CN116909737B

CN116909737B - Port safety monitoring edge computing task unloading scheduling method based on deep learning

Info

Publication number: CN116909737B
Application number: CN202310870617.8A
Authority: CN
Inventors: 阚继承
Original assignee: Hainan University of Science and Technology
Current assignee: Hainan University of Science and Technology
Priority date: 2023-07-14
Filing date: 2023-07-14
Publication date: 2024-02-13
Anticipated expiration: 2043-07-14
Also published as: CN116909737A

Abstract

The application discloses a port safety monitoring edge computing task unloading scheduling method based on deep learning, which is applied to a port safety monitoring edge computing task unloading scheduling edge server; the method comprises the following steps: collecting port environment state and edge equipment state information; preprocessing the port environment state; constructing a deep reinforcement learning model; developing deep reinforcement learning training; and (5) scheduling decisions in real time. The beneficial effects of this application are: the method comprises the steps of constructing a deep reinforcement learning model on an edge server, establishing an unloading decision and scheduling strategy for optimizing tasks of the edge equipment, improving the overall operation efficiency of the equipment and the accuracy of task allocation by unloading and scheduling the edge computing tasks, realizing data encryption, safe transmission and privacy protection, meeting the real-time performance and delay requirements of port safety monitoring processing tasks, being beneficial to improving the performance and effect of a port safety monitoring system and improving the coping capacity of port safety threats.

Description

Port safety monitoring edge computing task unloading scheduling method based on deep learning

Technical Field

The application relates to the field of artificial intelligence, in particular to a port safety monitoring edge computing task unloading scheduling technology based on deep learning.

Background

Along with the development and application of the internet of things and artificial intelligence technology, in some large application scenes, cloud edges are included, but edge equipment possibly has insufficient computing capacity and limited storage space, so that task disposal and scheduling are not timely, and the edge equipment or a server has strong computing capacity and storage space, so that tasks which cannot be timely processed by the edge equipment can be locally, completely or partially offloaded through establishing agents in the edge server, and the offloaded tasks are completed by the edge server through task scheduling, so that each task can be timely and efficiently completed, and the overall operation efficiency of the internet of things system is improved through intelligent decision. Also, the technology is widely applied to modern port logistics management systems.

1. Port intelligent safety monitoring

The intelligent port safety monitoring means that safety monitoring and management of port environment are realized by utilizing intelligent technology and a monitoring system. The following are some common port intelligent security monitoring technologies and applications:

(1) Video monitoring system: the video pictures of the port are monitored and recorded in real time by installing cameras in each key area of the port. The video monitoring system can be used for detecting abnormal events, monitoring the running condition of the port in real time, and carrying out video analysis and intelligent identification on important areas.

(2) Intelligent recognition technology: and intelligent identification is carried out on the port video monitoring image by utilizing computer vision and machine learning technology. For example, identifying and tracking specific persons, vehicles or objects, detecting abnormal behavior, unauthorized access areas, identifying dangerous objects, etc.

(3) Intrusion detection system: the perimeter and key area of the port are monitored by using the technologies of sensors, radars and the like, and abnormal intrusion events are discovered and alarmed in time. Intrusion detection systems can identify and locate potential intruders in a variety of ways, such as by sound, thermal imaging, radar, and the like.

(4) Environmental monitoring system: the sensor and the monitoring equipment are utilized to monitor the environmental factors of the port, such as meteorological conditions, air quality, water quality and the like in real time. The environmental monitoring system may help predict and address potential safety risks, such as weather disasters, pollution, and the like.

(5) Data analysis and early warning: and analyzing and processing the collected monitoring data, identifying abnormal modes and trends, and timely giving out early warning. The data analysis can help port management personnel predict and prevent security risks, and corresponding measures are taken to ensure port security.

