task information, and set the violation coefficient tolerance threshold of each process based on the fuel business process; A deep learning module is used to collect business process image data on the fuel plant side according to the fuel business process, and determine and predict violations based on the deep learning model according to the business process image data on the fuel plant side; A violation evaluation module is used to determine a violation evaluation coefficient and a behavior forgiveness coefficient according to predicted violation behaviors, and to modify the violation evaluation coefficient according to the behavior forgiveness coefficient; The monitoring and alarm module is used to monitor the workers' violation behaviors according to the revised violation evaluation coefficient and the violation coefficient tolerance threshold.
2. The deep learning-based intelligent supervision system for fuel plants according to claim 1 is characterized in that: The business process module is also used to: Obtain the fuel entry information, and extract the type data, specification data and quantity data of the fuel entering the factory based on the fuel entry information; Formulate fuel entry tasks based on the type, specification and quantity data of the fuel entering the factory, and generate corresponding fuel business processes based on the fuel entry tasks.
3. The deep learning-based intelligent supervision system for fuel plants according to claim 2 is characterized in that: The business process module is also used to: According to the fuel business process, determine the number of subtasks in each process and the average completion time of the process, calculate the ratio of the number of subtasks to the average completion time, and obtain the allowable violation coefficient; Determine whether the difference between the current allowable violation coefficient and the preset allowable violation coefficient is greater than a first preset threshold value, and if the difference between the current allowable violation coefficient and the preset allowable violation coefficient is greater than the first preset threshold value, set the preset first violation coefficient allowable threshold value as the violation coefficient allowable threshold value; If the difference between the current allowable violation coefficient and the preset allowable violation coefficient is less than or equal to the first preset threshold, determining whether the difference between the current allowable violation coefficient and the preset allowable violation coefficient is greater than a second preset threshold; If the difference between the current permissible violation coefficient and the preset permissible violation coefficient is greater than the second preset threshold, setting the preset second permissible violation coefficient threshold as the permissible violation coefficient threshold; If the difference between the current allowable violation coefficient and the preset allowable violation coefficient is less than or equal to the second preset threshold, the preset third violation coefficient allowable threshold is set as the violation coefficient allowable threshold.
4. The deep learning-based intelligent supervision system for fuel plants according to claim 1 is characterized in that: The Deep Learning module is also used to: Acquire historical worker motion image data on the fuel plant side, and establish a worker motion data set based on the historical worker motion image data on the fuel plant side; Establish a deep learning model and train it based on the worker action dataset to obtain a worker action prediction model; Determine the working area on the fuel plant side corresponding to the business process according to the fuel business process, and collect image data of the working area on the fuel plant side; Predicting worker actions of image data of the working area at the fuel plant side according to the worker action prediction model to obtain predicted worker actions, and extracting several predicted action key point data in the predicted worker actions; Obtain preset illegal action key point data, and calculate the correlation coefficient between the predicted action key point data and the illegal action key point data; Action image data whose correlation coefficient between predicted action key point data and illegal action key point data is less than a third preset threshold is screened out, and the action image data whose correlation coefficient is less than the third preset threshold is set as predicted illegal behavior.
5. The deep learning-based intelligent supervision system for fuel plants according to claim 4 is characterized in that: The Deep Learning module is also used to: Determine the motion trajectory of the predicted motion key point data according to the predicted motion key point data, calculate the trajectory correction coefficient according to the motion trajectory of the predicted motion key point data, and correct the motion trajectory according to the trajectory correction coefficient; Calculate the area value of the corrected action trajectory and the illegal action trajectory corresponding to the key point data of the illegal action to obtain the difference area value; The difference area values of each key point are added together to obtain the difference area cumulative added value, and the difference area cumulative added values of the predicted worker's action are added together to obtain the correlation coefficient between the predicted action key point data and the illegal action key point data.
6. The deep learning-based intelligent supervision system for fuel plants according to claim 5 is characterized in that: The Deep Learning module is also used to: Obtaining a motion trajectory of the predicted motion key point data within a preset period of time, and calculating a radian value of the motion trajectory according to the motion trajectory of the predicted motion key point data within the preset period of time; Determine the illegal action trajectory according to the illegal action key point data within a preset time period, calculate the arc value of the illegal action trajectory, and calculate the ratio of the arc value of the action trajectory to the arc value of the corresponding illegal action trajectory; The ratio of the arc value of the action trajectory to the arc value of the corresponding illegal action trajectory is normalized to obtain the trajectory correction coefficient.
7. The deep learning-based intelligent supervision system for fuel plants according to claim 5 is characterized in that: The Violation Evaluation Module is also used to: Acquire image data of predicted violation behaviors, determine the type of violation behaviors of the workers according to the action images of the violation behaviors of the workers, and set an initial violation index according to the type of violation behaviors; The correlation coefficient between the key point data of the action predicted by the violation and the key point data of the violation action is obtained, the initial violation index is corrected according to the correlation coefficient between the key point data of the action predicted by the violation and the key point data of the violation action, and the corrected violation index is set as the violation evaluation coefficient.
8. The deep learning-based intelligent supervision system for fuel plants according to claim 7 is characterized in that: The Violation Evaluation Module is also used to: Obtain the number of historical false alarms of the predicted violation, and determine a first violation forgiveness value according to the number of historical false alarms of the predicted violation; Obtain the number of historical violations of the worker corresponding to the predicted violation, and determine a second violation forgiveness value according to the number of historical violations of the worker corresponding to the predicted violation; The behavior forgiveness coefficient is determined according to the first violation forgiveness value and the second violation forgiveness value, and the behavior forgiveness coefficient is multiplied by the violation evaluation coefficient to obtain the corrected violation evaluation coefficient.
9. The deep learning-based intelligent supervision system for fuel plants according to claim 8 is characterized in that: The Violation Evaluation Module is also used to: The behavior forgiveness coefficient is calculated according to the behavior forgiveness coefficient calculation formula. The specific behavior forgiveness coefficient calculation formula is: ， in, F is the behavior forgiveness coefficient, , are the first weight value and the second weight value respectively, , They are the first violation forgiveness value and the second violation forgiveness value respectively.
10. The deep learning-based intelligent supervision system for fuel plants according to claim 1 is characterized in that: The monitoring module is used to: Determine the fuel business process corresponding to the predicted violation based on the predicted violation, and judge whether the violation evaluation coefficient after correction of the predicted violation exceeds the violation coefficient allowable threshold of the corresponding process; If the violation evaluation coefficient after correction of the predicted violation exceeds the violation coefficient allowable threshold of the corresponding process, an alarm reminder will be issued; If the corrected violation evaluation coefficient of the predicted violation does not exceed the violation coefficient allowable threshold of the corresponding process, the predicted violation will be recorded.