The application of the intelligent port safety monitoring technology can improve the safety of the port, reduce the occurrence of safety accidents, and improve the management efficiency and the operation level of the port. By combining a plurality of technical means and systems, a comprehensive and intelligent port security monitoring system can be constructed.

2. Task offloading scheduling for edge computation in port security monitoring scenarios

In the port security monitoring scene, the task unloading scheduling of the edge calculation refers to distributing and unloading the monitoring task from the central server to the edge equipment for processing, so as to improve the instantaneity, reduce the network delay and disperse the calculation load. The general flow of the edge computing task unloading scheduling in the port security monitoring scene is as follows:

(1) Task analysis: analyzing and classifying the port security monitoring tasks, and knowing the characteristics, resource requirements and priorities of the tasks. Such as video stream analysis, object recognition, anomaly detection, and the like.

(2) Edge device evaluation: edge devices within ports are evaluated, including computing power, storage capacity, network connections, etc. The characteristics and available resources of each edge device are known for task offloading scheduling.

(3) And (3) state monitoring: the state of the edge device is monitored in real time, including load conditions, network delays, available memory space, etc. This information is of great importance for task offloading scheduling decisions.

(4) Task offloading policies: and formulating a task unloading strategy according to the characteristics of the task and the resource and state information of the edge equipment. Factors such as the priority of tasks, resource requirements, load balancing of edge devices, network transmission cost and the like are considered.

(5) Task allocation: and distributing the monitoring task to the proper edge equipment according to the task unloading strategy. Taking into account computing power, storage capacity and network connection conditions of the edge device, it is ensured that tasks can be efficiently performed on the edge device.

(6) Task scheduling: scheduling task execution order and concurrency on edge devices. According to the priority and real-time requirements of the tasks, the execution sequence of the tasks is reasonably arranged, and the computing power and load balancing of the edge equipment are considered.

(7) And (3) real-time monitoring: and monitoring the execution condition of the edge equipment and the processing progress of the task. And acquiring monitoring data and results on the edge equipment in real time, and ensuring that tasks are executed on the edge equipment according to expectations.

(8) Dynamic adjustment: and dynamically adjusting a task unloading strategy and task allocation according to the state change of the edge equipment, the task priority and the system requirement. For example, when a certain edge device is too loaded, tasks may be reassigned to other idle devices.

When the edge computing task offload scheduling related to port security monitoring, in addition to considering the general flow of task allocation and scheduling, the requirements of port traffic and security monitoring need to be paid particular attention to the following aspects:

(1) The calculation requirements are as follows: port security monitoring involves computationally intensive tasks such as real-time video analysis, object recognition, anomaly detection, and the like. In task offloading scheduling, it is necessary to consider whether the computing power of the edge devices is sufficient to perform these tasks. The edge devices should have sufficient processing power and computational power to meet the computational requirements of the monitoring task.

(2) Task priority and timeliness: security monitoring tasks may have different priority and timeliness requirements. Some tasks, such as intrusion detection, emergency response, etc., require fast processing and timely response. In task unloading scheduling, tasks with high priority and timeliness sensitivity should be preferentially allocated to edge devices for processing according to the priority and timeliness requirements of the tasks.

(3) Data privacy and security: in port security monitoring, video and image data may relate to sensitive information and private content. In task offloading scheduling, it is necessary to ensure privacy and security of data. The data can be protected by means of data encryption, safe transmission and the like, and the transmission and storage of sensitive data are reduced as much as possible in the task unloading process, so that the safety and privacy of ports are protected.

(4) Task offloading decision strategy: in task offloading scheduling, a task offloading decision strategy needs to be formulated. This includes determining which tasks are suitable for offloading to the edge device, how to perform reasonable task allocation and scheduling based on the computational requirements of the tasks, timeliness requirements, and resource conditions of the edge device. The task unloading decision strategy should comprehensively consider the factors such as calculation requirement, priority, safety and the like.

(5) Real-time and latency requirements: port security monitoring requires real-time acquisition and processing of monitoring data, as well as rapid response to security events. In task unloading scheduling, the execution delay and real-time requirement of the task are required to be considered, so that the edge equipment can timely process and respond to the monitoring task, and the transmission delay and the system response time are reduced.

By considering the calculation requirement, task priority, data privacy and safety, real-time performance and delay requirement and other factors of the safety monitoring in a finer manner, the edge calculation task unloading scheduling in the port safety monitoring can be more effectively carried out. This will help to improve the efficiency and safety of the monitoring system to address port security challenges.

3. Problems and disadvantages of the prior art

(1) Task allocation imbalance: conventional task offload scheduling algorithms may not be able to effectively balance task load between edge devices. Some edge devices may take over too many tasks while others are in an idle state. This results in waste of resources and imbalance in performance.

(2) Data transmission and processing delays: during task offloading, a large amount of data transfer and processing is involved. The conventional algorithm may not be capable of effectively managing the delay of data transmission and processing, resulting in excessively long task execution time and failure to meet the real-time requirement.

(3) Insufficient data privacy and security protection: port security monitoring involves sensitive video and image data. Conventional algorithms may not have adequate measures to protect the privacy and security of data, making the data vulnerable to unauthorized access and attacks.

(4) Lack of flexibility and adaptability: traditional algorithms may lack adaptability to dynamic changes in port environments. With changes in port security monitoring tasks and new threats, traditional algorithms may not be able to adjust task offloading policies in time to accommodate new situations.

(5) Lack of comprehensive consideration of task priority and timeliness: conventional algorithms may lack comprehensive consideration of task priority and timeliness. Certain urgent and important tasks may not be prioritized, resulting in a reduced effectiveness of security monitoring.

The above problems and deficiencies indicate that the prior art has room for improvement in port security monitoring task offload scheduling. The algorithm based on deep reinforcement learning can be optimized in the aspects, and the performance and effect of task unloading scheduling are improved.

Disclosure of Invention

The application designs a port safety monitoring edge calculation task unloading scheduling method based on deep learning for solving the technical problems.

The technical scheme adopted for solving the technical problems is as follows:

a port security monitoring edge computing task unloading scheduling method based on deep learning is applied to a port security monitoring edge computing task unloading scheduling edge server, and comprises the following steps:

p100: collecting port environment state and edge equipment state information;

p200: preprocessing the port environment state;

p300, constructing a deep reinforcement learning model;

p400: developing deep reinforcement learning training;

p500: and (5) scheduling decisions in real time.

The port safety monitoring edge computing task unloading scheduling system comprises an edge server and edge equipment which is connected with the edge server in an information way; a deep reinforcement learning model is built on the edge server, and an unloading decision and a scheduling strategy for optimizing the tasks of the edge equipment are built; collecting port environment state and edge equipment state information, preprocessing the information, and defining the information as state representation of an intelligent agent: representations including port environmental status, edge device status, and task status;

the port safety monitoring edge computing task unloading scheduling method based on deep learning is characterized in that the deep reinforcement learning model is constructed on an edge server based on a deep neural network, and an agent is generated by interactive learning with the environment through the deep reinforcement learning model; the model comprises a plurality of hidden layers, and an activation function and an optimization algorithm are adopted; the input layer receives the port environment state, the edge equipment state and the task state, and the output layer outputs a decision of task unloading scheduling.

The port safety monitoring edge calculation task unloading scheduling method based on deep learning comprises the following steps of:

p310: initializing a deep reinforcement learning model, and setting training parameters and super parameters;

p320: for each edge device, acquiring the current load condition, network delay and available storage space information of the edge device;

p330: for each task, calculating priority and timeliness indexes of the task, and distributing a task ID;

p340: for each time step, the following is performed:

p341: collecting the current port environment state and the edge equipment state, and taking the current port environment state and the edge equipment state as the input of a deep reinforcement learning model;

p342: based on the current state, selecting an action by using a deep reinforcement learning model to represent an unloading scheduling decision of the task; the actions include at least one of: assigning tasks to specific edge devices, adjusting priorities of the tasks and/or adjusting execution sequences of the tasks;

p343: executing the selected action, and distributing the task to the appointed edge equipment for processing;

p344: monitoring the execution condition of a task, and recording rewards and updating the environmental state;

p345: storing the observed state, the selected action, the obtained reward, and the next state in an experience playback buffer;

p350: randomly extracting a batch of samples from the experience playback buffer area for training and updating the deep reinforcement learning model; calculating a training target, at least comprising a target Q value calculation method using a deep Q network;

p360: distributing the tasks to corresponding edge equipment for processing according to priority and timeliness indexes, data privacy and safety, real-time performance and delay requirements of the tasks and output actions of a deep reinforcement learning model; meanwhile, referring to the load condition of the edge equipment, the real-time requirement of the task and the data privacy requirement, adjusting the execution sequence and concurrency of the task;

p370: in task unloading scheduling, data encryption and security transmission measures are adopted for secret-related data, privacy protection measures are adopted for private data, and data security in the data transmission and processing processes is ensured;

p380: periodically updating the states of the edge equipment and the task according to the execution result of the task and the port environment state monitored in real time;

p390: repeating the steps P310 to P380 until all time steps are completed or a preset condition is reached;

p399: outputting a task unloading scheduling result; the task unloading schedule at least comprises the allocation condition of the tasks, the loading condition of the edge equipment and/or the execution sequence of the tasks.

The port security monitoring edge calculation task unloading scheduling method based on deep learning, wherein training parameters in the P310 comprise learning rate, batch size and training iteration times;

the super parameters in the P310 include the hidden layer number and the neuron number of the neural network, the discount factor of the reinforcement learning algorithm and the exploration rate.

The port security monitoring edge calculation task unloading scheduling method based on deep learning, wherein the priority and timeliness index calculation of each task in the P330 comprises the following steps:

each task refers to a plurality of computation-intensive tasks involved in port security monitoring, including real-time video analysis, target identification and anomaly detection;

when each task is unloaded and scheduled, decomposing the task and dividing the priority; the method comprises the steps of subdividing a video stream analysis task into image frame processing, target detection and behavior recognition subtasks;

priority division is carried out according to the emergency degree and importance of each task, so that the fineness of the scheduled and unloaded tasks is improved, and the requirements of different tasks are met;

when different security threats exist in different monitoring areas at the same time, adopting a multi-stage task unloading scheduling strategy; the method comprises the following steps: performing preliminary real-time video analysis and target recognition on each edge device, and rapidly filtering and processing conventional conditions; and sending the video data with the abnormality to a cloud or a central server for abnormality detection and analysis.

The port security monitoring edge computing task unloading scheduling method based on deep learning, wherein the data encryption, security transmission and privacy protection measures in the P370 comprise:

the data encryption measure is to encrypt and decrypt data related to sensitive information and privacy content in video and images by adopting a symmetrical or asymmetrical encryption algorithm;

the security transmission measure is to adopt IPsec VPN or SSL VPN technology to realize data transmission;

the privacy protection measures are to realize the protection of private data by adopting a differential privacy technology, a homomorphic encryption technology or an identity authentication technology.

The port security monitoring edge computing task unloading scheduling method based on deep learning, wherein port environment state information in P100 comprises one or more of the following information:

information of cargo in port: the type of cargo, the number of cargo, and the destination of the cargo;

ship information: ship position, ship cargo condition;

the edge device information in P100 includes one or more of the following combinations of information:

edge device status information: computing power, storage resources, and availability of each edge device;

edge device location information: including the location of each edge device within the port.

The port security monitoring edge computing task unloading scheduling method based on deep learning comprises the following steps of: comprising the following steps: and the edge server extracts key features according to the environment information and the edge equipment information and performs dimension reduction processing.

The port security monitoring edge computing task unloading scheduling method based on deep learning, wherein the deep reinforcement learning training in P400 comprises the following steps:

in the training stage, training a deep neural network in an edge server by using a reinforcement learning algorithm (the deep neural network in the edge server is trained to generate an agent by the reinforcement learning algorithm), so as to realize the maximization of the cumulative reward function;

the port security monitoring edge calculation task unloading scheduling method based on deep learning, wherein the real-time scheduling decision in P500 comprises the following steps:

in the test stage, the trained deep neural network (namely the intelligent agent) in the edge server makes a task unloading scheduling decision according to the current state, and the task is distributed to the edge server which is nearest and idle for processing.

1. Core idea

The algorithm is based on a deep reinforcement learning method, and a task unloading scheduling strategy is learned by training an agent. The intelligent agent takes the port environment state and the edge equipment state as input and outputs corresponding task unloading scheduling decisions. The intelligent agent continuously learns and optimizes the scheduling strategy through a deep reinforcement learning algorithm by interacting with the environment so as to maximize the cumulative rewards and realize efficient task unloading scheduling. The algorithm mainly considers the following aspects:

(1) Task priority and timeliness considerations:

in the algorithm, tasks are given different priorities and the timeliness requirement of the tasks is considered. This may be achieved by assigning each task a priority index and an timeliness metric index. During the training process, the priority and timeliness of the tasks are used as important reward signals to encourage the model to preferentially select tasks with more urgent timeliness requirements for unloading scheduling. This ensures that tasks of high priority and urgency are handled in time.

(2) Data privacy and security considerations:

in order to protect privacy and security of sensitive data involved in port security monitoring, the algorithm adopts measures such as data encryption, security transmission, privacy protection and the like. During task offloading, video data is encrypted and the data is transmitted to an edge device using a secure transmission protocol. In addition, differential privacy techniques may also be employed to protect the privacy of data to prevent unauthorized disclosure of data.

(3) Real-time and latency requirements considerations:

port security monitoring has high requirements for real-time and low latency. In the algorithm, the real-time requirement of the task is considered as an important reward signal. The deep reinforcement learning model is used for preferentially selecting tasks meeting the real-time requirement to unload through learning task unloading scheduling strategies. In addition, by interacting with the state of the edge device, the algorithm can dynamically adjust the execution sequence and concurrency of the tasks to reduce the delay of task processing.

2. Key point is that

When designing a task offloading scheduling algorithm based on deep reinforcement learning for the specificity of port security monitoring, the following aspects are key points:

(1) Task decomposition and prioritization: port security monitoring involves a number of computationally intensive tasks, such as real-time video analysis, object recognition, and anomaly detection. When the task is unloaded and scheduled, the task can be appropriately decomposed and prioritized. For example, video stream analysis tasks are subdivided into subtasks such as image frame processing, object detection, and behavior recognition, and prioritized according to the urgency and importance of the task. Thus, the tasks can be scheduled and offloaded more finely, and the special requirements of different tasks are met.

(2) Multi-stage task offloading scheduling: port security monitoring typically requires processing large amounts of video data and may present different security threats in different monitored areas. For this case, a multi-level task offload scheduling policy may be employed. First, preliminary real-time video analysis and object recognition are performed on the edge devices to quickly filter and process the conventional cases. Then, the video data suspected of abnormality is sent to a cloud or a central server for deeper abnormality detection and analysis. Therefore, the data transmission and calculation pressure can be effectively reduced, and the accuracy of anomaly detection is improved.

(3) Real-time and low latency requirements: port security monitoring has high requirements for real-time and low latency. When the task is unloaded and scheduled, the real-time requirement of the task needs to be considered, and the real-time is taken as an important reward signal. Timeliness metrics may be introduced and corresponding rewards mechanisms set in the deep reinforcement learning model to encourage the model to prioritize tasks with more urgent timeliness requirements for offloading scheduling.

(4) Data privacy and security: port security monitoring involves sensitive video and image data, and it is important to protect the privacy and security of the data. In task offloading scheduling, data privacy and security requirements should be considered. Measures such as data encryption, safe transmission and privacy protection can be taken, and the data security in the data transmission and processing process is ensured.

(5) Network bandwidth and stability: port security monitoring typically requires task offloading and data transfer between distributed edge devices. In task offloading scheduling, network bandwidth and stability needs to be considered to avoid task overload and network delay. The network state and the bandwidth utilization condition can be monitored, and a deep reinforcement learning algorithm is combined, so that proper edge equipment is selected for task unloading, and network load balancing and efficient task execution are ensured.

The above-mentioned at least one technical scheme that this application embodiment adopted can reach following beneficial effect:

compared with the prior art, the port safety monitoring edge computing task unloading scheduling algorithm based on deep reinforcement learning has the following innovation points and improvement effects:

1. the efficiency and the accuracy of task unloading and scheduling are improved: according to the algorithm, the task is distributed more accurately according to the port environment state and the edge equipment resource by training the agent to learn the task unloading scheduling strategy, so that the task can be distributed rapidly and executed efficiently. Compared with the traditional method, the method can better balance task loads among the edge devices and avoid the condition of unbalanced task distribution.

2. Protecting data privacy and security: the algorithm adopts measures such as data encryption, safe transmission, privacy protection and the like in the task unloading and scheduling process, and ensures the safety of sensitive data in the transmission and processing process. The confidentiality of the data is protected by the data encryption, unauthorized access is prevented by the secure transmission, and the privacy of the data is further guaranteed by the privacy protection technology.

3. The real-time performance and delay requirements of the task are met: the algorithm prioritizes tasks with urgency by taking the real-time needs of the tasks as important reward signals. Meanwhile, the execution sequence and concurrency of the tasks are dynamically adjusted, so that the delay of task processing is effectively reduced. Therefore, the requirements of port safety monitoring on real-time performance and low delay can be better met.

4. Enhanced task priority and timeliness considerations: the algorithm performs task unloading scheduling according to the priority and timeliness index of the task, and ensures that the task with high priority and urgency is processed in time. Through training and optimizing the deep reinforcement learning model, the priority and timeliness of the tasks can be better comprehensively considered, and the quality and effect of task unloading and scheduling are improved.

Compared with the prior art, the port safety monitoring edge computing task unloading scheduling algorithm based on deep reinforcement learning has the advantages of improving efficiency and accuracy, protecting data privacy and safety, meeting the real-time performance and delay requirements of tasks and the like. These advantages will help to improve the performance and effectiveness of the port security monitoring system and to improve the capability of coping with port security threats.

Drawings

The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this application, illustrate embodiments of the application and together with the description serve to explain the application and do not constitute an undue limitation to the application. In the drawings:

FIG. 1 is a functional block diagram of the present application;

FIG. 2 is a schematic diagram of a deep reinforcement learning model design of the present application.

Detailed Description

For the purposes, technical solutions and advantages of the present application, the technical solutions of the present application will be clearly and completely described below with reference to specific embodiments of the present application and corresponding drawings. It will be apparent that the described embodiments are only some, but not all, of the embodiments of the present application. All other embodiments, which can be made by one of ordinary skill in the art without undue burden from the present disclosure, are within the scope of the present disclosure.

The following describes in detail the technical solutions provided by the embodiments of the present application with reference to the accompanying drawings.

In order to facilitate understanding of the embodiments of the present application, as shown in fig. 1, a method for unloading and scheduling a port security monitoring edge computing task based on deep learning is applied to a port security monitoring edge computing task unloading and scheduling edge server, and the method includes:

p100: collecting port environment state and edge equipment state information;

p200: preprocessing the port environment state;

p300, constructing a deep reinforcement learning model;

p400: developing deep reinforcement learning training;

p500: and (5) scheduling decisions in real time.

by establishing an intelligent body on the edge server and establishing a deep reinforcement learning model in the intelligent body, the intelligent body and the environment perform interactive learning, and an unloading decision and scheduling strategy for optimizing the tasks of the edge device is established, so that the problems of insufficient computing capacity and limited storage space of the edge device can be effectively solved, the computing capacity of the edge server can be fully utilized by reasonably unloading the tasks required to be executed by the edge device and performing scientific scheduling, the storage space of the edge server is fully utilized, the overall operation efficiency of the device can be improved by balancing the task allocation of each edge server, the network load balance and the efficient execution of the tasks are ensured, and the requirements of port safety monitoring on real-time and low delay are met.

The port safety monitoring edge computing task unloading scheduling method based on deep learning is characterized in that the deep reinforcement learning model is constructed on an edge server based on a deep neural network, and an agent is generated by interactive learning with the environment through the deep reinforcement learning model; the model comprises a plurality of hidden layers, and an activation function and an optimization algorithm are adopted; the input layer receives the port environment state, the edge device state and the task state, and the output layer outputs the decision of task unloading scheduling, as shown in fig. 2.

p340: for each time step, the following is performed:

p360: distributing the tasks to corresponding edge equipment for processing according to priority and timeliness indexes, data privacy and safety, real-time performance and delay requirements of the tasks and output actions of a deep reinforcement learning model; meanwhile, referring to the load condition of the edge equipment, the real-time requirement of the task and the data privacy requirement, adjusting the execution sequence and concurrency of the task; the method comprises the steps of establishing weight indexes for priority and timeliness indexes, data privacy and safety, real-time performance and delay requirements of tasks, wherein the higher the priority and timeliness indexes, the data privacy and safety, the real-time performance and delay requirements of the tasks are, the larger the weight values are, and the higher the priority is on equipment use, data transmission and task execution.

P370: in the task unloading scheduling, data encryption, safe transmission and privacy protection measures are adopted to ensure the data safety in the data transmission and processing process; the video data is encrypted by using an encryption technology, and a secure transmission protocol is used in the data transmission process to ensure that the data is not acquired by unauthorized persons.

p399: outputting a task unloading scheduling result; the task unloading schedule at least comprises the task allocation condition, the load condition of the edge equipment and/or the task execution sequence, as shown in fig. 2.

The algorithm established based on the work flow of the deep reinforcement learning model comprises the following steps:

input: a port security monitoring task set and an edge equipment set;

and (3) outputting: unloading a scheduling result of the task;

by the improved algorithm, the port safety monitoring edge calculation task unloading scheduling based on deep reinforcement learning can better consider factors such as priority and timeliness of tasks, data privacy and safety, instantaneity and delay requirements.

These parameters and super parameters need to be adjusted and optimized according to the specific situation to obtain the best algorithm performance.

and carrying out priority division according to the urgency degree and importance of each task, improving the fineness of the scheduled and offloaded tasks, and meeting the requirements of different tasks.

Task decomposition and prioritization: port security monitoring involves a number of computationally intensive tasks, such as real-time video analysis, object recognition, and anomaly detection. When the task is unloaded and scheduled, the task can be appropriately decomposed and prioritized. For example, video stream analysis tasks are subdivided into subtasks such as image frame processing, object detection, and behavior recognition, and prioritized according to the urgency and importance of the task. Thus, the tasks can be scheduled and offloaded more finely, and the special requirements of different tasks are met.

Multi-stage task offloading scheduling: port security monitoring typically requires processing large amounts of video data and may present different security threats in different monitored areas. For this case, a multi-level task offload scheduling policy may be employed. First, preliminary real-time video analysis and object recognition are performed on the edge devices to quickly filter and process the conventional cases. Then, the video data suspected of abnormality is sent to a cloud or a central server for deeper abnormality detection and analysis. Therefore, the data transmission and calculation pressure can be effectively reduced, and the accuracy of anomaly detection is improved.

The algorithm ensures the data security in the data transmission and processing process by taking data encryption, security transmission and privacy protection measures. The improvement improves the performance and the safety of the algorithm, so that the algorithm can process port safety monitoring tasks more effectively, protect the privacy of sensitive data and meet the real-time performance and delay requirements of the tasks.

ship information: ship position, ship cargo condition;

It will be appreciated by those skilled in the art that embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.

The foregoing is merely exemplary of the present application and is not intended to limit the present application. Various modifications and changes may be made to the present application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc. which are within the spirit and principles of the present application are intended to be included within the scope of the claims of the present application.

Claims

1. The port safety monitoring edge computing task unloading scheduling method based on deep learning is applied to a port safety monitoring edge computing task unloading scheduling edge server and is characterized in that: the method comprises the following steps:

p100: collecting port environment state and edge equipment state information;

p200: preprocessing the port environment state;

p300, constructing a deep reinforcement learning model;

p400: developing deep reinforcement learning training;

p500: real-time scheduling decision making;

the deep reinforcement learning model is constructed on an edge server based on a deep neural network; the model comprises a plurality of hidden layers, and an activation function and an optimization algorithm are adopted; the input layer receives the port environment state, the edge equipment state and the task state, and the output layer outputs a decision of task unloading scheduling;

the deep reinforcement learning model workflow is:

p340: for each time step, the following is performed:

p360: distributing the tasks to corresponding edge equipment for processing according to priority and timeliness indexes, data privacy and safety, real-time performance and delay requirements of the tasks and output actions of a deep reinforcement learning model;

2. The port security monitoring edge computing task offloading scheduling method based on deep learning as set forth in claim 1, wherein: the training parameters in the P310 at least comprise a learning rate, a batch size and training iteration times;

the super parameters in the P310 at least comprise the hidden layer number and the neuron number of the neural network, the discount factor and the exploration rate of the reinforcement learning algorithm.

3. The port security monitoring edge computing task offloading scheduling method based on deep learning as set forth in claim 1, wherein: the calculation of the priority and timeliness index of each task in the P330 includes:

each task refers to a plurality of computation-intensive tasks related to port safety monitoring, and at least comprises real-time video analysis, target identification and anomaly detection;

when each task is unloaded and scheduled, decomposing the task and dividing the priority; at least comprising sub-dividing the video stream analysis task into image frame processing, object detection and behavior recognition sub-tasks;

4. The port security monitoring edge computing task offloading scheduling method based on deep learning as set forth in claim 1, wherein: the data encryption, secure transmission and privacy protection measures in P370 include:

5. The port security monitoring edge computing task offloading scheduling method based on deep learning as set forth in claim 1, wherein: the port environmental status information in the P100 at least comprises one or more of the following information:

ship information: ship position, ship cargo condition;

the edge device information in the P100 at least includes one or more of the following information combinations:

6. The port security monitoring edge computing task offloading scheduling method based on deep learning as set forth in claim 1, wherein: pretreatment in P200: comprising the following steps: and the edge server extracts key features according to the environment information and the edge equipment information and performs dimension reduction processing.

7. The port security monitoring edge computing task offloading scheduling method based on deep learning as set forth in claim 1, wherein: the deep reinforcement learning training in P400 includes:

and using the deep neural network in the edge server trained by the reinforcement learning algorithm to realize the maximum cumulative reward function.

8. The port security monitoring edge computing task offloading scheduling method based on deep learning as set forth in claim 1, wherein: the real-time scheduling decision in P500 includes:

and the deep neural network in the edge server makes a task unloading scheduling decision according to the current state, and distributes the task to the edge server which is nearest and idle for processing.